C Y B E R AT TA C K P R E V E N T I O N A N D D E T E C T I O N F O R E L E C T R I C V E H I C L E C H A R G I N G Vom Fachbereich Informatik der Technischen Universität Darmstadt genehmigte dissertation zur Erlangung des akademischen Grades Doktor-Ingenieur (Dr.-Ing.) von dustin kern Erstreferent: Prof. Dr.-Ing. Matthias Hollick Korreferent: Prof. Dr. Christoph Krauß Korreferent: Prof. Dr. Stefan Katzenbeisser Darmstadt 2025 Hochschulkennziffer D17 Dustin Kern, Cyber Attack Prevention and Detection for Electric Vehicle Charging, Dissertation, Darmstadt, Technische Universität Darmstadt, 2025. Fachgebiet Sichere Mobile Netze Fachbereich Informatik Darmstadt, Technische Universität Darmstadt Jahr der Veröffentlichung: 2025 Tag der mündlichen Prüfung: 26. März 2025 URN: urn:nbn:de:tuda-tuprints-298239 © Urheberrechtlich geschützt https://rightsstatements.org/page/InC/1.0/?language=de In Copyright https://rightsstatements.org/page/InC/1.0/ https://nbn-resolving.org/urn:nbn:de:tuda-tuprints-298239 https://rightsstatements.org/page/InC/1.0/?language=de https://rightsstatements.org/page/InC/1.0/ A B S T R A C T The increasing adoption of Electric Vehicles (EVs) is transforming the automotive landscape, driven by the need for a more sustainable transportation sector. To support the widespread use of EVs, an effi- cient and reliable charging infrastructure is essential. For this, several related communication protocols and backend systems have been es- tablished to manage power delivery, authorization, and billing. The main goals are the security of charge session payments and a power grid-friendly scheduling of EV charging loads. The charging of EVs, however, also involves various security risks in associated use-cases. On the one hand, regarding the use-case of charge authorization and billing, existing protocols fail to protect against relevant adversaries. As a result, backend operators are exposed to the risk of significant financial damages and EV users are exposed to severe privacy risks. On the other hand, regarding the use-case of charge session power control and load balancing, existing processes can be manipulated by compromised systems. As a result, adversaries may be able to cause severe physical damage to involved systems and potentially harm power grid operations. In this dissertation, we address selected security risks. For charge authorization and billing, we present three solutions to enhance the preventive security of EV charging protocols. More specifically, we present concepts (i) for the integration of crypto-agility and the use of Post-Quantum Cryptography (PQC), (ii) for the use of Self-Sovereign Identities (SSIs) to enhance EV user privacy, and (iii) for the adop- tion of a standardized authorization framework to reduce existing complexity. Regarding manipulations of charge session control, we present several concepts for the analysis, detection, and mitigation of related attacks. More specifically, we present (i) a feasibility analysis of resulting attacks on grid stability and a related co-simulation frame- work, (ii) different anomaly detection concepts for either large-scale coordinated attacks on the grid or attacks in individual charging ses- sions, (iii) approaches for improving detection performance, including a Generative Adversarial Network (GAN)-based Intrusion Detection System (IDS) optimization and a combination of large-scale and ses- sion-based detection, and (iv) methods for attack mitigation based on IDS outputs. All presented concepts are implemented and evaluated with regard to relevant criteria. Concepts for EV charging protocol security are evaluated regarding performance/usability criteria based on proof- of-concept implementations and regarding security/privacy criteria based on formal protocol analyses using the Tamarin prover. Concepts iii for the analysis, detection, and mitigation of session control-related at- tacks are implemented with simulation-based approaches to evaluate their effect on involved systems and their detection/mitigation perfor- mance. Used Tamarin models and simulation data are published for reproducibility and future use in related studies. Overall, our results show the presented concepts can provide a significant benefit to the security of EV charging in the sector’s future. Z U S A M M E N FA S S U N G Die zunehmende Verbreitung von Elektrofahrzeugen (EVs) verändert die Autolandschaft, angetrieben durch die Notwendigkeit eines nach- haltigeren Transportsektors. Um den weit verbreiteten Einsatz von EVs zu unterstützen, ist eine effiziente und zuverlässige Ladeinfrastruktur unerlässlich. Zu diesem Zweck wurden verschiedene Kommunikati- onsprotokolle und Backend-Systeme entwickelt, um Energieversor- gung, Autorisierung und Abrechnung zu verwalten. Hauptziele sind die Sicherheit der Abrechnung von Ladevorgängen und eine strom- netzfreundliche Planung der Ladevorgänge von EVs. Das Laden von EVs birgt jedoch eine Reihe von Sicherheitsrisiken in den damit verbundenen Anwendungsfällen. Einerseits sind die bestehenden Protokolle für die Autorisierung und Abrechnung von Ladevorgängen nicht in der Lage, sich gegen relevante Angreifer zu schützen. Dies hat zur Folge, dass die Backend-Betreiber dem Risiko erheblicher finanzieller Schäden ausgesetzt sind und die Nutzer von EVs ein großes Risiko für ihre Privatsphäre eingehen. Im Hinblick auf den Anwendungsfall der Steuerung von Ladevorgängen und des Lastausgleichs können bestehende Prozesse durch kompromittierte Systeme manipuliert werden. Dies kann dazu führen, dass Angreifer schwere physische Schäden an den betroffenen Systemen verursachen und den Betrieb des Stromnetzes beeinträchtigen können. In dieser Dissertation befassen wir uns mit ausgewählten Sicherheits- risiken. Für die Ladeautorisierung und -abrechnung stellen wir drei Lösungen vor, um die präventive Sicherheit von EV-Ladeprotokollen zu verbessern. Genauer gesagt stellen wir Konzepte vor (i) für die Integration von kryptographischer Agilität und die Verwendung von Post-Quantum-Kryptographie (PQC), (ii) für die Verwendung von Self-Sovereign Identities (SSIs) zur Verbesserung der Privatsphäre von EV-Nutzern und (iii) für die Einführung eines standardisierten Autorisierungsframeworks zur Reduzierung der bestehenden Komple- xität. Im Hinblick auf die Manipulationen der Ladevorgangssteuerung stellen wir verschiedene Konzepte zur Analyse, Erkennung und Be- grenzung entsprechender Angriffe vor. Genauer gesagt, präsentieren wir (i) eine Machbarkeitsanalyse der daraus resultierenden Angriffe iv auf die Netzstabilität und ein entsprechendes Co-Simulations-Fra- mework, (ii) verschiedene Konzepte zur Erkennung von Anomalien, entweder für groß angelegte koordinierte Angriffe auf das Netz oder für Angriffe in einzelnen Ladesitzungen, (iii) Ansätze zur Verbesse- rung der Erkennungsleistung, einschließlich der Optimierung eines Intrusion Detection Systems (IDS) auf der Basis eines Generative Ad- versarial Network (GAN) und einer Kombination aus groß angelegter und sitzungsbasierter Erkennung, und (iv) Methoden zur Angriffsbe- grenzung auf der Grundlage der IDS-Ausgaben. Alle vorgestellten Konzepte werden implementiert und im Hin- blick auf relevante Kriterien bewertet. Konzepte für die Sicherheit von EV-Ladeprotokollen werden anhand von Proof-of-Concept-Imple- mentierungen hinsichtlich der Kriterien Performance/Nutzbarkeit und anhand von formalen Protokollanalysen mit dem Tamarin-Prover hinsichtlich der Kriterien Sicherheit/Privatsphäre bewertet. Konzepte für die Analyse, Erkennung und Begrenzung von Angriffen auf die Ladevorgangssteuerung werden mit simulationsbasierten Ansätzen implementiert, um ihre Auswirkungen auf die beteiligten Systeme be- ziehungsweise ihre Erkennungs- und Begrenzungsleistung zu bewer- ten. Die verwendeten Tamarin-Modelle und Simulationsdaten werden zur Reproduzierbarkeit und zur zukünftigen Verwendung in verwand- ten Studien veröffentlicht. Insgesamt zeigen unsere Ergebnisse, dass die vorgestellten Konzepte einen erheblichen Beitrag zur Sicherheit des Ladens von EVs in der Zukunft leisten können. v A C K N O W L E D G M E N T S Throughout my research career, I had the pleasure of working with several excellent researchers, the collaboration, guidance, and support of whom greatly aided my academic journey. In this section, I would like to express my sincere gratitude towards these people. First, I would like to thank my colleagues, co-authors, and supervisors from my time at Fraunhofer SIT, who first introduced me to the field of academic research and working on scientific publications. Specifically, I would like to thank Christian Plappert, Daniel Zelle, Maria Zhdanova, Andreas Fuchs, and Christoph Krauß. Their continued guidance enabled me to have a smooth and successful start in academia. Next, I would like to thank Christoph Krauß for generously offering me a PhD student position at Darmstadt University of Applied Sciences. His dependable support, helpful counsel, and trust provided a significant con- tribution to the development of my academic career. Additionally, I would like to thank the co-authors and colleagues, whose collaboration during my time at Darmstadt University of Applied Sciences aided the creation of this cumulative dissertation and furthered my academic journey. Specifically, I would like to thank Timm Lauser, Christoph Krauß, Nouri Alnahawi, Alexan- der Wiesmaier, Ruben Niederhagen, Matthias Hollick, Adrian Kailus, Jonas Primbs, Michael Menth, Viet Ha The, and Phat Nguyen Tan. Next, I would like to thank Matthias Hollick for kindly accepting the su- pervision of my doctorate. The invaluable feedback and insightful guidance of my supervisors greatly helped in the improvement of my research and further shaped my academic process. Similarly, I would like to thank Christoph Krauß and Stefan Katzenbeisser for the co-supervision of my doctorate. Additionally, I would like to thank Björn Scheuermann and Florian Steinke for joining my supervisors on the doctoral committee. It is an honor to hold my PhD defense in front of such a highly distinguished committee. Finally, I would like to thank my family for their continued support throughout the years. vii C O N T E N T S Abstract iii Zusammenfassung iv Acknowledgments vii List of Figures xiii List of Tables xiii List of Abbreviations xiii List of Publications xvii Collaborations and My Contribution xxi i Synopsis 1 Introduction 3 1.1 Context on EV Charging . . . . . . . . . . . . . . . . . . 3 1.2 Motivation and Problem Statement . . . . . . . . . . . . 4 1.3 Methodology and Contributions . . . . . . . . . . . . . 6 1.4 Structure of the Thesis . . . . . . . . . . . . . . . . . . . 8 2 Background 11 2.1 EV Charging Actors, Protocols, and Standards . . . . . 11 2.2 Detailed System Models . . . . . . . . . . . . . . . . . . 12 2.2.1 Charge Authorization and Billing . . . . . . . . 13 2.2.2 Charge Control and Load Balancing . . . . . . . 14 2.3 Adversary Models . . . . . . . . . . . . . . . . . . . . . 16 2.3.1 Charge Authorization and Billing . . . . . . . . 17 2.3.2 Charge Control and Load Balancing . . . . . . . 18 3 Contributions 21 3.1 Prevention of Attacks on EV Charging by Enhancing Protocol Security . . . . . . . . . . . . . . . . . . . . . . 21 3.1.1 Discussion of Contributions . . . . . . . . . . . . 22 3.1.2 Comparison with Related Work . . . . . . . . . 27 3.2 Analysis and Detection of Attacks on EV Charging Ses- sion Control . . . . . . . . . . . . . . . . . . . . . . . . . 31 3.2.1 Discussion of Contributions . . . . . . . . . . . . 31 3.2.2 Comparison with Related Work . . . . . . . . . 40 4 Conclusion 45 4.1 Contributions of this Thesis . . . . . . . . . . . . . . . . 45 4.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . 47 Bibliography 51 ii Publications 5 QuantumCharge: Post-Quantum Cryptography for Electric Vehicle Charging 69 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 70 ix x contents 5.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . 73 5.3 System and Attacker Model . . . . . . . . . . . . . . . . 76 5.4 Requirements . . . . . . . . . . . . . . . . . . . . . . . . 77 5.5 Security Concept . . . . . . . . . . . . . . . . . . . . . . 78 5.6 Formal Security Verification . . . . . . . . . . . . . . . . 84 5.7 Implementation and Practical Feasibility Evaluation . . 87 5.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 90 5.9 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 5.10 References . . . . . . . . . . . . . . . . . . . . . . . . . . 92 6 Self-Sovereign Identity for Electric Vehicle Charging 97 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 98 6.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . 99 6.3 Related Work . . . . . . . . . . . . . . . . . . . . . . . . 102 6.4 System Model and Requirement Analysis . . . . . . . . 104 6.5 SSI Concept . . . . . . . . . . . . . . . . . . . . . . . . . 106 6.6 Implementation . . . . . . . . . . . . . . . . . . . . . . . 113 6.7 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . 113 6.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 120 6.9 References . . . . . . . . . . . . . . . . . . . . . . . . . . 120 7 Streamlining Plug-and-Charge Authorization for Electric Ve- hicles with OAuth2 and OIDC 125 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 126 7.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . 128 7.3 Related Work . . . . . . . . . . . . . . . . . . . . . . . . 132 7.4 System-/Adversary Model and Requirements . . . . . 134 7.5 Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 7.6 Implementation . . . . . . . . . . . . . . . . . . . . . . . 141 7.7 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . 142 7.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 147 7.9 References . . . . . . . . . . . . . . . . . . . . . . . . . . 148 7.10 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 8 Analysis of E-Mobility-based Threats to Power Grid Resilience155 8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 156 8.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . 157 8.3 e-Mobility System Model . . . . . . . . . . . . . . . . . 157 8.4 Adversary Model . . . . . . . . . . . . . . . . . . . . . . 158 8.5 Framework for e-Mobility-Based Grid Attack Analysis 159 8.6 Implementation and Case Studies . . . . . . . . . . . . 161 8.7 e-Mobility-Based Protections . . . . . . . . . . . . . . . 164 8.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 165 8.9 References . . . . . . . . . . . . . . . . . . . . . . . . . . 165 9 Detection of e-Mobility-Based Attacks on the Power Grid 169 9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 170 9.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . 171 9.3 System Model . . . . . . . . . . . . . . . . . . . . . . . . 172 9.4 Adversary Model . . . . . . . . . . . . . . . . . . . . . . 173 contents xi 9.5 e-Mobility-Based IDS Concept . . . . . . . . . . . . . . 174 9.6 Implementation and Evaluation . . . . . . . . . . . . . . 177 9.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 181 9.8 References . . . . . . . . . . . . . . . . . . . . . . . . . . 182 10 Detection of Anomalies in Electric Vehicle Charging Sessions 185 10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 186 10.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . 187 10.3 System Model . . . . . . . . . . . . . . . . . . . . . . . . 187 10.4 Adversary Model . . . . . . . . . . . . . . . . . . . . . . 188 10.5 IDS Concept for EV Charging Session Anomalies . . . 189 10.6 Implementation and Evaluation . . . . . . . . . . . . . . 191 10.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 196 10.8 References . . . . . . . . . . . . . . . . . . . . . . . . . . 196 11 Attack Analysis and Detection for the Combined Electric Vehicle Charging and Power Grid Domains 199 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 200 11.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . 201 11.3 System Model . . . . . . . . . . . . . . . . . . . . . . . . 202 11.4 Adversary Model . . . . . . . . . . . . . . . . . . . . . . 202 11.5 Cyberattack Analysis and Detection Concept for the e-Mobility Domain . . . . . . . . . . . . . . . . . . . . . 204 11.6 Implementation . . . . . . . . . . . . . . . . . . . . . . . 206 11.7 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . 207 11.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 210 11.9 References . . . . . . . . . . . . . . . . . . . . . . . . . . 210 12 Improving Anomaly Detection for Electric Vehicle Charging with Generative Adversarial Networks 213 12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 214 12.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . 215 12.3 EV Charging System Model . . . . . . . . . . . . . . . . 216 12.4 Adversary Model . . . . . . . . . . . . . . . . . . . . . . 216 12.5 GAN-Based IDS for EV Charging . . . . . . . . . . . . . 217 12.6 Implementation and Evaluation . . . . . . . . . . . . . . 219 12.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 222 12.8 References . . . . . . . . . . . . . . . . . . . . . . . . . . 222 13 Anomaly Detection and Mitigation for Electric Vehicle Charging- Based Attacks on the Power Grid 225 13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 226 13.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . 227 13.3 System Model . . . . . . . . . . . . . . . . . . . . . . . . 227 13.4 Adversary Model . . . . . . . . . . . . . . . . . . . . . . 228 13.5 EV Charging-Based Attack Detection and Mitigation . 228 13.6 Implementation . . . . . . . . . . . . . . . . . . . . . . . 230 13.7 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . 231 13.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 235 13.9 References . . . . . . . . . . . . . . . . . . . . . . . . . . 235 xii contents Erklärung zur Dissertationsschrift 237 L I S T O F F I G U R E S Figure 1 High-Level EV Charging System Model . . . . 4 Figure 2 Detailed Charge Authorization and Billing Sys- tem Model . . . . . . . . . . . . . . . . . . . . . 13 Figure 3 Detailed Charge Control and Load Balancing System Model . . . . . . . . . . . . . . . . . . . 15 L I S T O F TA B L E S Table 1 Performance Comparison of Different Approaches 26 Table 2 Comparison with Related Work on Charging Protocol Security/Privacy . . . . . . . . . . . . 30 Table 3 Comparison of Different IDS Evaluations . . . 39 Table 4 Comparison with Related Work on Charging Attack Analysis/Detection . . . . . . . . . . . . 44 L I S T O F A B B R E V I AT I O N S AUC Area Under the Curve BMS Battery Management System CAN Controller Area Network CCH Contract Clearing House CL Camenisch-Lysyanskaya CP Charge Point CPO Charge Point Operator CPS Certificate Provisioning Service CSR Certificate Signing Request DAA Direct Anonymous Attestation DICE Device Identifier Composition Engine DID Distributed Identifier DER Distributed Energy Resource DH Diffie Hellman xiii xiv list of abbreviations DNP3 Distributed Network Protocol 3 DoS Denial of Service DSA Digital Signature Algorithm DSO Distribution System Operator DSRP Design Science Research Process DTR Decision Tree Regressor ECC Elliptic Curve Cryptography ECU Electronic Control Unit eMAID e-Mobility Account Identifier eMIP eMobility Interoperation Protocol eMSP e-Mobility Service Provider EV Electric Vehicle FDI False Data Injection FPR False Positive Rate GAN Generative Adversarial Network GDPR General Data Protection Regulation HSM Hardware Security Module IDS Intrusion Detection System IoT Internet of Things LOF Local Outlier Factor Mad Manipulation of demand MitM Man-in-the-Middle ML Machine Learning MLP Multilayer Perceptron MV Medium Voltage NEVI National Electric Vehicle Infrastructure NIST National Institute of Standards and Technology OEM Original Equipment Manufacturer OCHP Open Clearing House Protocol OCPI Open Charge Point Interface OCPP Open Charge Point Protocol OICP Open InterCharge Protocol OSCP Open Smart Charging Protocol PCID Provisioning Certificate Identifier PII Personally Identifiable Information PKI Public Key Infrastructure list of abbreviations xv PnC Plug-and-Charge PQC Post-Quantum Cryptography PUF Physical Unclonable Function PV Photovoltaic RAR Rich Authorization Request RF Random Forest ROC Receiver Operating Characteristic RSA Rivest-Shamir-Adleman SoC State of Charge SSI Self-Sovereign Identity TLS Transport Layer Security TPM Trusted Platform Module TPR True Positive Rate V2G Vehicle to Grid V2H Vehicle to Home L I S T O F P U B L I C AT I O N S During the course of my academic career, I co-authored several papers that I list below. Nine of these publications are part of this thesis as indicated in the list below. conference papers [A] Andreas Fuchs, Dustin Kern, Christoph Krauß, and Maria Zh- danova. “TrustEV: Trustworthy Electric Vehicle Charging and Billing.” In: Proceedings of the 35th ACM/SIGAPP Symposium on Applied Computing SAC 2020. ACM, 2020. doi: 10.1145/ 3341105.3373879. Not part of this thesis. [B] Andreas Fuchs, Dustin Kern, Christoph Krauß, and Maria Zh- danova. “Securing Electric Vehicle Charging Systems through Component Binding.” In: 39th International Conference on Com- puter Safety, Reliability and Security, SAFECOMP. Springer, Sept. 2020. doi: 10.1007/978-3-030-54549-9_26. Not part of this thesis. [C] Andreas Fuchs, Dustin Kern, Christoph Krauß, and Maria Zh- danova. “HIP: HSM-Based Identities for Plug-and-Charge.” In: Proceedings of the 15th International Conference on Availability, Reliability and Security. ARES ’20. Virtual Event, Ireland: Asso- ciation for Computing Machinery, 2020. isbn: 9781450388337. doi: 10.1145/3407023.3407066. Not part of this thesis. [D] Andreas Fuchs, Dustin Kern, Christoph Krauß, Maria Zh- danova, and Ronald Heddergott. “HIP-20: Integration of Vehicle-HSM-Generated Credentials into Plug-and-Charge In- frastructure.” In: Computer Science in Cars Symposium. CSCS ’20. Feldkirchen, Germany: Association for Computing Machinery, 2020, pp. 1–10. isbn: 9781450376211. doi: 10.1145/3385958. 3430483. Not part of this thesis. [E] Daniel Zelle, Timm Lauser, Dustin Kern, and Christoph Krauß. “Analyzing and Securing SOME/IP Automotive Services with Formal and Practical Methods.” In: Proceedings of the 16th Inter- national Conference on Availability, Reliability and Security. ARES ’21. Vienna, Austria: Association for Computing Machinery, 2021. isbn: 9781450390514. doi: 10.1145/3465481.3465748. Not part of this thesis. Best Paper Award. [F] Dustin Kern and Christoph Krauß. “Analysis of E-Mobility- based Threats to Power Grid Resilience.” In: Proceedings of xvii https://doi.org/10.1145/3341105.3373879 https://doi.org/10.1145/3341105.3373879 https://doi.org/10.1007/978-3-030-54549-9_26 https://doi.org/10.1145/3407023.3407066 https://doi.org/10.1145/3385958.3430483 https://doi.org/10.1145/3385958.3430483 https://doi.org/10.1145/3465481.3465748 xviii list of publications the 5th ACM Computer Science in Cars Symposium. CSCS ’21. Ingolstadt, Germany: Association for Computing Machinery, 2021. isbn: 9781450391399. doi: 10.1145/3488904.3493385. Part of this thesis. [G] Dustin Kern, Timm Lauser, and Christoph Krauß. “Integrating Privacy into the Electric Vehicle Charging Architecture.” In: Proceedings on Privacy Enhancing Technologies 3 (2022), pp. 140– 158. doi: 10.56553/popets-2022-0066. Not part of this thesis. [H] Dustin Kern, Christoph Krauß, Timm Lauser, Nouri Alna- hawi, Alexander Wiesmaier, and Ruben Niederhagen. “Quan- tumCharge: Post-Quantum Cryptography for Electric Vehicle Charging.” In: Applied Cryptography and Network Security. Ed. by Mehdi Tibouchi and XiaoFeng Wang. Cham: Springer Na- ture Switzerland, 2023, pp. 85–111. isbn: 978-3-031-33491-7. doi: 10.1007/978-3-031-33491-7_4. Part of this thesis. [I] Dustin Kern and Christoph Krauß. “Detection of e-Mobility- Based Attacks on the Power Grid.” In: 2023 53rd Annual IEEE/I- FIP International Conference on Dependable Systems and Networks (DSN). 2023, pp. 352–365. doi: 10.1109/DSN58367.2023.00042. Part of this thesis. [J] Dustin Kern, Christoph Krauß, and Matthias Hollick. “Detec- tion of Anomalies in Electric Vehicle Charging Sessions.” In: Proceedings of the 39th Annual Computer Security Applications Conference. ACSAC ’23. Austin, TX, USA: Association for Com- puting Machinery, 2023, pp. 298–309. isbn: 9798400708862. doi: 10.1145/3627106.3627127. Part of this thesis. [K] Adrian Kailus, Dustin Kern, and Christoph Krauß. “Self- Sovereign Identity for Electric Vehicle Charging.” In: Applied Cryptography and Network Security. Ed. by Christina Pöpper and Lejla Batina. Cham: Springer Nature Switzerland, 2024, pp. 137– 162. isbn: 978-3-031-54776-8. doi: 10.1007/978-3-031-54776- 8_6. Part of this thesis. Best Student Paper Award. [L] Dustin Kern, Christoph Krauß, and Matthias Hollick. “Attack Analysis and Detection for the Combined Electric Vehicle Charg- ing and Power Grid Domains.” In: Proceedings of the 19th Interna- tional Conference on Availability, Reliability and Security. ARES ’24. Vienna, Austria: Association for Computing Machinery, 2024. isbn: 9798400717185. doi: 10.1145/3664476.3664512. Part of this thesis. [M] Timm Lauser, Daniel Zelle, Dustin Kern, Christoph Krauß, and Lars Völker. “Security Protocols for Ethernet-Based In- Vehicle Communication.” In: 2024 IEEE Vehicular Networking https://doi.org/10.1145/3488904.3493385 https://doi.org/10.56553/popets-2022-0066 https://doi.org/10.1007/978-3-031-33491-7_4 https://doi.org/10.1109/DSN58367.2023.00042 https://doi.org/10.1145/3627106.3627127 https://doi.org/10.1007/978-3-031-54776-8_6 https://doi.org/10.1007/978-3-031-54776-8_6 https://doi.org/10.1145/3664476.3664512 list of publications xix Conference (VNC). 2024, pp. 148–155. doi: 10.1109/VNC61989. 2024.10575984. Not part of this thesis. [N] Viet Ha The, Dustin Kern, Phat Nguyen Tan, and Christoph Krauß. “Improving Anomaly Detection for Electric Vehicle Charging with Generative Adversarial Networks.” In: Proceed- ings of the 40th ACM/SIGAPP Symposium on Applied Computing SAC 2025. Accepted for publication. To appear. Catania, Italy: Association for Computing Machinery, 2025. isbn: 979-8-4007- 0629-5/25/03. doi: 10.1145/3672608.3707823. Part of this thesis. [O] Dustin Kern, Christoph Krauß, and Matthias Hollick. “Anomaly Detection and Mitigation for Electric Vehicle Charging-Based Attacks on the Power Grid.” In: Proceedings of the 40th ACM/SI- GAPP Symposium on Applied Computing SAC 2025. Accepted for publication. To appear. Catania, Italy: Association for Com- puting Machinery, 2025. isbn: 979-8-4007-0629-5/25/03. doi: 10.1145/3672608.3707802. Part of this thesis. preprint papers [P] Jonas Primbs, Dustin Kern, Michael Menth, and Christoph Krauß. Streamlining Plug-and-Charge Authorization for Electric Vehicles with OAuth2 and OIDC. 2025. doi: 10.48550/arXiv. 2501.14397. arXiv: 2501.14397 [cs.CR]. Part of this thesis. https://doi.org/10.1109/VNC61989.2024.10575984 https://doi.org/10.1109/VNC61989.2024.10575984 https://doi.org/10.1145/3672608.3707823 https://doi.org/10.1145/3672608.3707802 https://doi.org/10.48550/arXiv.2501.14397 https://doi.org/10.48550/arXiv.2501.14397 https://arxiv.org/abs/2501.14397 C O L L A B O R AT I O N S A N D M Y C O N T R I B U T I O N This thesis is based on nine publications that I had the pleasure to co- author with several excellent researchers: Nouri Alnahawi (Darmstadt University of Applied Sciences), Viet Ha The (Bosch Global Software Technologies), Matthias Hollick (Technical University of Darmstadt), Adrian Kailus (DB Systel GmbH), Christoph Krauß (Darmstadt Uni- versity of Applied Sciences), Timm Lauser (Darmstadt University of Applied Sciences), Michael Menth (University of Tübingen), Phat Nguyen Tan (Bosch Global Software Technologies), Ruben Niederha- gen (Academia Sinica; University of Southern Denmark), Jonas Primbs (University of Tübingen), and Alexander Wiesmaier (Darmstadt Uni- versity of Applied Sciences). I thank all co-authors for their invaluable contributions. In the following, I outline the author contributions for every publication that is part of this cumulative dissertation:1 Chapter 5 is based on the paper “QuantumCharge: Post-Quantum Cryptography for Electric Vehicle Charging” by Dustin Kern, Christoph Krauß, Timm Lauser, Nouri Alnahawi, Alexander Wiesmaier, and Ruben Niederhagen [H]. I contributed the background and related work analysis on Electric Vehicle (EV) charging security. Nouri Alna- hawi and Ruben Niederhagen contributed the background and related work analysis for Post-Quantum Cryptography (PQC) as well as the motivation for the importance of PQC. I contributed the system model design. All authors contributed to the adversary model design and to the requirement analysis. I contributed the concept for integrating crypto-agility with PQC support for EV charging. Christoph Krauß contributed an analysis of the quantum security of existing algorithms to the concept. I contributed the proof-of-concept implementation and the corresponding practical feasibility evaluation. Furthermore, I contributed to the formal security verification, by modeling rele- vant EV charging communication processes in Tamarin. Timm Lauser contributed to the formal security verification, by modeling relevant aspects of crypto-agility and refining the final Tamarin model for proof generation. Alexander Wiesmaier and Christoph Krauß contributed with feedback throughout the paper’s creation process. All authors contributed with discussions and proof-reading. Chapter 6 is based on the paper “Self-Sovereign Identity for Elec- tric Vehicle Charging” by Adrian Kailus, Dustin Kern, and Christoph Krauß [K]. The paper is based on the Master Thesis by Adrian Kailus of which Christoph Krauß and I were co-supervisors. Hence, Adrian Kailus contributed to the paper by creating the original Thesis, in- cluding background, related work, system model, adversary model, 1 References in this chapter refer to my list of publications given on Pages xvii to xix. xxi xxii collaborations and my contribution requirements, concept, implementation, and evaluation, which are all relevant to the paper. Christoph Krauß and I contributed to this Thesis with feedback throughout its creation process. For the paper, Adrian Kailus further contributed by transferring the Thesis’ related work, system model, adversary model, requirements, concept, and implementation sections. Christoph Krauß contributed by transferring the introduction, background, and conclusion sections. I contributed to the paper by transferring the evaluation and by adapting the concept to use Anoncreds for privacy protection. Additionally, I contributed the formal security and privacy analysis using the Tamarin prover. All authors contributed with discussions and proof-reading. Chapter 7 is based on the paper “Streamlining Plug-and-Charge Authorization for Electric Vehicles with OAuth2 and OIDC” by Jonas Primbs, Dustin Kern, Michael Menth, and Christoph Krauß [P]. I con- tributed the background and related work analysis on EV charging security. Jonas Primbs contributed the background and related work analysis on OAuth 2. I contributed the system-/adversary model de- sign and to the requirement analysis. Jonas Primbs contributed by designing the concept for the use of OAuth for EV credential manage- ment. I contributed to the concept design with insights on EV charging standards, processes, and requirements. Jonas Primbs contributed the proof-of-concept implementation and performance evaluation. I con- tributed the formal security analysis using the Tamarin prover as well as the requirements discussion. Michael Menth and Christoph Krauß contributed with feedback throughout the paper’s creation process. All authors contributed with discussions and proof-reading. Chapter 8 is based on the paper “Analysis of E-Mobility-based Threats to Power Grid Resilience” by Dustin Kern and Christoph Krauß [F]. I contributed the background and analysis of related work. Christoph Krauß contributed the first introduction and system model drafts. I contributed further details to the system model and the adver- sary model design. Additionally, I contributed the concept, implemen- tation, and evaluation. Christoph Krauß contributed with feedback throughout the paper’s creation process. All authors contributed with discussions and proof-reading. Chapter 9 is based on the paper “Detection of e-Mobility-Based Attacks on the Power Grid” by Dustin Kern and Christoph Krauß [I]. I contributed the background and analysis of related work. Additionally, I contributed the system-/adversary model design. Furthermore, I contributed the concept, implementation, and evaluation. Christoph Krauß contributed with feedback throughout the paper’s creation process. All authors contributed with discussions and proof-reading. Chapter 10 is based on the paper “Detection of Anomalies in Elec- tric Vehicle Charging Sessions” by Dustin Kern, Christoph Krauß, and Matthias Hollick [J]. I contributed the background and analysis of related work. Additionally, I contributed the system-/adversary collaborations and my contribution xxiii model design. Furthermore, I contributed the concept, implementation, and evaluation. Christoph Krauß and Matthias Hollick contributed with feedback throughout the paper’s creation process. All authors contributed with discussions and proof-reading. Chapter 11 is based on the paper “Attack Analysis and Detection for the Combined Electric Vehicle Charging and Power Grid Domains” by Dustin Kern, Christoph Krauß, and Matthias Hollick [L]. I con- tributed the background and analysis of related work. Additionally, I contributed the system-/adversary model design. Furthermore, I contributed the concept, implementation, and evaluation. Christoph Krauß and Matthias Hollick contributed with feedback throughout the paper’s creation process. All authors contributed with discussions and proof-reading. Chapter 12 is based on the paper “Improving Anomaly Detection for Electric Vehicle Charging with Generative Adversarial Networks” by Viet Ha The, Dustin Kern, Phat Nguyen Tan, and Christoph Krauß [N]. I contributed the related work analysis on EV charging Intrusion Detection Systems (IDSs). Viet Ha The contributed the related work analysis on Generative Adversarial Networks (GANs). I contributed the system-/adversary model excluding specifics of the GAN-based adversary, which were added by Viet Ha The. Viet Ha The contributed by designing the concept for a GAN-based IDS optimization. I con- tributed to the concept design with insights on its applicability to the EV charging IDS use-case. Viet Ha The contributed the implementa- tion and evaluation of the concept. Phat Nguyen Tan contributed to the implementation and aided with data collecting. I contributed to the implementation and evaluation by providing the original datasets and IDS for the GAN adversary to run against. Christoph Krauß con- tributed with feedback throughout the paper’s creation process. All authors contributed with discussions and proof-reading. Chapter 13 is based on the paper “Anomaly Detection and Mitiga- tion for Electric Vehicle Charging-Based Attacks on the Power Grid” by Dustin Kern, Christoph Krauß, and Matthias Hollick [O]. I con- tributed the background and analysis of related work. Additionally, I contributed the system-/adversary model design. Furthermore, I contributed the concept, implementation, and evaluation. Christoph Krauß and Matthias Hollick contributed with feedback throughout the paper’s creation process. All authors contributed with discussions and proof-reading. Part I S Y N O P S I S 1 I N T R O D U C T I O N Electric Vehicles (EVs) provide several advantages over traditional internal combustion engine vehicles. For one, EVs can provide a re- duction in greenhouse gas emissions throughout a vehicle’s life-cycle, especially if powered by low-carbon energy sources [50]. Additionally, an increasing share of EVs can have a positive impact on air quality by reducing pollution [83]. Hence, EVs are commonly considered to be an important aspect of the transition to a more sustainable mobility sector. Due to the many advantages of EVs, governments around the world are enacting regulations and incentives to accelerate the adoption of EVs [92]. In fact, the global number of EVs has already exceeded 40 million in 2023 and is expected to grow to 250 million in 2030 based on stated government policies [58]. 1.1 context on EV charging EVs are “refueled” by charging at a Charge Point (CP). This process involves a variety of backend systems/actors as well as communication protocols to enable the dynamic exchange of relevant information between the different actors. A high-level overview of the EV charging system model is shown in Figure 1. Besides EV and CP, this system model includes three actors: (i) the e-Mobility Service Provider (eMSP), who has contractual relations with EV users for the billing of charging sessions, (ii) the Charge Point Operator (CPO), who manages their CPs for the metering and control of charging sessions, and (iii) the Distribution System Operator (DSO), who ensures that power grid operation maintains within its capacity. This system model implements two main use-cases regarding the charging of EVs, namely the billing and control of charging sessions: charge billing : For the billing of charging sessions, an EV user has a contract with an eMSP. Additionally, the user’s EV is provisioned with digital credentials (key pair and certificate) that are generated by the eMSP and uniquely identify the user’s contract. The EV can then use these credentials to authenticate itself at a CP, which serves as the basis of charge authorization and billing. Hereby, the process of provisioning credentials into an EV requires communication from the eMSP over CPO and CP to the EV and the processes of charge authorization/-billing 3 4 introduction Charge Point Operator (CPO) Charge Point (CP) Electric Vehicle (EV) e-Mobility Service Provider (eMSP) EV User Distribution System Op- erator (DSO) Distribution Grid Communication Energy C om m u - ni ca ti on Communication En er gy Communication C om m un ic at io n e-Mobility Service Contract Figure 1: High-Level EV Charging System Model requires communication from the CP over CPO to the eMSP. Overall, this process of automatic EV authorization/billing is referred to as Plug-and-Charge (PnC). charge control : For the control of charging sessions, the DSO communicates with distributed measurement devices of the power grid. The received measurements serve to build an accu- rate estimation of the grid’s state. Based on this estimate, the DSO communicates group capacity allotments to CPOs that de- fine the maximum allowed power consumption over time. A CPO can then distribute the allotted capacity among its CPs. Once an EV is connected to a CP, they can negotiate a consump- tion profile based on the allotted capacity, physical charging limits, and tariff information. Overall, this process of controlling charging sessions can be used to avoid peaks in EV power de- mand and ensure a more grid-friendly charging behavior, i.e., to implement load balancing. 1.2 motivation and problem statement The outlined use-cases of charge billing and -control also serve to show the associated cybersecurity risks of EV charging. For one, the PnC-based automated authorization and billing process opens several potential threat vectors that may be exploited by an adversary with the risk of significant financial damages. Second, while the existing control methods for load balancing can have a positive effect on power grid stability if used appropriately, the potential abuse of these control methods by an adversary results in the threat of charging load-based attacks on grid resilience. More specifically, in the context of this thesis, we identify the fol- lowing security risks in the existing PnC processes: fixed cryptographic algorithms : Current communication pro- tocols do not provide crypto-agility, but define fixed signature 1.2 motivation and problem statement 5 and encryption algorithms [75]. This significantly increases the negative impact of a security vulnerability in one of the used algorithms as the migration to new algorithms is complex and would require changes in existing protocol standards. One very specific vulnerability in current algorithms is their insecurity against quantum computers. While current quantum computers do not possess the required capacity to pose an immediate threat, experts generally see an increasing risk of practical attacks over the next years [99]. To address this risk, the United States have already defined the year 2035 as the target for completing the mi- gration of federal systems to Post-Quantum Cryptography (PQC) [98]. Considering the long lifespans of relevant systems (up to 35 years for EVs [63]), this motivates the importance of crypto- agility and PQC in EV charging. lack of data protection : Current communication protocols do not include appropriate privacy-preserving measures [67]. In- stead, with existing protocols different actors simply exchange a variety of personal data, such as unique user identifiers, charging times, and locations [76]. Hereby, the main issue is the linkability between different PnC authentications of the same user across locations by all involved backend actors. This circumstance en- ables the creation of movement profiles and the inferring of consumer habits and motivates the importance of considering privacy-preserving alternatives to the current PnC authentication process. complex/proprietary security architecture : Current com- munication protocols are based on a complex centralized Public Key Infrastructure (PKI) and require proprietary implementa- tions of security-relevant processes [114]. The high level of com- plexity resulted in critique by relevant stakeholders [14] and shows the importance of complexity-reducing measures [29]. Additionally, relevant parts of the PnC process are out-of-scope for existing protocols, for example, the method of provisioning eMSP-generated credentials into an EV, requiring operators to implement a patchwork of different solutions (e.g., [132]). This motivates the consideration of comprehensive alternatives based on standardized solutions to streamline the underlying security architecture. Addressing these outlined risks would significantly contribute to the security of EV charging protocols and the related authorization, billing, and load balancing processes. However, even with secure protocols, a relevant risk is posed by the potential compromise of individual systems. This risk is especially relevant regarding attacks on load balancing since compromised systems could manipulate related data without the need to handle any additional protocol security 6 introduction measures [71]. The threat of system compromise is arguably most prevalent in EVs and CPs, which are commonly left unattended in public places posing a significant risk of physical compromise [76]. Published reports on locally-/remotely exploitable vulnerabilities in related systems further motivate the importance of considering the consequences of system compromise; for example, in vehicles [15, 126] or CPs [64, 146]. One method of addressing the threat of attacks on load balancing processes by compromised systems is through prevention, e.g., by means of secure system/software design or by employing system security measures such as a secure-/measured boot based on trusted hardware [38–41]. However, since no preventative measure can guaran- tee perfect security, attack detection is an important aspect of ensuring the resilient operation of cyber-physical systems in the presence of an active adversary [117]. This circumstance motivates the consideration of attack detection measures related to load balancing processes of EV charging. More specifically, in the context of this thesis, we identify the fol- lowing security risks in the existing load balancing processes: small-scale attacks : By manipulating the charging control pro- cess of an individual session, an adversary may cause physical damage to local components [72]. For example, an adversary- controlled EV/CP may ignore dangerous operating conditions or manipulate charging behavior to cause physical damage to the EV’s battery and potentially start a fire [118]. large-scale attacks : By manipulating the charging control pro- cess of several sessions at the same time, an adversary may cause significant harm to power grid stability [71]. Several simulation- based analyses investigate charging load-based attacks on the grid [2, 31, 70, 146], showing a potential impact that ranges from frequency instability over line outages to large-scale blackouts. 1.3 methodology and contributions In the context of this thesis, we have the goal of investigating meth- ods for addressing the previously outlined threats. For this, related contributions are split over nine publications [48, 67, 70–75, 114]. As is common practice in the field, these publications methodologically follow the Design Science Research Process (DSRP) model [112], mean- ing we start by analyzing a specific system model and related work under consideration of an assumed adversary with defined goals and capabilities. This analysis serves the purpose of problem identification and motivation by deriving relevant threats to the system. Afterwards, the objectives of a solution are defined in the form of requirements that address the derived threats. Following this, we design a security 1.3 methodology and contributions 7 concept to meet the identified threats/requirements. The concept is implemented and evaluated to assess the effectiveness of its security measures. Finally, the results of the work are communicated in the form of a peer-reviewed scientific publication along with a presenta- tion and discussions at a scientific conference. More specifically, our contributions can be categorized into (i) pre- ventive security measures that propose changes/extensions to existing communication protocols in order to address identified security threats and (ii) measures for the analysis and detection of attacks on the con- trol of EV charging sessions. Regarding preventive security measures, our evaluations consider performance/usability aspects based on a proof-of concept implementation running on resource-constrained de- vices as well as security aspects based on a symbolic protocol analysis using the Tamarin prover [95]. Regarding analysis/detection measures, our evaluations are based on simulated attacks considering various aspects of EV charging and its relation to power grid operation. The main contributions of our publications can be summarized as follows: (i) We design, implement, and evaluate a concept for the integration of crypto-agility and PQC support into existing EV charging and billing processes (cf. [75]; Chapter 5). (ii) We design, implement, and evaluate a concept for privacy- preserving charge authorization and billing based on a decen- tralized Self-Sovereign Identity (SSI) system (cf. [67]; Chapter 6). (iii) We design, implement, and evaluate a concept that streamlines the existing security architecture based on OAuth 2, representing a widely accepted and standardized solution (cf. [114]; Chap- ter 7). (iv) The security of all three concepts [67, 75, 114] is shown by formal analyses using the Tamarin prover. For [67], a privacy analysis is additionally performed using the Tamarin prover. All our Tamarin models are published for reproducibility of the auto- mated proofs and for reusability of the used modeling concepts. (v) We perform a simulation-based analysis of EV charging load- based attacks on the grid, identifying high-impact attack strate- gies and showing the risk of large-scale power outages as EV adoption increases (cf. [70]; Chapter 8). (vi) We design, implement, and evaluate a distributed hybrid Intrusion Detection System (IDS) for the detection of large-scale coordi- nated charging-based attacks on the grid, combining a rule- based IDS with regression-based anomaly detection (cf. [71]; Chapter 9). 8 introduction (vii) We design, implement, and evaluate a hybrid IDS for the detec- tion of anomalies in individual charging sessions, combining regression-based charging load forecasting with classification- and novelty-based anomaly detection (cf. [72]; Chapter 10). (viii) We design, implement, and evaluate a co-simulation concept covering relevant physical- and communication aspects of EV charging and the power grid, enabling the analysis of, among other things, novel/stealthy attack vectors (cf. [73]; Chapter 11). Additionally, we design, implement, and evaluate an IDS for the detection of these novel/stealthy large-scale attacks, incorporat- ing features from both the EV charging and power grid domains (cf. [73]; Chapter 11). (ix) We design, implement, and evaluate a Generative Adversarial Network (GAN)-based IDS training method for EV charging that can help to eliminate gaps or biases in existing IDSs (cf. [48]; Chapter 12). (x) We design, implement, and evaluate a two-step IDS that com- bines large-scale with session-specific detection, further enhanc- ing detection performance (cf. [74]; Chapter 13). Additionally, we design, implement, and evaluate attack mitigation measures that either correct manipulated data or counteract malicious changes in charging load based on the different IDS outputs (cf. [74]; Chapter 13). (xi) The used data sets of all IDS papers [48, 71–74] are published for reproducibility and for future use in related studies. Similarly, the IDS code of [48, 72–74] is published. Additionally, the code of the co-simulation environment for [73] is published. 1.4 structure of the thesis The remainder of this thesis is structured as follows: Chapter 2 presents the relevant background on EV charging, its billing-related processes, and its relation to power grid opera- tion. The chapter starts with an overview of existing EV charging actors and their relations as well as relevant communication protocols and standards in Section 2.1. Afterwards, we derive detailed system models for the two main use-cases of charge authorization/-billing and charge control/load balancing in Sec- tion 2.2. Finally, related adversary models are motivated and discussed in Section 2.3. Chapter 3 discusses our contributions. The chapter starts with a discussion of our work that focuses on the prevention of attacks on EV charging authorization/-billing in Section 3.1. Afterwards, 1.4 structure of the thesis 9 we discuss our work that focuses on the analysis, detection, and mitigation of attacks on charge control/load balancing in Sec- tion 3.2. For both subject areas, we highlight the motivations, main contributions, and key findings of our individual publica- tions. Additionally, we provide comparisons of our publications among themselves and within the context of related work. Chapter 4 provides concluding remarks. We summarize the main contributions and key takeaways of this thesis in Section 4.1. Afterwards, we discuss potential topics for future work in Sec- tion 4.2. Chapters 5 to 7 comprise papers that present concepts for the prevention of attacks on EV charging by enhancing protocol security. In Chapter 5, we present a concept for adding crypto- agility and PQC support to existing EV charging communication. In Chapter 6, we propose an approach that uses SSIs for EV charging authorization and protects the privacy of EV users. In Chapter 7, we design a method to streamline EV credential management while enabling more fine-grained authorizations based on OAuth. Chapters 8 to 13 comprise papers that present concepts for the analysis and detection of attacks on EV charging session con- trol. In Chapter 8, we present an analysis of the impact of EV charging-based attacks on the power grid. In Chapter 9, we design an IDS for the detection of large-scale coordinated EV charging-based attacks. In Chapter 10, we propose an IDS that detects anomalies in individual EV charging sessions. In Chap- ter 11, we develop a co-simulation concept for the combined EV charging and grid domains, which is used for the analysis of advanced/stealthy attacks and for the design of a corresponding detection approach. In Chapter 12, we present a method for op- timizing EV charging IDSs utilizing GAN-generated attack data. In Chapter 13, we propose the combination of large-scale and charging-based detection systems and develop attack mitigation methods that use the respective IDS outputs. 2 B A C K G R O U N D The following sections provide the relevant background for this thesis. Section 2.1 provides a brief background on EV charging actors, pro- tocols, and standards. Afterwards, Section 2.2 describes the derived system model for the authorization, billing, and control of EV charging sessions in detail. Finally, Section 2.3 describes the assumed adversary model. 2.1 EV charging actors , protocols , and standards EV charging primarily involves an EV receiving energy from a CP. Besides these primary actors, several secondary/backend actors play a role in the authorization, billing, and control of charging sessions. For authorization and billing, the most important backend actor is the eMSP, with whom the EV user is registered. The eMSP and EV user are in a contractual relationship for the billing of sessions, i.e., the user pays the eMSP and the eMSP pays whoever provided the energy. In practice, this is implemented by providing eMSP-generated credentials (key pair and certificate) to the EV, which serves as the basis for a secure authorization/billing process. Besides the eMSP, the current EV charging infrastructure involves a number of additional backend actors for charge authorization/billing. First, the CPO, who manages their CPs and is the only backend actor that directly communicates with the CPs. That is, all authorization/ billing relevant traffic needs to be forwarded over the CPO. Second, to handle relations between eMSPs and CPOs and to reduce the number of required connections, an intermediary backend actor is usually introduced, namely the Contract Clearing House (CCH). The role of the CCH in authorization/billing is mainly just to forward traffic be- tween associated eMSPs and CPOs. Third, an additional backend actor, the Certificate Provisioning Service (CPS), is introduced to serve as a trusted third party between eMSP and EV for the process of installing eMSP-generated credentials into the EV. Finally, the EV’s Original Equipment Manufacturer (OEM) also plays a role in the authoriza- tion/billing process by providing the EV with initial credentials that can be used to secure the process of installing eMSP credentials into the EV. Notably, overlapping between roles is possible. For instance, an OEM may also provide eMSP services to its EV users. Additionally, an eMSP may also act as CPO and/or implement the functions of the CPS. 11 12 background For the control of charging session/load balancing the most im- portant backend actor is the DSO, who monitors/manages the local distribution-level power grid to ensure that grid operations stay within its normal operation bounds. The implementation of load balancing requires the operation of a number of remote stations in the grid for the collection of relevant grid measurements, enabling grid state esti- mations. Based on these estimations, desirable load balancing profiles can be defined, e.g., to reduce charging loads during peak demand times and reduce the overall stress on the grid. To put these profiles into action, the DSO needs to communicate them to CPOs who can distribute them to CPs. Over the past years, several communication protocols/standards have been developed that enable the required interactions between the different actors [28]. ISO 15118 [61–63] is an international stan- dard that defines a communication interface between an EV and a CP. The communication between a CPO and their CPs is usually im- plemented using the de-facto standard protocol Open Charge Point Protocol (OCPP) [106–108]. Several protocols exist to handle the com- munication between eMSP, CCH, and CPO [68], e.g.: Open Clearing House Protocol (OCHP) [123] (with its extension OCHPdirect [122]), Open InterCharge Protocol (OICP) [54, 55], eMobility Interoperation Protocol (eMIP) [46], or Open Charge Point Interface (OCPI) [105]. Between DSO and CPOs, the Open Smart Charging Protocol (OSCP) [109] can be seen as an internationally representative protocol [49]. Finally, to handle the management of grid components by the DSO, different options exist [10], e.g., IEC 60870-5-104 [59] (in Europe) or Distributed Network Protocol 3 (DNP3) [21] (in the US). 2.2 detailed system models Based on the existing actors/protocol definitions for EV charging, we derive detailed system models for the core use-cases of charge autho- rization/billing and charge control/load balancing in the following sub-sections. These detailed system models provide the basis of the respective models we used for the various publications that this thesis is based on. Hereby, we chose different abstraction levels per paper, depending on the respective paper’s required level of detail, to reduce unnecessary complexities for the reader. Specifically, we may leave out actors that exclusively forward data and, thus, have no effect on a paper’s research problem/concept design. In Section 2.2.1 we derive the detailed system model of the charge authorization/billing use-case, which is relevant to the papers in Chapters 5 to 7 [67, 75, 114]. In Section 2.2.2 we derive the detailed system model of the charge control/load balancing use-case, which is relevant to the papers in Chapters 8 to 13 [48, 70–74]. 2.2 detailed system models 13 Contract Clearing House (CCH) Charge Point Operator (CPO) Charge Point (CP) Electric Vehicle (EV) e-Mobility Service Provider (eMSP) Certificate Provisioning Service (CPS) Original Equipment Manufac- turer (OEM) Provisioning Credential 1 Credential Installation Req. Credential Installation Req. Credential Installation Req. 2 3 5 Data Forwarding 4 9 14Gen. and Enc. Contract Credentials Sign Enc. Data 67 Contract Credentials Contract Credentials Contract Credentials 8 1011 PnC Auth. Billing Data Billing Data 12 13 15 Figure 2: Detailed Charge Authorization and Billing System Model 2.2.1 Charge Authorization and Billing On a high-level, the required processes for charge session autho- rization and billing start during EV manufacturing where the EV is provided with provisioning credentials by its OEM. These provision- ing credentials are used to install eMSP contract credentials into the EV, which can later be used for EV authentication at a CP. In more detail (cf. Figure 2), the different process steps can be summarized as follows: preparation : The EV comes pre-equipped with provisioning cre- dentials (based on elliptic-curve cryptography) from its OEM (cf. Figure 2; Step 1). The respective credential certificate in- cludes a Provisioning Certificate Identifier (PCID), that uniquely identifies the EV. This PCID is provided by the EV user to the eMSP during the conclusion of a charging contract such that the eMSP can link between the user’s EV and contract (not shown in Figure 2). credential installation : Once the EV is connected to a CP for the first time it can use its provisioning credentials to request the installation of eMSP contract credentials (cf. Figure 2; Step 2). For this, the EV generates a credential installation request and signs it with the provisioning credential private key. The installation request contains the provisioning credential certificate and is forwarded over CP/CPO/CCH to the eMSP (cf. Figure 2; Steps 3– 5). The eMSP validates the provisioning credential certificate based on the respective OEM root of trust and verifies the EV’s signature based on the provisioning credential certificate’s public key. If the verification is successful, the eMSP generates contract credentials for the EV (cf. Figure 2; Step 6). Hereby, the respective credential certificate includes an e-Mobility Account Identifier 14 background (eMAID), that uniquely identifies the user’s charging contract. The contract credentials are encrypted for the EV based on the provisioning credential certificate’s public key and the encrypted data is signed by the CPS and appended with the CPS’ certificate (cf. Figure 2; Step 7). The results are sent over CCH/CPO/CP to the EV (cf. Figure 2; Steps 8–11), which validates the CPS’ certificate based on a locally installed Vehicle to Grid (V2G) root and verifies the signature based on the CPS’ certificate public key. Afterwards, the contract credentials are decrypted based on the provisioning credential private key. credential usage : If the EV has valid contract credentials, it can authenticate itself at a CP via ISO 15118’s PnC processes (cf. Figure 2; Step 12). For this, a challenge-response protocol is ex- ecuted between EV and CP, which starts with the EV sending its contract credential certificate to the CP. The CP responds with a nonce, which the EV signs using its contract credential private key. The CP can validate the contract credential certifi- cate based on the respective eMSP root and verifies the EV’s signature based on the contract credential certificate’s public key. If the verification is successful, the EV is cleared to charge. Dur- ing the charging session, the CP periodically receives measure- ments of the consumed energy by its meter. The CP can request signatures over these measurements from the EV (not shown in Figure 2). For these signatures, the EV again uses its contract credential private key. Once the charging session is completed, the CP forwards all billing relevant (eMAID, times, measure- ments, optional signatures, etc.) data to the CPO (cf. Figure 2; Step 13). The CPO forwards this data over the CCH to the eMSP (cf. Figure 2; Steps 14–15). This way, the CPO can bill the eMSP for the user’s consumed energy and the eMSP can bill the user. 2.2.2 Charge Control and Load Balancing On a high-level, the required processes for charge control and load balancing start with the DSO monitoring the grid and deriving state estimations. Afterwards, the estimated capacities can be distributed to different controllable loads, including CPOs. CPOs then split their allotted capacity among their CPs and finally, specific charging pa- rameters are negotiated between an EV and a CP. In more detail (cf. Figure 3), the different process steps can be summarized as follows: charge control inputs : A CPO can receive charge control-rele- vant input from two sources, namely from the respective eMSP and DSO. The DSO is generally responsible for the monitoring and management of the distribution grid (cf. Figure 3; Steps 1– 2). This includes the measurement of industrial or aggregated 2.2 detailed system models 15 Charge Point Operator (CPO) Charge Point (CP) Electric Vehicle (EV) e-Mobility Backend Systems (for billing etc.) Distribution System Operator (DSO) Distribution Grid Measurement-/ Control Devices Distributed Loads (residential, industrial, etc.) Distributed Energy Resources (PV, wind, etc.)En er gy Energy Energy Energy 8 Control Signals Measure- ments 21 Group Capacity Forecast Group Measurements Request New Capacity 3 10 11 Charging Profiles Session Updates 59 Tariff Data 4 Charge Parameters Profile Selection 6 7 Figure 3: Detailed Charge Control and Load Balancing System Model residential loads as well as the measurement and control of Distributed Energy Resources (DERs) (Photovoltaic (PV), wind, etc.). Based on the various measurement inputs, the DSO gen- erates grid state estimations to deal with missing/delayed data and potential measurement inaccuracies and obtains a view of the grid’s state that is as accurate as possible. State estimates in combination with historic data can then be used to forecast the expected available capacity over time. These capacity fore- casts are then divided across CP groups and distributed to the respective CPOs (cf. Figure 3; Step 3). Besides capacity limits, another method of implementing load balancing is through price incentives in order to nudge consump- tion into the desired direction (e.g., cheaper charging during off- peak hours) [27, 43, 90]. For this purpose, an eMSP could define specific tariffs that incentivize grid-friendly charging behavior and provide these tariffs to the CPO (cf. Figure 3; Step 4). charging profile distribution and negotiation : Based on the received inputs, the CPO generates charging profiles for their CPs that define the maximum allowed consumption over time and can optionally include associated tariff data (cf. Fig- ure 3; Step 5). Charging profiles can be defined either session- independent or in relation to an individual charging session and multiple overlapping profiles per CP are possible. Once an EV is connected to a CP (and authorized to charge), the two systems perform a charge profile negotiation (cf. Figure 3; Steps 6–7). The EV starts by sending its supported current/volt- age limits, an estimate of its required energy, and optionally its planned departure time. The CP can forward the received data to 16 background the CPO, for purposes of session planning and session-specific profile generation, and responds to the EV with its own sup- ported current/voltage limits and its allowed charging profiles. Afterwards, the EV responds with its selected profile and option- ally its planned energy consumption over time (if the planned consumption is less than the allowed maximum). Finally, the charging process can start (cf. Figure 3; Step 8). charging session : During a charging session, the CP receives pe- riodic measurements from its electricity meter. Measurements can be reported to the CPO (cf. Figure 3; Step 9) either as clock- aligned- or session-event-related meter values and the CP can indicate different types of measurands (e.g., energy import/ex- port, voltage, current, EV State of Charge (SoC)) with optional signatures. During the charging process, both involved parties can always initiate a re-negotiation of charging parameters/profiles. Addi- tionally, ISO 15118 in its newest version [63] also offers support for bidirectional charging. Bidirectional charging can be used in the Vehicle to Home (V2H) context, e.g., by charging the EV during times with low energy prices and using the stored energy during high prices to reduce the owner’s energy bill. Moreover, bidirectional charging can be used in the V2G context, e.g., by using EVs as a distributed energy storage to compensate for existing fluctuations in energy demand/generation especially regarding PV and wind sources. group measurements and capacity adjustment : Finally, the CPO sends periodic metering aggregates to the DSO indicating the combined energy consumption of defined groups of CPs (cf. Figure 3; Step 10). This data may be used by the DSO as input to future state estimations and capacity forecasts. Additionally, if the DSO’s capacity forecast is not sufficient for the CPO’s CP consumptions, the CPO can request additional capacities from the DSO (cf. Figure 3; Step 11). 2.3 adversary models Based on the derived system models for EV charging (cf. Section 2.2), we describe and motivate different adversary models for the core use-cases of charge authorization/billing and charge control/load bal- ancing in the following sub-sections. These adversary models provide the basis of the respective models we used for the various publications that this thesis is based on. Hereby, we use different sub-sets per paper, depending on the respective paper’s focus (e.g., security vs. privacy). Specifically, we may leave out actors with that exclusively forward 2.3 adversary models 17 data and thus, have no effect on a paper’s research problem/concept design. In Section 2.3.1 we discuss the adversary model of the charge autho- rization/billing use-case, which is relevant to the papers in Chapters 5 to 7 [67, 75, 114]. In Section 2.3.2 we discuss the adversary model of the charge control/load balancing use-case, which is relevant to the papers in Chapters 8 to 13 [48, 70–74]. 2.3.1 Charge Authorization and Billing A successful attack on EV charging-related authorization and billing data/processes could cause significant financial damages [33] to oper- ators (e.g., if adversaries can charge for free) and/or users (e.g., if an adversary can charge on the account of a benign user). Additionally, attacks on EV charging communication protocols can threaten user privacy [76], charging safety [118], and ultimately power grid stability [2, 31, 70, 146]. Hence, it is important to consider a wide variety of competent adversaries. When looking at protocol security/privacy, we generally consider the following two types of adversaries: network adversary : We consider the threat of a remote adver- sary with full control over the used communication channels based on the Dolev-Yao model [25]. That is, an adversary with a Man-in-the-Middle (MitM) position on all communication pro- cesses who can intercept, modify, or drop any existing messages and arbitrarily create new messages. The adversary, however, cannot break any of the deployed cryptography unless they have access to the used private keys or in the case of [75] have access to a sufficiently powerful quantum computer that can break conventional cryptographic primitives. adversary with access to leaked backend data : We consi- der the threat of a remote adversary who has gained access to security-relevant user-specific data from the backend systems. Specifically, this concerns the contract credentials of all users, which are currently (cf. Section 2.2.1) generated in the backend systems of eMSPs. This setup enables the risk of a large-scale compromise of billing-relevant user data. For security-focused threat analyses [75, 114], we additionally con- sider the possibility of local adversaries with physical access to sys- tems: compromised EV: We consider the threat of a local adversary with physical access to an EV, who can physically tamper with, add, or remove included components/Electronic Control Units (ECUs). This adversary could be the owner of the EV or someone who has gained temporary access to it (e.g., in the car-sharing 18 background use-case). The adversary can use this access to modify or extract any local data (e.g., the user’s contract credentials) unless the data is kept in a secure storage (e.g., in a Hardware Security Module (HSM)). compromised CP: We consider the threat of a local adversary with physical access to a CP. Similarly to the compromised EV case, the adversary can physically tamper with, add, or remove in- cluded components and modify or extract any local data unless it is tamper-protected. Finally, for a privacy-focused threat analysis [67], we additionally consider the potential of ill-intentioned operators: honest-but-curious operators : We consider the privacy risk that results if Personally Identifiable Information (PII) is unneces- sarily exposed to various backend operators. Modeling operators as active adversaries is often considered to be too restrictive of an adversary model as operators are bound by regulations, audits, and the desire to maintain reputation, leading to the honest-but- curious operator model [111]. In this model, existing operators try to learn as much information as possible from legitimately received messages while not deviating from the defined proto- cols. Additionally, we assume that the different operators do not collaborate beyond the defined protocols, e.g., to exchange additional user-related data. 2.3.2 Charge Control and Load Balancing The previous adversaries/threats from Section 2.3.1 mainly cover the security of communication protocols and the involved authorization- /billing-relevant data. However, even if the respective protocols/data were to be secured, a significant threat remains, namely, the threat of compromised systems realizing adversarial changes in the physical charging behavior. Specifically, an adversary with control over CPs and/or EVs could use this access to manipulate the affected device’s charging behavior to cause physical damage to components (e.g., damaging the EV’s battery and starting a fire [118]). Additionally, an adversary with control over a large number of CPs and/or EVs could use this access to conduct demand-side attacks on the grid similar to [125] as evaluated in several publications [2, 31, 70, 146]. Notably, while attacks on the power grid may be considered as low-probability high-impact events, the consideration of such events is arguably very important in the context of critical infrastructure re- silience [85]. The power grid is a critical infrastructure where possible disruptions can lead to vast consequences ranging from significant economic damages [65] to severe harm to human life [7]. This circum- stance makes the grid an attractive target for sophisticated attacks 2.3 adversary models 19 [53] with an increasing risk of cyber attacks by advanced persistent threat actors like nation-state adversaries [45]. Besides evaluations in research [2, 31, 70, 146], the importance of considering EV charging- based attacks on the grid is further demonstrated by related reflections of industry stakeholders [49, 120] and by real-world examples of cy- ber attacks being used to cause wide-spread power outage in the Ukrainian grid in 2015/16 [13, 79]. An adversary could achieve the compromise of EVs/CPs via locally and/or remotely exploitable vulnerabilities; see [15, 126] for examples of vulnerabilities in vehicles or [64, 146] for CPs. Specific examples of real-world exploits include: the compromise of CPs via an insecure local interface [22], the remote compromise of a vehicle’s internal systems via its cellular interface [96], the remote compromise of CPs via their web interfaces [31], or the compromise of EV charging via an insecure charge control protocol [147]. Especially remotely exploitable vulnerabilities enhance the scalability of attacks and thus increase the potential threat posed to power grid resilience. On a base level, the adversary’s potential attack strategies can be grouped into two categories: Manipulation of demand attacks : Manipulation of demand (Mad) attacks generally describe the malicious alteration of a system’s power demand [125]. In the context of this thesis, Mad attacks could be executed by an adversary with control over the EV and/or CP involved in a charging session. More specifically, an adversary with control over the EV can send manipulated current/voltage limits as well as estimated energy needs and expected departure time to affect the resulting charging speeds. Additionally, the EV can adversely affect the charging profile negotiation process with its profile selection, e.g., to counteract price incentive-based load balancing. Similarly, an adversary with control over the CP can send manipulated current/voltage limits to affect charging speeds. Moreover, the CP can send arbitrary charging profiles to the EV in order to conduct Mad attacks, as profiles are not authenticity protected by the CPO. Furthermore, an adversary with control over both the EV and CP involved in a charging session can almost arbitrarily (within physical limits, limits of safety devices, etc.) change the charging process without having to rely on the constraints of current communication protocols. These kinds of attacks could be used to cause overload scenarios or a significant imbalance in power demand and generation and thus harm the grid [125]. False Data Injection attacks : False Data Injection (FDI) attacks generally describe the malicious alteration of data used for state estimation [86]. In the context of this thesis, FDI attacks could be executed by an adversary with control over the EV and/or 20 background CP involved in a charging session. More specifically, an adver- sary with control over the EV can send manipulated values for its estimated energy needs, expected departure time, and planned power consumption over time. Similarly, an adversary with control over the CP can manipulate any of the EV’s load balancing-relevant data that it forwards to the backend, as this data is not authenticity protected by the EV. Moreover, the CP can send arbitrary metering data to the backend, if these values are not securely authenticity protected by the meter. If any of these values are used in state estimation or capacity forecast- ing, a manipulation could deteriorate estimations leading to a larger divergence in assumed- and actual power demand, thus, potentially degrading grid stability. For more advanced analyses and detection designs [48, 73], we additionally consider the following attack strategies: combined attacks : For [73] we introduce the possibility of an ad- versary with control over EVs and/or CPs to execute attacks combining Mad and FDI. This attack strategy could be used to execute stealthy attacks, posing a challenge to detection sys- tems, for example, by executing a Mad attack to increase power demand while at the same time reporting “normal” metering values via an FDI attack in order to hide the existence of the attacker. Additionally, this attack strategy could be used to in- crease the potency of attacks, for example, by executing a Mad attack to increase power demand while at the same time, report- ing reduced demand via an FDI attack in order to increase the divergence in assumed- and actual power demand even further. Generative Adversarial Network attacks : For [48] we in- troduce the possibility of two GAN-specific attack strategies. These two strategies generally target an IDS that is deployed to detect related Mad/FDI attacks in the EV charging context. First is a poisoning attack [17], which targets the training process of an IDS by generating adversarial samples that are injected into the IDS’ training data (e.g., during periodic re-training). For this, a GAN can be used to generate charging data samples that appear authentic but include minor deviations to degrade IDS performance. Second is an evasion attack [143], which targets the operational phase of an IDS by generating adversarial samples that evade detection. This way, a GAN can be used to find spe- cific Mad/FDI attack implementations that current systems fail to detect. 3 C O N T R I B U T I O N S The following sections provide an overview of the contributions that are offered by our publications. More specifically, we discuss key find- ings and contextualize our publications with regard to related work. Section 3.1 deals with the parts of this work that focus on securing charge authorization/billing use-cases by designing preventive mea- sures that improve protocol security (cf. Chapters 5 to 7 [67, 75, 114]). Section 3.2 deals with the parts of this work that focus on the analy- sis, detection, and mitigation of attacks on the charge control/load balancing use-cases (cf. Chapters 8 to 13 [48, 70–74]). 3.1 prevention of attacks on EV charging by enhancing protocol security As shown in our discussions of the charge authorization/-billing system- and adversary model Sections 2.2.1 and 2.3.1, existing proto- cols and standards already provide a basic level of security. However, in the presence of advanced adversaries, they fail to meet vital secu- rity/privacy requirements. For instance, while the use of Transport Layer Security (TLS) – which is generally either supported by the considered protocols (e.g., OCPP) or required (e.g., for PnC with ISO 15118) – can generally prevent attacks by a network adversary, it does not protect the security of billing-relevant data from a com- promised EV/CP or protect user privacy from an honest-but-curious operator. Several security/privacy issues with the existing EV charging pro- tocols have already been discussed/addressed by related work (cf. Section 3.1.2). However, we were able to identify three specific issues with room for further research, which we try to address with our pub- lications in Chapters 5 to 7 [67, 75, 114]. Specifically, we identified: (i) a lack of crypto-agility and no consideration of post-quantum secure cryptographic algorithms, (ii) a strictly centralized security architec- ture posing a single point of failure and raising privacy concerns, and (iii) a complex PKI with proprietary credential management process and remaining gaps in standardization. Details on our contributions and their comparison to related work are discussed in the following sub-sections. 21 22 contributions 3.1.1 Discussion of Contributions Our contributions with regards to the security of charge authorization and billing processes include: (i) a protocol extension to ISO 15118 which enables crypto-agility and PQC support (cf. [75]; Chapter 5), (ii) an architecture for the use of a decentralized SSI system that enables privacy-preserving charge authorization and billing (cf. [67]; Chapter 6), and (iii) a method for streamlining credential management based on the widely accepted and standardized OAuth 2 solution (cf. [114]; Chapter 7). Details are as follows: integrating crypto-agility and pqc into EV charging : Considering the long lifespan of EVs (up to 35 years [63]) and CPs (10+ years [124]), the long-term security of related protocols is of vital importance. A significant aspect of long-term security is a protocol’s resilience to the discovery of vulnerabilities in the used cryptographic algorithms. One such vulnerability may be created by the continuous advancement in the field of quantum computing, since a large-enough quantum computer could be used to break the conventionally used asymmetric cryptographic primitives, e.g., for Rivest-Shamir-Adleman (RSA), Digital Signa- ture Algorithm (DSA), Diffie Hellman (DH), and Elliptic Curve Cryptography (ECC) based on Shor’s algorithm [121]. Since ISO 15118 defines the use of ECC-based cryptography, its au- thorization and billing process would be vulnerable to a large- enough quantum computer. Additionally, since no concept for crypto-agility is included, there is no clear method of recovering the system to a secure state if a vulnerability is found in one of the used algorithms. In Chapter 5 [75], we present a solution to address this issue. For this, we integrate an algorithm negotiation process into existing ISO 15118 messages, where the EV presents a priority- order list of its supported algorithms and the CP makes the final selection (similar to TLS). The algorithm negotiation process supports conventional and PQC algorithms and offers backward compatibility, e.g., for legacy systems that cannot be upgraded to support PQC due to hardware limitations. Additionally, our concept includes the generation of all EV key pairs locally within an HSM in the EV, protecting private keys from both backend data leaks and local adversaries with physical access to EVs. We further include changes to the contract credential installation process as well as the PnC authorization/billing processes to enable the verification of strong formal security properties. To evaluate our solution regarding feasibility/usability aspects, we implement it as a proof-of-concept on resource-constrained hardware. Specific PQC algorithms are selected from the re- 3.1 prevention of attacks on EV charging by enhancing protocol security 23 spective National Institute of Standards and Technology (NIST) standardization process [103]. Security requirements are verified based on a formal analysis using the Tamarin prover. Specific security properties are defined based on the notion of injective agreement, which is a well-established and strong authentication property originally defined by Lowe [87] and commonly used in current research (e.g. [11, 44, 76, 135]). The results of our evaluation identify suitable PQC algorithms for the use-case and verify that the desired security properties hold. Specifically, we show that the PQC signature algorithms Dilithium, FALCON, and SPHINCS+ (using the fast f-parameter sets) can meet the necessary timing requirements. Additionally, our Tamarin-based security analysis shows that the proposed solution provides the desired strong authentication guarantees for the processes of credential installation, charge authorization, and billing. using Self-Sovereign Identities for EV charging : Current EV charging protocols require the operation of a complex cen- tralized PKI. Hereby, different certificate chains have to be es- tablished for the various systems (CPOs, eMSPs, CPSs, etc.), imposing trust relations and requiring processes for certificate management. Additionally, the current security architecture ex- hibits a lack of privacy considerations. Specifically, various kinds of EV user-related PII are shared with actors that do not require this data for their operation, which poses risks regarding com- pliance with current privacy regulations like the EU’s General Data Protection Regulation (GDPR) [76]. For instance, with cur- rent protocols, several backend actors receive unique identifiers of individual users along with a charge session’s location and timing data, which enables the creation of movement profiles and may allow for the inference of consumer habits. In Chapter 6 [67], we present a solution to address these issues. For this, we propose an approach for using a decentralized SSI infrastructure to replace the current PKI and using Anoncreds for a secure and privacy-preserving charge authorization and billing process. SSIs give a user full control over their digital identities, based on verifiable credentials and a distributed ledger technol- ogy (replacing the traditional PKI). Identities are implemented via Distributed Identifier (DID) documents (replacing the tra- ditional certificates) that are stored on the ledger and where encryption/authentication is enabled via an associated DID key pair. Anoncreds can be seen as a special kind of DID that enable privacy-preserving authentications based on zero-knowledge proofs with Camenisch-Lysyanskaya (CL)-based credentials and paring-based revocation [77]. 24 contributions In our concept, we assume a public permissioned ledger where a steward can grant second-level write permissions to OEMs and eMSPs. The OEM can use this permission to issue EV-generated provisioning DID records to the ledger. Later, the EV can gener- ate a contract Anoncred and use the provisioning DID to securely request an eMSP to issue the Anoncred to the ledger, including an update to the necessary revocation information. Finally, the EV can use its contract Anoncred to create a zero-knowledge proof for the existence of a charging contract with a specific eMSP and a proof of non-revocation using its credential master secret. Overall, with our concept, the CPO/CP can only link a charging session/location to a specific eMSP (instead of a specific user) and the eMSP can only link a user’s session to a specific CPO (instead of a CP/location). This way, the user’s privacy can be protected even against the honest-but-curious operator model. To evaluate our solution regarding feasibility/usability aspects, we implement it as a proof-of-concept. Specific SSI operations are implemented using the Hyperledger Indy SDK [57]. Security and privacy requirements are verified based on a formal ana- lysis using the Tamarin prover. Specific security properties are defined based on the notion of injective agreement [87]. Privacy properties are defined based on the notion of symbolic unlinkabil- ity, commonly defined as the adversary’s inability to distinguish between a scenario where the same user is involved in multiple protocol runs with a scenario that involves different users per protocol run [23]. Privacy properties are verified with Tamarin’s observational equivalence mode. The results of our practical evaluation show that the concept of SSIs can be transferred to the EV charging use-case. Our perfor- mance results show that the added computational- and commu- nication overhead remains within acceptable bounds and meets existing timing requirements. Additionally, our Tamarin-based security analysis shows that the proposed solution provides the desired strong authentication guarantees for the processes of credential installation, charge authorization, and billing. Fur- thermore, our Tamarin-based privacy analysis shows that the desired non-traceability and non-linkability properties can be guaranteed for EV users and their charging sessions. streamlining EV charging with oauth 2: Current EV charg- ing protocols rely on a complex security architecture involving proprietary processes. For instance, the complex ISO 15118 PKI already resulted in critique by relevant stakeholders [14] and an importance for complexity-reducing measures [29]. Additionally, relevant parts of the ISO 15118 contract credential installation process are out-of-scope for the standard, resulting in other or- 3.1 prevention of attacks on EV charging by enhancing protocol security 25 ganizations trying to fill the gap (see, e.g., the related proposal by the German VDE [132]). These complexity issues and gaps in standardization can result in security-relevant problems for real-world implementations and thus pose an unnecessary threat to the underlying security architecture of EV charging. In Chapter 7 [114], we present a solution to address this issue. For this, we propose the use of OAuth methods for the credential management for EV charging. OAuth represents a widely ac- cepted and standardized solution and thus a significant improve- ment over the current patchwork of proprietary solutions. With our concept, the eMSP acts as OAuth authorization server and allows the EV user to perform an app-guided cross-device au- thorization flow to grant an EV the permission to install contract credentials via a Certificate Signing Request (CSR). Additionally, our concept utilizes Rich Authorization Requests (RARs) to en- able a fine-grained authorization process and implement, e.g., limits on charging duration/cost. To evaluate our solution regarding feasibility/usability aspects, we implement it as a proof-of-concept on resource-constrained hardware. The eMSP’s OAuth server interface is implemented using the cloud-based Authlete authorization server API [8]. Se- curity requirements are verified based on a formal analysis using the Tamarin prover. This analysis is especially noteworthy since an existing IETF draft [69] that proposes security best practices for the use of OAuth cross-device flows explicitly recommends conducting a formal security analysis. Our Tamarin-based ana- lysis meets these best practices and specific formal security properties are defined based on the notion of injective agreement [87]. The results of our practical evaluation show that OAuth can be used to streamline credential management for EV charging. Our performance results show that the added overhead remains minimal and we argue that the app-guided cross-device au- thorization flow can enhance usability by providing a uniform solution based on familiar interaction concepts. Additionally, our Tamarin-based security analysis shows that the proposed solution provides the desired strong authentication guarantees for the credential management processes. Since the discussed contributions of Chapters 5 to 7 [67, 75, 114] provide separate solutions to different problems, a cross-paper com- parison is not straightforward. However, we can nonetheless identify (dis-)advantages and trade-offs between the solutions. In terms of privacy, Chapter 6 [67] is the only one of our concepts that includes privacy-preserving measures for the EV user. In terms of security, all three concepts offer the same strong security properties 26 contributions based on injective agreement. Notably, while Chapters 5 and 6 [67, 75] look at both credential installation and charge authorization/billing, Chapter 7 [114] only focuses on the credential installation/manage- ment aspect. Since the default ISO 15118 EV authorization process does not provide the same level of strong authentication properties [75, 76], providing a consistent level of security when using Chapter 7 [114] would still require adjustments in the authentication process. This could, for instance, be achieved by a straightforward adoption of the process from Chapter 5 [75], which as discussed already includes the required changes for injective agreement during charge authoriza- tion/billing. Regarding the performance of the different solutions, Table 1 pro- vides a high-level comparison based on respective measurements. Specifically, we compare overall computational overhead (time) and communication overhead (size) created by the different approaches. As a baseline for comparison, we use the measurements for the default ISO 15118 process with the secp256r1 ECC curve from Chapter 5 [75]. Note that the timing measurements for the default process are relatively slow due to its un-optimized Java implementation. For the measurements of Chapter 5 [75], we list the values of the Dilithium2 PQC algorithm, which offered the fastest computational times of our comparison. Overall, we can see that all solutions were able to pro- vide good performance results and can meet existing limits like the ISO 15118 message timeouts of 5 seconds for receiving a contract credential installation response and 2 seconds for verifying an EV authorization. Additionally, we see that even the accumulated mes- sage sizes per process remain well within the ISO 15118 size limit of 4,294,967,295 bytes. Table 1: Performance Comparison of Different Approaches Overhead Approach Process Time [ms] Size [byte] ISO 15118 [63] Credential Installation 3,631.1 5,732 Charge Authorization 539.45 2,332 Chapter 5 [75] Credential Installation 885.8 39,129 Charge Authorization 97.47 14,997 Chapter 6 [67] Credential Installation 2,786.91 14,962 Charge Authorization 487.5 7,660 Chapter 7 [114] Credential Installation 1,514.0 8,686 Charge Authorization N/A N/A 3.1 prevention of attacks on EV charging by enhancing protocol security 27 3.1.2 Comparison with Related Work As mentioned previously, several security/privacy shortcomings within the existing EV charging protocols have already been discussed/ad- dressed by related work. In order to provide a clear impression of the distinct contributions and novelties that our publications offer, this section provides an overview of related work and discusses our con- tributions within this context. Table 2 shows an overview and details are discussed in the following: papers with security focus : Different publications already iden- tify the issue that existing communication protocols mainly rely on TLS for security while application layer data is often forwarded across multiple TLS channels, thus, lacking end-to- end security. The authors of [12] propose a middleware-based architecture to provide end-to-end security for EV charging data. Their approach discusses data-centric security, focusing on billing and session control, based on a standard middle- ware solution. In [131], a concept is presented for providing non-repudiation and end-to-end security for EV charging. Their solution uses a tree-based signature scheme and additional en- cryptions, focusing on billing security while also protecting user privacy through data minimization. Several papers propose security concepts for EV charging based on HSMs for secure credential storage/usage. In [40], the authors discuss a solution for the direct import of eMSP-generated cre- dentials into a Trusted Platform Module (TPM) [130] (a specific kind of HSM) in the EV. Hereby, the credential can be bound to a TPM authorization policy to ensure that the associated private key can only be used if the EV booted into a secure software state (based on a measured boot). An architecture for secure EV charging is presented in [39] that implements component binding between an EV’s communication controller and Battery Management System (BMS) based on a TPM and Device Identi- fier Composition Engine (DICE). The concept protects against counterfeit/malicious hardware modifications and ensures that the EV’s charging system is in a manufacturer-approved state in order to enhance the security of billing and charge session control. In [38, 41] another TPM-based concept for the security of EV charging credentials is presented. The solution enables an EV to generate its own contract credentials in its TPM and then request a corresponding certificate from the eMSP. Hereby, the certificate can again be bound to a TPM authorization policy. Other publications exist that focus on different use-cases. For instance, the authors of [5] propose a blockchain-based secu- rity architecture for peer-to-peer charging systems, enabling 28 contributions users to share their private CPs with other EV users. They present a blockchain-based management system for ensuring the authenticity of participating CPs and EVs. Additionally, a cryptocurrency-based payment system for secure billing is pre- sented. Another use-case is dynamic wireless charging, whereby an EV is charged while driving over on-road charging pads. A Physical Unclonable Function (PUF)-based lightweight authenti- cation protocol for dynamic wireless EV charging is presented in [9]. The concept safeguards against numerous attacks while producing minimal overhead. Security and privacy properties are formally analyzed in the random oracle model and an addi- tional Tamarin-based security verification is performed. In [102], a scheme for physical layer security and secure authentication for dynamic wireless charging is proposed. The scheme uses low-latency message authentication codes for the application layer and an artificial noise-based scheme for the physical layer. papers with privacy focus : Several publications already propose the use of privacy-enhancing technologies for EV charging. The authors of [52] discuss existing privacy issues in ISO 15118, in- cluding the excessive use of PII and the possibility of backend operators to create movement profiles of EV users. Additionally, they propose privacy enhancements that use anonymous cre- dentials with group signatures and require the introduction of a trusted third party. In [34] a formal privacy analysis of [52] is con- ducted that identifies weaknesses and suggests improvements. In [81, 82], the authors propose a pseudonymous authentication protocol for dynamic wireless charging. The approach offers low- overhead authentication while ensuring location privacy. A smart card-based multi-user EV roaming protocol is proposed in [101]. The authors propose the use of different EV user pseudonyms per charging session to protect user privacy and a formal security verification is perform