Klir, Stefan (2024)
Modeling of individual lighting preferences depending on various influencing parameters.
Technische Universität Darmstadt
doi: 10.26083/tuprints-00026389
Ph.D. Thesis, Primary publication, Publisher's Version
Text
2023-11-28_Klir_Stefan.pdf Copyright Information: CC BY-SA 4.0 International - Creative Commons, Attribution ShareAlike. Download (32MB) |
Item Type: | Ph.D. Thesis | ||||
---|---|---|---|---|---|
Type of entry: | Primary publication | ||||
Title: | Modeling of individual lighting preferences depending on various influencing parameters | ||||
Language: | English | ||||
Referees: | Khanh, Prof. Dr. Tran Quoc ; Dörsam, Prof. Dr. Edgar | ||||
Date: | 15 January 2024 | ||||
Place of Publication: | Darmstadt | ||||
Collation: | xxviii, 211 Seiten | ||||
Date of oral examination: | 28 November 2023 | ||||
DOI: | 10.26083/tuprints-00026389 | ||||
Abstract: | Inspired by the evolutionary adapted visual systems of humans towards the natural changing sunlight of a day and year, office lighting should likewise not be static. Due to the possible changeability of illuminance and color temperature of multi-channel luminaires, such dynamic conditions can be realized in offices. Since the light is designed for people, their preferences should additionally be addressed in the selection process of a dynamic light scenario. However, by the fact that each person has a different perception of the current environment through different psychological and physiological states, the dynamic preference must be determined with the associated influencing parameters. The questions that arise are, how does a system have to be designed to model the individual user light preferences based on certain influencing parameters and in particular what are the impacting parameters. To address these questions, this doctoral thesis introduces a 32 months long-term field study with 30 in the research phase developed floor lamps. Through the specialized hard- and software, participants are able to insert the current state-of-mind as well as the current light preference with an illuminance and color temperature selection out of 100 combinations within a web interface. Furthermore, environmental data of the room as well as the prevailing weather condition are gathered during the period of use. In order to enhance the psychological rated light spectra space, each day a new set of 100 white light settings out of a collection with 5,781 spectra is automatically distributed to the floor lamp users. More diverse light spectra are able to be rated in a so-called training process, in which the participants perform a self-conducted subject study each day with five diverse light settings. The maximized space of rated light settings in combination with the current mood, indoor sensor values and weather data forms the ground truth data set for the evaluation. During the evaluation, a novel static and dynamic preference lighting models (PLMs) were defined. The static PLM introduces two equations, one for illuminance and one for correlated color temperature (CCT) that can be fitted with a low amount of data to the preferences of each individual participant. Whereby the illuminance preference is modelled only with the time of day and in contrast the CCT preference integrate time of day, week of year, indoor temperature and humidity. Thus, these models incorporate the dynamic light preference behavior of a user with certain environmental parameters and can be personalized with few data points. Whereas, the dynamic PLM is based on a data-driven approach. A contextual multi armed bandit (CMAB) is stated as the main component with 32 environmental input features. Since multi armed bandits include only the current environment in the prediction of a light setting, the sensor and weather input features are classified into cluster-labels based on the time-series characteristics of the last six hours. Light spectra with the highest preference are thus predicted for a certain environment and individual user based on the ground truth data set gathered in the long-term field study. A novel reward function for the lighting domain forms the main component for a user rating estimation. This estimation of non-existing user ratings enables a fully automatic learning process, and a generalized user model which is able to react to preference changes. Furthermore, a novel preference rating function is defined and enables the comparability between users by abstracting the absolute light rating with quantile ranges and including usage parameters. In particular, the three involved parameters are: (i) Ten quantile ranges of the individual user ratings, (ii) the illumination durations of each spectrum as well as (iii) the frequency of how often a spectrum was illuminated to abstract each user’s individual rating behavior in a utility function. This likewise enables, by discarding the absolute user ratings from the equation, to obtain a fully usage based preference rating, and a full automated learning process without any user interaction is made possible. As application, this preference rating function is included in a user-based collaborative filtering approach that suggests light ratings based on other participants with a similar light behavior in a certain environment. Therefore, knowledge about the light similarity between the participants in environments are revealed. As result, light preferences of other users with a similar light preference in a given environment can thus be suggested to enhance the learning process of the dynamic PLM. The fitted static PLM equations of the combined ten users from the long-term floor lamp study resulted in a high coefficient of determination for both the illuminance and CCT equations of R2 = 0.97 and R2 = 0.98 respectively and demonstrated the meaningfulness of these functions. A four-month evaluation experiment with the trained dynamic PLM stated that high quality light suggestions are predicted, which have a higher median rating as the manual adjusted lights settings previous to the evaluation study for spectra that are illuminated for longer than ten minutes. Manual light adjustments during the study have a similar median rating as the smart predictions by the CMAB. The satisfaction of the participants as well as the meaningfulness of the predictions in a real environment could therefore be demonstrated. The feasibility of the user-based collaborative filtering approach, which was enabled by the high amount of 24,261 gathered light rating data, is presented with light suggestions for the participants in a certain environment. In 21 most rated environments, on average five light rating predictions are suggested for the ten users. This approach enables an active learning process of new, unseen light settings for users of the same light preference group per environment. With the presented framework and approaches of this doctoral thesis, it is now possible to predict dynamic changing illuminances and color temperatures with respect to influencing environmental parameters even with very little knowledge of a user based on the static PLM. If a large data set with multiple users exists, the user-based collaborative filtering approach can be consulted to actively enhance the suggested light settings. As soon as a sufficient amount of data is present, the introduced dynamic preference lighting model (PLM) is able to predict preferred light settings based on more environmental impacts and learn dynamically varying light preferences per user. Due to the novel light preference rating function which can also integrate only the usage of the luminaire, a minimal invasive system is introduced in which subjects only have to use the lamp with the provided predictions or manually adjusting it for unique conditions and the PLM dynamically adapts to these new preferences. This results in a comprehensive framework for individual light spectra predictions per user with respect to dynamically changing environmental parameters. |
||||
Alternative Abstract: |
|
||||
Status: | Publisher's Version | ||||
URN: | urn:nbn:de:tuda-tuprints-263890 | ||||
Classification DDC: | 600 Technology, medicine, applied sciences > 620 Engineering and machine engineering 600 Technology, medicine, applied sciences > 621.3 Electrical engineering, electronics |
||||
Divisions: | 18 Department of Electrical Engineering and Information Technology > Adaptive Lighting Systems and Visual Processing | ||||
Date Deposited: | 15 Jan 2024 13:03 | ||||
Last Modified: | 18 Jan 2024 07:26 | ||||
URI: | https://tuprints.ulb.tu-darmstadt.de/id/eprint/26389 | ||||
PPN: | 514753536 | ||||
Export: |
View Item |