Learning desk fan usage preferences for personalised thermal comfort in shared offices using tree-based methods

Abstract

Recent studies in the built environment lead a paradigm shift from conventional, one-fit-for-all thermal comfort models to more localised and personalised comfort models that cater to personal preferences as well as open up significant energy savings potential. In this project, desk fan usage preferences are collected using a wireless sensor and actuator network (WSAN) in two shared offices along with indoor and outdoor environmental conditions as well as user presence information. To predict the personal preferences of desk fans usage, the fan state is modelled as a classification problem and the fan speed as a regression problem considering only the instances when the fan state is ON. Tree-based methods such as decision tree, random forests and boosted trees are investigated and random forests method is found to achieve the best performance with an average test set accuracy of 97.73% in predicting the fan state among six users on a dataset of total size (192021, 10). It also achieved 95.42% average test accuracy for instances where user is present and the fan speed is estimated with average root-mean-squared error (RMSE) of 15.83 of the fan setting. The most important variables are also identified and it is noted that these variables vary among the users indicating diversity in preferences. Finally, the real-world deployment of the methods are discussed along with insights for future data collection.