Abstract

With climate change, low carbon space-cooling approaches are becoming more important. The adaptive comfort model for mixed mode operation can be a promising approach to the cooling energy challenge. However, adaptive models use indoor operative temperature, which requires the measurement of air temperature, air velocity, and globe temperature in a space. Collecting real-time and long-term data for these is difficult. We used machine learning to predict operative temperature with minimum measurement equipment, for controls to optimise fan operation and minimise AC energy use. Field measurements and Energy Plus simulation were used to create large datasets, 75 % of which were used to train the machine learning algorithm to predict operating temperature, and the remaining 25% were used for testing the algorithm. The random forest model in R from the tidy models library using the ranger engine proved successful. The Operative Temperature prediction Root Means Square Error value is 0.090°C. The algorithm classified data for being in/out of the comfort band, and 0.88% of the values are misclassified. While this paper demonstrates the machine learning approach that makes this possible, future work will demonstrate the implementation of the control algorithm and its testing.