Real-time data based thermal comfort prediction leading to temperature setpoint control

The different thermal comfort indices such as Predictive Mean Vote (PMV), Standard Effective Temperature (SET), and Thermal Sensations (TS) have been used to predict occupants’ thermal comfort in a building. The advances in the machine learning approach help overcome the challenges of predicting current traditional thermal indices in a real-time environment. The different indices have different types of data samples (continuous/labelled). Therefore, while considering the machine learning technique in developing the models of the predictive thermal indices, it is essential to select the vital features, the proper learning type, the algorithm, and the evaluation method to establish the models of the predictive thermal comfort. The main focus of this paper is on the development of the ML model and the evaluation technique that helps in selecting the best model in predicting the thermal indices. This work proposes the new neighbourhood-component-analysis Bayesian-optimization-algorithm-based artificial-neural-network to develop a predictive model for the thermal indices. Here, we have proposed a regression-based model to predict PMV, SET and a classification-based model to predict 7-point TS. The statistical-testing results specify that the ANN model's performance is highly accurate and more reliable in predicting the thermal perception in a real-time environment. The performance of the selected model is validated using subjective measures. This prediction leads to the pre-emptive control of the setpoint temperature of the air-conditioning unit, hence resulting in energy efficiency and comfort.