A human comfort prediction method for indoor personnel based on time-series analysis

In buildings, the heating ventilation and air conditioning system (HVAC) creates a comfortable environment for indoor occupants by setting a temperature strategy. However, this approach leads to unreasonable indoor environmental comfort and wasted energy because it does not dynamically adjust to changes in environmental and has a long response time. In this study, a high-precision human comfort prediction method for indoor personnel based on time-series analysis is proposed as the control strategy for HVAC systems. The method includes the data pre-processing module, the class imbalance processing module, and the human comfort network model module. We propose the Human-Comfort Bi-directional Long Short-Term Memory (HC-BiLSTM) network to achieve a better human comfort prediction, and the Synthetic Minority Oversampling Technique for Time-series (SMOTE-TS) algorithm to solve the class imbalance problem in human comfort dataset. A public dataset collected in Pennsylvania, USA, was selected for this study to validate the performance of the proposed method. The experimental results show that the human comfort prediction method proposed in this study achieves 0.9482 and 0.9659 on Macro-averaging and Micro-averaging, respectively, which is the highest accuracy in the known related research.