The Parallel Temporal Convolutional Neural Network Based Building Energy Consumption Prediction Model

Abstract Building energy consumption (BEC) prediction aims to forecast the future energy consumption distribution, which plays a significant part in achieving energy conservation control, formulating energy conservation strategies, and quantifying the potential of energy conservation. The relevant characteristics of BEC, such as indoor environmental parameters, time information and weather, have a highly significant influence on the predictive accuracy of energy consumption data. Therefore, the relevant characteristics of BEC and historical energy consumption data are typically employed as the database of prediction models. To enhance the rationality of the prediction results and improve the prediction accuracy, a depth BEC prediction model based on the parallel temporal convolutional neural network (PTCN) is proposed in this paper. The PTCN can fully leverage the feature information of two time series, indoor factors and outdoor environmental factors, extract the features respectively for energy consumption prediction, and ultimately combine them through linear fitting. Its parameters are optimised through the least square method to better map the nonlinear relationship between input parameters and energy consumption data. In comparation with other prediction methods, the forecasting accuracy is higher and the performance is improved by more than 20%, which is suitable for BEC prediction.