Temporal Convolutional Neural Network-Based Cold Load Prediction for Large Office Buildings

Abstract

In heating, ventilation, and air conditioning (HVAC) systems for large office buildings, accurate cooling load prediction facilitates the elaboration of energy-efficient and energy-saving operation strategies for the system. In this paper, a hybrid prediction model based on gray relational analysis-improved black widow optimization algorithm-temporal convolutional neural network (GRA-IBWOA-TCN) is proposed for cold load prediction of large office buildings. First, the factors influencing cold load in large office buildings were analyzed, with GRA used to identify key features and reduce input data dimensionality for the prediction model. Second, three improvement strategies are proposed to enhance optimization performance at different stages of the black widow optimization algorithm, aimed at establishing a prediction model for optimizing TCN hyper-parameters through IBWOA. Finally, the algorithm optimization and prediction model comparison experiments were conducted with the intra-week dataset (T1) and the weekend dataset (T2) of a large office building as the study samples, respectively. The results show that the mean absolute percentage error values of the GRA-IBWOA-TCN model for the prediction results of the T1 and T2 datasets are 0.581% and 0.348%, respectively, which are 81.1% and 88.3% lower compared to the TCN model, and exhibit the highest prediction accuracy in optimizing the results of the TCN model and the prediction models, such as backpropagation, support vector machine, long short-term memory, and convolutional neural network, with multiple algorithms, good stability, and generalization ability. In summary, the hybrid prediction model proposed in this paper can provide effective technical support for the energy-saving management of HVAC systems in large office buildings.