Heat Load Prediction of District Heating Systems Based on SCSO-TCN

Abstract

Heat load prediction is crucial to the heat regulation of district heating systems (DHS). In heat load forecasting tasks, deep learning can frequently achieve more accurate model building. A deep learning algorithm, the temporal convolutional network (TCN), has been used for DHS heat load prediction. However, there are many hyperparameters for TCN. Manually tuning the TCN parameters cannot make the model have good performance. This study presents a hybrid method based on sand cat swarm optimization (SCSO) and TCN. The SCSO is used to optimize the hyperparameters (number of filters, filter size, dropout rate, and batch size) of TCN. To verify the effectiveness of SCSO-TCN, another two hybrid models, particle swarm optimization with TCN and the sparrow search algorithm with TCN, are established for comparison. The historical heat load data of three heat exchange stations in Tianjin is utilized for the testing experiments. The findings demonstrate that SCSO-TCN has higher predictive accuracy and better generalization ability than the PSO-TCN and SSA-TCN models.