Heating load prediction of residential district using hybrid model based on CNN

Abstract

Optimal control of heat network in residential district requires accurate short-term heat load prediction value of each building. However, the current research on residential heat load prediction mainly relies on historical data related to heat load and does not consider the influence of building ontology parameters on different building types, resulting in low prediction accuracy. The objective of this study is to obtain accurate short-term heat load prediction values of different buildings in residential district. Based on this, the building ontology parameters were introduced, and the Lasso method was adopted to comprehensively analyze and select the influencing factors of heat load. A hybrid short-term heat load prediction model using adaptive T-distributed Satin Bowerbird (tSBO) algorithm to optimize convolutional neural network (CNN) was proposed. The actual operation data of established heating system in 10 residential districts were collected to test the performance of the model. The results show that building ontology parameters have a significant impact on the prediction accuracy of future heat load. After introducing the building ontology parameters, the mean absolute percentage error (MAPE) and root mean square error (RMSE) of the predicted values of the hybrid model were reduced by about 29.61% and 22.00%, respectively. Compared with other prediction models, the proposed hybrid model achieves an average reduction of 18.08% and 16.26% in MAPE and RMSE.