Building Load Prediction Model Based on Integration of Mechanism and Data in District Heating Systems

Abstract

For the building load prediction in district heating systems, the traditional methods cover the mechanism models and the data-driven model. However, the data of the district heating systems was often of low quality or partly missing, and thus the traditional models cannot present high accuracy in load prediction due to the insufficient data available for training. This paper proposed a building load prediction model based on the integration of mechanism and data in district heating systems. The proposed model considered the heat transfer mechanism process and was trained by historical data. As for the training results in the scenario of insufficient data, the proposed prediction model presented a remarkable improvement by decreasing the MAPE and RMSE to 7.34% and 7.74%, respectively, on the same test dataset, compared with the data-driven model. The mechanism process integrated into the proposed model served as an additional feature and enhanced the data quality.