Prediction of short-term heat load of office buildings based on GNN-LSTM modeling

To address the accuracy and real-time requirements of heat load forecasting in current office buildings, a hybrid forecasting framework (hybrid GNN LSTM model (HGLM)) based on hybrid graph neural network and long and short-term memory network is proposed, and the interrelationships between heat load feature nodes are fully considered in the graph data design to improve the accuracy of the forecasting target. In order to evaluate the performance of the model, the dataset was first constructed by simulating the data with weather factors using DeST software, and compared with several conventional forecasting models such as GNN, LSTM, MLP and RNN. The experimental results show that the accuracy of the coefficient of determination (R2) of the proposed HGLM model is improved by 20.7%, 1.8%, 10.6%, and 35.9% for short-term prediction (168 hours) The results show that the model performs well in the prediction of thermal loads in office buildings.