Comparison study of short-term heating load predicting models for district heating stations
Short-term heating load prediction is studied based on the measured operation data and meteorological data of a centralized heating station in Shiyan,Hubei Province.Five methods of random forest(RF),extreme gradient boosting(XGBoost),BP neural network,support vector regression(SVR),and long short term memory(LSTM)neural network are used for model training and testing respectively.Based on particle swarm optimization algorithm(PSO),the hyper-parameters of each predicting models are optimized,and the optimal models are obtained.On this basis,a comparative study on the performance of different models in different short-term load prediction scenarios is carried out.The research results show that:in the future 24-hour prediction scenario,the prediction accuracy of random forest and XGBoost models is the highest,and their mean absolute errors(MAE)are 0.84 W/m2 and 1.00 W/m2 respectively.In the future 1-hour prediction scenario,the prediction accuracy of SVR model is the highest,with a MAE of 0.18 W/m2.
district heatingload predictionrandom forestXGBoostBP neural networksupport vector regressionLSTM