| At present,China’s winter heating energy consumption accounts for a large proportion of the total social energy consumption,and heating and energy saving are imperative.Accurate prediction of heating load can help operation and maintenance personnel grasp the energy demand of the building in advance and carry out precise control,thereby avoiding unnecessary energy waste.With the popularization of metering devices and the development of artificial intelligence technology,machine learning technology has developed rapidly in the prediction of building energy consumption.Taking a residential building in a cold area as an example,this paper uses the tree model and the neural network model to predict the short-term and ultra-shortterm prediction of heating load.In this paper,the correlation factors are used to initially screen the impact factors.The screening targets are real-time monitoring data of outdoor meteorological parameters.Considering the thermal inertia of buildings,the influence of historical time values of outdoor variables is also taken into account.Then,the obtained influence factors are processed by principal component analysis to obtain model input parameters.Finally,the input parameters are input into the tree model and the neural network model to establish a load forecasting model.For the tree model,the DT model,GBDT model and XGBoost model are used to predict the heating load.The study found that the XGBoost model performs best in both short-term load forecasting and ultra-short-term load forecasting.For the neural network model,the LSTM model and its variants(Dropout-LSTM model,Bidirection-LSTM model)and TCN model are used to predict the heating load.The study found that the neural network model predicts the ultra-short-term load better than the short-term load.For the LSTM model,the Bidirectional-LSTM model with two-way neural network mechanism has the best effect on ultra-short-term load forecasting,and the RMSE is 22.1% lower than the LSTM model.The prediction of ultra-short-term heating load by the TCN model has reached expectations,which is comparable to the LSTM model,which proves that the convolutional neural network can be used for the prediction of heating load.In order to comprehensively evaluate the short-term heating load forecast with relatively small sample size and relatively stable change in a short period of time,it is recommended to use the XGBoost model;the ultra-short-term heating load forecast with large sample size and rapid load change in a short time,the Bidirectional-LSTM model and the TCN model are recommended. |
