Electric Heating Load Prediction Based on TCN-LSTM Hybrid Neural Network

The traditional load prediction is based on linear modeling, which does not consider the uncertainty of load and has no small errors. Deep learning methods are used for load prediction because of their strong data feature extraction and fitting ability. From the accuracy of electric heating load prediction, an electric heating load prediction model based on TCN-LSTM hybrid neural network is proposed. Several influencing factors with the strongest correlation with electric heating load are selected as input features by Pearson correlation analysis among multiple features, and weakly correlated meteorological features are filtered out; then CEEMDAN decomposition is used to decompose the historical electric heating load time series into multiple eigenfunctions as well as a residual term to obtain the historical input feature time series corresponding to each input feature; finally, the TCN and LSTM hybrid models are finally used for electric heating load prediction. The simulation results show that the electric heating load prediction model based on the TCN-LSTM hybrid neural network can extract effective information from the historical load data, realize the dimensionality reduction processing, and improve the operation rate and accuracy of the artificial network.