Short-Term Residential Load Forecasting Method Based on Combined Deep Learning Model

The residential load has the characteristics of strong randomness and high meteorological coupling, and the traditional prediction method is difficult to effectively capture its complex variation law. Deep learning provides ideas for short-term load forecasting of residents, but it still faces the bottleneck problems of lack of basis for historical data set selection and model parameters relying on manual experience adjustment. Therefore, based on the selection of similar days, a short-term residential load forecasting method based on WOA optimized CNN-GRUAttention is proposed. Firstly, the influencing factors of load are selected by Pearson correlation coefficient, and the similar days are selected by weighted Euclidean distance based on the influencing factors, and the historical data set is constructed. Secondly, the combination prediction model of convolutional neural network and gated recurrent unit optimized by whale algorithm is constructed, and the attention mechanism is introduced to strengthen the attention to the important information of GRU and improve the prediction accuracy of the model. Finally, this method is used to predict the actual residential load in a certain area of southwest China, and compared with multiple models to verify the effectiveness of the proposed method. The example analysis shows that the combined model is superior to other comparative models in the four evaluation indexes of MAE, RMSE, MAPE and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}$</tex>, and the proposed method has higher prediction accuracy.