Integrated Energy System Load Forecasting Based on GSO-LSTM and Feature Selection

In order to realize the dynamic balance between the power supply of the integrated energy system and the change of multiple loads, a specific park is selected as the object to carry out the research on short-term forecasting methods of power loads. First, an extreme gradient boosting algorithm is used to select the effective influences on the electrical load in each region. Then, an integrated GSO-LSTM load forecasting method based on feature selection is proposed, combining the golden search optimization algorithm and the long short-term memory neural network. For the overall electrical load of the integrated energy park, the effective influencing factors selected by the XGBoost algorithm are introduced as feature variables, and the input features are constructed to realize the “integrated prediction” of the electric load of the integrated energy park, which solves the problem of ignoring the differences in the degree of influence of different factors during full feature prediction. Finally, the effectiveness of the proposed model was verified through example.