Medium-term load prediction based on GA-LightGBM- LSTM

Achieving accurate and fast mid-term load forecasting is of great significance to ensure timely power supply and reduce resource waste. In the medium-term load forecasting, Pearson correlation coefficient has the problem of only analyzing data with linear relationship, and LSTM forecasting medium-term load has the problem of inaccurate late negative forecasting due to the long forecasting step. To address the above problems, a GA-LightGBM-LSTM based medium-term load prediction model is proposed. The features with a high impact on load are selected by screening with LightGBM, and the parameters such as the learning rate and the maximum depth of the tree of LightGBM are optimized by genetic algorithm to select the best parameters. Then, the screened features with high impact on load are used as the input of LSTM, and the LSTM model is used for medium-term load prediction. Finally, a comparison is made with CatBoost, SVR, and LSTM. The results show that the accuracy of the model proposed in this paper is 95.35%, 91.54%, and 93.15% for large industry, general industry, and non-general industry, respectively, which effectively improves the accuracy of the predicted loads compared with the other three models.