LAUSR

Recently, neural networks based on intelligent algorithms have been widely used in short-term power load prediction. However, these algorithms have poor reproducibility in the case of repetition. Aiming at the shortcomings of the gray wolf optimizer (GWO) algorithm, such as slow convergence speed and easy to fall into local optimum, an improved gray wolf optimizer (IGWO) was proposed. In order to better extract features from time series data, the improved grey wolf optimizer-bidirectional long short-term memory (IGWO-BILSTM) power load prediction model is established by bidirectional long short-term memory (BILSTM). Pearson correlation analysis uses the actual power load data set of a certain region and selects the highly correlated factors as the input of the network. The BILSTM model was used to train the preprocessed data, the IGWO algorithm was used to find the optimal solution parameters during the training, and finally, the optimal prediction results were output to the test set. The experimental results show that compared with the prediction model established by the traditional optimization algorithm, the mean absolute percentage error (MAPE) value of the IGWO-BILSTM model is 1.63, the root mean square error (RMSE) value is 1.47, and the mean absolute error (MAE) value is 1.17. The convergence speed is faster and the prediction result is more accurate. The accuracy of multidimension load forecasting is greatly improved.