Short-Term Power Load Prediction Based on Improved Quadratic Mode Decomposition and BiLSTM-Attention
In order to solve the problems of lack of effective optimization of parameter selection of variational mode decomposition in short-term power load prediction and the weakening of long-term information when using long short-term memory neural network prediction, a prediction method based on improved quadratic mode decomposition and redistribution of input weights in neural network by using attention mechanism was proposed. Firstly, the decomposition parameters in the traditional quadratic mode decomposition were optimized by the evaluation standard of decomposition loss. Then, on the basis of feature selection, the attention mechanism and the forward and reverse memory layers were added to the long-term and short-term neural network, and the training prediction was carried out for each modal component separately. Finally, the subsequence prediction results were reconstructed and output. The analysis of the case shows that the proposed method solves the problems of parameter selection and long-term information weakening of variational mode decomposition in prediction, effectively reduces the decomposition loss, and has higher prediction accuracy.
attention mechanismslong-term and short-term neural networksdecomposition lossesquadratic modal decompositionshort-term power load forecasting
万方数据
维普
