Short-Term Load Forecast Based on Feature Reconstruction and Bidirectional LSTM

Short term load forecasting is a guarantee for the safe, stable, and economic operation of the power system. With the large-scale integration of distributed new energy, load fluctuations increase, and the input feature data dimension increases during forecasting. Therefore, it is particularly important to screen and process input features reasonably when using deep neural networks for prediction. In response to the varying degrees of influence of various feature factors at different time points, this paper proposes a feature screening and input feature reconstruction method based on the maximum information coefficient of MIC, which achieves feature impact analysis, screening, and input feature reconstruction at sampling point granularity. The BI-LSTM neural network with stronger information mining ability is introduced to achieve high-precision short-term load forecasting. The results show that using the proposed input feature reconstruction strategy and assigning input feature weights according to the degree of influence significantly improves the prediction accuracy of deep neural networks. The method proposed in this paper has good adaptability to load fluctuations..