Short-Term Load Forecasting for Regional Power Grids Based on Correlation Analysis and Feature Extraction

Currently, a high proportion of new energy is connected to the distributed power grid, mainly wind power and photovoltaic, which increases the uncertainty of the forecasting environment. For this specific environment, this paper proposes a short-term load forecasting (STLF) method for regional power grids to meet the demand of enterprises for forecasting accuracy. The method improves forecasting accuracy through three stages: data correlation analysis, feature extraction, and load forecasting. More specifically, in the correlation analysis stage, this paper uses the Pearson correlation coefficient (PCC) to explore the correlation between meteorological factors and electric load. The strongly correlated meteorological variables are selected based on the value of PCC. In the feature extraction stage, this paper uses a convolutional neural network (CNN) as a feature extractor to extract more representative feature information from the input data. In the load forecasting stage, a long-short term memory (LSTM) model is established. It has a unique gate structure to limit the impact of the input data on the LSTM's parameters, thus improving the model's prediction performance. Finally, we demonstrate that the proposal has high STLF accuracy.