Short-Term Load Forecasting Method for Distribution Networks Based on Sample Entropy Secondary Decomposition and Deep Learning

In order to address the difficulty in accurately predicting the load of distribution networks due to strong load fluctuations and high randomness of user behavior, this paper proposes a short-term load forecasting method for distribution networks based on sample entropy (SE) secondary decomposition and deep learning. First, the random forest (RF) algorithm is used to extract features from external factors affecting load data. Then, time-varying filtering empirical mode decomposition (TVFEMD) is applied to preliminarily decompose the load sequence into subsequences. Combining SE and singular spectrum analysis (SSA), highly complex subsequences are further decomposed to improve the quality of the sequences as inputs to the prediction model. Next, the subsequences and load influencing factors are input into a bidirectional long short-term memory (BiLSTM) model, whose hyperparameters are optimized by an improved dung beetle optimization (IDBO) algorithm, for load forecasting. Finally, the predicted results of the output subsequences are superimposed to obtain the final prediction result.