Research on Source-Load Rolling Prediction Method Based on Decomposition and Prediction Integration

Due to the intermittent nature of new energy and the randomness of load demand, traditional forecasting methods are difficult to meet the requirements of high accuracy. To solve this problem, a rolling forecasting method based on the integration of set empirical decomposition technique (EEMD) and bidirectional long-short term memory network (BiLSTM) is adopted in this paper. Firstly, EEMD is used to decompose the complex energy and load time series into different frequency bands to reduce the complexity of data. Then, BiLSTM is introduced to predict the decomposed time series, so as to utilize its advantages in the prediction of their respective time series to achieve higher prediction accuracy. The results of numerical examples show that the proposed prediction method effectively improves the reliability of source-load prediction.