Short-term Load Forecasting Based on CNN-LSTM with Quadratic Decomposition Combined

Abstract

Short-term power load has strong randomness and volatility, in order to improve the accuracy of load forecasting, this paper proposes a combined forecasting model based on quadratic decomposition, convolutional neural (CNN) network and long short-term memory (LSTM) neural network. Firstly, the original load series was decomposed into several intrinsic mode components and residuals by complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN). Then, the sample entropy and K-means (SE-K-means, SK) were introduced to reconstruct the sub-sequences obtained by decomposition into three sequences, and the strong non-stationary sequences in the reconstructed components were decomposed twice by using variational mode decomposition. A CNN-LSTM model was established to predict each sub-sequence obtained by decomposition. Finally, the forecasting results were superimposed to achieve effective load forecasting. By using the actual load data for verification, it can be seen that from the perspective of four evaluation indicators: R2, mean absolute error, root mean square error and mean absolute percentage error, the proposed model has higher fitting and prediction accuracy than XGBoost, LSTM, CEEMDAN-LSTM and CEEMDAN-CNN-LSTM models.

Keywords

• short-term load forecasting
• sample entropy reconstruction
• quadratic decomposition
• convolutional neural network
• long short-term memory network