Decomposition-Based Optimal Temporal Convolutional Networks Applied for Load Forecasting

Short-term power load forecasting can provide huge economic value. However, the high volatility and nonlinearity of the load curve pose a challenge for forecasting. In this paper, we propose a prediction model integrating CEEMDAN, LSTM and TCN. In the integrated model, we use orthogonal experiments and segmentation to optimize the model structure. Besides, we propose a sample entropy-based sequence fluctuation analysis method to integrate different models. Our model can significantly improve the prediction accuracy and reduce the training time compared to single models.