Short-term Electricity Load Forecasting Based on CVMD-TCN-BiLSTM

Short-term load data is often mixed with different types of noise and high volatility. Traditional sequence decomposition methods do not consider the influence of high-frequency noise when extracting sequence features. To address the above problems of accurate forecasting, a combined CVMD-TCN-BiLSTM forecasting method with decomposition denoising and reconstructive decomposition is introduced. The original electric load data are decomposed into a set of relatively smooth subsequences using the complementary ensemble empirical mode decomposition(CEEMD), and the high frequencies component with noise is denoised jointly with the wavelet thresholding method, and the low frequency components with signal are retained for cumulative decomposition. Then, the denoised data are subjected to secondary signal feature extraction using variational modal decomposition(VMD) to obtain a set of components with strong smoothness and containing different frequencies. Finally, the prediction of each component is performed using a temporal convolutional network-bi-directional long- and short-term memory neural network(TCN-BiLSTM), and the prediction results are iterated to obtain the complete prediction results. By analyzing the load data of a place in Australia as an example, compared with the traditional load forecasting algorithm, the effectiveness of the proposed model is verified.