A novel short-term household load forecasting method combined BiLSTM with trend feature extraction

Aiming to reduce the short-term household load prediction error caused by small load scale and differently residential electricity consumption behavior, a novel hybrid forecasting model based on wavelet threshold denoising (WTD), variational mode decomposition (VMD) and bidirectional long short-term memory (BiLSTM) network is proposed in this paper. In this hybrid model, WTD is used to denoise the original data firstly. Then trend feature is extracted by VMD. Finally the trend feature and historical load data are put into BiLSTM model for training and testing. The prediction effect of the proposed model is demonstrated with experiment using total household electricity consumption in the United States in a region, and comparison with common short-term household load forecasting models are presented. The experiment results show that the hybrid model improves the accuracy of short-term household load forecasting by providing more stable and more precise forecast under trend feature extraction.