Year: 2025
Type: article
Abstract: Enhancing the accuracy of electricity consumption forecasting is essential for the safe and stable operation of power systems and the strategic planning of grid companies. To achieve high-precision short-term electricity consumption forecasting, this paper proposes a VMD-xLSTM forecasting model that combines Variational Modal Decomposition (VMD) with an extended Long Short-Term Memory network (xLSTM). This novel hybird model firstly uses the VMD decomposition algorithm to decompose the original electricity consumption time series into multiple modal components and a residual component. Subsequently, leveraging the powerful long-term dependency capturing capabilities of the xLSTM neural network, each decomposed component is independently established the model and forecasted. Finally, the prediction results of each component are superimposed to reconstruct the overall predicted value of electricity consumption. Experimental results demonstrate that the proposed VMD-xLSTM model performs outstandingly, with a Symmetric Mean Absolute Percentage Error (SMAPE) of only 0.86%. This is notably lower than that of the traditional Long Short-Term Memory network (LSTM) at 2.33%, the standalone xLSTM model at 2.15%, and the VMD-LSTM model at 0.90%. Additionally, all other prediction error evaluation metrics of this model are comprehensively superior to the compared models, clearly verifying the VMD-xLSTM model's effectiveness in short-term electricity consumption forecasting tasks through comparative experiments.
Authors: Chen Zhi, Han Zhong-kuan, Xun Chao, Shen Yu, Liu Lin, Chen Yan-tao, Zeng Shuai
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4414584947