What does this research mean for the field?

A hybrid short-term load forecasting model combining similar day theory, Variational Mode Decomposition (VMD), and neural networks outperforms standard deep learning models like LSTM, GRU, CNN, TCN, and Transformer in both accuracy and computational efficiency. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

May 6, 2025Energies9 citationsOpen Access

Short-Term Load Forecasting Based on Similar Day Theory and BWO-VMD

QCQi ChengState Key Laboratory of Advanced Electromagnetic Engineering and Technology JSJing ShiQingdao University SCSiwei ChengNational Taiwan University

Key Points

Key points are not available for this paper at this time.

Abstract

Short-term power load forecasting at the regional level is essential for maintaining grid stability and optimizing power generation, consumption, and maintenance scheduling. Considering the temporal, periodic, and nonlinear characteristics of power load, a novel short-term load forecasting method is proposed in this paper. First, Random Forest importance ranking is applied to select similar days and a weighted eigenspace coordinate system is established to measure similarity. The daily load sequence is then decomposed into high-, medium-, and low-frequency components using Variational Mode Decomposition (VMD). The high-frequency component is predicted using the similar day averaging method, while neural networks are employed for the medium and low-frequency components, leveraging historical and similar-day data, respectively. This multi-faceted approach enhances the accuracy and granularity of load pattern analysis. The final forecast is obtained by summing the predictions of these components. The case study demonstrates that the proposed model outperforms LSTM, GRU, CNN, TCN and Transformer, with an RMSE of 660.54 MW and a MAPE of 7.81%, while also exhibiting fast computational speed and low CPU usage.

问 AI

Bookmark

View Full Paper