Short-Term Load Forecasting for Microgrid Energy Management Using Variational Mode Decomposition and Adaptive Neuro-Fuzzy Inference System

Abstract

With the increasing penetration of Renewable Energy Resources (RESs) in Microgrids (MGs), accurate short-term load forecasting plays a vital role in energy management and operational planning. However, the non-stationary and highly variable characteristics of load time series pose significant challenges to forecasting models. To address these issues, this study proposes a hybrid forecasting framework based on Variational Mode Decomposition (VMD) and Adaptive Neuro-Fuzzy Inference System (ANFIS). VMD is first employed to decompose the original load series into several intrinsic mode functions (IMFs), enabling the extraction of frequency-related features at different time scales. Subsequently, an individual ANFIS model is developed for each IMF component, and the final forecasting results are obtained through reconstruction. A dataset containing 35,048 samples is used to evaluate the proposed framework. Comparative results with SVR, ANN, and conventional ANFIS models demonstrate that the proposed VMD–ANFIS framework achieves the highest forecasting accuracy. Under the same experimental conditions, the MAPE values of SVR, ANN, ANFIS, and VMD–ANFIS are 5.5394%, 2.9168%, 2.7191%, and 0.3847%, respectively. The results indicate that VMD effectively enhances the forecasting capability of ANFIS by decomposing complex load signals into relatively stable components, providing an effective solution for short-term load forecasting in microgrid energy management applications.