A New Hybrid Framework Based on Grey and Neuro-Fuzzy Inference System for Energy Demand Forecasting in Vietnam

Accurate energy consumption forecasting faces two major challenges: limited historical data and complex consumption patterns. To address these challenges, this study proposes a new hybrid framework named the Decomposition-based Grey-Neuro-Fuzzy Architecture (DeGNA). The model first uses the Denton method to convert limited annual records into high-frequency monthly data. Next, it applies STL decomposition to separate the data into trend, seasonal and residuals components. A rolling-window GM(1,1) model is then used to predict the main growth trend, while a GWO-optimized ANFIS model uses economic indicators (IIP and FDI) to forecast complex seasonal changes. This study evaluates the performance of the DeGNA model using Vietnam’s energy data from 2009 to 2024. The results show that DeGNA achieves high accuracy with a MAPE of 4.11%, performing competitively alongside modern machine learning (XGBoost) and deep learning (LSTM, BiLSTM) models. More importantly, a 5-fold cross-validation reveals that while autoregressive models such as SARIMAX perform well during stable growth periods, DeGNA provides more consistent performance under macroeconomic disruptions. While SARIMAX struggles with sudden demand shocks (maximum error of 22.23), DeGNA maintains a much tighter error distribution (maximum 15.39) and the lowest variance (±1.09%). Therefore, DeGNA provides a highly reliable tool for national energy planning under data constraints.