Long-Term Electricity Demand Forecasting for Thailand’s Small Business Sector: An LSTM-Based Approach

Accurate electricity demand forecasting is essential for Thailand’s energy system planning, particularly for the small business sector, whose consumption exhibits high variability (coefficient of variation: 26.34%). This study develops a Long Short-Term Memory (LSTM) model to forecast monthly electricity consumption of Thailand’s small business sector over a 12-month horizon from September 2025 to August 2026. The analysis is based on 284 months of historical consumption data (January 2002–August 2025) obtained from official national statistics. The forecasting framework employs Min–Max normalization and a supervised learning formulation with a 12-month lookback window. The dataset is chronologically divided into training (80%) and testing (20%) subsets, and model performance is evaluated using Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). The optimized LSTM model achieves strong forecasting accuracy, with a training RMSE of 93.93 and MAPE of 5.45%, and a testing RMSE of 143.68 and MAPE of 6.25%. These results meet the criterion for highly accurate forecasting (MAPE < 10%), demonstrating the model’s ability to capture long-term trends and seasonal patterns while generalizing well to unseen data. The findings highlight the suitability of LSTM-based models for long-term electricity demand forecasting in high-volatility small-business sectors and underscore their practical relevance for energy planning and policy development in Thailand.