Forecasting short-term electricity prices in Malaysia using AI and machine learning approaches

Accurate short-term electricity price forecasting is critical for ensuring market efficiency, grid reliability, and cost optimization in liberalized power markets such as Malaysia's. With the restructuring of the Malaysian Electricity Supply Industry (MESI 2.0) and increasing renewable energy penetration, price volatility has become more pronounced, creating new challenges for conventional forecasting techniques. This study explores the application of artificial intelligence (AI) and machine learning (ML) approaches—including Long Short-Term Memory (LSTM) networks, XGBoost, and hybrid deep learning models—for forecasting day-ahead System Marginal Prices (SMP) within Malaysia's Single Buyer Market framework. Historical pricing data, load demand, fuel mix, temperature, and solar irradiance were used as input features. The models were trained and validated using real SMP data from the Single Buyer Data Catalogue, with performance evaluated using RMSE, MAE, and MAPE metrics. Results indicate that LSTM outperforms other models in capturing non-linear temporal dependencies and sudden price spikes, achieving a forecasting accuracy improvement of up to 18% over baseline autoregressive models. The findings underscore the potential of AI-powered models to support grid operators, electricity retailers, and energy traders in managing risks, enhancing bidding strategies, and optimizing demand response participation. Furthermore, the study emphasizes the importance of model interpretability and data granularity in ensuring trustworthiness and operational deployment in Malaysia's evolving electricity market.