Long-term Electricity Price Forecasting Using a Random Forest-based Machine Learning Approach

Abstract

Electricity price forecasting are important in optimizing energy trading, consumption scheduling, and operational planning within smart grid infrastructures. This study proposes a data-driven approach using a Random Forest (RF) regression model implemented in MATLAB for accurate electricity price prediction. Unlike conventional models, the RF model is evaluated under both open-loop and closed-loop forecasting scenarios to assess its short-term accuracy and long-term stability. The model is trained on time-series electricity pricing data enriched with lagged variables and temporal features, allowing it to learn from past behaviors and predict future price fluctuations. Open-loop forecasting utilizes actual historical values at each time step, enabling the model to demonstrate its pattern recognition capabilities with minimal cumulative error. Conversely, the closed-loop approach relies on recursive self-generated predictions to simulate real-world deployment, where future data is unavailable. Despite expected error propagation, the model maintains trend fidelity and captures peak patterns effectively across all three data channels. Performance evaluation using RMSE (0.534), MAE (~0.0276), and R² (~0.783) confirms the model’s accuracy, robustness, and generalization ability across multiple channels. Additionally, the consistent sensitivity score highlights the model’s responsiveness to price changes. The results underscore the RF model’s suitability for reliable electricity price forecasting, offering a balance between predictive accuracy and computational efficiency. This research supports the advancement of intelligent forecasting tools for dynamic electricity markets and reinforces the feasibility of integrating RF-based prediction systems in both academic research and industrial energy management applications.