Document Type : Original Article

Author

• hanif heidari

Iranian Research Institute for Information Science and Technology (IranDoc)

Abstract

The electricity market is dynamic, competitive, and unregulated, experiencing rapid growth in the past decade due to liber-alization and increased demand driven by technological advancements and digitalization. Forecasting monthly electricity prices is vital for governments, decision-makers, and investors to optimize profits and make informed choices. However, accurate predictions are challenging because electricity prices are influenced by various unpredictable factors, including weather patterns, holiday seasons, and technological progress. This study proposes an effective method for predicting future monthly average electricity prices with minimal error. It employs support vector regression (SVR) as the primary forecasting technique and introduces four error correction procedures: mean error modification, neural network approximation, poly-nomial approximation using particle swarm optimization and polynomial approximation using least square optimization. To evaluate the effectiveness of these methods, we analyzed the average wholesale monthly prices in Bulgaria, Greece, Hungary, Netherlands, Norway and Romania. Our findings demonstrate that the proposed approaches significantly reduce prediction errors. Among the methods tested, first-order polynomial approximation with least square optimization yields the highest accuracy and computational efficiency.

Keywords

• Virtual error points
• Support vector regression (SVR)
• Polynomial approximation
• Particle swarm optimization
• Least square optimization