Electricity price forecasting using artificial neural networks

Abstract

A model for forecasting the electricity price in Colombia using artificial neural networks is proposed in this work. Two neural networks structures including the price series in the first and the price series plus the water reserve levels in the latter are used. The results are compared with a Generalized Autorregresive Conditional Heteroskedastic Model (GARCH) model, which shows better adjustment inside the training period, but the neural networks have better performance forecasting outside the training sample. Historical data was supplied by the Company XM belonging to ISA Group, where 120 days were used as training patterns and the next 31 days were left to test the next month forecast.

Keywords

• Forecasting
• electricity price
• artificial neural networks
• time series models