Electricity Spot Prices Forecasting for MIBEL by using Deep Learning: a comparison between NAR, NARX and LSTM networks

Abstract

In the latest times, power markets in Europe, including the Spanish one called MIBEL (Mercado Ibérico de Electricidad), are being deregulated and coupled. As a result, electricity can be easily purchased and sold across further areas and countries. On the other hand, trying to guarantee renewable projects profitability, Power Purchase Agreements and Options contracts are arising as a feasible solution. The problem arises when the power plant owners have to negotiate the purchase electricity price in order to optimize risks and profits, as well as make future plans. Thus, several methods for Electricity Price Forecasting (EPF) have been developed and presented, showing different results, as market spot prices suffer from strong seasonality, spikes and high volatility. In this paper, three methods, based on Deep Learning Dynamic Neural Networks (NAR, NARX and LSTM) applied to forecast MIBEL electricity spot prices are discussed in order to evaluate their adequacy, accuracy and reliable horizon.