Neural Networks Method to Forecast Electricity Price for Markets with High Volatility

摘要

This paper proposes a novel and practical approach to forecast spot electricity price with large volatility based on neural networks. In the electricity market, the spot price is fluctuating every day, and the economic impact is large. Therefore, it is important to reduce the risk related to the electricity price volatility by forecasting the electricity price daily. Since the spot electricity price has a strong correlation with power demand, it is necessary to estimate the power demand firstly. The proposed method has two forecasting steps, the power demand forecasting and the electricity price forecasting. Firstly, the peak power demand of the next day is predicted using neural networks with data observed for appropriate periods in the past. Secondly, based on the obtained power demand, a forecasting method of electric power prices at the peak time of the next day is presented. The proposed forecasting methods are applied to the actual power market in Japan and verified to be effective to predict electricity prices.