Seasonal Models for Forecasting Day-Ahead Electricity Prices

Prediction algorithms are increasingly popular techniques within the energy industry to assist in forecasting future prices and to help reduce costs. When predicting day-ahead energy prices several significant external factors that influence electricity prices need to be considered. Initially we use the transparent Nonlinear AutoRegressive Moving Average model with eXoge-nous inputs (NARMAX) to model the relationship of factors that contrib-ute to energy costs and day-ahead electricity prices. Energy data from 2017 were analyzed separately for Spring, Summer, and Autumn periods to ob-serve the effect of factors on seasonality. The seasonal NARMAX models identify important input factors and their correlated lagged values such as price, energy demand, generation type, and environmental temperature to be significant factors in accurately predicting day-ahead prices for all three seasons. The identified factors are used to refine a Seasonal AutoRegressive Integrated Moving Average model with eXogenous variables (SARIMAX). To conclude the seasonal NARMAX models proved useful for removing in-significant external factors and the seasonal refined SARIMAX models permitted accurate prediction of day-ahead prices.