A Forecast Combination Method For Electricity market price modeling

Electricity market clearing price (MCP) forecasting is very important for electricity market participants and the system operator. Accurate forecasting is essential for designing bidding strategy, risk management, and market operation. There have been continuous efforts in electricity price forecasting research in recent years and various techniques have been proposed in literature, from classic linear time series models, such as ARMA to modern machine learning based nonlinear techniques, such as SVM. However there is still no clear consensus on which approach is the preferred one. Given these two different approaches to forecasting, it is natural to ask whether combining them may produce forecast results more reliable than the results obtained by either individual approach respectively. In this paper, we use combined model which is based on ARMAX and SVM to forecast day-ahead electricity prices. To find the “best” combination for electricity market data, we evaluate different combination schemes, such as optimal simple weighting scheme, and linear or nonlinear weighting scheme implemented by neural networks. The Australian National Electricity Market (NEM) data are used in the empirical study. Our results indicate that combined forecasts are more accurate than the original base models in terms of mean daily errors and other measures. Among the proposed three combination schemes, nonlinear weighting scheme has a better accuracy result than the other two for the Australian NEM MCP forecast.