Innovations-based Neural Network Seasonal Day-ahead Marginal Price Forecasting

Abstract

Successful bidding and operational strategies of electric power generators (GENCO) depend highly on the availability of accurate and timely load and price forecasts. Several techniques have been proposed and applied over the past few years to predict the marginal price of electricity in deregulated markets. To improve accuracy, these techniques apply time-consuming, complex, and hybrid methods requiring multiple inputs and large databases. This article introduces the first application of the method of “innovations” and a single artificial neural network to provide accurate forecasting results with mean absolute percentage error comparable to more complex and hybrid artificial neural network forecasting methods. The proposed model is applied to data of two seasons of Spain's power market operator (OMEL) marginal price data. The technique provided average accuracy improvement of 26% with overall mean absolute percentage error of 6.5%, which is reasonable considering the number of inputs and the simplicity of this model compared to other proposed models.