Electricity Demand and Supply Curve Estimation with Neural Network

Abstract

We used deep neural network technology to estimate the demand and supply curves of the JEPX day-ahead market. We studied relatively moderate market situations to focus on fundamental market characteristics. Weather forecast data are used as inputs and the disclosed clearing price, cleared quantity, and total volume of demand and supply bids were used as training data. No information about the actual curve shapes was disclosed, but the shapes of the estimated curves showed the responsiveness to the disclosed data sets. We evaluated the minimum distance between the estimated curves to the actual clearing price and cleared quantity. The distribution of this indicator showed that almost 90% of test cases were in the 10% error range. This was almost the same level of performance as regression of the disclosed clearing price determined using a support vector and convolutional neural network. The proposed method is inferior in terms of estimation of the clearing price, cleared quantity, and total volume of demand and supply bids, but we believe it can add intuitive understanding about the market situation to support bidding operations using this method with some other regression-based method.