Development of Preconditioned Deep Neural Network for Electricity Price Forecasting

This paper proposes an efficient method for electricity price forecasting. It is important to understand the behavior of electricity price in advance so that the profit is maximized while the risk is minimized through electric power trading in power markets. The behavior is related to uncertainties as well as high nonlinearity so that more sophisticated methods are required to forecast electricity prices. In this paper, a preconditioned Deep Neural Network (DNN) is proposed to evaluate better predicted values. As the preconditioned technique, k-means is employed to classify electricity prices into some clusters and DNN that consists of Autoencoder and MLP Multi-layer Perceptron (MLP) of Artificial Neural Network (ANN) is constructed at each cluster. Also, the data increase method with the Gaussian random numbers is presented to improve the precondition technique. The effectiveness of the proposed method is demonstrated for real data of ISO New England, USA.