Electricity price forecasting in Turkey with artificial neural network models

The electricity market has experienced significant changes towards deregulation with the aim of improving economic efficiency. The electricity pricing is a major consideration for consumers and generation companies in deregulated electric markets, so that offering the right price for electricity has become more important. Various methods and ideas have been tried for electricity price forecasting. Artificial neural networks have received much attention with its nonlinear property and many papers have reported successful experiments with them. This paper introduces artificial neural network models for day-ahead electricity market in Turkey. Using gradient descent, gradient descent with momentum, Broydan, Fletcher, Goldfarb and Shanno (BFGS) and Levenberg-Marquardt algorithm with different number of neuron and transfer functions, 400 different models are created. Performances of different models are compared according to their Mean Absolute Percentage (MAPE) values; the most successful models MAPE value is observed as 9.76%.

Keywords: electricity price forecasting, neural networks, day-ahead electricity market, Turkey.
JEL Classification: C02, C13, C45, C53