Predictions of System Marginal Price of Electricity Using Recurrent Neural Network

Abstract

The accuracy of system marginal price (SMP) is important for bidding of generation companies. Based on analyzing characteristic of SMP, electrical load, historical value of SMP corresponding time and tendency of current SMP are regarded as three main influencing factors in estimating the next numeric value of SMP. A recurrent neural network is also introduced to forecast the SMP, because it has an ability of mapping dynamic system and SMP is regarded as a result of dynamic power market run. Aiming at the difficulty of determining neural network's structure and weights, the GA optimization algorithm is used to get them by previously combining binary encoding and real encoding. The history data of American California showed this method is effective and the forecast model is accurate