Influence of data preprocessing on prediction of complex valued load time series

Exploitation of modern power systems requires prediction of the electrical load time series for operation of power utilities and load estimates for market operation and system planning. Increase of energy produced from renewable sources and deregulation of electrical energy market makes load prediction more important nowadays. By its nature, electrical load time series are highly non linear and require modeling in the complex domain. Therefore, neural network based models, with fully complex activation functions, are appropriate choice for prediction of electrical load time series. However, their performance can be affected by input data preprocessing. Due to that cause, the paper analyses influence of data preprocessing on prediction of complex valued load time series. The analysis is performed on metered load data, that represents fifteen minutes average of active and reactive power, obtained from the medium voltage grid and with application of simple predictor structures, i.e. neural adaptive filters, applied to the one step ahead prediction tasks. Keywordscomplex valued load; time series prediction; data preprocessing; neural adaptive filter