Day Ahead Wind Power Forecasting Using Complex Valued Neural Network

Abstract

Wind power forecast is one of the daily processes performed by Wind Power Plants (WPPs). It is very important to provide the generation-consumption balance one-day in advance for electric power system. In this study a day ahead wind power forecast in hourly bases is carried out for seven WPPs. The data used in this forecast is composed of the generation data of seven WPPs and the numerical weather forecasts of these WPP site. While the train data consist of 12-month data, the test data consist of 6-month data. Complex Valued Neural Network (CVNN), a special kind of artificial neural network (ANN), are preferred as the forecast method and compared with Real Valued Neural Network (RVNN). While hour, wind speed forecasts and wind direction forecasts are used as the system inputs, the output is forecasted wind power. Since the CVNN works with complex number, the non-complex inputs are converted to complex values. Normalized Mean Absolute Error (NMAE) and Normalized Root Mean Square Error (NRMSE) are preferred to show the forecast accuracy. While RVNN has an average of 12.S2% NMAE and 16.S% NRMS, CVNN has 11.75% NMAE and 15.77% NRMSE. It is seen that CVNN method is more successful with the lower error rates than RVNN. Therefore, CVNN can be used as an effective tool for wind power forecast.