PREDICTIVE ANALYSIS OF SHORT-TERM WIND POWER FORECASTING USING VARIOUS NEURAL NETWORKS MODEL

Abstract

This study proposes a intelligent method for short-term wind power forecasting and uncertainty analysis. In practice, the power output of a wind turbine is a direct function of wind speed. Owing to the intermittent and irregular nature of wind, the wind power generation is not easily dispatched and the prediction of wind power is highly uncertain. To allow a procedure for more accurate forecasting, few wind power prediction models like: NARX (Non-Linear Exogenous Inputs), NLIO (non-Linear Input& Output), RNN (Recurrent Neural Network), LSTM (Long Short Term Memory), GBM(Gradient Boosting Machine), ANN/MLP (Artificial Neural Network/ Multi Layer Perceptron), Linear Regression, GA based SVM (Genetic Algorithm based Support Vector Machine) are established and discussed in case studies form. The historical wind speed and wind power generation data for year 2015 of wind power plant located in Kolkata, available on Ninja portal has been used for short term, dayahead wind power forecasting. The models are mainly evaluated on basis of following parameters , they are : MAE (Mean Absolute Error) ; MAPE (Mean Absolute Percentage Error) ; MSE (Mean Square Error) ; RMSE (Root Mean Square Error), Variance. So, on basis of comparison of predictive models over the above parameters, we have concluded in each case studies, that which model performs best under the given circumstances for that case study. Overall, from whole analysis we can conclude that LSTM performs the best while RNN is quite close to it. In Daily average methodology, RNN performs the best having close competitor as GA based SVM.