Wind Power Forecasting with Deep Learning Networks: Time-Series Forecasting

Studies have demonstrated that changes in the climate affect wind power forecasting under different weather conditions. Theoretically, accurate prediction of both output and using statistics-based models is difficult. In practice, traditional machine learning can perform long-term with a mean absolute percentage error (MAPE) 10% to 17%, which does not meet engineering requirements for our renewable energy project. Deep networks (DLNs) been employed obtain correlations between meteorological features generation multilayer neural convolutional architecture gradient descent algorithms minimize estimation errors. This has wide applicability field forecasting. Therefore, this study aimed at (24–72-h ahead) an MAPE less than by Temporal Convolutional Network (TCN) algorithm DLNs. experiment, we performed TCN model pretraining historical data outputs turbine from Scada plant Turkey. The experimental results indicated 5.13% 72-h prediction, adequate within constraints Finally, compared performance four DLN-based forecasting, namely, TCN, long short-term memory (LSTM), recurrent network (RNN), gated recurrence unit (GRU) models. We validated outperforms other three terms input volume, stability reduction, forecast accuracy.