A NOVEL APPROACH FOR BUILDING WIND POWER OUTPUT PREDICTION BASED ON DEEP LEARNING RNN AND CNN MODEL

Abstract

Wind power has come out as a quickly growing source of renewable energy now a days. Wind is intermittent in nature due to wind speed alterations. Accurate prediction of wind power is essential for efficient operation of wind power system, in return providing power network management and control. Modern wind turbines have Supervisory Control and Data Acquisition (SCADA) systems are installed for wind power forecast. In this research, LSTM, GRU and CNN-LSTM based deep learning models are constructed. To evaluate the prediction performance of neural network-based models, a novel comparative analysis is performed utilizing the lookback parameter. In the predictive model, wind speed, nacelle orientation, yaw error, blade pitch angle, and ambient temperatures were considered as input features, while wind power was assessed as an output feature. Input features were directed in models based on wind turbine physical process. The deep learning models have been given training, testing and validation against SCADA measurements. The study of the results reveals GRU gives minimum MAPE value of 0.074 having slight difference with MAPE values of Hybrid (CNN-LSTM)’s MAPE of 0.0751 and Bi-LSTM giving MAPE of 0.078. The simulation result showed that proposed GRU model gives suitable MAE values of validation and MAPE values of predictions in comparison to Hybrid (CNN-LSTM) and Bi-LSTM retaining more accuracy with less computational cost.