Recurrent neural network models for wind power forecasting
Abstract
The historical values of wind power generation are generally utilized in most forecasting models and methods in the literature. Unfortunately, the operational parameters such as wind speed, pitch angle, ambient temperature, nacelle position, and wind direction may affect the forecasting results. Therefore, the primary objective of this paper is to propose recurrent neural network models for wind power generation forecasting considering these operational parameters. In this study, the wind power generation data and the associated operational parameters from the wind turbine generator 05 of a wind farm are investigated. The data from July 1 st , 2024 to July 31 st , 2024 is collected from the SCADA system. Firstly, the nonlinear autoregressive neural network with external input is applied to make the wind power generation prediction. Secondly, the layer recurrent neural network model is employed to forecast wind power generation. Thirdly, the distributed delay neural network model is implemented to predict wind power generation. Fourthly, the time delay neural network model is trained to estimate the wind power. Finally, these recurrent neural network models are compared to determine the better wind power generation forecasting model in terms of mean absolute error, mean absolute percent error, and root mean square error.