Time Series Forecasting of Generated Power from Texas Wind Turbine

As the demand for renewable energy rises, optimizing wind power as an energy source is crucial. Wind power is one of the cleanest forms of energy available, and understanding the energy that wind turbines generate over time is necessary for building a better foundation for wind energy reliance. Previous research has explored using Long Short-Term Memory (LSTM) and LSTM-based algorithms for practical wind turbine data analysis and predictions. In those cases, the LSTM-based forecasts showed the most robust wind turbine prediction rates based on various variables, mainly wind speed and direction, and generated active power. This research implements LSTM, Nonlinear Autoregressive (NAR), and Nonlinear Autoregressive Exogenous (NARX) networks on simulated Texas wind turbine data to compare the techniques that produce better predictions. The data are normalized using correlation analysis techniques on the following data features: system power generated, wind speed, wind direction, pressure, and air temperature. The data are separated for training and testing and run through the LSTM, NAR, and NARX algorithms. After obtaining the mean squared error (MSE) of the testing data, the algorithms are compared to determine the best-predicting algorithm. The results show which algorithm on time series data holds the most robust prediction of generated energy from wind turbines.