Wind Power Forecasting Using LSTM-Grid Search

Various countries adopt Renewable Energy Sources (RES) like wind power in their generation mix to meet in-creasing energy demand and achieve net-zero targets. However, wind generation is highly uncertain and creates challenges to system operations. Accurate wind power forecasting enables grid operators and energy utilities to anticipate fluctuations in wind power output. This helps to integrate wind energy into the overall energy mix effectively. Various types of forecasting models like time series, machine learning, and deep learning models are used to obtain accurate wind power forecasts. Deep learning models like LSTM show superior performance over other types of models. However, the performance of the LSTM model heavily relies on the selection of its hyperparameters. This necessitates a suitable optimization algorithm for hyperparameter optimization. Therefore, this paper proposes a hybrid LSTM model called LSTM-Grid Search (LSTM_GS) to obtain accurate wind power forecasts. The Grid Search algorithm shows superior performance due to its exhaustive search capability across multiple parameter dimensions. Further, data pre-processing, seasonal analysis, and seasonal adjustment are performed to enhance the forecasting accuracy. The performance of the hybrid methodology is evaluated by data collected from the Bonneville Power Administration (BPA). The proposed methodology is compared with traditional LSTM and ANN models. The results demonstrate that the hybrid methodology outperforms the reference models.