Reinforcement Learning Model for Wind Power Prediction and Grid Stability Enhancement

Abstract

Wind Power Forecasting and Reliability of Power Grids is the integration of wind energy into modern power systems with difficulties due to its intermittent and unpredictable nature; thus, enhancements are necessary to overcome these issues. In the case of transmission system operators overseeing large-scale wind farms, precise forecasting is the utmost importance for facilitating economic decision-making, efficient energy balancing, and dependable grid operation. This work suggests a hybrid data-driven strategy in which hourly mean wind speed data is preprocessed using a technique that removes non-Gaussian distribution and daily nonstationarity. To improve representation learning, a SDAE is used in conjunction with batch normalisation to execute deep feature extraction. In order to make accurate forecasts about wind power, LSTM networks use these extracted properties. In addition, numerical weather forecast data is processed using DBSCAN clustering to eliminate outliers, which improves the training sample quality and overall model efficiency. The results show that the suggested SDAE-DBSCAN-LSTM model achieves a high prediction accuracy with a R2 value of 0.9538%, surpassing numerous current techniques. Finally, the suggested system helps to stabilise the grid and integrate renewable energy sources reliably while also greatly improving the accuracy of wind power forecasts.