Wind Speed prediction using sparse SCADA data of an offshore wind farm of Pakistan using different machine learning techniques

• Wind speed is forecasted using SCADA data. • Optimization of machine learning models for wind speed forecasting. • Features determination that have the greatest influence on wind speed prediction. • Promotes clean energy policy based on wind energy. The amount of wind power installed globally is steadily rising as a result of global efforts to replace fossil fuels and slow the rise in the average global temperature. This study aims to investigate wind speed prediction using sparse SCADA data of local wind farm located at Jhampir Thatta Pakistan. The Machine learning(ML) based feature selection method like gradient boosting is used to determine the most pertinent features, helping optimize wind speed systems. Different advanced ML and Deep learning (DL) algorithms have been applied for model development of wind speed prediction namely linear regression, random forest, AdaBoost, support vector machine, XGBoost , LSTM & GRU depends on the selection of features. These models have taken raw data from Jan-2021to Dec-2021with sample rate of 10 minutes . Morever, these models were evaluated using metrics such as mean absolute error (MAE), root mean squared error (RMSE) & coefficient of determination (R²). The comparative analysis of the results shows that ensemble-based models perform better than conventional regression methods, with XGBoost and Random Forest with the R 2 of 0.99 and RMSE of 0.04 and 0.05 respectively. These models proved to be consistent with higher R 2 score and lower RMSE. The precise wind forecast with sparse data, based on modern forecasting techniques require sufficient computational power which would greatly help to harness the intermittent wind energy resource viably.