1. Federated Learning (FL) Model of Wind Power Prediction
- Author
-
Amal Alshardan, Sidra Tariq, Rab Nawaz Bashir, Oumaima Saidani, and Rashid Jahangir
- Subjects
Wind power prediction ,federated learning (FL) ,linear regression (LR) ,support vector regression (SVR) ,random forest regression (RFR) ,extreme gradient boosting regression (XGBR) ,Electrical engineering. Electronics. Nuclear engineering ,TK1-9971 - Abstract
Wind power is a cheap renewable energy that plays an important role in the economic development of a country. Identifying potential locations for energy production is challenging due to the diverse relationship between wind power potential and the weather characteristics of a location. Many machine learning models were proposed to predict the wind power production level for different locations. There is also a need for a global machine-learning model to enable wind power prediction of multiple locations with a single global model. A Federated Learning (FL) based model is proposed to train and evaluate the global model of wind power prediction of different locations using wind speed and wind direction. The proposed wind power prediction model is implemented in Pakistan to forecast the wind power of four distinct locations in Pakistan, using Linear Regression (LR), Support Vector Regression (SVR), Random Forest Regression (RFR), Extreme Gradient Boosting Regression (XGBR), and Multilayer Perceptron Regression (MLPR) models. The evaluation of the model from 30% of the test dataset reveals that RFR outperformed with a coefficient of determination (R2) of 0.9717, a Mean Squared Error (MSE) of 0.0007 kW, a Root Mean Squared Error (RMSE) of 0.0256 kW, and a Mean Absolute Error (MAE) of 0.018 kW. The XGBR model also performed well with R2 of 0.9681, MSE of 0.0007 kW, RMSE of 0.0270 kW, and MAE of 0.0129 kW. The accuracy of global models demonstrated the ability of the FL approach to deal with the heterogeneity of diverse weather characteristics of multiple locations for wind power prediction.
- Published
- 2024
- Full Text
- View/download PDF