Content Details | NUML Online Research Repository

Title ANALYSIS OF FACTORS AFFECTING WIND TURBINE ENERGY OUTPUT USING MACHINE LEARNING

Author(s) Ghulam Murtaza

Abstract Electrical energy generated by wind turbines is stochastic in nature due to its dependency on various factors. Such randomness raises barriers in adjusting the energy stocks of the power systems according to need. Multiple approaches have been proposed to predict the energy output of wind turbines and to meet the corresponding energy demands. This thesis investigates variables (also called factors or features) that affect the wind turbine’s output. The energy obtained from turbines varies as it relies on factors. Some of the important factors or features are turbine blade area, wind speed, temperature, air density, humidity, tower height, the angular position of the blades, pressure, etc. All such features are required to be investigated, analyzed, and evaluated with the help of state-of-the-art Machine Learning (ML) models to identify their importance and significance in predicting the wind turbine output. ML techniques used are Categorical Boosting (CatBoost), Extreme Gradient Boosting (XGBoost), Decision Tree (DT), Random Forest (RF), Gradient Boosting Regression (GBR), Light Gradient Boosting Model (LGBM), Extra Tree and Adaptive Boosting (AdaBoost). Evaluation of the ML methods and analysis of the factors are carried out on three different latest and publically available datasets. The experimental results show that CatBoost compared to all other methods demonstrates the lowest Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) in predicting the energy output. The wind speed is identified as the most significant factor in predicting the energy output by all of the ML methods. Additionally, a new method is proposed which ensembles CatBoost, RF and AdaBoost methods. The proposed method is named as X-CRA, in which the prediction s of CatBoost, RF and AdaBoost are fed into XGBoost through a stacking approach and final energy output is obtained. The experimental results show that X-CRA outperforms CatBoost and all other ML methods in predicting the wind energy output.

Type Thesis/Dissertation MS

Faculty Engineering and Computer Science

Department Engineering

Language English

Publication Date 2024-01-10

Subject Electrical Engineering

Publisher NUML

Contributor(s) Ghulam Murtaza

Format

Identifier

Source

Relation

Coverage

Rights