چكيده به لاتين
Wind turbines, as a renewable and clean energy source, contribute to reducing greenhouse gas emissions, preserving natural resources, and decreasing dependency on fossil fuels. This technology, with low operational costs, offers long-term economic savings and energy sustainability, ensuring energy security for future generations. Additionally, wind turbines create job opportunities and foster technological advancements in renewable energy, serving as an effective solution to combat climate change and provide sustainable energy.
Failure prediction in wind turbines is essential to enhance efficiency and reduce maintenance costs. This process, by early detection of potential issues, prevents unexpected downtimes and severe damages, thereby extending the lifespan of equipment. Accurate failure prediction also optimizes maintenance scheduling and maintains stable energy production, ultimately lowering operational costs and increasing trust in renewable energy systems.
Therefore, in this study, the Grey Wolf Optimization (GWO) algorithm was employed to select key and influential features to improve the accuracy of fault detection in wind turbines using five machine learning methods: Artificial Neural Network, Support Vector Machine, Decision Tree, AdaBoost, and XGBoost. The dataset for this research was obtained from a 3 MW wind turbine supplying power to a large production facility near the southern coast of Ireland. The data, extracted from the turbine's SCADA system, include real-time measurements and operational warnings recorded in 10-minute intervals over an 11-month period from May 2014 to April 2015.
The findings of this study indicate that the Grey Wolf Optimization algorithm successfully identified and selected influential features within the datasets for fault detection, fault diagnosis, and fault prognosis, which are fewer in number than the total features. This approach improved evaluation metrics. In some cases, the evaluation metrics in both scenarios (with and without GWO-based feature selection) were relatively similar. However, given that these results were achieved with fewer features in the GWO-based feature selection scenario, this highlights the advantages of using this method for identifying key features.
