Integrating Machine Learning and MLOps for Wind Energy Forecasting: A Comparative Analysis and Optimization Study on Türkiye’s Wind Data

Creative Commons License

Oyucu S., AKSÖZ A.

Applied Sciences (Switzerland), cilt.14, sa.9, 2024 (SCI-Expanded) identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 14 Sayı: 9
  • Basım Tarihi: 2024
  • Doi Numarası: 10.3390/app14093725
  • Dergi Adı: Applied Sciences (Switzerland)
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Agricultural & Environmental Science Database, Applied Science & Technology Source, Communication Abstracts, INSPEC, Metadex, Directory of Open Access Journals, Civil Engineering Abstracts
  • Anahtar Kelimeler: latency, machine learning, MLOps, RMSE, wind energy
  • Sivas Cumhuriyet Üniversitesi Adresli: Evet

Özet

This study conducted a detailed comparative analysis of various machine learning models to enhance wind energy forecasts, including linear regression, decision tree, random forest, gradient boosting machine, XGBoost, LightGBM, and CatBoost. Furthermore, it developed an end-to-end MLOps pipeline leveraging SCADA data from a wind turbine in Türkiye. This research not only compared models using the RMSE metric for selection and optimization but also explored in detail the impact of integrating machine learning with MLOps on the precision of energy production forecasts. It investigated the suitability and efficiency of ML models in predicting wind energy with MLOps integration. The study explored ways to improve LightGBM algorithm performance through hyperparameter tuning and Docker utilization. It also highlighted challenges in speeding up MLOps development and deployment processes. Model performance was assessed using the RMSE metric, conducting a comparative evaluation across different models. The findings revealed that the RMSE values among the regression models ranged from 460 kW to 192 kW. Focusing on enhancing LightGBM, the research decreased the RMSE value to 190.34 kW. Despite facing technical and operational hurdles, the implementation of MLOps was proven to enhance the speed (latency of 9 ms), reliability (through Docker encapsulation), and scalability (using Docker swarm) of machine learning endeavors.