Automated classification of power quality disturbances in a SOFC&PV-based distributed generator using a hybrid machine learning method with high noise immunity. (26th May 2022)

• Record Type: Journal Article
• Authors: Yılmaz, Alper; Küçüker, Ahmet; Bayrak, Gökay
• Is Part Of: International journal of hydrogen energy. Volume 47:Number 45(2022)
• Page Start: 19797
• Page End: 19809
• Publication Date: 2022-05-26
• Subjects: Hydrogen energy -- Power quality -- Machine learning -- SOFC -- Distributed generation; Hydrogen as fuel -- Periodicals; Hydrogène (Combustible) -- Périodiques; Hydrogen as fuel; Periodicals; 665.81
• DOI: 10.1016/j.ijhydene.2022.02.033
• Languages: English
• ISSNs: 0360-3199

Abstract

Abstract: In this study, a new hybrid machine learning (ML) method is developed to classify the power quality disturbances (PQDs) for a hydrogen energy-based distributed generator (DG) system. The proposed hybrid ML method uses a new approach for the feature extraction by using a pyramidal algorithm with an un-decimated wavelet transform (UWT). The pyramidal UWT method is used and investigated with the Stochastic Gradient Boosting Trees (SGBT) classifier to classify PQD signals for a Solid Oxide Fuel Cell & Photovoltaic (SOFC&PV)-based DG. The overfitting problem of SGBT in noisy signals is eliminated with the features extracted by pyramidal UWT. Mathematical, simulative and real data results confirm that the developed UWT-SGBT method can classify PQDs with high accuracy of up to 99.59%. The proposed method is also tested under noisy conditions, and the pyramidal UWT-SGBT method outperformed other ML with wavelet transform (WT)-based methods in the literature in terms of noise immunity. Highlights: A hybrid ML method is proposed for classifying PQDs with high noise immunity. Pyramidal UWT is used to feature extraction for PQDs in a SOFC-based DG. The SGBT classifier performance is dramatically improved with pyramidal UWT. The pyramidal UWT is used to solve the overfitting problem of SGBT.