A Real-Time Improved ML Method for PQD Classification of a PV-Powered EV Charging Station
Alper Yılmaz, Tolga Ateşci, Hasan Meral, Gökay Bayrak
Abstract
The installation of electric vehicle charging stations (EVCSs) that are powered by renewable energy sources has been growing rapidly. However, this has raised a crucial issue regarding the quality of power supplied to these stations. Due to the intermittent nature of renewable energy sources and the high-power requirements of EV charging, power quality disturbances (PQDs) occur more. This study proposes a new intelligent PQD classification method that considers feature extraction/selection based on pyramidal undecimated wavelet transform (p-UWT) and minimum redundancy maximum relevance (mRMR). The feature vector, derived through the application of mRMR, comprises a mere ten elements. The p-UWT-mRMR combination overcomes the problem of noise sensitivity in WTs. In addition, Bayesian optimization and UWT-mRMR have addressed hyperparameter selection difficulties and overfitting in support vector machine models. The proposed method demonstrated an impressive classification accuracy of 99.55% when faced with 30-dB noise. A prototype test platform is developed with EVCS-integrated PV systems in the laboratory to verify the performance of the proposed method in real-time cases. Dynamic analysis revealed that all PQDs have runtimes ranging from 5 to 10 ms in experiments. The proposed method has been validated on a dataset of over 20 000 real-world signals with a test accuracy of 99.11%.