Fault detection and diagnosis in a grid-connected photovoltaic system

Abstract

Photovoltaic (PV) power generation has been an active research topic in the recent few years. One of the main goals of researchers in grid integrated PV systems is to improve the performance of the system in terms of efficiency, availability and reliability. For this reason, it is crucial to develop efficient methods for PV system’s fault detection and diagnosis. In this report, an automatic fault detection and diagnosis approach is proposed for a grid connected PV system. The main objective is to improve the classification accuracy and reduce the detection time. This method merges the benefits of machine learning (ML) technique and statistical process monitoring approaches. The analytic methods were first investigated for fault detection due to their quick implementation time. Kernel based independent component analyses KICA techniques was developed to overcome the shortcomings of principle component analysis PCA based fault detection. The support vector machines SVM classifier was built mainly for fault diagnosis and classification, such that feature extraction step is done using both KICA and PCA to optimize the best model. In this work the “one to one” classification SVM algorithm is used. To validate our method, fault detection and diagnosis of a lab implemented grid-connected PV system was performed. In this experiment, 7 typical PV systems faults were injected. The experiments were carried out for about 15 seconds in each fault scenario, and several measurements were recorded. Data samples were filtered, smoothed then processed through PCA and KICA to set the thresholds for fault detection, then the obtained reduced data sets were used to train the multi-layer SVM classifiers.