Novel Application of Heterogeneous Ensemble Learning in Fault Diagnosis of Photovoltaic Modules

The existing fault diagnosis methods of photovoltaic modules have some limitations: they only consider the influence of the area size of fault array when determining the fault level, and ignore the impact of the strength of fault factor itself. Addressing the issues, and in order to surmount the limited performance caused by the reliance on a single method simultaneously, this study proposes a novel fault diagnosis method of photovoltaic modules based on heterogeneous ensemble learning using current-voltage characteristic curves and ambient conditions. Moreover, a selection strategy considering both accuracy and diversity comprehensively is used to screen base learners to acquire superior diagnostic performance. The optimal integration members are incorporated adopting the probabilistic strategy and stacking algorithm respectively. In order to validate the effectiveness of the proposed method, two datasets are obtained based on a laboratory experiment platform and the corresponding simulation model respectively. The results demonstrate that the ensemble model based on probabilistic strategy proposed in this paper achieves more comprehensive diagnosis ability compared with the individual classifiers and the ensemble model based on stacking algorithm.