A novel hybrid approach combining kernel extreme learning machine and model-agnostic meta-learning for photovoltaic fault diagnosis with limited samples

To address the issue of limited photovoltaic fault samples, this study proposes a novel fault diagnosis approach combining kernel extreme learning machine with an improved model-agnostic meta-learning. Firstly, kernel extreme learning machine effectively processes strong nonlinear features through its kernel techniques and enhances learning speed. Secondly, the introduction of model-agnostic meta-learning enables the model to quickly adjust its parameters with a few samples, adapting to new types of faults. And during the iterative exploration process, 10-fold cross-validation is used to obtain the optimal parameter combination with the highest accuracy. The effectiveness of this method is separately validated on datasets containing simulated and real-world photovoltaic system faults. The diagnostic performance in different low-sample scenarios is also analyzed. The results indicate that the proposed method achieved an accuracy of 96.97%. In experiments with real-world data, this method improved accuracy by approximately 13.21%, 13.73%, 1.32%, 4.93%, and 1.44% compared to KELM, CNN, ABC-KELM, MAML, and MLELM, respectively. In terms of diagnostic speed, it is about 99 s faster than traditional CNN networks. And it exhibits better performance than RFC and WOA-ELM. This provides a new perspective and tool to address the sample limitation issue in photovoltaic fault diagnosis.