Digital Twin Based Kernel RF for Fault Detection and Diagnosis in Photovoltaic Systems

Ensuring the reliability, efficiency, and economic viability of photovoltaic (PV) systems requires effective fault detection and diagnosis, which remains a significant challenge. In this work, we propose a novel and integrated framework that leverages Digital Twin (DT) technology together with the RK-RF algorithm for fault detection and classification in grid-connected PV systems. The DT serves as a dynamic reference model, capturing real-time system behavior and generating sensitive error indicators for anomaly detection. These indicators are then efficiently processed by the RK-RF algorithm, which incorporates Kernel Principal Component Analysis (KPCA) for dimensionality reduction, enhancing classification accuracy while reducing computational costs compared to traditional machine learning or deep learning methods. Experimental validation on a grid-connected PV system emulator, including one healthy state and five representative fault scenarios, demonstrates the high effectiveness of the proposed method. Specifically, the DT enables rapid and reliable anomaly detection, while the RK-RF achieves near-perfect classification accuracy ( \approx 100\%≈100%≈100% ), outperforming benchmark techniques such as support vector machines (SVM), k-nearest neighbors (KNN), neural networks (NN), and recurrent neural networks (RNN). Overall, the proposed DT-RK-RF framework provides a robust, scalable, and interpretable solution for predictive maintenance and performance optimization in PV systems, with potential applicability to other complex energy and industrial infrastructures. This work emphasizes both computational efficiency and real-time reliability, addressing key limitations of existing DT+AI approaches.