<p>Online monitoring and fault detection play important roles in ensuring the healthy performance of grid-connected PV power stations. In this work, both linear and nonlinear multivariate statistical analyses based on PCA and Kernel PCA were utilized to achieve this. Additionally, Hotelling’s T², Q-static, and KDE techniques have been incorporated into the algorithm. The proposed real-time fault detection system was tested on a 7 kWp grid-connected PV station in Adrar, Algeria’s Saharan region. Various fault experiments, including open circuit, short circuit, and partial shading at different degrees, were conducted to validate the algorithm thoroughly. The experimental results show that the detection performance varies from 34% for short-circuit faults to 99% for open-circuit faults, depending on fault type and severity. Notably, KPCA consistently outperformed PCA, particularly in nonlinear conditions such as partial shading: for instance, detection improved from 22.9 to 61.3% under 2/6 shading, and from 14.2 to 74.2% under 3/6 shading. This research demonstrates the efficacy of the developed online fault detection system in enhancing the operational efficiency and reliability of grid-connected PV power stations.</p>