Abstract: This paper proposes an improved Sand Cat Swarm Optimization algorithm (ISCSO) to optimize the penalty factor and kernel function parameter of Support Vector Machine (SVM) for photovoltaic fault monitoring. The algorithm is enhanced by introducing Logistic mapping for population initialization, and together with the Lévy flight strategy and a nonlinear parameter strategy. The proposed ISCSO algorithm is benchmarked against the Sand Cat Swarm Optimization (SCSO), Grey Wolf Optimizer (GWO), and Improved Grey Wolf Optimizer (IGWO). Experimental results reveal that the ISCSO-SVM model achieves average recognition accuracy improvements of 0.74%, 0.54%, 3.52% and 14.28%, respectively, compared with SCSO-SVM, IGWO-SVM, GWO-SVM and the original SVM. In terms of convergence rate, the proposed ISCSO converges at the 3rd iteration on the Schwefel's Problem 1.2 function and the 5th iteration on the Griewank function, which substantially surpasses SCSO, IGWO, and GWO. In addition, ISCSO attains the maximum generalization validation accuracy of 99.45%. Compared with SVM, GWO-SVM, SCSO-SVM, and IGWO-SVM, the proposed method raises the average accuracy by 2.28%, 0.96%, 0.64%, and 0.5%, respectively. This study verifies the superiority of the proposed ISCSO algorithm in effectiveness, stability and feasibility, offering theoretical support for deploying ISCSO-tuned SVM for photovoltaic fault monitoring tasks.