This paper examines the performance of supervised machine learning algorithms for anomaly detection in photovoltaic (PV) power systems. Given a net-labeled and balanced dataset, we converted the problem into a binary classification task. We tested three models on the set-Random Forest, XGBoost, and LightGBM—by different metrics: precision, recall, and F1 score. All the models achieved perfect performance (100%) on the test set and showed strong generalization capabilities, confirmed by learning curves, cross-validation, and randomized label tests. The results demonstrate that these supervised methods can robustly differentiate between normal and abnormal system behavior. Compared with previous work based on deep or unsupervised learning, our approach is distinguished by its simplicity, interpretability, and stability. These results support the deployment of such models in real-time industrial environments for intelligent monitoring and predictive maintenance. Future directions include extension to larger datasets, exploration of hybrid models, and generalization to other areas of renewable energy.