As the scale and complexity of electrical grids continue to expand, the necessity for robust fault detection techniques becomes increasingly urgent. This paper seeks to address the limitations in traditional fault detection approaches, such as the dependence on human experience, low efficiency, and a lack of logical relationships. In response, this study presents a cascaded model that leverages the Random Forest classifier in combination with knowledge reasoning. The proposed method exhibits a high efficiency and accuracy in identifying six basic fault types. This approach not only simplifies fault detection and handling processes but also improves their interpretability. The paper begins by constructing a power fault simulation model, which is based on the IEEE 14-bus system. Subsequently, a Random Forest classification model is developed and compared with other commonly used models such as Support Vector Machines (SVMs), k-Nearest Neighbor (KNN), and Naïve Bayes, using metrics such as the F1-score, accuracy, and confusion matrices. Our results reveal that the Random Forest classifier outperforms the other models, particularly in small-sample datasets, with an accuracy of 90%. Then, we apply knowledge mining technology to create a comprehensive knowledge graph of power faults. At last, we use the transE model for knowledge reasoning to enhance the interpretability to assist decision making and to validate its reliability.