Intelligent fault diagnosis in power systems: A comparative analysis of machine learning-based algorithms

Background: The electric power transmission line plays a vital role in ensuring a consistent and reliable supply of electrical power. However, due to their long length, these lines are prone to faults, which can lead to power outages or disturbances in the network, thereby reducing the overall reliability of the supply to consumers. Hence, a prompt restoration is required to ensure a reliable power supply. It necessitates the accurate fault detection and classification (FDC) in the early stages of transmission line fault occurrences. Methods: This paper proposes an intelligent approach based on decision tree (DT) to address the FDC problem in transmission lines. The study is carried out on Western System Coordinating Council (WSCC) 9-bus test system, which includes the optimal placement of phasor measurement units (PMUs) for effective FDC. Multiple faults are induced in the WSCC 9-bus system by varying parameters such as fault distance, fault resistance, and fault inception angle using MATLAB/Simulink. In these fault scenarios, the PMU-assisted wide area measurement system determines the post-fault bus voltage, current magnitudes, and phase angles. The data collected from these simulations are used for training and testing the proposed DT-based model. Results: The proposed scheme achieves an average fault classification accuracy of 99.95%, detection accuracy of 100%, and the average response time to detect the fault with maximum fault resistance of 800 Omega is 15.712 ms. The results indicate that the proposed approach outperforms other machine learning-based FDC techniques, including support vector machines, K-nearest neighbor, ensemble methods, and neural networks. The proposed FDC scheme shows its superiority in FDC accuracy for all test cases considered. In addition, incorporating phase angle measurements enhances the FDC accuracy compared to the system considers the voltage and current magnitude features alone. Conclusions: The experimental results demonstrate that the proposed algorithm surpasses other state-of-the-art algorithms in quantitative assessments, leading to enhanced accuracy.