Data-driven fault detection and classification in power system using multivariate time-domain features

Consistent and quick fault detection is critical for stability while operating on security, the contemporary system, especially with an emergence for a growing system complexity and nonlinear operating circumstances. The paper introduces a three-phase power system classification and fault detection data-driven framework, build over a ML (machine learning) method. To simulate realistic operating conditions and a broad variety of fault conditions, power-system multi-machine model was developed with MATLAB/Simulink, to LG, LL, LLG, LLL, LLLG faults and normal operating conditions. The measured data of voltage and current of three phase system take place, consequently used to create a labeled dataset to detect binary faults and classify multiclass faults. Standardized preprocessing, cross-validation and performance measures were used to evaluate multiple machine learning models, namely Logistic Regression, Decision Tree, Random Forest, SVM and XG-Boost. The experimental findings are used to show that models based on are much more functional compared to the linear and single-tree classifiers. XG-Boost was best in terms of both binary fault detection and multiclass classification with 98.9 and 95.4 percent accuracy respectively, and is still stable in the presence of an imbalanced class distribution and noisy data. The importance of the analysis of feature importance implemented significant analysis of the role played by phase currents and voltages in fault discrimination. The suggested framework is a strong, interpretable, and computationally efficient approach to intelligent fault diagnosis and protection of contemporary power plants, which is why it can be applied to real-time monitoring and smart grids.