Smart grids integrating renewable energy sources, distributed generation, and complex control systems face mounting challenges in fault detection and diagnosis, particularly during transitions between grid-connected and islanded operation modes. Traditional single-model approaches struggle with the dynamic behavior and changing fault characteristics inherent in modern power systems. Researchers have now proposed a hybrid machine learning framework that addresses these limitations through a two-stage diagnostic process.
The system employs a random forest model as its first stage, equipped with dual output layers. The first layer determines overall grid status—normal operation or fault condition—while the second identifies whether the smart grid is operating in grid-connected or islanded mode. Once a fault is detected, the system routes to one of two specialized convolutional neural networks: one classifies the faulty bus location, while the other categorizes the fault type from among 11 predefined short circuit scenarios.
Validation testing used a modified IEEE 14-bus test system enhanced with renewable generation resources, simulated in MATLAB/Simulink under diverse fault conditions. Results demonstrate exceptional performance metrics: 98% accuracy in fault detection, 95% accuracy in mode identification, 80% accuracy in faulty bus localization, and 87% accuracy in fault type classification. These figures substantially exceed performance of conventional approaches including support vector machines, k-nearest neighbors, and recurrent neural networks.
The hybrid architecture's key advantage lies in its ability to handle the complexity of modern grids where operational modes shift dynamically and fault signatures vary accordingly. By separating fault detection from classification tasks and incorporating mode awareness, the framework reduces misclassification rates common in single-model systems. This scalable solution enables utilities to implement real-time automated fault diagnosis, supporting faster response times and improved grid resilience. As renewable penetration increases globally, such AI-driven diagnostic systems become essential infrastructure for maintaining reliability and accelerating fault isolation in increasingly complex power networks.
