Research Review on AI-Driven Fault Prediction and Diagnosis Methods for Power Systems

Abstract

In the current era of deep integration between energy transition and digitalization, power systems as critical infrastructure are facing increasingly complex operational challenges. The rapid development of artificial intelligence (AI) technology has injected new theoretical momentum and methodological support into the field of fault prediction and diagnosis. This paper systematically reviews the evolution from traditional physics-based diagnostic mechanisms to data-driven intelligent paradigms, with a focus on analyzing the integration pathways of cutting-edge AI methods such as deep learning, graph neural networks, and transfer learning in power equipment condition monitoring, fault feature extraction, and causal inference. The research demonstrates that AI technology can significantly enhance the representation capabilities of high-dimensional, non-stationary power system data, enabling early warning of potential faults and accurate identification of multi-source faults, while showing distinct advantages in addressing uncertainties brought by renewable energy integration. However, current research still faces challenges including poor model interpretability, inadequate adaptability to small-sample scenarios, and incomplete cross-domain knowledge transfer mechanisms. Future research should focus on constructing hybrid intelligent diagnostic frameworks combining physical constraints with data-driven approaches, exploring lightweight model deployment strategies for edge computing, and strengthening interdisciplinary theoretical integration to promote the autonomous and intelligent development of power system fault management.