• Proposal of a Causal AI System to Overcome DL Limitations • Dual Innovation in Causal Reasoning and Knowledge Adaptation • Quantified Performance Gains in Robustness and Accuracy The transmission system operates in an open environment, where unknown failure modes can lead to significant performance degradation due to shifts in data distribution. Traditional deep learning methods, which rely on the assumption of independent and identically distributed data, often overlook the essential causal relationships underlying faults, resulting in poor generalization in real-world scenarios. To address this issue, this study proposes a cross-scenario robust verification system (RVS) that integrates causal reasoning with dynamic knowledge updating. The system employs a deep confidence calibration module to quantify decision risk. In addition, an intervention engine based on a structural causal model (SCM) is utilized to perform counterfactual reasoning on observed data. This process strips away superficial correlations and reveals the fundamental causal mechanisms of faults. Meanwhile, graph neural networks (GNNs) are adopted to enable dynamic topology reconfiguration, allowing new and existing knowledge to coexist and reinforce each other. Experimental results demonstrate that the proposed system achieves an area under the curve (AUC) of 0.914 in unknown fault detection tasks, representing a 5% improvement over baseline methods. The distribution stability index (DSI) reaches 0.941, and diagnostic accuracy remains stable at 86.3% under the worst-case scenario. These performance gains are attributed to causal feature extraction - which resists spurious correlations - and structured knowledge storage, which effectively preserves experiential knowledge. This research provides a novel paradigm for transmission system operation and maintenance, one that integrates causality and knowledge evolution.
Wang et al. (Sun,) studied this question.