Enhanced Fault Diagnosis Method for Cable Terminals in Distribution Networks Based on MIC and Improved Deep Belief Network

The stable operation of cable terminals in distribution networks is crucial for power systems. However, traditional fault diagnosis models suffer from inefficiency and insufficient reliability. To achieve precise fault diagnosis of cable terminals, this paper proposes a novel method integrating the Maximum Information Coefficient (MIC) and an Improved Archimedes Optimization Algorithm (IAOA)-optimized Deep Belief Network (DBN). First, the MIC theory is applied to reduce dimensionality and extract key features from the dissolved gas concentration in insulating oil, identifying the most critical diagnostic indicators. Next, the selected features are fed into the DBN model. To address the challenge of hyperparameter selection in DBN, the IAOA is introduced for optimization. To enhance the global search capability of the Archimedes Optimization Algorithm (AOA) and prevent premature convergence, three improvement strategies are implemented: chaotic initialization, arithmetic crossover operator, and cosine-based density reduction factor. Finally, a fault simulation test platform for distribution cable terminals is established to collect sample data under different fault conditions and construct labeled datasets for validation. Experimental results demonstrate that the proposed method achieves an accuracy of 98.33% on the test set. Compared with conventional fault diagnosis models, the proposed approach exhibits superior stability and higher recognition accuracy.