Abstract
The increase in the scale and voltage level of urban power grids has made cable lines the main way of power transmission. However, their insulation systems are prone to Partial Discharge (PD) caused by local defects, leading to insulation degradation and even catastrophic power outages. Traditional PD detection methods are limited by sensor installation, environmental noise, and electromagnetic interference, which makes it challenging to balance sensitivity, real-time performance, and scalability. There is an urgent need to develop more efficient and reliable online monitoring technologies. Therefore, this study proposes a cable online PD detection and state assessment framework that integrates an improved Deep Belief Network (DBN) with an Inertial Krill-Herd Algorithm with Differential Bat Adaptation (IKHA-DBA). The DBN is improved by embedding DropConnect and elastic weight consolidation techniques. The IKHA-DBA algorithm is introduced to optimize the DBN to enhance the accuracy, real-time performance, and state evaluation capability of cable PD detection. Experiments have shown that the improved DBN model improved precision, recall, and AUC by 5.9%, 6.4%, and 6.0% compared to ResNet50-SE on the B0005 dataset. At the same time, the final model showed further improvements of 1.6%, 0.3%, and 1.4% in various indicators compared to the improved DBN. On the B0006 dataset, the final model's accuracy was 3.09% higher than the GA-DBN model and 5.13% higher than the improved DBN model. In summary, the proposed final framework has achieved comprehensive improvements in accuracy, robustness, and real-time performance in cable PD online detection and state evaluation, providing reliable technical support for the safe operation of urban power grid cable assets.