Simultaneous Partial Discharge Diagnosis and Localization in Gas-Insulated Switchgear via a Dual-Task Learning Network

Diagnosis and location of partial discharge (PD) are the basis for ensuring the reliable operation of gas-insulated switchgear (GIS). Current PD diagnosis and localization are implemented as two separate tasks and the connection between them is ignored, which is challenging and time-consuming for improving the performance of the model. To address this issue, we propose a dual-task network (DTN) to realize GIS PD diagnosis and localization simultaneously. First, an attention bidirectional gated recurrent unit is constructed as a feature extractor to effectively mine temporal dependencies while extracting discriminative features. Then, the multigate mixture-of-experts (MMoE) is adopted to learn the difference and coupling relationship between PD diagnosis and localization tasks. Finally, the PD diagnosis and localization results are output by merging the various weight parameters of the MMoE expert network. In addition, a homoscedastic uncertain loss function is introduced to automatically adjust the subtask weights to optimize the DTN proposed. The experimental results demonstrate that the PD diagnosis accuracy of the DTN proposed reaches 97.56% and that the localization error is <10 cm. DTN offers significantly improved performance compared with single-task and other methods, offering a creative approach to the diagnosis and localization of GIS PD.