Current methods for assessing the state of partial discharge (PD) in gas-insulated switchgear (GIS) often treat the diagnosis and severity evaluation tasks independently, limiting model performance and ignoring defect types in the PD development stages. To address these issues, we propose a novel multi-task learning network for the condition evaluation of PD. Firstly, we introduce a lightweight attention combined with an expert network module to explore the correlations between different tasks while retaining their unique features. Secondly, we incorporate a CNN-Transformer module to capture both local and global features of GIS PD signals. Lastly, we implement a dynamic weight averaging method for the multi-task network, which adaptively adjusts the loss weights based on the loss variation rate of each sub-task. Experimental results demonstrate that the proposed multi-task network significantly improves the GIS state assessment performance for each sub-task, advancing the development stage evaluation of PD for various insulation defects.

partial discharge,GIS,condition assessmen,multi-task learning,CNN-Transformer