A Novel Transformer-LSTM-Based Multitask Network for Partial Discharge Condition Assessment in Gas-Insulated Switchgear

作者：Jing Yan, Yanxin Wang, Jianhua Wang, Yingsan Geng, Dipti Srinivasan · 年份：2024 · DOI：10.23919/cmd62064.2024.10766217 · 被引用次数：5 · 研究领域：High voltage insulation and dielectric phenomena、Power Transformer Diagnostics and Insulation、Elevator Systems and Control

Gas-insulated switchgear (GIS) partial discharge (PD) condition assessment is crucial for ensuring the safe operation of power systems. Traditional methods often struggle to accurately diagnose the complex PD patterns present in GIS equipment, and research on localization and risk assessment is limited. Additionally, current methods overlook the guidance provided by PD type and location information for risk assessment and the inter-task relationships, as well as historical information guidance. In this study, we propose a novel approach for GIS PD condition assessment using a Transformer-LSTM-based multitask network. The proposed network architecture combines the strengths of Transformer and LSTM models to effectively capture temporal and spatial dependencies in PD signals and extract historical dependency knowledge to improve model performance. The multitask learning framework allows the network to simultaneously perform multiple PD-related tasks, such as PD type classification, severity assessment, and localization. Moreover, by exploring the inter-task relationships, particularly the guidance provided by location and type for severity assessment, the performance of each task is effectively enhanced. Experimental results demonstrate that the proposed Transformer-LSTM based multitask network achieves superior performance compared to traditional methods, with improved accuracy and robustness in GIS PD condition assessment. This research presents a promising solution for enhan...