电工技术学报2026，Vol.41Issue(9)：3129-3140,12.DOI:10.19595/j.cnki.1000-6753.tces.251049
基于物理信息神经网络的充油设备高能电弧故障压力泄放装置结构优化方法
Optimization Method for Pressure Relief Device Structure in Oil-Filled Equipment under High-Energy Arcing Faults Based on Physics-Informed Neural Network
摘要
Abstract
Recently,oil-filled equipment explosions caused by arcing faults have successively occurred.Conventional pressure relief devices are inadequate to meet the actual pressure relief demands of large-capacity,high-voltage oil-filled equipment.Due to the lack of theoretical guidance and experimental data,advancements in pressure relief technology have heavily relied on engineering experience and trial-and-error approaches.To this end,this paper proposes an optimization method for pressure relief devices based on physics-informed neural network(PINN).The proposed method combines accurate bubble-fluid-solid multiphysics simulations with the efficient predictive capability of the PINN model,achieving global optimization of structural parameters of pressure relief device with reduced computational cost.It advances the traditional experience-driven design for component structures into an efficient and quantitative phase. First,a bubble-fluid-solid multiphysics coupled model capable of accurately describing the pressure during arcing faults in oil-filled equipment was established to generate training datasets for the PINN.Second,the monotonic relationships between oil pressure and key input variables were incorporated into the PINN model.This allowed the model to accurately predict peak oil pressure while strictly adhering to physical constraints,thereby improving its interpretability and generalization.Finally,the particle swarm optimization algorithm was employed to obtain the global optimal structural parameters of the pressure relief device under the objective of minimizing the oil pressure peak.The effectiveness of the optimized scheme was validated through both simulations and experiments. Simulation results indicate that the optimized scheme effectively enlarges the pressure relief area and creates a streamlined venting path.The maximum flow rate of the optimized device reaches 248.2 L/s,representing a 53%improvement over the conventional design.Meanwhile,the oil discharge volume within 60 ms increases by 71%.Owing to its enhanced oil discharge capability,the oil pressure peak is reduced by 127 kPa.As the fault current increases,the mitigation effect becomes more pronounced.At 50 kA,the oil pressure peak is reduced by 226 kPa.Furthermore,arcing faults at depths of 250 mm,500 mm,and 750 mm show that the peak pressures are decreased by 38.1%,31.1%,and 30.4%,respectively.A greater pressure relief effect is observed when the fault occurs closer to the device.Subsequently,an optimized pressure relief device(PRD)prototype was manufactured and tested on the established on-site experimental platform.Under peak fault currents of 12 kA,15 kA,and 20 kA,the measured peak pressures with the optimized device are 145 kPa,177 kPa,and 254 kPa,respectively,representing reductions of 18.5%,26.9%,and 34.0%compared to the conventional design.These results demonstrate the superior performance of the optimized pressure relief device in mitigating arc-induced oil pressure. The following conclusions can be drawn from the experimental and calculation analysis:(1)By adjusting the structural parameters,the optimized scheme effectively mitigates the limitations of conventional designs,including the limited venting area and restricted oil discharge capacity.(2)Compared with traditional devices,the optimized prototype significantly reduces the peak oil pressure and moderates its growth rate with increasing fault current,demonstrating superior performance under different fault conditions and locations.(3)The developed PINN accurately approximates complex multiphysics simulation results within millisecond-level time.Combined with the global search capability of particle swarm optimization,the proposed approach overcomes the drawbacks of relying solely on numerical simulations for optimization,including low computational efficiency and the inability to guarantee global optimality.关键词
电弧故障/充油设备/压力泄放装置/物理信息神经网络Key words
Arcing fault/oil-filled equipment/pressure relief device/physics-informed neural network分类
信息技术与安全科学引用本文复制引用
刘浩,闫晨光,桑凡雅,陈一悰,梁栋..基于物理信息神经网络的充油设备高能电弧故障压力泄放装置结构优化方法[J].电工技术学报,2026,41(9):3129-3140,12.基金项目
国家自然科学基金资助项目(52277125). （52277125）
