地下变电站变压器室爆炸行为分析

Analysis of Explosion Behavior in Transformer Room of Underground Substation

针对地下变压器室存在爆炸问题,根据变压器油电弧产气爆炸行为特性,本文提出构建递归神经网络变分自编码器的潜在因子分析方法,获得变压器电弧产气行为的低维特性并刻画爆炸行为。首先,利用实验和计算流体力学软件FLACS对某220 kV全地下变电站的主变压器室内变压器油释放可燃气体引发的爆炸进行模拟计算,比较不同可燃气体量与不同泄放面积下地下变电站内的压力温度分布;进一步利用递归神经网络变分自编码器方法提取不同气体爆炸时的温度变化和压力变化数据潜在因子,从低维动态轨迹和稳定流形平面角度探测产生爆炸行为特性。分析表明,计算范围内室外最大泄爆压力随泄放面积增大保持相对稳定略有上升,可燃气体当量体积增大也会提升室外最大泄爆压力。本文为变压器产气爆炸特征研究提供了一种新的途径。

In view of the explosion problem in an underground transformer room,a latent factor analysis method of re⁃current neural network(RNN)-based variational auto-encoder(VAE)is proposed to analyze the explosion behavior ac⁃cording to explosion characteristics of gas from transformer oil by arc.First,the software FLACS is used to simulate the explosion caused by the released combustible gas produced from transformer oil in the main transformer room of a 220 kV underground substation.The pressure and temperature distributions in the underground substation under different vol⁃umes of combustible gas and different discharge areas are compared.Furthermore,the RNN VAE method is used to ex⁃tract latent factors for the variation data of temperature and pressure under different gas explosions,and the explosion behavior characteristics are detected from the low-dimensional dynamic trajectory with the stable manifold plane.It is found that the outdoor maximum explosion pressure will keep relatively stable and slightly increase with the increase in discharge areas,and the increase in the equivalent volume of combustible gas will also increase the outdoor maximum explosion pressure.The research in this paper provides a novel way to study the explosion characteristics of gas pro⁃duced by arc from transformer oil..