基于神经网络的高寒地区CF_(4)和SF_(6)/CF_(4)检测

Neural Network-based CF_(4) and SF_(6)/CF_(4) Detection in High Altitude and Extreme Cold Regions

高寒地区须携带多台仪器以满足3种不同量级SF_(6)气体中CF_(4)气体浓度的检测需求,现场运维效率低且仪器购置成本高。为此,首先设计了一种基于热释电检测技术的SF_(6)气体中CF_(4)气体浓度检测仪器,可自动选择不同的放大电阻以实现多量程切换。然后提出了BP和PSO-BP2种神经网络温度-压力协同补偿模型,并通过搭建高效模拟实验平台为模型预测提供数据支撑,预测结果表明,PSO-BP神经网络优于BP神经网络。最后将PSO-BP神经网络温度-压力协同补偿模型内置于多量程检测仪器CF_(4)气体浓度检测仪器。模拟实验结果表明,该检测仪器在不同温度和压力下,小量程和大量程检测误差和重复性分别不超过±2%和1.6%,混合比量程下误差和重复性分别不超过±0.5%和0.2%,对高寒地区电网运维检修具有重要作用。

In extreme cold regions,the need to carry multiple instruments to meet the demands for detecting varying concentration levels of CF_(4) gas within SF_(6) gas leads to inefficient field operations and high costs for instrument acquisition.To overcome this,an SF_(6) gas CF_(4) concentration detector utilizing pyroelectric detection technology was initially developed,capable of automatically switching among different ranges by selecting appropriate amplification resistances.Subsequently,two neural network models for temperature-pressure collaborative compensation,BP and PSO-BP,were introduced.Data for model predictions were supported by an effective simulated experimental platform,with results indicating the PSO-BP neural network's superiority over the BP network.The PSO-BP neural network's temperature-pressure collaborative compensation model was then embedded within the multi-range detection instrument for CF_(4) gas concentration.Simulation experiments demonstrated that the instrument maintains a detection error and repeatability within ±2% and 1.6% across small and large ranges,and within ±0.5% and 0.2% for mixed ratio ranges,respectively,under varying temperatures and pressures.This technological advancement significantly enhances maintenance operations within the power grids of cold regions.