飞机货舱火灾CO浓度神经网络补偿算法研究  

Study on compensation algorithm of CO concentration in aircraft cargo fire based on neural network

作　　者：王海斌 瞿忱 张志慧 WANG Haibin;QU Chen;ZHANG Zhihui(College of Civil Aviation Safety Engineering,Civil Aviation Flight University of China,Guanghan 618307,Sichuan,China)

机构地区：[1]中国民用航空飞行学院民航安全工程学院,四川广汉618307

出　　处：《安全与环境学报》2023年第10期3606-3612,共7页Journal of Safety and Environment

基　　金：四川省院省校合作项目(2022YFSY0048);中国民用航空飞行学院研究生科研创新计划项目(XSY2022-04)。

摘　　要：针对飞机货舱火灾探测误报率偏高且响应速度较慢的问题,采用电化学式一氧化碳传感器来代替传统民机所用的光电式烟雾探测器来探测飞机货舱火灾,并提出了一种基于粒子群算法(Particle Swarm Optimization,PSO)优化长短期记忆(Long Short-Term Memory,LSTM)神经网络的一氧化碳浓度补偿模型。首先在自搭建试验平台采集密闭空间火灾的多项试验数据,然后用PSO优化LSTM的隐藏层神经元个数和学习率,提高了LSTM的预测精度。通过与其他3种神经网络对比,PSO改进LSTM模型在基于时间序列的火灾一氧化碳检测中具有更好的补偿效果。通过浓度补偿,可以使电化学式一氧化碳探测器在飞机货舱火灾发生的早期阶段进行更准确的探测预警。Aiming at the problem of high false alarm rate and slow response speed of aircraft cargo fire detection,this paper briefly analyzes the causes of these problems.Electrochemical carbon monoxide sensors are used to replace the photoelectric smoke detector used in traditional civil aircraft to detect aircraft cargo fire,and a carbon monoxide concentration compensation model based on Particle Swarm Optimization(PSO)to optimize Long Short-Term Memory(LSTM)neural network.First of all,the test data such as carbon monoxide concentration,temperature and humidity of three typical solid combustibles,beech,cotton rope,and corrugated paper are collected on the self-built test platform in case of fire in the confined space,and the carbon monoxide concentration measured by the flue gas analyzer is used as the benchmark for the measurement value of the electrochemical carbon monoxide sensor.Then,PSO is used to optimize the number of neurons in the hidden layer of LSTM and the learning rate.And the prediction error of the model is taken as the fitness value of the particles,and the velocity and position of the particles are updated as the global extremum.The architecture flow of the PSO LSTM compensation model is introduced in MATLAB.Then,the carbon monoxide concentration compensation model is obtained by training the test data.The training results show that the prediction accuracy of LSTM can be improved.By comparing and analyzing the concentration compensation effect of PSO LSTM and LSTM,Back Propagation(BP),Genetic Algorithms(GA)BP neural networks,and integrating R2,average absolute percentage error,root mean square error,and other evaluation indicators,it is concluded that PSO improved LSTM model has better compensation effect in the time-series-based fire carbon monoxide detection.Through concentration compensation,the electrochemical carbon monoxide detector can detect and warn other side gases more accurately in the early stage of aircraft cargo fire than the photoelectric smoke on traditional civil aircraft,and has higher

关 键 词：安全工程 火灾探测 粒子群算法 长短期记忆神经网络 浓度补偿 

分 类 号：X932[环境科学与工程—安全科学]