基于改进BP网络的弹药库房有害气体检测  

Detection of Harmful Gas in Ammunition Warehouse Based on BP Neural Network Improved by Particle Swarm Optimization

作　　者：刘建国[1] 安振涛[1,2] 张倩[1] 赵志宁[1] 

机构地区：[1]军械工程学院弹药工程系,石家庄050003 [2]军械工程学院弹药保障与安全性评估军队重点实验室,石家庄050003

出　　处：《计算机测量与控制》2014年第1期125-127,130,共4页Computer Measurement &Control

摘　　要：通过气体采样分析,确定了弹药库房需要重点检测的有害气体种类,在此基础上构建了基于传感器阵列与粒子群优化BP网络的气体检测系统;针对线性惯性权重调整的粒子群优化BP神经网络算法的不足,提出了一种新的非线性惯性权重调整方法,既保证了算法在运行前期具有较快的搜索速度和全局搜索能力,又保证了后期具有较高的搜索精度;为了克服算法在运行后期的局部收敛,在算法运行过程中引入速度变异算子,使算法摆脱了易陷入局部最优点的束缚;最后,通过实验对气体检测系统的性能进行了检验。结果表明,该气体检测系统速度快、精度高,能够较好地实现对弹药库房有害气体的检测。Through the analysis of the gas sample, the species of harmful gas was defined in the ammunition warehouse. The gas analysis system composed of sensor array and BP neural network was established. An advanced BP algorithm based on improved particle swarm opti- mization with non--linear adjustment of inertia weight was proposed by analyzing the shortcomings of BP algorithm using linear adjustment of inertia weight. The algorithm can not only maintain the characteristic of fast speed and global searching in the early convergence phase, but also keep a high precision in the later convergence phase. In order to escape from the local minimum basin of attraction of the later phase, a mutation operator of velocity is added to the Particle Swarm Optimization （PSO） algorithm. In the end, the performance of the algorithm was tested by the experiment. The experimental result indicates that the system has a rapid detection speed and a high precision. It is an effective method for harmful gas detection.

关 键 词：改进的粒子群算法 非线性惯性权重 速度变异算子 气体检测 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]