摘要
本文建立了基于误差反向传播算法(BP)的多层感知器(MLP)结构的神经网络进行储罐区泄漏源的实时定位,提出了现场传感器阵列优化布置的原则,设计了输入空间和解空间,应用计算机仿真获得的泄漏浓度场对建立起来的神经网络进行了训练,并成功进行了验证,分析了隐层节点数、权值和阈值的取值范围和符号、学习因子等因素对网络性能的影响。研究表明利用神经网络的非线性映射能力对储罐区泄漏源进行实时定位是可行的,单隐层的MLP可以被训练来反向分析泄漏浓度场。
The BP algorithm based MLP neutral network applied to real- time leak position in tank area is developed. The principle of field sensors array optimized collocation is presented and the input and output solution space is designed. The concentration field simulated by computer is utilized to train the proposed neural network which is validated successfully. The influence of number of nodes in hidden layer, the sign and range of threshold and the study factor to the performance of neural network is analyzed. It is concluded that using nonlinear mapping of neural network to locate the leaking source is feasible. MLP with one hidden laver can be trained in backward analvsis of the dispersion.
出处
《中国安全生产科学技术》
CAS
2006年第4期79-83,共5页
Journal of Safety Science and Technology
基金
国家十五科技攻关滚动课题(编号:2004BA803B05)资助
关键词
人工神经网络
泄漏源定位
传感器阵列
重大危险源
artificial neutral network
leak position
sensor array
major hazard installation
