一种新型BP神经网络模型在火灾探测信息处理中的应用  

Improved neural network model application in fire detection

作　　者：赵望达[1] 李卫高[1] 熊涵予 韩柯柯 

机构地区：[1]中南大学土木工程学院,湖南长沙410075

出　　处：《铁道科学与工程学报》2015年第5期1212-1218,共7页Journal of Railway Science and Engineering

基　　金：浙江省交通运输厅科技计划项目(2010H01);中南大学研究生自由探索资助项目(2014zzts245);中南大学大学生创新训练资助项目(CL14115)

摘　　要：现阶段,神经网络模型在火灾探测信息处理应用中存在以下缺陷:选取火灾特征组合具有主观性;选取的神经网络类型缺乏对比;缺乏大量实验数据对神经网络泛化能力的验证。利用NIST机构所做一系列火灾探测研究实验数据样本,通过信息熵理论在火灾信号选取中的应用获取火灾复合探测信号特征选取的组合形式,并在此基础上建立火灾探测信息处理神经网络初始模型。经过一系列Matlab仿真实验,分析神经网络的模型结构、传递函数和训练函数对仿真结果的影响,提出一种基于trainbr训练函数、tansig传递函数的3-7-1结构BP神经网络模型。采用网络训练时间、探测点、误报率和网络输出区间进行网络性能分析,验证所提出模型在火灾探测中应用具有训练速度快,结果稳定可靠,探测灵敏的特点。In the processing of fire detection,some defects are contained in the use of neutal network model. These defects are the subjectivity of the selection of fire feature,the lack of comparison among the selected neural network types and the lack of adequate experiments.Based on a series of data samples of fire detection experi-ments given by NIST,the combination of signal feature selection for composite fire detection through the applica-tion of information entropy theory in the selection of fire signal is obtained,and the initial model of neural net-work for the fire detection information processing is established.After a series of exploratory experiments simula-ted by Matlab,the effects of the structure of the neural network model,transfer function and the training function on the simulation results are analyzed,and an improved BP neural network model based on trainbr training func-tion,tansig transfer function and the structure of 3 -7 -1 is proposed found.Through the network performance analysis of network training time,probe points,the false positive rate,and network output interval,the proposed model is fast,reliable and senstitive in fire detection.

关 键 词：火灾探测 神经网络 信息熵 

分 类 号：TK121[动力工程及工程热物理—工程热物理]