LM算法在火灾早期探测中的应用  

Research of earlier fire detection based on LM arithmetic

作　　者：刘二丽[1] 张认成[1] 崔长彩[1] 刘庆[1] 

机构地区：[1]华侨大学机电及自动化学院,福建泉州362021

出　　处：《消防科学与技术》2010年第1期54-57,共4页Fire Science and Technology

基　　金：基于智能无线传感器网络的火灾探测技术及系统研究(2009J01290);基于多传感器融合的电力设备局部放电在线监测诊断技术及系统研究(2009 H0031)

摘　　要：针对传统火灾探测方法在外界干扰下难以快速、准确地识别火灾,提出基于过程特征信息的神经网络火灾探测方法。选用CO和CO2气体体积分数比值、比值上升速率和加速度作为火灾过程特征参数,设计三层误差反向传播(BP)人工神经网络模型,采用改进的BP算法-LM算法进行网络学习训练。实验包括6种真实火灾源材料及蜡烛、香烟、液化气3种虚假火灾源材料。结果表明,基于LM算法的人工神经网络火灾探测系统能有效识别火灾,提高火灾探测的可靠性与准确性。Due to the weakness of traditional fire detection methods in detecting the fire, such as slow and inaccuracy under the interference of environment, an identification method for fire smoke based on information of the process of feature and the neural network method of fire detection is discussed. CO and CO2 volume fraction ratio, the ratio of increase in rates and characteristics of the acceleration are chosen as characteristic parameters of the process of fire. By designing a model of three layers back--propagation （BP） artificial neural network and using improved BP arithmetic--LM arithmetic, the network is learned and trained. 6 materials of true fire and 3 materials of fake fire like candle, cigareiLte and LPG were tested. The experimental results show that, fire detection system based on BP artificial neural network can effectively identify the fire and improve the reliability of fire detection, as well as accuracy.

关 键 词：火灾早期探测 LM算法 过程特征 神经网络模型 

分 类 号：X924.3[环境科学与工程—安全科学] TP277[自动化与计算机技术—检测技术与自动化装置]