神经网络在智能火灾预警系统的应用  

Application of neural network in intelligent fire alarm system

作　　者：丁承君[1] 赵泽羽 朱雪宏[1] 冯玉伯 

机构地区：[1]河北工业大学机械工程学院,天津300130

出　　处：《传感器与微系统》2018年第1期154-156,160,共4页Transducer and Microsystem Technologies

基　　金：天津市科技支撑计划资助项目(14ZCDZGX00811;13ZCZDGX01200;15ZXHLGX0210);天津市产学研合作项目(14ZCZDSF00025);天津市"863"成果转化项目(13RCHZGX01116;14RCHZGX00862)

摘　　要：提出了综合利用多种传感器准确判断火焰状态的算法。在神经网络算法下,根据一氧化碳传感器、烟雾传感器和温度传感器数据,对火焰状态进行分类。使用有限内存的拟牛顿(L-BFGS)优化算法对传统反向传播(BP)算法进行改进,加快了神经网络的收敛过程,有效提高了系统精度。实验验证了算法能够有效降低火灾预警系统的误报率,增强系统安全报警的灵敏度和可靠性,实现了实时有效的火灾预警。An algorithm for judging flame state by using a variety of sensors is proposed. According to carbon monoxide sensor,smoke sensor and temperature sensor data,flame state is classified under the neural network algorithm. And use the L-BFGS optimization algorithm to improve the traditional BP algorithm to speed up the convergence process of neural network,and effectively improve precision of system. The experiment proves that this algorithm can effectively reduce the false alarm rate of the fire alarm system,enhance the sensitivity and reliability of the system security alarm,and realtime and effective fire warning can be carried out.

关 键 词：多传感器数据融合 火灾预警 神经网络 有限内存的拟牛顿算法 

分 类 号：TP202[自动化与计算机技术—检测技术与自动化装置]