摘要
对前期大量试验采集的火灾气体数据进行特征提取,找出能够代表火灾整体特征的过程特征信息.通过体积分数曲线拟合分析,提取出体积分数、速度和加速度估值等火灾特征信息参量,建立适合于火灾早期探测的学习向量量化(LVQ)神经网络算法.通过对比分析证明,该算法比传统火灾探测器报警时间提前3～21min,且对于真假火灾可进行准确识别,实现火灾早期探测预警的目标.
The process feature information standing for the whole feature of fire can be found, after the feature is extracted from the fire gas data collected in a great number of the previous experiments. And the fire feature parameters such as the estimated values of gas concentration, velocity and acceleration can be extracted through the analysis of the curve fitting of the gas concentration to establish the algorithm of learning vector quantization （LVQ） neural network suitable for early fire detection. The comparison analysis has proven that the alarm time of the algorithm can be advanced 3 to 21 minutes compared with that of the traditional fire detectors, so that it could detect precisely whether the fire is true or not, to attain early fire detection alarm.
出处
《华侨大学学报（自然科学版）》
CAS
北大核心
2011年第6期607-610,共4页
Journal of Huaqiao University(Natural Science)
基金
福建省自然科学基金资助项目(2009J01290)
国务院侨办科研基金资助项目(09QZR04)
福建省厦门市科技计划项目(3502Z20103028)
关键词
火灾探测
红外光谱
学习向量量化
神经网络
早期预警算法
fire detection
infrared spectroscopy
learning vector quantization
neural network
early alarm algorithm
