基于模糊神经网络的火灾识别算法

Forest Fire Recognition Algorithm Based on Fuzzy Nerual Network

火灾自动识别能够及时准确预报火情。在森林大空间的环境中,由于火灾信号具有非线性和不确定性,将采集的探测信号做简单的分析与比较,误报率比较高。如何融合几个传感器的信号进行有效地火灾识别是一个难点。为提高预测的准确性,针对传统的森林火情预测系统误报率高的缺点,提出一种基于模糊神经网络的火灾识别算法。首先,将模糊控制和神经网络以串联的方式结合,将采集的传感器信号进行处理后送入三层前馈BP网络进行处理,输出明火概率、阴燃火概率、无火概率,然后,将它们作为模糊控制系统的输入,模糊化后进行模糊推理,最后去模糊化得出火灾概率大小。并利用MATLAB工具箱对构建的算法模型进行仿真分析,仿真结果表明,本文的方法能够有效地融合多个火灾探测传感器的信号,快速而准确的判断出火情的大小,提高火灾识别的准确率,减少误报率。

Fire automatic recognition can efficiently and accurately forecast fires. Because forest scene is a large space field, fire detection signals are non - liner and uncertain. False alarm rate of fire recognition increases if detec- tion signals are simply analyzed and compared. Therefore, the difficulty is how to fuse the signals of several senors in order to detect fires. Considering the high false alarm rate of traditional forest fire forecast systems, this paper presen- ted a forest fire recognition method based on Fuzzy Nerual Network. It combined fuzzy control with neural network in a serial way for fire recognization. Firstly, the signals collected by sensor were put into the three - layer feed - for- ward BP network, which then recognized and output the fire probabilities, smoldering fire probabilities, and no_fire probabilities. Afterwards, the three probabilities were transferred into fuzzy control system, and processed by fuzzy membership function. After fuzzy reasoning we can obtain the fire probability. Finally, MATLAB toolbox was used to simulate and analyze the proposed model. The results show that the system can effectively fuse the signals of several senors, judge the situation of the fire more quickly and accurately, and reduce the rate of false alarm.