神经网络在矿井胶带机火灾探测中的应用

综合考虑温度、温度变化率、一氧化碳浓度、一氧化碳浓度变化率为胶带机火灾探测系统的参数 ,采用三层前馈 BP神经网络对胶带输送机火灾探测进行研究 ,经过 81组数据样本训练得到合适的网络结构及网络参数 .仿真结果表明 :当隐层节点数为 1 0 ,动量因子为 0 .95时 ,训练时间最短 ,误差下降最快 ;对 PVC胶带机的火灾燃烧试验探测结果表明 ,当输出报警限取 0 .5时 ,经一定的时间延时后 ,运用该网络可在燃烧发生 35 0 s时进行报警 ,从而有效实现了矿井胶带输送机火灾的早期报警 ,并增强了系统的抗干扰能力及对环境的适应性能

Application of neural network to early detection of the fire of rubber belt conveyer in mine

The detection of the fire of rubber belt conveyer is studied by means of a forward neural network with three layers, and temperature, rate of temperature change, concentration of carbon monoxide and rate of carbon monoxide concentration change are taken as the parameters of the network. The appropriate parameters and architecture of network are obtained after the network is trained with 81 sets of data. The simulation result shows that training time is the shortest and error reduces most rapidly when the neurons in hidden layer are ten and momentum coefficient is equal to 0.95. The test result of PVC rubber belt conveyer shows that when 0.5 is regarded as the alarm threshold and certain time delay is employed, the neural network fire detecting system will alarm as fire takes place about 350 s. So it can effectively implement early alarm; at the same time, the it greatly enhances the suitability of an alarm system to environmental variation and anti-interference capability, and the intelligent degree of an alarm system is also improved.