热辐射引发设备二次损坏概率的BP神经网络预测

Prediction of probability of equipment secondary damage caused by thermal radiation based on BP neural network

为提高化工园区火灾热辐射风险预测精度,预防设备二次损坏事故的发生,通过建立3层反向传播(BP)神经网络模型来预测设备损坏概率,研究二次损坏概率与影响因素参数(设备容积,视角系数,与爆炸点的距离)之间关系。以用多米诺风险理论计算出的设备二次损坏概率为样本集,对建立的BP神经网络模型进行训练、测试和误差分析。结果表明,设备损坏概率与神经网络预测的概率之间最小误差值为9.962 5×10-3。设备二次损坏概率随距离的增大而减小,且随设备容积、视角系数的增大而增大。其中,视角系数对损坏概率的影响最明显。

For the sake of improving chemical industry park fire thermal radiation risk prediction accura- cy and preventing equipment secondary damage accident, a BP neural network model was built for predic- tion the damage probability. Relationships between secondary damage probabilities and parameters, such as volume of equipment, view factor and distance, were discussed. Damage probabilities related to differ- ent parameters were calculated based on theory of Domino risk and selected as samples for training and tes- ting the network. Difference between prediction and actual results was also discussed. The results show that minimum mean squared error is 9. 962 5×10^-3, that the damage probability decreases with increase of distance and increases with increase of view factor, and that the influence of view factor on damage probability is most obvious.