基于GA-BP神经网络的化工园区应急救援能力可靠性分析

Reliability Analysis for Emergency Resue Capacity in the Chemical Industry Park Based on GA-BP Neural Network

提出一种基于遗传算法优化的BP神经网络(GA-BP)的化工园区应急救援能力可靠性分析模型。通过事故树分析对16家化工园区应急救援能力的可靠性进行量化,并作为GA-BP神经网络模型训练的输出值;以事故树中的28项基本事件为依据进行分类总结,建立化工园区应急救援能力层次分析评估指标体系,在日常生产状态下的应急系统维护与事故时的应急处置能力这两个准则层下分为要素层,包括应急系统硬件维护、应急救援人员管理、应急管理机构、应急预案与演练和信息传递、应急人员动员、现场处置、事故后恢复能力,指标层元素分别对应事故树的基本事件,并计算指标层元素相对于目标层的复合权重,再以调查问卷的方式邀请专家对化工园区的指标层元素进行打分,将每一园区的各项要素得分与复合权重相乘作为GA-BP神经网络模型的输入值;从样本组中选取12组作为训练样本、4组作为测试样本,验证建立的GA-BP神经网络模型的可行性,并与传统BP神经网络的分析数据进行对比。结果显示:GA-BP神经网络输出数据的平均误差为3.83%,均方误差为0.002;而BP神经网络输出数据的平均误差为8.13%,均方误差为0.004;GA-BP神经网络的分析结果与事故树的分析结果更为接近,且相对于事故树分析减少了复杂的建树过程,具有更高的易用性。

Based on the genetic algorithm optimized BP neural network （GA-BP）, this paper proposes a reliability a- nalysis model for the response capacity in the chemical industry park. According to the fault tree analysis, the paper quantifies the reliability of the emergency rescue capability of 16 chemical industry parks and takes the quantitative results as GA-BP neural network output values. On the basis of 28-basic events in the fault tree,the paper classi- fies, summarizes and establishes the hierarchy analysis index system for the chemical industry park. Under the two criteria layer, one is the emergency system maintenance in the daily production, the other is the emergency response capacity in the accidents, and the paper divides the hierarchy analysis index system into elements, including emer- gency system hardware maintenance, emergency rescue personnel management, emergency management agencies, e-mergency plans and exercises and information delivery, emergency personnel mobilization, on-site disposal, and abili- ty to recover after an accident. The index elements correspond to the basic fault tree event, and the paper calculates the composite weights for the target layer. The paper invites experts to indicate for the rate layer element in the chemical industry park in the questionnaire, and multiplies the score of every element with the weights in every park,as the input values for GA-BP neural network. From the sample group, the study selects 12 groups as the training sample,4 group as the test samples, to verify the feasibility on the genetic algorithm optimization neural network model based on reliability analysis model, and compares with the traditional BP neural network analysis data. The results show that the average error is 3.83 %, mean square error is 0. 002 based on the GA-BP neural network. But, the average error is 8.13 %, mean square error is 0. 004 based on the BP neural network. So, the anal- ysis results of the GA-BP neural network are closer with the fault tree analysis, and it reduces the complexity process comparison with the fault tree analysis, and is greater useful.