基于模糊人工神经网络的安全风险评估模型

A risk assessment model based on the fuzzy artificial neural network(FANN)

针对安全风险的复杂性、不确定性和不可预见性特征愈加明显的发展趋势,提出了一种基于模糊人工神经网络的风险评估模型。在传统事故树分析方法的基础上,引入模糊集理论和层次分析法,利用三角模糊数收集、整理专家的专业性判断语言,然后通过聚类分析和去模糊化处理得到事故树中基本事件的发生概率。将基本事件作为人工神经网络的输入层神经元,顶上事件作为输出层神经元,形成事故树嵌入人工神经网络的方法,提出该人工神经网络的构建、训练及测试方法。以沿海航行船舶的沉船事故作为风险评估案例应用该风险评估模型,结果表明,该模型的计算误差(均方误差)在0.013~0.014,精度符合应用要求,与传统事故树计算方法得到的结果相比,二者差值绝对值的平均值、最大值及均方误差分别为0.0093、0.0196和0.0052,二者计算结果的一致性较好。所构建的风险评估模型在自我优化、使用便捷性和动态适用性方面优势明显。

The paper intends to propose a risk assessment model based on fuzzy artificial neural network(FANN)in view of the increasing trend of the complexity,uncertainty and unpredictability of the safety risks.On the basis of the traditional accident tree analysis method,we have introduced the fuzzy set theory and the analytic hierarchical process by collecting the professional judgment and the sorted language of experts via the triangular fuzzy numbers,so as to obtain the occurrence probability of the basic events in the accident tree through the cluster analysis and de-fuzzing procedure.And,then,taking the basic events as the input layer neuron of the artificial neural network and the top event as the output layer neuron,we have managed to formulate and work out the method of embedding the accident tree into the artificial neural network.By taking the coastal navigation in the sinking of the ship as a case of risk assessment,and,by applying the risk assessment model,the results show that the calculation error of the model(i.e.the mean square error model)between 0.013-0.014,indicating that the calculation accuracy satisfies the application requirements,precision in comparison with the traditional fault tree method to get the results,the difference between the average absolute value,the maximum and mean square error has been worked out at 0.0093,0.0196 and 0.0052,respectively,which proves the calculated results turn out to be in nice consistency.Moreover,the risk assessment model built up in our paper enjoys obvious advantages in self-optimization,the purposeful application and dynamic adaptation.