基于GMM-PNN模型的海底油气管道风险等级评价  

On the risk assessment of the subsea level oil and gas pipeline based on the GMM-PNN model

作　　者：张新生[1] 杨青 ZHANG Xin-sheng;YANG Qing(School of Management,Xi’an University of Architecture and Technology,Xi'an 710000,China)

机构地区：[1]西安建筑科技大学管理学院,西安710000

出　　处：《安全与环境学报》2021年第3期935-942,共8页Journal of Safety and Environment

基　　金：国家自然科学基金项目(41877527)。

摘　　要：为了提高海底油气管道风险评价的准确率,保证管道安全运行,利用高斯混合模型(GMM)和概率神经网络(PNN)对管道进行风险评价。在传统PNN的基础上做出两点改进:一是改变PNN的结构模型,在原有网络结构上增加一个特征层,目的是通过线性变化的方式增强输入维度之间的联系;二是将全局单一参数改为在模式层采用GMM,并用随机梯度下降法对参数进行更新。考虑海底管道在偶发因素下的风险,将相关指标量化,利用GMM-PNN模型划分等级,然后与PNN模型、人工神经网络、支持向量机进行对比。结果表明,GMM-PNN模型对训练样本数量要求较低且准确率高于其他3种模型,能够更加准确地对海底管道进行风险等级评价。The paper intends to introduce the Gaussian mixture model and the probabilistic neural network to the evaluation practice of the risk gradation of the submarine oil and gas pipelines. It is just to improve the accuracy of the risk assessment of the submarine oil and gas pipelines that it would be necessary to guarantee the safety operation of the said pipeline. And,so,the paper would like to present 2 improvements based on the traditional probabilistic neural network. One of the improvements is to transform the constitutional structure of the traditional probabilistic neural network model by adding a feature layer on the original network structure,whose purpose is to add the feature layer to heightening of the connection between the input dimensions via the linear change. Whereas the other is to change the traditional global single variable into a Gaussian mixture model in the pattern layer so as for the each mode to be an independent parameter,it is necessary to update the parameters to the authentic useful ones by using the stochastic gradient descent method. And,so,it would be possible to propose a GMM-PNN model. On account of the risks of the submarine pipeline under the accidental factors,it is necessary to quantify and qualify the relevant indicators via the "disturbance","the protective measures" and"the safety attributes"as the criteria so as to raise and classify the levels through the GMM-PNN model. And,in the case for analysis practice,it would be necessary for us to reduce the probability of the over-fitting model as an effective method so as to determine the number of neurons in the feature layer and the parameter p of the number of similar neurons in the mode layer to heighten the model accuracy. And,so,through comparison with the PNN model,it would be possible for us to regulate or adjust the parameter m,so as to find that the GMM-PNN model for the artificial neural network to be able to support the vector machine,which can be said as the secret of the method. Thus,it can be seen that the results w

关 键 词：安全管理工程 风险评价 偶发因素 高斯混合模型 概率神经网络 

分 类 号：X937[环境科学与工程—安全科学]