补偿模糊神经网络在埋地管道重构的应用

Application of Compensated Fuzzy Neural Network to Reconstruction of Buried Pipeline

使用ANSYS有限元分析软件建立埋地钢质管道模型,模拟计算管道在不同影响因素(电压、埋深、管径、土壤电阻率以及破损半径)下的电位峰值和电位均值,得到960组数据,以此构建了6输入(电压、埋深、管径、土壤电阻率、电位峰值以及电位均值)1输出(破损半径)的补偿模糊神经网络模型。对补偿模糊神经网络模型进行训练,探讨补偿模糊神经网络模型中各因素对训练误差收敛的影响。对两种数据标准化方法(min-max标准化方法和x/max标准化方法)进行对比,发现采用x/max标准化方法对收敛更有效。当其他因素一定时,初始输入隶属函数宽度对训练误差存在一定的影响,其值为0.1时更有利于收敛,而初始输出隶属函数宽度对收敛没有明显的影响。通过正交试验的方法确定了神经网络的最优因素组合,对其进行了性能验证,输出破损半径与实际破损半径之间的相对误差在1%以下,表明该补偿模糊神经网络输出的数据能够反映实际情况。采用补偿模糊神经网络模型,输入6因素计算输出破损半径。在ANSYS中输入管径、管长、破损位置(实际检测中可以探测得到)、破损半径等信息,输出埋地钢质管道三维重构图,为检修人员是否进行开挖修复提供更详细的管道信息。

Buried steel pipeline model was established by using ANSYS finite element analysis software,and the potential peak value and potential mean value of the pipeline under different influencing factors(voltage,buried depth,pipe diameter,soil resistivity and damage radius)?were simulated and calculated.960 sets of data were obtained to construct a compensated fuzzy neural network model with 6 inputs(voltage,buried depth,pipe diameter,soil resistivity,potential peak value and potential mean value)and 1 output(damage radius).The compensated fuzzy neural network model was trained,and the influence of various factors in the compensated fuzzy neural network model on the convergence of training error is discussed.By comparing the two data standardization methods(minmax standardized method and x/max standardized method),it is found that the x/max standardized method is more effective for convergence.When other factors are constant,the initial input membership function width has a certain influence on the training error.When the value is 0.1,the value of the initial input membership function is more favorable for convergence and the initial output membership function width has no obvious influence on the convergence.The optimal factor combination of neural network was determined by the orthogonal test method,and its performance was verified.The relative error between the output damage radius and the actual damage radius is below 1%,which indicates that the output data of the compensated fuzzy neural network can reflect the actual situation.Using the compensated fuzzy neural network model,6 factors are inputted to calculate the output damage radius.In ANSYS,the information of the pipe diameter,the pipe length,the damage position(which can be detected in the actual test),the damage radius and so on are inputted,and the three-dimensional reconstruction map of the buried steel pipeline is outputted to provide a more detailed pipeline information for the maintenance personnel to carry out excavation and repair.