摘要
针对BP神经网络在油田注水管网系统故障诊断中收敛速度慢、精度低、易陷入局部最优等问题,提出一种改进的BP神经网络诊断模型。该模型将自适应差分进化算法(SDE)用于神经网络的数据训练中,用以确定最优权重和阈值,并将仿真模拟与神经网络的训练相结合,计算管网故障工况,从而得出注水站、注水井的压力、流量异常和管段的故障情况。该方法能准确诊断管网系统各类节点和管线的故障位置,以及对站启停泵、井欠过注和管道漏损、堵塞等故障类型进行判断,通过实验对比分析验证了方法的高效性,对故障点位置一级诊断正确率为100%,故障类型二级诊断的正确率达98.07%。
Aiming at the BP neural network’s slow convergence speed,low accuracy and being easy to fall into local optimum in fault diagnosis of oilfield water injection pipe network system,an improved BP neural network diagnosis model was proposed,which has the adaptive differential evolution algorithm(SDE)applied to the data training of the neural network to determine the optimal weights and thresholds,and then has the simulation combined with the training of the neural network to calculate fault conditions of the pipeline network so as to obtain the pressure and flow abnormality of both water injection stations and wells and the fault conditions of the pipe section.The results show that,this method can accurately diagnose fault locations of various nodes and pipelines in the pipeline network system,as well as judge such types of faults as starting and stopping pumps,under-and over-injection,pipeline leakage and blockage.The experimental results show that this method is efficient,and the accuracy of the first-level diagnosis of the fault location reaches 100%and the second-level diagnosis of the fault type reaches 98.07%.
作者
王妍
卢淑文
李杰
许彦飞
高胜
闻镜强
高望
WANG Yan;LU Shu-wen;LI Jie;XU Yan-fei;GAO Sheng;WEN Jing-qiang;GAO Wang(School of Mechanical Science and Engineering,Northeast Petroleum University)
出处
《化工机械》
CAS
2023年第5期749-757,共9页
Chemical Engineering & Machinery
基金
国家重点研发计划(批准号:2018YFE0196000)资助的课题
2019年黑龙江省省属本科高校基本科研业务经费东北石油大学引导性创基金(批准号:2019YDL-15)资助的课题。
关键词
注水管网
BP神经网络
差分进化
故障诊断
water injection pipe network
BP neural network
differential evolution
fault diagnosis