基于MIV分析的GA-BP神经网络闸门健康诊断  

GA-BP Neural Network Gate Health Diagnosis Based On MIV Analysis

作　　者：罗光宇 杨光明[1] LUO Guang-yu;YANG Guang-ming(College of Energy and Electric Engineering,Hohai University,Nanjing 210098,China)

机构地区：[1]河海大学能源与电气学院,江苏南京210098

出　　处：《水电能源科学》2021年第11期203-206,194,共5页Water Resources and Power

摘　　要：由于闸门结构在水利枢纽运行中担任拦截水流、控制水位、调节流量、排放泥沙和飘浮物等的任务,闸门的健康诊断具有重要意义。针对闸门健康诊断模型的输入变量较多,造成网络检测的稳定性和精确度不高的问题,提出一种基于MIV(mean impact value)算法的GA-BP神经网络的闸门健康诊断方法。通过对138组闸门历史检测数据进行MIV分析,得到输入变量(诊断指标)与输出变量(闸门的健康等级)相关程度和平均贡献度,通过变量筛选原则筛选出对输出变量相关程度最大的输入变量,利用遗传算法对筛选完输入变量后的网络进行初始权值和阈值的优化,实现网络的全局优化。仿真结果表明,该方法的检测正确率从BP、GA-BP的65.79%、76.32%提高到了89.47%,能够显著提高闸门健康诊断的稳定性与准确率。Due to the importance of gate structure in water conservancy and hydropower engineering,the health diagnosis of gate is of great significance.Aiming at the problem that the stability and accuracy of the network detection is not high due to the large number of input variables in the gate health diagnosis model,a GA-BP neural network method for gate health diagnosis based on mean impact value(MIV)algorithm is proposed in this paper.Through MIV analysis of138 groups of gate historical detection data,the correlation degree and average contribution degree of input variable(diagnostic index)and output variable(health level of gate)are obtained.The input variable with the largest correlation degree to the output variable is selected by variable screening principle.The initial weight and threshold value of the network after filtering input variables are optimized by genetic algorithm,and global optimization of the network is realized.The simulation results show that the detection accuracy of the method is improved from 65.79%and 76.32%of BP and GABP to 89.47%,which can significantly improve the stability and accuracy of gate health diagnosis.

关 键 词：闸门健康诊断 遗传算法 BP神经网络 MIV算法 

分 类 号：TV34[水利工程—水工结构工程] TP18[自动化与计算机技术—控制理论与控制工程]