一种GOA优化SOM神经网络的VP型倾斜仪故障智能诊断方法  

An Intelligent Fault Identification Method for VP Tiltmeter Using GOA-Optimized SOM Neural Network

作　　者：庞聪[1,2] 马武刚 李查玮[1,2,3] 龚燕民 刘晓磊 江勇[1,2] 廖成旺 PANG Cong;MA Wugang;LI Chawei;GONG Yanmin;LIU Xiaolei;JIANG Yong;LIAO Chengwang(Institute of Seismology,CEA,40 Hongshance Road,Wuhan 430071,China;Wuhan Gravitation and Solid Earth Tides,National Observation and Research Station,40 Hongshance Road,Wuhan 430071,China;Hebei Key Laboratory of Earthquake Dynamics,465 Xueyuan Street,Sanhe 065201,China;Yixian Seismic Station of Hebei Earthquake Agency,Yixian,Baoding 074200,China)

机构地区：[1]中国地震局地震研究所,武汉市430071 [2]武汉引力与固体潮国家野外科学观测研究站,武汉市430071 [3]河北省地震动力学重点实验室,河北省三河市065201 [4]河北省地震局易县地震台,河北省保定市074211

出　　处：《大地测量与地球动力学》2023年第3期322-326,共5页Journal of Geodesy and Geodynamics

基　　金：河北省地震动力学重点实验室开放基金(FZ202212);湖北省自然科学基金(2019CFB768);中国地震局地震研究所和应急管理部国家自然灾害防治研究院基本科研业务费(IS202236328,IS202226321);中国地震局地震科技星火计划(XH15030);中国大陆综合地球物理场仪器研发专项(Y201707);武汉引力与固体潮国家野外科学观测研究站开放基金(WHYWZ202208)。

摘　　要：提出一种VP型倾斜仪故障智能诊断方法。利用经验模态分解(EMD)将归一化故障信号分解为6个本征模态函数(IMF),分别计算其近似熵,构建EMD多尺度近似熵输入矩阵;结合蝗虫优化算法(GOA)对自组织特征映射(SOM)神经网络的参数进行优化,将得到的GOA最优值嵌入到SOM模型中,组建GOA-SOM诊断模型。应用诊断测试集得到诊断目标的聚类标签值,将其与训练集的聚类标签以及真实故障类型进行比对,得到故障诊断结果。结果证明,GOA-SOM模型在100次随机抽样条件下的诊断正确率均值和标准差分别为99.329 7%、1.218 8,优于传统诊断模型。We propose an intelligent fault identification method for the VP-type tiltmeter. Using empirical mode decomposition(EMD), we decompose the normalized fault signal into six intrinsic mode functions(IMF), and calculate the approximate entropy respectively to construct the EMD multiscale approximate entropy input matrix. Combined with the grasshopper optimization algorithm(GOA), we optimize the parameters of the self-organizing feature map(SOM) neural network. Then, we embed the obtained GOA optimal value, and form the GOA-SOM identification model. We use the identification test set to obtain the cluster label value of the target, and compare it with the cluster label of the training set and the real fault type, to obtain the fault identification result. The experiments show that the GOA-SOM model has a mean and standard deviation of the identification accuracy under 100 random sampling conditions of 99.329 7% and 1.218 8. These are better than traditional models.

关 键 词：倾斜仪故障诊断 经验模态分解 蝗虫优化算法 自组织特征映射神经网络 多尺度近似熵 

分 类 号：P315[天文地球—地震学]