遗传神经网络的瞬变电磁视电阻率求解算法  

Genetic Neural Network for Apparent Resistivity Solution of Transient Electromagnetic

作　　者：秦善强[1] 付志红[1] 朱学贵[1] 籍勇亮 

机构地区：[1]输配电装备及系统安全与新技术国家重点实验室(重庆大学),重庆400044 [2]国网重庆电力公司电力科学研究院,重庆401123

出　　处：《电工技术学报》2017年第12期146-154,250,共10页Transactions of China Electrotechnical Society

基　　金：国家自然科学基金项目(51277189);输配电装备及系统安全与新技术国家重点实验室自主研究重点项目(2007DA10512714103);国家电网公司总部科技项目(基于瞬变电磁法的接地网状态检测及故障诊断新技术研究)资助

摘　　要：提出用遗传神经网络求解中心回线装置下瞬变电磁法(TEM)的视电阻率。根据中心回线方式的瞬变电磁响应关系式,设计出神经网络的输入输出关系和单输入单输出的三层网络结构。计算出神经网络的输入输出样本集,并通过尝试法确定隐含层的神经元个数。引入遗传算法优化神经网络结构的连接权值,得到了最优连接权值的GABP神经网络。用该神经网络对瞬变电磁响应的非线性方程进行拟合,得到以实测数据计算的核函数值所一一对应的瞬变场参数值,达到求视电阻率并成像的目的。通过对地下高阻块状异常体模型和电力系统钢制扁钢材料的接地网模型两个实例模型的仿真计算证明,得到视电阻率断面图,达到了求解反问题的效果。理论模型与实际数据计算表明,遗传优化的BP神经网络使得瞬变电磁视电阻率的计算时间大大缩短,是个实用的算法。该方法为瞬变电磁接地网故障实时诊断平台提供必要的技术基础。The genetic neural network is applied to compute the central-loop transient electromagnetic （TEM） apparent resistivity. According to the central loop TEM response, a relationship between inputs and outputs of an ANN as well as three-layer architecture with single input and single output for ANN is designed. The ANN sample sets are calculated, and the number of hidden layer neuron is determined through trial method. A genetic algorithm is introduced to optimize connection weights of the ANN and then a GABP neural network with the optimal connection weights. A nonlinear equation of the TEM response is fitted by GABP neural network, and transient parameter value is obtained. The transient parameter value and kernel value from measured data are corresponding one by one. Finally, the apparent resistivity is calculated. Two examples with GABP neural network are presented. The first one is a forward calculated model of high impedance abnormal body in uniform half space, the second one is the grounding grid with the structural shape ‘ ’ in a substation of power system. Pseudo-section of apparent resistivity is obtained, and good effects to solve inverse problem is achieved. The comparisons between the theoretical models and the measured data show that the GABP is a usefulness algorithm, which can reduce much computing time of TEM apparent resistivity. This method provides the base technology for a real-time fault diagnosis platform of grounding grids using TEM.

关 键 词：瞬变电磁场 人工神经网络 遗传算法 反向传播 视电阻率 

分 类 号：TM153.1[电气工程—电工理论与新技术]