GA优化LVQ网络的配电网接地故障选线方法

Ground Fault Line Selection Method in Distribution Network Using GA Optimal LVQ Network

针对配电网故障相电压过零点且高阻接地故障选线困难的问题,文中提出了应用遗传算法优化学习量量化神经网络的配电网单相接地故障选线方法。首先利用小波分析方法提取线路零序电流信号的模极大值,以此作为学习量量化神经网络的输入向量,采用局部搜索算子改进的遗传算法去优化神经网络的初始权值向量,解决了网络对初始权值的敏感性问题。加速网络的收敛过程,提高网络的聚类精度,实现对不同故障类型进行故障线路的快速、准确识别。仿真结果表明,该方法有效地减少了传统学习量量化神经网络选线的误判几率,提高了选线速度和精确度。

In order to solve the problem of crossing-zero faulty phase voltage and fault line selection difficulty in distribution network with high resistance ground,a single-phase ground fault line selection method based on learn vector quantization（LVQ）neural network optimized by genetic algorithm（GA）is proposed. Firstly,the modulus maxima of line zero sequence current signals is extracted by wavelet analysis method,which is considered as the input vector of LVQ. On the basis of the improved GA with local search operator to optimize the initial weight vector of the neural network,which solved the problem of the sensitive issues of initial weights,improves the convergence and the clustering precision,and confirms that fault line can be identified quickly and accurately for different fault types. The simulation results indicate that the proposed method can effectively decrease the probability of misjudge compared with the traditional LVQ neural network method,and the method increases the speed and accuracy of line selection.