一种新的遗传算法及其在变压器故障诊断中的应用

A novel genetic algorithm and its application to transformer fault diagnosis

提出了一种新的遗传算法,其基本思想是以网络权重和偏差的实数形式作为基因构成染色体向量,采用基因多点交叉和动态变异进行种群最优选择。研究结果表明,这种新的遗传算法是一种随机优化算法,克服了梯度下降法的不足,能够自动调节网络参数、网络的连接权重和偏差。在此基础上设计出一种基于遗传算法和溶解气体分析的变压器故障在线诊断系统。该系统只要将传感器测出的变压器中线圈电流、特征气体的含量作为输入参数,就能对信息进行融合分析,得到输入变量(线圈电流、溶解气体含量)与输出结果(故障类型、程度和部位)的复杂对应关系;能有效地减少输入层神经元的个数,改进网络内部结构,提高神经网络模型的学习效率和诊断的准确率,诊断精度高,漏报少,无误报现象。

A novel genetic algorithm is presented. The algorithm is intended to select the individuals with a smaller unfitness value in virtue of multi-point crossover and mutation in which gene vectors are composed of connection weights and bias terms of the neural networks. The results show that the system can ascertain complex corresponding relationship between input parameters (winding current, dissolved gas content) and outcome(fault type, severity, position) in virtue of windings current of power transformer, dissolved gas content from sensor. The random optimized algorithm overcomes the deficiency of grads descend algorithm and can automatically tune the network parameters, connection weights and bias terms of the neural networks, an online diagnosis system is set up based on the genetic algorithm and dissovled gas analysis (DGA). The algorithm can decrease the number of the network input nerve cells effectively and ameliorate network inner structural and improve the study efficiency and veracity. So the system brings about accurate diagnosis, less leaked diagnosis, and diagnosis without mistake.