基于改进微分进化优化神经网络的电力变压器故障诊断  

Fault diagnosis of power transformer based on improved differential evolution-neural network

作　　者：刘丽[1] 唐杰[1] 刘卓[2] 

机构地区：[1]邵阳学院电气工程系,湖南邵阳422000 [2]湖南大学电气与信息工程学院,湖南长沙410082

出　　处：《电力系统保护与控制》2010年第20期106-109,共4页Power System Protection and Control

摘　　要：提出了一种改进微分进化(IDE)与误差反向传播神经网络(BP)相结合的变压器故障诊断新方法。提出了不依赖优化问题的控制参数自适应调整策略,并动态监视种群适应度方差的变化,使IDE具有强劲的全局搜索能力,能很快地寻找到全局最优点。该算法能有效地克服常规BP神经网络权值和阈值收敛速度慢、易陷入局部极小值等缺点,实现了两种算法的取长补短。将该算法用于变压器故障诊断,并与基于其他算法的变压器故障诊断进行比较,仿真结果表明该算法具有收敛速度快、鲁棒性好、分类精度高的优点。The proposed model combining improved differential evolution（IDE） algorithm with BP algorithm is applied to fault diagnosis of power transformer.IDE algorithm adopts adaptive control parameters according to swarms＇ distribution condition and dynamic monitoring changes in population fitness variance.It has a strong global searching capability and can quickly find the global optimal point.The algorithm can effectively overcome defects of conventional BP algorithm,such as the slow convergence of weight and threshold and easy to fall local minimum value.And it uses the advantages of two kinds of algorithms.It is applied in power transformer fault diagnosis.Compared with other algorithms,simulation results show that the proposed method possesses the advantages of fast convergence performance,good robustness and high classification accuracy.

关 键 词：微分进化 神经网络 电力变压器 自适应 混沌搜索 

分 类 号：TM41[电气工程—电器] TP183[自动化与计算机技术—控制理论与控制工程]