一种模拟退火小波网络的电力变压器故障诊断

Simulated Annealing-Wavelet Network to Power Transformer Fault Diagnosis

关于电力变压器是电力系统的主要设备,变压器故障是危及电网安全的主要因素。在传统的故障诊断方法中,存在着易陷入局部最小和对初值依赖性较高的缺陷。根据油中溶解气体的故障诊断技术,为了准确及时检测电力变压器故障,提出对油中溶解的气体(H2,CH4,C2H2,C2H4,C2H6等)采用小波神经网络模型进行分析,对电力变压器进行故障诊断,同时引入了模拟退火算法对模型进行了结构和参数的优化,加快了训练收敛速度,避免了陷入局部极小值。进行仿真,结果表明了改进方法的有效性。

Power transformers are major power system equipments and their failures are the main factors for threatening the power system.The tranditional fault diagnosis methods are easy to fall into the local optima and the high dependence on the initial solution.This paper is based on diagnosis technology of dissolved gas analysis.Wavelet neural network（WNN） is used to analyze the dissolved gases,such as H2,CH4,C2H2,C2H4 and C2H6,and to diagnose faults of transformers.Simulated annealing algorithm（SA） is proposed to optimize the structure of the model and parameters.The proposed model can speed up the convergence rate and escape local minimum.A number of examples show that the proposed method is vilid.