基于改进PSO优化PNN网络的变压器故障诊断方法

Fault Diagnosis for Transformer Based on PNN Optimized By Improved PSO Algorithm

根据变压器产生故障时特征气体和故障类型的非线性关系,结合油中溶解气体分析方法,采用了基于改进粒子群-概率神经网络(PNN)的故障诊断方法。针对PNN网络平滑因子按照经验选取的不足,以及使用粒子群优化(PSO)该参数时搜索精度低、容易早熟收敛等缺点,改进粒子群引入遗传算法的变异操作,并在迭代中对惯性权重动态调整和加速因子的线性变化,并用于训练PNN神经网络平滑因子集合;然后将改进PSO-PNN神经网络应用于变压器故障诊断中,通过诊断测试验证了该方法的有效性。

According to the non-linear characteristics relationship between fault characteristic gas and fault types of transformers,a probabilistic neural network（PNN） fault diagnosis method based on improved particle swarm algorithm is designed with the data of dissolved gas analysis.To overcome the disadvantages of experience selection of smoothing factors,the low precision search and premature convergence produced by optimizing smoothing factors with PSO（particle swarm optimization）,the improved particle swarm optimization algorithm is used to train the smoothing factors of PNN by introducing the mutated operation of genetic algorithm,and the dynamic adjustment of inertia weight and linear change of acceleration factor during iterative process.The improved PSO-PNN neural network is applied to transformer fault diagnosis.The diagnostic tests show the effectiveness of the method.