基于Δu模式和RBF网络的局部放电模式识别

Partial Discharge Pattern Recognition Based on Voltage-difference Method and Radial Basis Function Neural Network

局部放电(PD)模式识别是诊断高压电气设备内绝缘缺陷的重要方法之一.采用了一种Δu模式参量作为局部放电的图谱特征,并采用不变矩作为放电特征;同时,采用了径向基函数神经网络(RBFNN)对局部放电Δu模式参量构成的图谱特征进行识别.结果表明采用正交最小二乘法(OLS)训练的RBFNN对Δu模式中的不变矩特征参量进行识别,RBFNN收敛速度快且稳定性强,识别率达到85.7%以上,能够很好地识别由5种人工绝缘缺陷模型产生的局部放电信号,在实际应用中具有良好的应用前景.

Partial discharge （PD） pattern recognition is widely regarded as a significant measure for diagnosis of dielectric defects in electrical equipment. Voltage-difference method, i. e. △u pattern, as a basis for PD pattern recognition is presented, and the invariant moments are studied as the features of PD in △u pattern. Furthermore, radial basis function neural network（RBFNN） is applied to PD pattern recognition according to the disadvantages of RBFNN. The results show that RBFNN has faster convergence speed and stronger stability, the performance of RBFNN is up to 85.7% with orthogonal least squares（ OLS）, and PD signals generated by five kinds of dielectric defects can be classified well with the method, so it is favorite to use in practice.