基于ReliefF特征量优化及BP神经网络识别的高压隔离开关故障类型与位置诊断方法

Fault Type and Position Diagnosis Method of High-voltage Disconnectors Based on Relief F Characteristic Quantity Optimization and BP Neural Network Recognition

针对高压隔离开关机械故障的类型和发生位置难以有效识别的问题,提出了一种将隔离开关多路振动特征利用Relief F算法进行优化然后利用BP神经网络进行融合决策的故障诊断方法。首先进行试验模拟故障,在隔离开关本体和操动机构上分布式安装振动传感器,采集不同位置振动信号;然后将多个传感器采集的振动信号进行经验模态分解,得到固有模态函数,分别计算能量距,并把多路信号的能量距进行融合。最后使用Relief F算法对其进行特征筛选,构成BP神经网络的输入特征向量,从而实现故障类型、位置诊断。试验结果表明,多路传感器融合特征向量相比于单路信号提取的特征,对隔离开关不同的故障具有较好的识别能力,可诊断出故障发生的种类及位置,提高了诊断准确率。

Aiming at the problem of the type and the location of mechanical failure of high-voltage disconnectors aredifficult to be effectively identified,a fault diagnosis method based on Relief F algorithm is proposed to optimize themulti-channel vibration characteristics and utilize BP neural network fusion decision. First,simulating the failurethrough the tests and the install multi-point vibration sensors in the disconnectors body and the operating mechanismto collect the vibration signals of different locations. Then the vibration signals collected by a plurality of sensors areconducted with EMD to obtain the intrinsic modal function. And calculating the energy moment and integrating theenergy-moment of the multi-channel signals. Finally,the Relief F algorithm is used for extracting the principal com-ponents to construct the input feature vector of the BP neural network,so as to realize the fault type and the positiondiagnosis. The experimental results show that the multi-channel sensor feature fusion has a better adaptability andclassification ability for the different faults of disconnectors than the single-channel signal feature extraction has,which can diagnose the type and the location of the fault and improve the diagnostic accuracy.