基于EEMD和卷积神经网络的高压断路器故障诊断

Research on Circuit Breaker Fault Diagnosis Based on EEMD and Convolutional Neural Network

高压断路器分合闸线圈的电流信号蕴含着丰富的断路器操动机构状态信息,对操动机构故障诊断具有重大意义。首先,文中通过集合经验模态分解(ensemble empirical mode decomposition,EEMD)具备的检测突变点性能确定有效分合闸线圈电流信号段,并对其进行EEMD自适应降噪处理。其次,运用时域求极值法对有效信号段进行信号处理,提取电流、时间复合特征量。最后,通过对复合特征量数据进行Kronecker张量积预处理,以便输入到卷积神经网络(convolutional neural network,CNN)中进行有监督地故障状态的辨识诊断。实验结果表明,文中所提分合闸线圈电流信号的电流、时间复合特征量提取方法有效、CNN卷积神经网络算法相比GA-BP和支持向量机(support vector machine,SVM)算法诊断精度更高,具有较高的实际应用价值。

The current signal of the opening and closing coil of the high-voltage circuit breaker contains a wealth of state information of the operating mechanism of the circuit breaker,which is of great significance for the fault diagnosis of the operating mechanism.Firstly,this paper determines the effective current signal segment by the detection of the breakpoint performance of EEMD,and performs EEMD adaptive noise reduction processing on it.Secondly,the time domain optimization method is used to process the effective signal segment,and the current and time composite feature quantities are extracted.Finally,the Kronecker tensor product preprocessing is performed on the composite feature quantity data to input into the CNN for the identification diagnosis of the supervised fault state.The experimental results show that the current and time composite feature extraction method of the closing coil current signal is effective,and the CNN convolutional neural network algorithm has higher diagnostic accuracy than GA-BP and SVM algorithms,and has higher practical application value.