基于EMWPE-Relief-LSTM的传感器故障诊断方法  

Sensor Fault Diagnosis Method Based on EMWPE-Relief-LSTM

作　　者：林涛 张达 王建君 LIN Tao;ZHANG Da;WANG Jian-jun(College of Artificial Intelligence and Data Science,Hebei University of Technology,Tianjin 300130,China)

机构地区：[1]河北工业大学人工智能与数据科学学院,天津300130

出　　处：《计算机仿真》2021年第9期414-419,共6页Computer Simulation

基　　金：河北省科技计划项目(17214304D)。

摘　　要：针对传感器故障信号非线性、非平稳性的特点,提出了基于经验多尺度加权排列熵与特征选择长短期记忆网络组合算法(EMWPE-Relief-LSTM)的传感器故障诊断方法。首先,采用集成经验模态分解(EEMD)将传感器输出的时间序列数据分解为本征模态分量(IMFs),利用峭度和方差这两个指标筛选出合适的本征模态分量;其次,计算本征模态分量的多尺度加权排列熵(EMWPE),将熵值作为特征向量;最后,通过Relief算法对故障特征向量降维,将降维后的故障特征向量输入长短期记忆网络(LSTM)分类器。实验结果表明,上述方法的传感器故障诊断准确率为99.3%,可实现传感器故障的精确诊断。Aiming at the non-linear and non-stationary characteristics of sensor fault signals, a sensor fault diagnosis method based on the EMWPE-Relief-LSTM algorithm was proposed.First, the Ensemble Empirical Mode Decomposition(EEMD) was used to decompose the time series data of the sensor into Intrinsic Mode Function(IMFs), and the two IMFs were screened using the kurtosis and variance indicators.Second, we calculated the multi-scale weighted permutation entropy(EMWPE) of the Intrinsic Mode Function and used the entropy value as the feature vector.Finally, the fault feature vector was reduced by the Relief algorithm, and the reduced feature vector was input into the Long Short-Term Memory(LSTM).The experimental results show that the accuracy of sensor fault diagnosis by this method is 99.3%, which can realize the accurate diagnosis of sensor faults.

关 键 词：传感器 故障诊断 集成经验模态分解 多尺度加权排列熵 长短期记忆网络 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置]