基于鲁棒输入训练神经网络的非线性多传感器故障诊断方法及其应用

针对非线性系统多传感器故障诊断时出现的检测准确性下降和数据重构产生的残差污染问题,提出了基于鲁棒输入训练神经网络非线性多传感器故障诊断模型.在目标函数中引入影响因子函数和可靠性系数,并通过计算机模拟和仿真确定最佳影响因子函数形式,抑制了多个含有显著误差故障数据的不良影响,并增加了具备高可靠性的重要数据影响权重,大大减小...

Nonlinear multi-sensor fault diagnosis method and its application based on robust input-training neural network

As fault detection accuracy reduces and residual contaminations appear in multi-sensor fault diagnosis,a nonlinear multi-sensor fault diagnosis model based on robust modified input-training neural network is proposed.The best influence factor function chosen from the results of computer simulation and reliability coefficients are introduced into the objective function for the purpose of inhibiting the influence of numerous failure data with significant errors and increasing the influence ratio of some high reliability data,which significantly reduces the residual contaminations and increases the accuracy and reliability of fault diagnosis.The case study was conducted to detect and validate some points from 1# high pressure heater in a 300MW power plant.The results verify the feasibility and accuracy of RITNN(robust input training neural network) in multi-sensor fault diagnosis.Compared to linear PCA(principal component analysis) and ITNN(input-training neural network),the RITNN method can be more effectively used in multi-sensor fault diagnosis and error data reconstruction.