Research On Data Preprocessing And Fault Diagnosis Methods For A Class Of Cyber Physics Systems

With the development of Internet technology,many key infrastructures in industrial production and social life have gradually been integrated into the Internet,forming Cyber Physical Systems(CPS).The result of the networked development of key infrastructure is to greatly improve its operational efficiency and automation.Meanwhile,the acquisition of physical device operation information is extremely convenient.However,system data has several drawbacks,such as large data volume,high dimensionality,various types,low value density and so on,which cause new challenges to data-based fault diagnosis.Taking the issues mentioned above,for the data preprocessing and intelligent fault diagnosis methods,this paper carries out new research.The main innovations and research contents are as follows:(1)K-Nearest Neighbor Fault Diagnosis Method Based on Element Importance Data Transform.Firstly,a network model for measuring the importance of variables is constructed by Means Impact Value,then based on the network model,the weight coefficient of element importance degree can be solved;Secondly,based on the importance evaluation model,the evaluation mechanism is further optimized and perfected so that the importance degree of the element obtained by the model is optimal.After that,the k-nearest neighbor method is combined for fault diagnosis processing.Finally,the simulation experiments on the UCI standard data set verify that the proposed data preprocessing method performs well in improving fault diagnosis performance.(2)Data preprocessing method and fault diagnosis based on information contribution evaluation function.No matter which way selected to solve the importance weight of an element,an accurate model is needed.First of all,a new method for updating the parameters of evaluation function is presented.The new method updates partial parameter of the model in real time without changing the other parameters,thus ensuring the accuracy of the model.Then,based on the above evaluation function,the feature weight can be obtained,and the weighted data is brought into the established model for fault diagnosis.Finally,the effectiveness of the algorithm is verified by a standard UCI data set.(3)Multi-degree-of-freedom broad neural network modeling analysis based on output error step-by-step compensation.Aiming at the problems of low fitting accuracy,poor classification effect,many parameters to be determined and time-consuming training of shallow neural network,combined with the existing research content of the broad neural network,a three-degree-of-freedom neural network structure based on error compensation is proposed in this chapter.The output of prediction error is compensated step by step to compensate the network output.The simulation results of the standard UCI dataset show that the proposed algorithm not only optimizes the complexity of the deep neural network algorithm,but also performs well in the classification performance of the network.