Research And Implementation Of Key Technologies For Data Preprocessing Of EMU Based On Digital Twin Model

With the continuous development of high-speed railway in China,the attention to the maintenance efficiency and quality of the EMU is increasing.At present,there are many problems in the operation and maintenance of EMUs,such as difficult to manage the whole life cycle data,inaccurate fault prediction and analysis,and poor self-healing ability.This thesis proposes a solution of EMU digital twin operation and maintenance system based on big data platform for the above problems.However,the realization of digital twin model is based on the high real-time and high fidelity of data.Unfortunately,the data collected by EMU has a series of problems,such as asynchronous,high dimension,incomplete,inaccurate of the data,which results in the digital twin model is difficult to build.This thesis focuses on the research of Affinity Propagation(AP)and the information entropy algorithms to solve the problem of asynchronous and incomplete real-time monitoring data of EMU.Finally,the algorithm is applied to the data preprocessing of EMU digital twin operation and maintenance system and achieve good results.The main work of this thesis includes the following three points:(1)A solution of EMU digital twin operation and maintenance system based on big data platform is proposed.This thesis expounds the overall framework of EMU digital twin model,and then use the least square support vector machine(LS-SVM)to realize the self-adjustment of data processing mode in the digital twin model,the operation flow of digital twin model is established.(2)This thesis studies the generation and main classification of multi-source data of EMU,analyzes several problems that need to be solved urgently in real-time monitoring data used in digital twin model,and puts forward the sequential pattern mining algorithm of AP and information entropy for data filling and trend prediction for the problem of data asynchrony and incompleteness which are difficult to solve.Through the comparison experiments,the algorithm proposed in this thesis obtains the better effect than the other similar algorithm.(3)The AP clustering algorithm is paralleled based on Map Reduce,which realizes the online preprocessing function of the monitoring data in the digital twin model of EMU.The efficiency and acceleration ratio of the preprocessing algorithm under the condition of single machine and cluster are compared through experiments,which proves the efficiency of the algorithm after paralleled in this thesis for monitoring data processing of EMU.