| As an outdoor power supply device erected along the railway,the overhead contact system(OCS)is the only non-backup subsystem in the entire high-speed railway traction power supply system.Its complex structure and harsh working environment are extremely prone to failures and are affected by uncontrollable external factors.At present,high-speed railway contact network maintenance is essentially a method of state detection and fault maintenance.It cannot predict in advance when a fault occurs,and little is known about the correlation between the fault and the occurrence of the fault.Therefore,on the premise of ensuring the safety of the OCS,data mining and fault prediction methods are used to realize the detection of the OCS fault symptoms,and then to realize the maintenance of the OCS status,and to achieve the purpose of scientific and reasonable health management.Maintenance and safe operation of electrified railways are of great significance.In this paper,based on the efficient use of sequence pattern mining and time series prediction models,the OCS faults are predicted and analyzed from the two dimensions of fault and fault,fault and time.This article mainly includes the following:1)Taking the fault data of the OCS as the research object,according to the high-speed rail OCS fault prediction and health management(PHM)platform construction requirements,referring to the "OCS inspection regulations" and "coding specifications",the catastrophe dictionary data contact dictionary Structural design and coding formulation,detailed recording of the location information,time information,climate information,collection method,fault cause and fault identification code where the fault occurred.After the text or picture type fault description is converted into coding,it is convenient for us to search the data,scroll up and down the fault according to geographic information or time information,etc.,for mining and regression analysis.2)In view of the low efficiency of the high utility sequence pattern mining algorithm,this article improves the HUSP-Miner algorithm and uses the utility list to replace the original database,which accelerates the sequence generation speed and the calculation speed of the upper limit of the utility,while avoiding many Scan the database twice.In terms of pruning strategy,in order to solve the problem that the efficient sequence mining algorithm no longer follows the downward closure like the frequent pattern algorithm,this paper compares the downward closure of several utility upper bounds based on the residual utility value,and proposes a PEU-based The new pruning strategy reduces the number of candidate sequences,narrows the search range of faults,and improves the efficiency of the algorithm.The mining results show that the OCS height difference fault is a very critical fault.It is susceptible to the influence of contact line height and positioning device,and it is also easy to cause the failure of the pull-out value exceeding the limit.It is easy to cause the height difference and the pullout value to exceed the limit,the failure of the support device and the occurrence of arcing.3)Aiming at the problem of predicting the fault intensity of OCS fault time series data,an ARIMAX prediction model combining the mining results of sequence patterns efficiently is proposed,and the idea of data mining is integrated into the OCS fault prediction model.First,introduced several common time series prediction models,introduced the structure and parameter selection of the model,analyzed its model characteristics and applicable data types,tested its rationality,and compared the prediction results of the ARIMAX model with the conventional model.The results show that the ARIMAX model combined with the highefficiency sequence pattern mining results has a better prediction effect. |
