Research On Train State Estimation Based On Multi-source Information Fusion

Railway is the important national transportation facility,which plays an important role in the political,economic,cultural and national defense construction and development.Therefore,it is of great significance to carry out research on accurate perception,fusion processing and scientific decision-making of internal and external environmental information,railway mobile equipment and fixed infrastructure,and to achieve a highly informationized,automated and intelligent train system,which will improve train operation safety,increase transport efficiency and reduce energy consumption.With the substantial increase of train speed and the development of sensing detection technology,the number of train ground detection equipment and on-board detection equipment is increasing,and the train state obtained is diverse,large in number and complex in relationship,which makes the state estimation based on a single sensor can no longer meet the increasing control and decision-making requirements of the train system.However,the estimation task,which is difficult to be completed by a single sensor or has high cost,can be solved through the multisensor information cooperation and fusion system composed of multiple autonomous sensors,the information complementation of different types sensors can be realized.Therefore,considering the inherent defects,uncertainties and limited sensing range of the sensor itself,as well as the complexity of the train’s actual operating environment and the variability of the operating conditions,train state estimation based on multi-source information fusion is studied in this paper,and the effective information in monitoring data is processed and mined to the maximum extent,so as to provide a solid foundation for ensuring the safe and efficient operation of the train.Specifically,the research work carried out in this paper mainly includes the following aspects.(1)Research on train state fusion estimation based on weight pre-allocationDuring the actual operation of the train,the sensor monitoring data will inevitably be affected by external noise interference and monitoring anomalies,and the accurate measurement results of the train state are difficult to obtain.At the same time,the hidden variable parameters that are difficult to measure in the train model make it difficult to accurately estimate the train state,which brings hidden dangers to the traffic safety.Therefore,based on the establishment of the braking model considering the external environment,in this paper,the distributed train state fusion estimation strategy based on weight pre-allocation is constructed,the maximum expectation online identification method is proposed,and the accurate estimation of system state and the real-time identification of hidden variable parameters are realized.Compared with traditional methods that only use single monitoring data,online state estimation and parameter identification based on multi-source fusion data can effectively reduce the impact of various adverse factors on the analysis results.(2)Research on multi model state estimation of train based on parallel fusion filteringConsidering the actual operation environment of the train and the disturbance of known or unknown external factors,various types of non-Gaussian noise will inevitably be mixed into the sensor monitoring data,and different types of sensors will also show different monitoring characteristics under the interference of different noise,which makes the state estimation results based on the monitoring data of a single sensor inaccurate.Therefore,in this paper,the train operation conditions are analyzed,and the train multi-modal model considering the actual operation environment is constructed.Combined with the characteristics of train data monitoring,a train state estimation method based on multi-sensor parallel fusion filtering is proposed,which realizes real-time identification of train operation conditions and real-time acquisition of train operation state.(3)Multi-rate asynchronous uniform sampling fusion estimation of train model with missing observationsIn the actual operation process of the system,it is difficult to keep the same sampling frequency of different sensors,even the same sensor may have different sampling rates,which makes it difficult to collect the system state at the same time during sensor data fusion.In addition,considering the data loss caused by network transmission interruption or communication failure,the situation of sensor monitoring data loss to be fused becomes more complex,ranging from intermittent loss of a small amount of monitoring data to continuous loss of a large number of monitoring data.Therefore,in this paper,a high-speed train state estimation method is proposed based on multi-rate asynchronous sensor fusion under the condition of missing observations,which realizes accurate fusion estimation of train multi-rate asynchronous sampling state under intermittent missing and continuity missing.(4)Multi-rate asynchronous non-uniform sampling fusion estimation of train model with observation driftWith the growth of the usage time of the sensor,affected by parts aging and faults,the performance degradation of the sensor will occur.Among them,the most common performance degradation is the drift of the measured value,and the drift degree will constantly change with the growth of time,which brings hidden dangers to the safe operation of the train.Therefore,it is very important to obtain accurate estimation of train state under the influence of drift.In this paper,the problem of optimal linear estimation for multi-rate sampling systems with multiple random drift measurements is studied,a non-augmented recursive linear state optimal filter is proposed,and the distributed fusion of train state based on covariance intersection fusion is realized.The problem of train state fusion estimation under the condition that sensor drift and its degree are difficult to measure is solved.