Wheel diameter dynamic correction method based on train multi-source fusion perception system

Abstract

The application discloses a wheel diameter dynamic correction method based on a train multi-source fusion perception system, and comprises the following steps: selecting a time T in a train cruising stage; measuring the fusion mileage L1 and the speed transmission L2 of the train by the multi-source perception equipment and the speed transmission equipment respectively within the time T; obtaining the correction diameter D of the train wheel based on the comparison error of the fusion mileage L1 and the speed transmission L2 g The step of measuring the fusion mileage L1 of the train by the multi-source perception equipment comprises the following steps: step one, acquiring observation data of various sensors of the train; step two, time registration is performed on the observation data of the various sensors; step three, wild value elimination is performed on the time-registered observation data; and step four, after the local filtering, the observation data are assigned different weights and then subjected to federal filtering, and the fusion result is output. The application greatly reduces the deployment of entity beacons, has no fixed installation position requirement limit, has high flexibility in calibration timing, can correct the wheel diameter parameter at any time, and has high calibration precision.

Description

Technical Field

[0001] This invention relates to the field of train control technology, specifically a method for dynamic wheel diameter correction based on a train multi-source fusion sensing system. Background Technology

[0002] As a key component of train operation, the wheelset experiences wear and tear during train operation due to friction between its tread and flange against the rails. This causes changes in the wheel diameter parameters. Consequently, speed and mileage calculations based on the previous wheel diameter will be inaccurate. Therefore, timely monitoring of changes in wheel diameter parameters is crucial for improving the accuracy of train speed measurement, distance measurement, and positioning.

[0003] Chinese patent discloses a train wheel diameter correction method and apparatus based on transponders (CN115056819A), relating to the field of rail transit technology. The method includes: upon receiving response information from the current wheel diameter correction transponder, searching for the adjacency relationship between the current wheel diameter correction transponder and the previous wheel diameter correction transponder on an electronic map based on the identifier in the response information of the current wheel diameter correction transponder and the identifier in the response information of the previous wheel diameter correction transponder; if an adjacency relationship is found, determining a train wheel diameter correction strategy based on the adjacency relationship; and correcting the train wheel diameter according to the wheel diameter correction strategy.

[0004] The aforementioned wheel diameter correction method selects the near-uniform speed coasting period during the ATO cruise phase of the train, comparing the wheel diameter parameters with the distances from two consecutive beacons or two beacons with a gap between them. To ensure positioning accuracy, this method requires the deployment of a large number of trackside beacons, resulting in high costs, significant construction and maintenance workload, and the inability to promptly correct wheel diameter parameters. Summary of the Invention

[0005] To address the shortcomings of the existing technology, this invention provides a method for dynamic wheel diameter correction based on a train multi-source fusion sensing system.

[0006] This invention adopts the following technical solution: a wheel diameter dynamic correction method based on a train multi-source fusion sensing system, comprising:

[0007] During the train's cruise phase, a time period T is selected;

[0008] Within time T, the fused mileage L1 and speed transmission L2 of the train are measured by multi-source sensing devices and speed transmission devices, respectively. The corrected diameter D of the train wheels is obtained based on the comparison error between the fused mileage L1 and speed transmission L2. g ;

[0009] Calculate the speed transmission information for each channel, calculate the number of speed transmission pulses P in time period T, and calculate the travel distance L2 in this period based on wheel diameter D and rotational speed R.

[0010]

[0011] After correcting the wheel diameter using the fusion mileage L1, the following is obtained:

[0012] Right now

[0013] Furthermore, the step of measuring the fused mileage L1 of the train using multi-source sensing devices includes,

[0014] Step 1: Acquire observation data from various sensors on the train;

[0015] Step two: Time registration of observation data from multiple sensors;

[0016] Step 3: Remove outliers from the time-registered observation data;

[0017] Step four: Perform local filtering on each observation data after outlier removal. After local filtering, assign different weights to each observation data and then perform federated filtering to output the fusion result.

[0018] The time registration includes

[0019] 1) Preprocess and analyze the observation data, and transmit the preprocessing and analysis results to the prior knowledge and registration requirements module;

[0020] 2) The prior knowledge and registration requirements module includes prior information for each sensor. The registration frequency and registration method are selected based on the prior knowledge and registration requirements module.

[0021] 3) Register the preprocessed observation data according to the selected registration frequency and registration method;

[0022] 4) Feed the registered observation data back to the prior knowledge and registration requirements module for parameter calibration;

[0023] 5) Output the registered observation data.

[0024] The preprocessing and analysis of the observation data includes obtaining information on the number of sensors, sensor sampling period, and sampling interval between sensors through preprocessing and analysis.

[0025] The selection of the registration frequency includes taking the average of all sensor sampling frequencies or taking the weighted average of all sensor sampling frequencies to calculate the sampling frequency.

[0026] The calculation formula is as follows when averaging all sensor sampling frequencies:

[0027]

[0028] When taking the weighted average of all sensor sampling frequencies, the calculation formula is as follows:

[0029]

[0030]

[0031] Among them, f i The sampling frequency of sensor i and the weight a of sensor i are... i Performance P of sensor i i The decision is made, where N is the total number of sensors.

[0032] The registration method employs the interactive multi-model Kalman filter time registration algorithm. The observation data of each sensor are used to obtain the registration value of each sensor at the registration point through the interactive multi-model Kalman filter time registration algorithm.

[0033] The interactive multi-model Kalman filter time registration algorithm consists of multiple motion models. By filtering multiple motion models and fusing these filtering results according to the model probabilities, the filtered output of the nearest sampling point to the registration time is obtained. Then, based on the state of the point and the time interval from the point to the registration time, the position of the time registration time is extrapolated to solve the problem when the time registration period and the sensor sampling period are not integer ratios.

[0034] The outlier removal process for each time-registered observation data includes...

[0035] The observation data from sensor i is used to detect outliers using an interactive multi-model Kalman filter time registration algorithm.

[0036] During one sampling period t in the fusion process, the valid bit value of sensor i is determined. i，t ;

[0037] If sensor i is currently in a valid state, i.e. i，t =True, then determine whether it is a continuous effective state or a fault recovery state; if it is a continuous effective state, that is, the number of consecutive fault cycles M = 0, sensor i is always in a normal state, and the observed data enters the subsequent fusion calculation process; if sensor i is in a fault recovery or start-up online state, that is, the number of consecutive fault cycles M > 0, it means that sensor i has recovered from the fault state to normal within this cycle, the data participates in the fusion with the observed data, and the number of consecutive faults M is cleared to zero.

[0038] If sensor i is currently in an invalid state, i.e., valid i，t=False, then determine the duration of the invalid state; if the duration of the invalid state is less than the set judgment threshold, that is, the number of consecutive fault cycles M < n, where n is the pre-set judgment threshold, wait for sensor i to recover, and use the interactive multi-model Kalman filter time registration algorithm to derive a one-step prediction value to participate in the fusion calculation, and maintain the fusion input of sensor i; if the duration of the invalid state is above the judgment threshold, that is, the number of consecutive fault cycles M ≥ n, give up waiting for sensor i to recover, and the observation data of sensor i will not participate in the fusion.

[0039] Step four includes multiple local filters and one main filter.

[0040] After the observation data from multiple sensors are used to perform state estimation for their respective local filters, the state estimates of each local filter are then combined. Together with covariance matrix L i The input information is passed to the main filter; the main filter fuses its own state estimate with the input information passed from each local filter, and finally outputs the state estimate of the main filter. Covariance Matrix L g The optimal solution;

[0041] The state and covariance matrix of each local filter are reset by feedback using the global fusion result of the master filter. The global optimal fusion result is then distributed to each local filter through the feedback reset process.

[0042] In open, enclosed, repositioning, and slippery track scenarios, different weighting factors β are assigned to each sensor. i , where 0≤β≤1.

[0043] The observation data is the raw data input from each sensor, including visual inertial navigation data, BeiDou RTK / quasar data, UWB radar data, and velocity sensor data.

[0044] The weighting factor β1 for visual inertial navigation sensors is 0 ≤ β1 ≤ 0.2; the weighting factor β2 for BeiDou RTK / quasar sensors is 0 ≤ β2 ≤ 0.5; the weighting factor β3 for UWB radar sensors is 0 ≤ β3 ≤ 0.3; and the weighting factor β4 for velocity sensors is 0 ≤ β4 ≤ 0.3.

[0045] The advantages of this invention are: it significantly reduces the deployment of physical beacons, lowers the workload of construction and maintenance, provides flexible and selective calibration timing, allows for on-the-spot correction of wheel diameter parameters, and offers high calibration accuracy. Increasing or decreasing the number of sensors does not affect the fusion effect, and it has the capability for real-time dynamic correction of wheel diameter. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of the merging of mileage L1 and speed transmission L2.

[0048] Figure 2 This is a flowchart of a train multi-source sensing positioning and speed measurement method.

[0049] Figure 3 This is a flowchart of the time registration process.

[0050] Figure 4 This is a flowchart of outlier removal.

[0051] Figure 5 This is a flowchart of the filtering and fusion process. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Combination Figure 1 As shown, a method for dynamic wheel diameter correction based on a train multi-source fusion sensing system is presented.

[0054] During the train's cruise phase, a time period T is selected;

[0055] Within time T, the fused mileage L1 and speed transmission L2 of the train are measured by multi-source sensing devices and speed transmission devices, respectively. The corrected diameter D of the train wheels is obtained based on the comparison error between the fused mileage L1 and speed transmission L2. R ;

[0056] Calculate the speed transmission information for each channel, calculate the number of speed transmission pulses P in time period T, and calculate the travel distance L2 in this period based on wheel diameter D and rotational speed R.

[0057]

[0058] After correcting the wheel diameter using the fusion mileage L1, the following is obtained:

[0059] Right now

[0060] In this embodiment, the method for measuring the fused mileage L1 of the train using multi-source sensing devices is as follows:

[0061] Combination Figure 2 As shown, step one involves acquiring observation data from various sensors on the train;

[0062] In this embodiment, the observation data is the raw data input from each sensor, including visual inertial navigation data, BeiDou RTK / quasar data, UWB radar data, and velocity sensor data.

[0063] Step two: Time registration of observation data from multiple sensors;

[0064] Combined Figure 3 As shown, time registration specifically includes:

[0065] 1) Preprocess and analyze the observation data to obtain information on the number of sensors, sensor sampling period, and sampling interval between sensors;

[0066] The preprocessing analysis results are transmitted to the prior knowledge and registration requirements module;

[0067] 2) The prior knowledge and registration requirements module includes prior information for each sensor. The registration frequency and registration method are selected based on the prior knowledge and registration requirements module.

[0068] The selection of the registration frequency in this embodiment includes taking the average of all sensor sampling frequencies or taking the weighted average of all sensor sampling frequencies to calculate the sampling frequency;

[0069] The calculation formula is as follows when averaging all sensor sampling frequencies:

[0070]

[0071] When taking the weighted average of all sensor sampling frequencies, the calculation formula is as follows:

[0072]

[0073]

[0074] Among them, f i The sampling frequency of sensor i and the weight a of sensor i are... i Performance P of sensor i i The decision is made where N is the total number of sensors, and P is the sensor performance. i The higher the value, the higher its weight a i The higher;

[0075] In this embodiment, the registration method is the Interactive Multi-Model Kalman Filter Time Registration Algorithm, abbreviated as IMM-KF time registration algorithm. The observation data of each sensor is used to obtain the registration value of each sensor at the registration point through the IMM-KF time registration algorithm.

[0076] The IMM-KF time registration algorithm consists of multiple motion models. By filtering multiple motion models and fusing these filtering results according to the model probabilities, the filtered output of the nearest sampling point to the registration time is obtained. Then, based on the state of the point and the time interval from the point to the registration time, the position of the time registration time when the time registration period and the sensor sampling period are not integer ratios is extrapolated.

[0077] 3) Register the preprocessed observation data according to the selected registration frequency and registration method;

[0078] 4) Feed the registered observation data back to the prior knowledge and registration requirements module for parameter calibration;

[0079] 5) Output the registered observation data.

[0080] Step 3: Remove outliers from the time-registered observation data;

[0081] Combined Figure 4 As shown, outlier removal specifically includes:

[0082] The observation data from sensor i are used for outlier detection using the IMM-KF time registration algorithm;

[0083] During one sampling period t in the fusion process, the valid bits of sensor i are determined. i，t ;

[0084] If sensor i is currently in a valid state, i.e. i，t =True, then determine whether it is a continuous effective state or a fault recovery state; if it is a continuous effective state, that is, the number of consecutive fault cycles M = 0, sensor i is always in a normal state, and the observed data enters the subsequent fusion calculation process; if sensor i is in a fault recovery or start-up online state, that is, the number of consecutive fault cycles M > 0, it means that sensor i has recovered from the fault state to normal within this cycle, the data participates in the fusion with the observed data, and the number of consecutive faults M is cleared to zero.

[0085] If sensor i is currently in an invalid state, i.e., valid i，tIf the value is False, then the duration of the invalid state is determined. If the duration of the invalid state is less than the set judgment threshold, i.e., the number of consecutive fault cycles M < n, where n is the pre-set judgment threshold, the system waits for sensor i to recover and uses the IMM-KF time registration algorithm to derive a one-step prediction value to participate in the fusion calculation, maintaining the fusion input of sensor i. If the duration of the invalid state is above the judgment threshold, i.e., the number of consecutive fault cycles M ≥ n, the system abandons waiting for sensor i to recover, and the observation data of sensor i does not participate in the fusion. This is because a one-step prediction using the IMM-KF time registration algorithm for a long time will lead to error accumulation, resulting in a large difference from the actual value and affecting the accuracy of other sensors.

[0086] Step 4: Perform local filtering on each observation data after outlier removal. After local filtering, assign different weights to each observation data and then perform federated Kalman filtering to output the fusion result.

[0087] Combination Figure 5 As shown, the specific fusion process in step four is as follows:

[0088] Use multiple local filters and one master filter.

[0089] After the observation data from multiple sensors are used to perform state estimation for their respective local filters, the state estimates of each local filter are then combined. Together with covariance matrix L i The input information is passed to the main filter; the main filter fuses its own state estimate with the input information passed from each local filter, and finally outputs the state estimate of the main filter. Covariance Matrix L g The optimal solution;

[0090] Figure 4 The middle dashed line signifies that the state and covariance matrix of each local filter are reset by using the global fusion result of the master filter. The global optimal fusion result is then distributed to each local filter through the feedback reset process, which plays an important role in improving the accuracy of the local filters.

[0091] In this embodiment, different weight allocation factors β are assigned to each sensor under different scenarios, such as open scene, closed scene, repositioning scene, and slippery track scene. i , where 0≤β i≤1; Weighting factor β1 for visual inertial navigation sensors, 0≤β1≤0.2; Weighting factor β2 for BeiDou RTK / quasar sensors, 0≤β2≤0.5; Weighting factor β3 for UWB radar sensors, 0≤β3≤0.3; Weighting factor β4 for velocity sensors, 0≤β4≤0.3. For example, in a specific open scene, β1=0.09, β2=0.48, β3=0.15, β4=0.28.

[0092] Combination Figure 2 As shown, this embodiment also includes a positioning beacon. The beacon's location information is obtained in advance through high-precision geographic mapping, generally with a position error of ±5cm. Compared to sensor positioning, the beacon's transponder position information is an absolutely reliable reference position. In this embodiment, the positioning beacon is only set up at the station and does not participate in the fusion process; it is used as a means to correct accumulated errors. When the train's onboard equipment reads the positioning beacon information, it can reset the beacon data based on the train's position.

[0093] The wheel diameter dynamic correction method of this invention can significantly reduce the deployment of physical beacons, lower the workload of construction and maintenance, and offers flexible and selective calibration timing, allowing for on-the-spot correction of wheel diameter parameters with high calibration accuracy. The multi-source sensing device of this invention performs fault-tolerant fusion of positioning and speed measurement information from visual inertial navigation, BeiDou RTK / quasar, UWB radar, and speed sensors, and uses positioning beacons to correct accumulated errors. By leveraging the strengths of each sensing device, highly reliable and accurate guidance information is obtained, leading to a more precise fused mileage L1 and the corrected diameter D of the train wheel. g .

[0094] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for dynamic wheel diameter correction based on a train multi-source fusion sensing system, characterized in that, include: During the train's cruise phase, a time period T is selected; Within time T, the fused mileage L1 and speed transmission distance L2 of the train are measured by multi-source sensing devices and speed transmission devices, respectively. Based on the comparison error between the fused mileage L1 and speed transmission distance L2, the corrected diameter D of the train wheels is obtained. g ; Calculate the number of speed transmission pulses P in time period T for each channel, and calculate the speed transmission distance L2 in this period based on the wheel diameter D and the number of revolutions R. After correcting the wheel diameter using the fusion mileage L1, the following is obtained: , Right now ; The steps for measuring the fused mileage L1 of a train using multi-source sensing devices include: Step 1: Acquire observation data from various sensors on the train; Step two: Time registration of observation data from multiple sensors; Step 3: Remove outliers from the time-registered observation data; Step four: Perform local filtering on each observation data after outlier removal. After local filtering, assign different weights to each observation data and then perform federated filtering to output the fusion result.

2. The wheel diameter dynamic correction method based on a train multi-source fusion sensing system according to claim 1, characterized in that: The time registration includes 1) Preprocess and analyze the observation data, and transmit the preprocessing and analysis results to the prior knowledge and registration requirements module; 2) The prior knowledge and registration requirements module includes prior information for each sensor. The registration frequency and registration method are selected based on the prior knowledge and registration requirements module. 3) Register the preprocessed observation data according to the selected registration frequency and registration method; 4) Feed the registered observation data back to the prior knowledge and registration requirements module for parameter calibration; 5) Output the registered observation data.

3. The wheel diameter dynamic correction method based on a train multi-source fusion sensing system according to claim 2, characterized in that: The preprocessing and analysis of the observation data includes obtaining information on the number of sensors, sensor sampling period, and sampling interval between sensors through preprocessing and analysis.

4. The wheel diameter dynamic correction method based on a train multi-source fusion sensing system according to claim 2, characterized in that: The selection of the registration frequency includes taking the average of all sensor sampling frequencies or taking the weighted average of all sensor sampling frequencies to calculate the sampling frequency. The calculation formula is as follows when averaging all sensor sampling frequencies: ， When taking the weighted average of all sensor sampling frequencies, the calculation formula is as follows: ， in, The sampling frequency of sensor i and the weight of sensor i are... Performance of sensor i The decision is made, where N is the total number of sensors.

5. The wheel diameter dynamic correction method based on a train multi-source fusion sensing system according to claim 2, characterized in that: The registration method employs the interactive multi-model Kalman filter time registration algorithm. The observation data of each sensor are used to obtain the registration value of each sensor at the registration point through the interactive multi-model Kalman filter time registration algorithm. The interactive multi-model Kalman filter time registration algorithm consists of multiple motion models. By filtering multiple motion models and fusing these filtering results according to the model probabilities, the filtered output of the nearest sampling point to the registration time is obtained. Then, based on the state of the point and the time interval from the point to the registration time, the position of the time registration time is extrapolated to solve the problem when the time registration period and the sensor sampling period are not integer ratios.

6. The wheel diameter dynamic correction method based on a train multi-source fusion sensing system according to claim 1, characterized in that: The outlier removal process for each time-registered observation data includes... The observation data from sensor i is used to detect outliers using an interactive multi-model Kalman filter time registration algorithm. During one sampling period t in the fusion process, the effective bit of sensor i is determined. ; If sensor i is currently in an active state, i.e. Then determine whether it is a continuously effective state or a fault recovery state; If it is a continuously effective state, that is, the number of consecutive failure cycles. If sensor i remains in normal condition, the observed data will be used in the subsequent fusion calculation process; if sensor i is in a fault recovery or restart state, i.e., the number of consecutive fault cycles... If this is true, it means that sensor i has recovered from a fault state to normal within this cycle, the data is fused with the observation data, and the number of consecutive faults M is cleared to zero. If sensor i is currently in an invalid state, i.e. Then, determine the duration of the invalid state; if the duration of the invalid state is less than the set judgment threshold, i.e., the number of consecutive fault cycles. Where n is a pre-set judgment threshold, after sensor i recovers to normal, a one-step prediction value is derived through the interactive multi-model Kalman filter time registration algorithm to participate in the fusion calculation, maintaining the fusion input of sensor i; if the duration of the invalid state is above the judgment threshold, i.e., the number of consecutive fault cycles. If the sensor i fails to recover, its observation data will not be used for fusion.

7. The wheel diameter dynamic correction method based on a train multi-source fusion sensing system according to claim 1, characterized in that: Step four includes multiple local filters and one main filter. After the observation data from multiple sensors are used to perform state estimation for their respective local filters, the state estimates of each local filter are then combined. Together with the covariance matrix The input information is passed to the main filter; the main filter fuses its own state estimate with the input information passed from each local filter, and finally outputs the state estimate of the main filter. Concordance Matrix The optimal solution; The state and covariance matrix of each local filter are reset by feedback using the global fusion result of the master filter. The global optimal fusion result is then distributed to each local filter through the feedback reset process.

8. The wheel diameter dynamic correction method based on a train multi-source fusion sensing system according to claim 7, characterized in that: Different weighting factors are assigned to each sensor in open, enclosed, repositioning, and slippery track scenarios. ,in .

9. The wheel diameter dynamic correction method based on a train multi-source fusion sensing system according to claim 8, characterized in that: The observation data is the raw data input from each sensor, including visual inertial navigation data, BeiDou RTK / quasar data, UWB radar data, and velocity sensor data. The weighting factor β1 for visual inertial navigation sensors is 0 ≤ β1 ≤ 0.2; the weighting factor β2 for BeiDou RTK / quasar sensors is 0 ≤ β2 ≤ 0.5; the weighting factor β3 for UWB radar sensors is 0 ≤ β3 ≤ 0.3; and the weighting factor β4 for velocity sensors is 0 ≤ β4 ≤ 0.3.

Images