Railcar iterative optimization positioning system, method and program product

Abstract

The application discloses a kind of track car iterative optimization positioning system, method and program product, including the training dataset of phase difference and distance is constructed, phase difference is input, distance is output, the training dataset of phase difference and distance is combined to construct Gaussian process prior equation, the nonlinear mapping relationship of phase difference and distance in Gaussian process prior equation is determined, the optimization of Gaussian process prior equation is realized, and the Gaussian process prior equation after optimization is formed, with the goal of maximizing log marginal likelihood to construct marginal likelihood maximization function, finite memory quasi-Newton optimization algorithm is used to solve marginal likelihood maximization function, and the optimal hyperparameter of Gaussian process prior equation after optimization is obtained, based on the nonlinear mapping relationship of phase difference and distance and optimal hyperparameter Gaussian process model is formed, and the distance between track car and destination is predicted using Gaussian process model and according to the real-time phase difference between track car and two preset positioning points.

Description

Technical Field

[0001] This invention relates to the field of precision positioning technology in industrial automation, specifically to an iterative optimization positioning system, method, and program product for a railcar. Background Technology

[0002] With the continuous improvement of industrial automation, the adoption of automated testing production lines has become a mainstream trend in the product manufacturing and quality inspection industries. In a typical automated testing unit, automated guided vehicles (AGVs) or rail-mounted automated guided vehicles (AGVs) are usually responsible for transporting products sequentially to the loading station, testing station, and sorting and warehousing station. Achieving rapid, accurate, and reliable docking between the product and the precision testing interface is the core link and technical challenge of the automated testing process. Even the slightest positioning deviation of the railcar can lead to a leaky test interface, causing media leakage and inaccurate test pressure, which not only affects the accuracy of the test results but may also pose serious safety hazards.

[0003] Currently, positioning is mainly achieved using technologies such as photoelectric encoders, lidar, and ultra-wideband (UWB). However, all three have certain drawbacks. Photoelectric encoders suffer from cumulative errors, resulting in decreased accuracy over long-term use and requiring recalibration after power failure. LiDAR is expensive and sensitive to water mist and dust in the testing environment, making it susceptible to interference. UWB systems are complex to deploy and prone to positioning blind spots in metallic environments. These shortcomings affect the repeatability, safety, and overall system efficiency of the test. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the aforementioned background technology and provide a low-cost, interference-resistant, high-precision, and adaptively optimized iterative optimization positioning system, method, and program product for rail vehicles.

[0005] To achieve this objective, the iterative optimization positioning system for railcars designed in this invention includes:

[0006] The training dataset construction module is used to construct a training dataset of phase difference and distance based on the phase difference between the track vehicle and two preset positioning points on the track and the distance between the track vehicle and the destination.

[0007] The mapping relationship construction module is used to construct a Gaussian process prior equation by taking the phase difference as input and the distance as output, and combining the training dataset of phase difference and distance. The nonlinear mapping relationship between phase difference and distance in the Gaussian process prior equation is determined by the Gaussian process smoothing kernel function, thereby optimizing the Gaussian process prior equation and forming the optimized Gaussian process prior equation.

[0008] The hyperparameter optimization module is used to construct a marginal likelihood maximization function with the objective of maximizing the log marginal likelihood. The marginal likelihood maximization function is solved using a finite-memory quasi-Newton optimization algorithm to obtain the optimal hyperparameters of the optimized Gaussian process prior equation.

[0009] The positioning distance calculation module is used to form a Gaussian process model based on the nonlinear mapping relationship between phase difference and distance and the optimal hyperparameters. Using the Gaussian process model and based on the real-time phase difference between the track vehicle and two preset positioning points, the distance between the track vehicle and the destination is predicted.

[0010] Furthermore, the expression for the prior equation of the Gaussian process is as follows: ), where D is the set of distances between the railcar and the destination; The phase difference between the track vehicle on the track and the two preset positioning points in the i-th acquisition; =0, which is the zero-mean function; This represents the nonlinear mapping relationship between phase difference and distance.

[0011] Furthermore, the method for determining the nonlinear mapping relationship between phase difference and distance in the prior equation of a Gaussian process using a Gaussian process smoothing kernel function includes: determining the nonlinear mapping relationship between phase difference and distance using the Matrn 5 / 2 kernel function, the expression for which is:

[0012] Where r is the Euclidean distance between two phase differences in the input space. ; For phase difference set Any phase difference in; The signal variance; This is a length-scale hyperparameter.

[0013] Furthermore, methods for constructing a marginal likelihood maximization function with the objective of maximizing the log marginal likelihood include: constructing the covariance matrix and the marginal likelihood maximization function;

[0014] The method for constructing the covariance matrix includes: calculating the covariance matrix. The i-th row and j-th column, ,in, , These are the phase differences between the i-th and j-th data points in the training dataset, respectively. , where is the Kronecker function; To observe the noise variance; traverse the set of phase differences. , build Symmetric positive definite covariance matrix K;

[0015] The method for constructing the marginal likelihood maximization function includes: based on the expression of the marginal likelihood maximization function... Construct a marginal likelihood maximization function, where T is the matrix transpose. Describe the determinant of matrix K. It is the natural logarithm.

[0016] Furthermore, the method for obtaining the optimal hyperparameters of the optimized Gaussian process prior equation by solving the marginal likelihood maximization function using a finite-memory quasi-Newton optimization algorithm includes: solving the problem using a finite-memory quasi-Newton method with boundary constraints. The optimal hyperparameters of the optimized Gaussian process prior equation are obtained. .

[0017] Furthermore, a Gaussian process model is formed based on the nonlinear mapping relationship between phase difference and distance and the optimal hyperparameters. The method for predicting the distance between the track vehicle and the destination using this Gaussian process model and based on the real-time phase difference between the track vehicle and two preset positioning points includes: calculating the mean distance between the track vehicle and the destination based on the Gaussian process posterior distribution and the collected phase difference; and using this mean distance as the predicted distance between the track vehicle and the destination. The expression is ,in,

[0018] , = , To predict the distance of the railcar from the destination, This represents the phase difference acquired.

[0019] Furthermore, the iterative optimization positioning system for the railcar also includes a positioning result judgment module. The positioning result judgment module is used to calculate the uncertainty of the prediction result, and judge the validity of the predicted distance between the railcar and the destination based on the uncertainty of the prediction result. Based on the validity, the Gaussian process model is iteratively optimized until the uncertainty of the prediction result meets the threshold requirement.

[0020] Furthermore, the method of calculating the uncertainty of the prediction result and judging the validity of the predicted distance between the track vehicle and the destination based on the uncertainty of the prediction result, and iteratively optimizing the Gaussian process model based on the validity until the uncertainty of the prediction result meets the threshold requirement, includes: calculating the variance of the distance between the track vehicle and the destination, the variance of the distance between the track vehicle and the destination... The expression is:

[0021] ,Will As the prediction uncertainty; a threshold for prediction uncertainty is set. ,like If the predicted distance between the railcar and the destination is not found, the prediction is considered invalid, and the result is marked as such. The number of consecutive incorrect predictions is then determined. Increment by 1, otherwise set the number of consecutive errors to 1. ;like Then the railcar continues to run, among which, If the error is a continuous error threshold, otherwise it is considered that the error between the predicted distance of the track vehicle to the destination and the actual distance of the track vehicle to the destination exceeds the threshold, an alarm is issued, and P sets of historical data closest to the current time are selected from the training dataset for hyperparameter optimization to obtain new optimal hyperparameters, which replace the original optimal hyperparameters in the Gaussian process model.

[0022] Furthermore, a method for iterative optimization positioning of a railcar based on the aforementioned iterative optimization positioning system includes:

[0023] A training dataset for phase difference and distance is constructed based on the phase difference between the track vehicle and two preset positioning points and the distance between the track vehicle and the destination.

[0024] Using the phase difference as input and the distance as output, a Gaussian process prior equation is constructed by combining the training dataset of phase difference and distance. The nonlinear mapping relationship between phase difference and distance in the Gaussian process prior equation is determined by the Gaussian process smoothing kernel function, thereby optimizing the Gaussian process prior equation and forming the optimized Gaussian process prior equation.

[0025] A marginal likelihood maximization function is constructed with the objective of maximizing the log marginal likelihood. The marginal likelihood maximization function is solved using a finite-memory quasi-Newton optimization algorithm to obtain the optimal hyperparameters of the optimized Gaussian process prior equation.

[0026] A Gaussian process model is formed based on the nonlinear mapping relationship between phase difference and distance and the optimal hyperparameters. The distance between the track vehicle and the destination is predicted using the Gaussian process model and the real-time phase difference between the track vehicle and two preset positioning points.

[0027] Furthermore, a computer program product includes a computer program / instructions that, when executed by a processor, implement the steps of the method.

[0028] The beneficial effects of this invention are as follows: First, this invention constructs a training dataset of phase difference and distance by using the phase difference between the track vehicle and two preset positioning points, and the distance between the track vehicle and the destination. Using phase difference as the positioning input feature, it eliminates the cumulative error defects of traditional photoelectric encoders, the environmental sensitivity defects of lidar, and the blind zone defects of ultra-wideband positioning in metallic environments. Even in complex working conditions such as water mist, dust, and dense metal equipment in industrial sites, it can still stably collect effective signals, ensuring the reliability and stability of positioning input from the data source. Simultaneously, the hardware only requires two near-field communication receivers in conjunction with the track positioning points to achieve data acquisition, eliminating the need for complex wiring and high-precision installation and debugging, significantly reducing system deployment costs and subsequent maintenance difficulty. Second, this invention constructs a Gaussian process prior equation with phase difference as input and distance as output, and uses a Gaussian process smoothing kernel function to determine the nonlinear mapping relationship between phase difference and distance. This can accurately fit the complex nonlinear correlation between phase difference and distance in industrial scenarios, breaking through the accuracy bottleneck of traditional linear positioning models and laying a solid model foundation for subsequent high-precision distance prediction. Then, this invention uses maximizing the logarithmic marginal likelihood as the optimization objective and employs a finite-memory quasi-Newton optimization algorithm to solve for the optimal hyperparameters of the Gaussian process prior equations. While ensuring the accuracy of hyperparameter optimization, it also considers algorithm efficiency and memory usage, adapting to the hardware computing power of industrial controllers. This allows for rapid model optimization, ensuring the Gaussian process model maintains optimal predictive performance and further improving the accuracy of distance calculations. Furthermore, based on nonlinear mapping relationships and optimal hyperparameters, this invention forms a complete Gaussian process model. During real-time positioning, it can not only output the precise distance between the track vehicle and the destination but also simultaneously calculate prediction uncertainties, enabling quantitative evaluation of the positioning results. This solves the problem of traditional positioning schemes that can only output positioning values ​​and cannot determine the reliability of the positioning, significantly improving the controllability and security of the positioning results. Finally, this invention continuously supplements the training dataset with valid positioning data to dynamically expand the sample size through positioning validity judgment and online iterative optimization mechanism. When continuous positioning is invalid, the latest historical data is automatically selected to re-optimize hyperparameters and update the model, so that the positioning system has online self-learning and self-optimization capabilities. During long-term operation, the positioning accuracy will not decrease with environmental changes and equipment wear and tear, but will continue to improve, with strong scene adaptability and robustness.

[0029] In summary, this invention deeply integrates near-field communication phase difference acquisition, Gaussian process nonlinear modeling, hyperparameter intelligent optimization, and online iterative updates to form a complete iterative optimization positioning scheme for track vehicles. This scheme ensures micron-level precise docking between the track vehicle and the test interface, effectively avoiding safety and quality issues such as poor sealing, media leakage, and inaccurate test pressure caused by positioning deviations. It comprehensively improves the positioning accuracy, operational stability, operational safety, and testing efficiency of automated testing production lines. Attached Figure Description

[0030] To more clearly illustrate the technical solutions of the embodiments disclosed in this invention, the accompanying drawings of the embodiments will be briefly described below. These drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention.

[0031] Figure 1 This is a module connection diagram of the iterative optimization positioning system for the railcar designed in this invention;

[0032] Figure 2 The flowchart shows the iterative optimization positioning method for the track vehicle designed in this invention. Detailed Implementation

[0033] The technical solutions (including preferred technical solutions) of the present invention will be further described in detail below with reference to the accompanying drawings and by way of listing some optional embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0034] Example 1

[0035] like Figure 1 As shown, this embodiment provides a track vehicle iterative optimization positioning system, which can be applied to the precise docking positioning scenario of track-guided vehicles in automated testing production lines for pressure relief valves. The system hardware layer includes dual near-field communication signal receivers, preset track positioning points, a central processing unit, a data storage device, an alarm module, and a track vehicle drive control unit. The software layer is functionally divided into a training dataset construction module, a mapping relationship construction module, a hyperparameter optimization module, a positioning distance calculation module, and a positioning result judgment module. Each module works together to complete the entire process of data acquisition, model construction, parameter optimization, real-time positioning, and iterative updates. The specific structure and working method are as follows:

[0036] The training dataset construction module is electrically connected to the dual near-field communication signal receivers and the data storage device. It is used to control the track car to run at a constant speed along the pressure relief valve interface track. The dual near-field communication signal receivers collect the phase difference data between the track car and two preset positioning points on the pressure relief valve interface track in real time. Simultaneously, the actual distance data between the track car and the test interface is acquired. Multiple sets of phase difference data and distance data are matched and bound one by one and stored in the data storage device to construct a training dataset covering the phase difference and distance of the entire track journey, providing standardized sample support for subsequent model construction.

[0037] The mapping relationship construction module communicates with the training dataset construction module. Using the phase difference in the training dataset as the input variable and the distance as the output variable, a Gaussian process prior equation is constructed. The Matern 5 / 2 kernel function in the Gaussian process smoothing kernel function is selected to determine the nonlinear mapping relationship between phase difference and distance. This accurately fits the nonlinear law of phase difference changing with distance in industrial field environment, making up for the shortcomings of linear models that cannot adapt to complex scenarios.

[0038] The hyperparameter optimization module is connected to the mapping relationship construction module. With the goal of maximizing the log marginal likelihood, it first constructs a covariance matrix with observation noise based on the kernel function and the training dataset, then generates the marginal likelihood maximization function, and uses a finite memory quasi-Newton optimization algorithm to solve the optimal hyperparameters of the Gaussian process prior equation. Under the premise of adapting to the computing power of industrial controllers, it can quickly and accurately optimize the model hyperparameters and improve the overall prediction accuracy of the model.

[0039] The positioning distance calculation module is connected to the hyperparameter optimization module and the dual near-field communication signal receivers. Based on the nonlinear mapping relationship between phase difference and distance and the optimal hyperparameters, a complete Gaussian process model is built. The module receives the on-site phase difference data collected by the dual near-field communication signal receivers in real time, inputs it into the Gaussian process model, and calculates the real-time mean distance of the track vehicle distance test interface through the Gaussian process posterior distribution. It also outputs the uncertainty variance of the distance prediction simultaneously, realizing the synchronous output of positioning results and reliability.

[0040] The positioning result judgment module is connected to the positioning distance calculation module, alarm module, and data storage. It presets an uncertainty threshold and a continuous invalid positioning threshold, compares the real-time calculated prediction uncertainty with the threshold, and determines the validity of the positioning result. When the positioning is valid, the current phase difference and distance data are added to the training dataset to expand the sample. When the positioning is invalid, abnormal data is marked and the number of consecutive invalidities is accumulated. If the number of consecutive invalidities reaches the threshold, the alarm module is immediately triggered to issue a prompt signal. At the same time, the latest sets of historical data in the data storage are retrieved to re-optimize the hyperparameters, replace the original hyperparameters of the model to complete the iterative update, and ensure that the model maintains the best positioning performance in the long term.

[0041] Example 2

[0042] like Figure 2As shown, this embodiment provides an iterative optimization positioning method for railcars, which implements the entire process based on the system described in Embodiment 1. Taking an automated testing production line for pressure relief valves as an example: Pressure relief valves are the last line of defense for ensuring the safe operation of critical pressure-bearing equipment such as pressure vessels, boilers, and energy transmission pipelines. Their reliability directly affects the safety of industrial systems and personnel. Therefore, each valve must undergo rigorous testing for sealing, opening and closing pressure, and operational characteristics before leaving the factory to ensure that the pressure relief valve can accurately and reliably open to release pressure when overpressure occurs and promptly close after the pressure returns to normal. Traditional manual testing methods suffer from drawbacks such as low efficiency, poor consistency, high labor intensity, and potential safety risks.

[0043] The iterative optimization positioning method for the track vehicle applied to the automated testing production line of pressure relief valves includes data acquisition, model building, hyperparameter optimization, real-time positioning, validity determination, and model iterative optimization. The specific steps are as follows:

[0044] Step 1: Constructing a phase difference and distance training dataset: Control the railcar to run along the pressure relief valve interface track. Collect the phase difference between the railcar and two preset positioning points on the track using onboard dual near-field communication receivers. Simultaneously collect the actual distance between the railcar and the interface test point. Pair and bind multiple sets of phase differences and distances to construct a phase difference and distance training dataset covering the entire track travel distance. Fix a near-field communication tag on the railcar. Two near-field communication receivers are arranged on the interface rail of the pressure relief valve. and The label plane remains parallel to the track plane, with a fixed relative distance between them; the track vehicle is controlled to move back and forth on the track from the far end to the destination. Temporarily deployed laser ranging equipment synchronously records the actual position of the track vehicle and the distance to the destination at each moment, forming a distance set D, i.e. Two receivers synchronously acquire tag signals and extract the phase difference, i.e. .in, and These are the phase data obtained from the two near-field communication receivers during the i-th communication. The phase difference set is obtained. Build a training dataset .

[0045] Step 2: Constructing the Gaussian process prior equation and nonlinear mapping relationship: Using the phase difference as the input variable and the distance between the track vehicle and the destination as the output variable, construct the Gaussian process prior equation, expressed as: , The distance data between the railcar and the destination corresponding to the phase difference collected in the i-th time; The phase difference data between the track dual positioning points and the track vehicle is collected in the i-th time. Identify the distribution of Gaussian processes; The mean function is used; in this embodiment, the zero mean function is adopted, i.e. =0; It is a nonlinear mapping kernel function between phase difference and distance. The nonlinear mapping relationship is determined using the Matrn 5 / 2 kernel function, expressed as follows:

[0046] r is the Euclidean distance between two phase differences in the input space. ; Phase difference data set Any set of phase difference data in the data; The variance of the signal represents the amplitude of the fluctuation in the phase difference signal. This is a length scale hyperparameter that controls the smoothness of the kernel function.

[0047] Step 3: Construct the covariance matrix and marginal likelihood maximization function: Based on the kernel function and the training dataset, construct a covariance matrix with observation noise. The element in the i-th row and j-th column of the matrix is... , The element in the i-th row and j-th column of the covariance matrix K; , These are the i-th and j-th phase difference data points in the training dataset, respectively. To observe the noise variance and characterize the noise interference during the data acquisition process; For the Kronecker function, Iterate through all phase difference data and construct... A symmetric positive definite covariance matrix K; constructing a marginal likelihood function with the objective of maximizing the logarithmic marginal likelihood: , Let log be the marginal likelihood function. hyperparameter set D represents the distance data between the railcar and the destination. Let D be the transpose of the distance dataset D; It is the inverse of the covariance matrix K; Let K be the determinant of the covariance matrix K; The operation is the natural logarithm; N is the total number of samples in the training dataset.

[0048] Step 4: Solve for the optimal hyperparameters of the Gaussian process prior equations: Use a finite-memory quasi-Newton optimization algorithm (based on the logarithmic marginal likelihood function). For the objective function, for the signal Length scale l and standard deviation of observation noise (Perform iterative optimization until the objective function converges, thus obtaining the optimal hyperparameters of the Gaussian process model.) , The optimal set of hyperparameters is defined as , which represents the final set of hyperparameters used in the model. For hyperparameters Let the function be a variable. The operator for finding the minimum value.

[0049] Step 5: Real-time positioning calculation and prediction uncertainty assessment: based on optimal hyperparameters A complete Gaussian process model is formed by combining the nonlinear mapping relationship with the real-time acquired phase difference data. The input model calculates the mean and variance of the predicted distance from the track vehicle to the destination using a Gaussian process posterior distribution. The formula for calculating the mean predicted distance (final positioning distance) is as follows: The formula for calculating prediction variance (location uncertainty) is as follows: . Real-time phase difference The predicted distance between the corresponding railcar and the destination; The variance of the predicted distance characterizes the positioning uncertainty; This is the kernel function vector representing the real-time phase difference and the phase difference of the training set. , This refers to the phase difference data between the two positioning points on the track and the track vehicle, which are collected in real time.

[0050] Step 6: Determining the effectiveness of localization and setting iterative optimization models: based on uncertainty thresholds With continuous invalid location threshold The comparison is performed, and iterative optimization is executed. If... If the location is deemed valid, the data will be compared. Add to the training dataset and update the sample library. That is ;like If the location is invalid, mark the abnormal data and set the number of consecutive invalid occurrences E = E + 1; if the number of consecutive invalid occurrences E < If E≥ If an alarm is triggered, the latest P groups of historical data in the training dataset are selected to re-execute hyperparameter optimization, and the original parameters of the model are replaced with the new optimal hyperparameters to complete the iterative update of the Gaussian process model. This is the square root of the predicted variance, i.e., the quantified value of the location uncertainty; To preset a positioning uncertainty threshold, The value ranges from 0.001 m to 0.01 m, with 0.005 m being preferred; E represents the number of consecutive invalid positioning attempts. To preset a threshold for continuous invalid location, The value of is in the range of 3 to 5, preferably 3; P is the number of historical data samples selected for iterative optimization, and the value of P is in the range of 20 to 50, preferably 30.

[0051] Example 3

[0052] This invention provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the iterative optimization positioning method for a track vehicle described in Embodiment 2.

[0053] In summary, this invention deeply integrates near-field communication phase difference acquisition, Gaussian process nonlinear modeling, hyperparameter intelligent optimization, and online iterative updates to form a complete iterative optimization positioning scheme for the railcar. This scheme ensures micron-level precise docking between the railcar and the pressure relief valve test interface, effectively avoiding safety and quality issues such as poor sealing, media leakage, and inaccurate test pressure caused by positioning deviations. It comprehensively improves the positioning accuracy, operational stability, operational safety, and testing efficiency of the automated pressure relief valve testing production line.

[0054] It should be noted that the above description of the technical solutions is exemplary, and this specification may be embodied in different forms and should not be construed as limiting it to the technical solutions set forth herein. Rather, providing these descriptions will ensure that the disclosure of this invention is thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art. Furthermore, the technical solutions of this invention are defined only by the scope of the claims.

[0055] Finally, it should be noted that the above embodiments are merely representative examples of the present invention. Obviously, the present invention is not limited to the above embodiments and many variations are possible. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention should be considered within the protection scope of the present invention.

Claims

1. A track vehicle iterative optimization positioning system, characterized in that: include: The training dataset construction module is used to construct a training dataset of phase difference and distance based on the phase difference between the track vehicle and two preset positioning points on the track and the distance between the track vehicle and the destination. The mapping relationship construction module is used to construct a Gaussian process prior equation by taking the phase difference as input and the distance as output, and combining the training dataset of phase difference and distance. The nonlinear mapping relationship between phase difference and distance in the Gaussian process prior equation is determined by the Gaussian process smoothing kernel function, thereby optimizing the Gaussian process prior equation and forming the optimized Gaussian process prior equation. The hyperparameter optimization module is used to construct a marginal likelihood maximization function with the objective of maximizing the log marginal likelihood. The marginal likelihood maximization function is solved using a finite-memory quasi-Newton optimization algorithm to obtain the optimal hyperparameters of the optimized Gaussian process prior equation. The positioning distance calculation module is used to form a Gaussian process model based on the nonlinear mapping relationship between phase difference and distance and the optimal hyperparameters. Using the Gaussian process model and based on the real-time phase difference between the track vehicle and two preset positioning points, the distance between the track vehicle and the destination is predicted.

2. The iterative optimization positioning system for railcars as described in claim 1, characterized in that: The expression for the prior equation of the Gaussian process is: ), where D is the set of distances between the railcar and the destination; The phase difference between the track vehicle on the track and the two preset positioning points in the i-th acquisition; =0, which is the zero-mean function; This represents the nonlinear mapping relationship between phase difference and distance.

3. The iterative optimization positioning system for railcars as described in claim 2, characterized in that: The method for determining the nonlinear mapping relationship between phase difference and distance in the prior equation of a Gaussian process using a Gaussian process smoothing kernel function includes: determining the nonlinear mapping relationship between phase difference and distance using the Matrn 5 / 2 kernel function, and the expression for the nonlinear mapping relationship between phase difference and distance is: Where r is the Euclidean distance between two phase differences in the input space. ; For the set of phase differences Any phase difference in; The variance of the signal; This is a length scale hyperparameter.

4. The iterative optimization positioning system for railcars as described in claim 3, characterized in that: Methods for constructing a marginal likelihood maximization function with the objective of maximizing the log marginal likelihood include: constructing the covariance matrix and the marginal likelihood maximization function; The method for constructing the covariance matrix includes: calculating the covariance matrix. The i-th row and j-th column, ,in, , These are the phase differences between the i-th and j-th data points in the training dataset, respectively. , where is the Kronecker function; To observe the noise variance; traverse the set of phase differences. , build Symmetric positive definite covariance matrix K; The method for constructing the marginal likelihood maximization function includes: based on the expression of the marginal likelihood maximization function... Construct a marginal likelihood maximization function, where T is the matrix transpose. Describe the determinant of matrix K. It is the natural logarithm.

5. The iterative optimization positioning system for railcars as described in claim 4, characterized in that: The method for solving the marginal likelihood maximization function using a finite-memory quasi-Newton optimization algorithm to obtain the optimal hyperparameters of the optimized Gaussian process prior equation includes: solving the problem using a finite-memory quasi-Newton method with boundary constraints. The optimal hyperparameters of the optimized Gaussian process prior equation are obtained. .

6. The iterative optimization positioning system for railcars as described in claim 5, characterized in that: A Gaussian process model is formed based on the nonlinear mapping relationship between phase difference and distance and optimal hyperparameters. The method for predicting the distance between the track vehicle and the destination using this model and the real-time phase difference between the track vehicle and two preset positioning points includes: calculating the mean distance between the track vehicle and the destination based on the Gaussian process posterior distribution and the collected phase difference; and using this mean distance as the predicted distance. The expression is ,in, , = , To predict the distance of the railcar from the destination, This represents the phase difference acquired.

7. The iterative optimization positioning system for railcars as described in claim 6, characterized in that: It also includes a positioning result judgment module, which is used to calculate the uncertainty of the prediction result, and judge the validity of the predicted distance between the track vehicle and the destination based on the uncertainty of the prediction result. Based on the validity, the Gaussian process model is iteratively optimized until the uncertainty of the prediction result meets the threshold requirement.

8. The iterative optimization positioning system for railcars as described in claim 7, characterized in that: The method for calculating the uncertainty of the prediction result, judging the validity of the predicted distance between the track vehicle and the destination based on the uncertainty of the prediction result, and iteratively optimizing the Gaussian process model based on the validity until the uncertainty of the prediction result meets the threshold requirement includes: calculating the variance of the distance between the track vehicle and the destination, and the variance of the distance between the track vehicle and the destination. The expression is: ,Will As the prediction uncertainty; a threshold for prediction uncertainty is set. ,like If the predicted distance between the railcar and the destination is not found, the prediction is considered invalid, and the result is marked as such. The number of consecutive incorrect predictions is then determined. Increment by 1, otherwise set the number of consecutive errors to 1. ;like Then the railcar continues to run, among which, If the error is a continuous error threshold, otherwise it is considered that the error between the predicted distance of the track vehicle to the destination and the actual distance of the track vehicle to the destination exceeds the threshold, an alarm is issued, and P sets of historical data closest to the current time are selected from the training dataset for hyperparameter optimization to obtain new optimal hyperparameters, which replace the original optimal hyperparameters in the Gaussian process model.

9. A method for iterative optimization positioning of a railcar based on the iterative optimization positioning system for railcars according to claim 1, characterized in that: include: A training dataset for phase difference and distance is constructed based on the phase difference between the track vehicle and two preset positioning points and the distance between the track vehicle and the destination. Using the phase difference as input and the distance as output, a Gaussian process prior equation is constructed by combining the training dataset of phase difference and distance. The nonlinear mapping relationship between phase difference and distance in the Gaussian process prior equation is determined by the Gaussian process smoothing kernel function, thereby optimizing the Gaussian process prior equation and forming the optimized Gaussian process prior equation. A marginal likelihood maximization function is constructed with the objective of maximizing the log marginal likelihood. The marginal likelihood maximization function is solved using a finite-memory quasi-Newton optimization algorithm to obtain the optimal hyperparameters of the optimized Gaussian process prior equation. A Gaussian process model is formed based on the nonlinear mapping relationship between phase difference and distance and the optimal hyperparameters. The distance between the track vehicle and the destination is predicted using the Gaussian process model and the real-time phase difference between the track vehicle and two preset positioning points.

10. A computer program product comprising a computer program / instructions, characterized in that: When the computer program / instructions are executed by the processor, they implement the steps of the method of claim 9.

Images