A vehicle state estimation method, system, and device based on hybrid kernel functions

By employing a dual Cauchy hybrid kernel function and an adaptive kernel width in Kalman filtering, the performance degradation problem of traditional Kalman filtering in non-Gaussian noise environments is solved, achieving higher filtering accuracy and robustness.

CN118094099BActive Publication Date: 2026-11-10NANJING UNIV OF INFORMATION SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410193062.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-02-21
Publication Date
2026-11-10
Estimated Expiration
2044-02-21

AI Technical Summary

Technical Problem

Traditional Kalman filters suffer from performance degradation when faced with non-Gaussian noise, and are particularly sensitive to heavy-tailed impulse noise. Furthermore, inappropriate selection of the fixed kernel width can lead to a decline in filtering performance.

Method used

An adaptive kernel maximum correlation entropy Kalman filter method based on a hybrid kernel function is adopted. The dual Cauchy hybrid kernel function is used as the cost function, and the kernel width is adaptively updated by Cauchy-Schwarz divergence to adapt to non-Gaussian noise environment.

Benefits of technology

It improves the filtering accuracy and robustness of vehicle state estimation, adapts to various non-Gaussian noise environments, and achieves better filtering performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118094099B_ABST
    Figure CN118094099B_ABST
Patent Text Reader

Abstract

This invention discloses a vehicle state estimation method, system, and device based on a hybrid kernel function, belonging to the field of data processing. It includes: constructing a linear system model of the vehicle state using state equations and observation equations; characterizing process noise and measurement noise in the system using a Gaussian mixture non-Gaussian noise statistical model; and employing the maximum correlation entropy criterion instead of the traditional MSE criterion for KF filtering based on the linear system model of the vehicle state. The kernel function in the maximum correlation entropy criterion is not a traditional Gaussian kernel function, but rather a convex combination of two Cauchy kernel functions with certain coefficients. The kernel width in the Cauchy kernel function is not a traditional fixed size that is unsuitable for unstable or unknown noise, or time-varying noise; instead, it is an adaptive kernel width criterion obtained based on the system model and the Cauchy-Schwarz divergence of the noise distribution. Based on the above kernel function and adaptive kernel width criterion, a kernel adaptive maximum correlation entropy KF filtering algorithm using a dual Cauchy hybrid kernel function is proposed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of data processing, and specifically to a vehicle state estimation method, system, and device based on hybrid kernel functions. Background Technology

[0002] Traditional Kalman filtering uses the minimum mean square error (MMSE) as a metric based on the second-order error moments of the mean and variance. However, when the noise signal is non-Gaussian, especially when the system is affected by heavy-tailed impulse noise, the performance of Kalman filtering (KF) can be severely degraded because the MMSE criterion is particularly sensitive to large outliers. In recent years, in the fields of signal processing and machine learning, for systems affected by non-Gaussian noise (especially heavy-tailed impulse noise), the use of the MCC criterion based on higher-order error moments instead of the aforementioned MMSE criterion based on second-order error moments has effectively improved the robustness of the algorithm and has been widely applied. Correlation entropy is a measure of local similarity and is insensitive to outliers.

[0003] On the one hand, Gaussian kernel functions are typically chosen as the cost function in MCC-based filtering algorithms. However, this is not always the optimal choice because Gaussian kernel functions are highly sensitive to kernel width, and currently there is no optimal kernel width selection standard. An inappropriate kernel width can lead to a decrease in expected filtering performance. On the other hand, in MCC-based filtering algorithms, kernel width, as a scaling parameter, controls the convergence speed and robustness of correlation entropy filtering against impulse disturbances or large outliers, as well as the existence of an optimal value. If the kernel width is not chosen appropriately, the superiority of correlation entropy index filtering will be lost. To address the problem that a fixed kernel width determined based on a specific noise level may result in poor estimation performance under real non-Gaussian noise conditions, and to solve the problem that using a fixed kernel width often fails to achieve the optimal result, this invention proposes a vehicle state estimation method, system, and device based on hybrid kernel functions. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a vehicle state estimation method, system, and device based on hybrid kernel functions. For cases where process noise and / or measurement noise in actual systems are non-Gaussian, this invention significantly improves filtering accuracy and robustness, enhances state estimation performance, and greatly expands the application scope of maximum correlation entropy Kalman filtering.

[0005] The objective of this invention can be achieved through the following technical solutions: Firstly, this application discloses a vehicle state estimation method based on a hybrid kernel function, comprising: For linear systems based on vehicle state, an adaptive kernel maximum correlation entropy Kalman filter method based on hybrid kernel functions is used to estimate the vehicle state, thereby obtaining the state estimate of the vehicle state. The adaptive kernel maximum correlation entropy Kalman filter method based on hybrid kernel functions includes: The cost function is a convex combination of the double Cauchy mixture kernel function targeting the residual term of the observation equation and the prediction error term of the state equation. The cost function is maximized to obtain the vehicle state estimate and the state estimate error covariance at the current time. The kernel widths of the kernel function for the residual term of the observation equation and the kernel function for the prediction error term of the state equation are adaptively updated based on the residual term of the observation equation and the Cauchy-Schwarz divergence.

[0006] In some embodiments, using a convex combination of two Cauchy mixture kernel functions as the cost function for the residual term of the observation equation and the prediction error term of the state equation includes the following steps: The residual term of the observation equation is expressed as: ; The prediction error term of the state equation is expressed as: ; in express The output of the observation at any moment, express The observation matrix at time, express Vehicle status at any given time; express The state transition matrix at time t; express Vehicle state estimation at any given time; The cost function is expressed as: (7) in express Vehicle state estimation at any given time. express Vehicle state estimation at any given time; express The nominal covariance of the state estimation error at time t. The nominal covariance represents the prediction error; This represents the nominal covariance of the measurement noise; the superscript "-1" indicates the inverse. , These are the Cauchy kernel function terms of the observation equation; , It is the Cauchy kernel function term of the prediction error of the state equation; The kernel width is The Cauchy kernel function, The kernel width is The Cauchy kernel function, This is called the heavy-tailed adjustment factor; e represents the error; for cases with covariance vector , , Represents the 2-norm; This represents the mixing coefficient.

[0007] In some embodiments, according to Vehicle state estimation at time -1 and The state estimation error covariance at time -1 is obtained by performing a one-step prediction according to the following prediction equation. Predicted vehicle state estimates and prediction error covariance at time: (5) (6) In the formula, express Predicting vehicle status at any given moment. express Vehicle state estimation at any given time. This represents the prediction error covariance; express The state estimation error covariance at time t. express The state transition matrix at time t; express The nominal covariance matrix of the noise at each time step; superscript T This indicates transpose.

[0008] In some embodiments, the kernel widths of the kernel function for the residual term of the observation equation and the kernel function for the prediction error term of the state equation are adaptively updated based on the residual term of the observation equation and the Cauchy-Schwarz divergence, as follows: (8) In the formula, express Vehicle state estimation at any given time; It is the Cauchy-Schwarz divergence that characterizes the two probability distributions in the observation noise; The closed-form solution of the Cauchy-Schwarz divergence is: (9) like , These are positive definite square elements. ,and Given a diagonal matrix, the closed-form solution of the Cauchy-Schwarz divergence can be expressed as: (10).

[0009] In some embodiments, maximizing the cost function to obtain the vehicle state estimate and the state estimate error covariance at the current time includes the following steps: Regarding the cost function Vehicle state estimation at time 1 Find the derivative and set it to 0. This can be expressed as: (11) Approximation in kernel function ,get: (12) (13) (14) In the formula, express 3D identity matrix Indicates the filter gain. express The state estimation error covariance at time t. Indicates intermediate variables. .

[0010] In some embodiments, the linear system state equation and observation equation of the vehicle state are constructed; the vehicle state includes vehicle position and vehicle speed; the noise in the system is described using a non-Gaussian noise statistical model.

[0011] In some embodiments, the construction of the linear system includes the following steps: Let the previous moment be... -1 time, current time is time; The state equation is expressed as: (1) In the formula, express Vehicle status at any given time. express The state transition matrix at time t, express Vehicle status at time -1 express The process noise at time t; the process noise follows a pattern with a mean of 0 and a covariance matrix of... is a non-Gaussian distribution, where , Indicates the expectation operation, superscript T Indicates transpose; The observation equation is expressed as: (2) In the formula, express The output of the observation at any moment, express The observation matrix at time, express Measurement noise at time; the measurement noise follows a pattern with a mean of 0 and a covariance matrix of... is a non-Gaussian distribution, where .

[0012] In some embodiments, the noise in the system is described using a non-Gaussian noise statistical model, including the following steps: The hybrid distribution consisting of L components can be represented as (3) For the weighting coefficients, satisfying ; ; The mean is covariance is The normal distribution; The heavy-tailed noise is characterized by a mixture model of two Gaussian density functions, expressed as: (4).

[0013] Secondly, this application discloses a vehicle state estimation system based on hybrid kernel functions, including; Linear module: A linear system that constructs vehicle states based on state equations and observation equations; the vehicle states include vehicle position and vehicle speed; Noise module: Based on a non-Gaussian noise statistical model, it describes the process noise in the state equation and the measurement noise in the observation equation of a linear system; Estimation Module: For linear systems based on vehicle state, an adaptive kernel maximum correlation entropy Kalman filter method based on a hybrid kernel function is used to estimate the vehicle state, thereby obtaining the estimated vehicle state value. The adaptive kernel maximum correlation entropy Kalman filter method based on a hybrid kernel function includes: using a convex combination of two Cauchy hybrid kernel functions for the residual term of the observation equation and the prediction error term of the state equation as a cost function; maximizing the cost function to obtain the vehicle state estimate at the current time and the state estimate error covariance at the current time. Kernel width self-updating module: Adaptively updates the kernel widths of the kernel function for the residual term of the observation equation and the kernel function for the prediction error term of the state equation based on the residual term of the observation equation and the Cauchy-Schwarz divergence.

[0014] Thirdly, this application discloses a terminal device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the memory stores the computer program capable of running on the processor, and when the processor loads and executes the computer program, it employs a vehicle state estimation method based on a hybrid kernel function as described in the first aspect.

[0015] The beneficial effects of this invention are: To address the issue that process and / or measurement noise is mostly heavy-tailed non-Gaussian noise, the method of this invention uses Kalman filtering based on the correlation entropy index of the higher-order moments of the error to reduce the interference of non-Gaussian noise. This method can better extract information from the error vector, improve the accuracy of state estimation and system robustness, and achieve better filtering accuracy. Since heavy-tailed non-Gaussian noise distributions are mostly characterized by homogeneous Gaussian mixtures or heterogeneous mixtures of Gaussian and other distributions, the method of this invention uses the Cauchy kernel function, which has heavy-tailed characteristics and is not sensitive to the kernel width of the Gaussian kernel function, to form a hybrid kernel function by convex combination of certain coefficients in the kernel function of the correlation entropy index, thereby achieving better filtering performance. To address the real-world scenario where system noise is unknown and unstable, this invention employs adaptive kernel width for correlation entropy filtering. The kernel width is adaptively updated by comprehensively considering the accessible parameters of the system model, variance propagation, and the Cauchy-Schwarz divergence of the noise distribution. Compared to a fixed kernel width, this approach better reflects the characteristics of noise uncertainty in the state equations, achieving superior filtering performance. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings.

[0017] Figure 1 This is a general framework diagram of this application; Figure 2 This is a schematic diagram of shot noise in the process noise of this application; Figure 3 For the state variables of this application A schematic diagram of the root mean square error under shot noise; Figure 4 For the state variables of this application A schematic diagram of the root mean square error under shot noise; Figure 5 This is a schematic diagram of the pulse heavy-tailed non-Gaussian noise of this application; Figure 6 For the state variables of this application A schematic diagram of the root mean square error under pulse-tailed non-Gaussian noise. Figure 7 For the state variables of this application A schematic diagram of the root mean square error under pulse heavy-tailed non-Gaussian noise. Detailed Implementation

[0018] 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.

[0019] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0020] Traditional Kalman filtering based on the mean squared error (MSE) index achieves optimal estimation under the assumption of Gaussian noise. However, when the noise in the actual state equation is non-Gaussian, the filtering performance of traditional Kalman filters deteriorates. This embodiment addresses the reality that process noise and / or measurement noise are non-Gaussian. Based on the statistical characteristics of non-Gaussian noise, the cost function index is replaced with a correlation entropy index based on the higher-order moments of the error signal. Furthermore, the single-kernel Gaussian function in the index is replaced with a double-Cauchy hybrid kernel function, which is inherently heavy-tailed and less sensitive to kernel width compared to the Gaussian kernel function. This hybrid kernel function extracts more information from the non-Gaussian error signal. The kernel width is the only free parameter in the Maximum Correlation Criterion (MCC) and plays a decisive role in filtering performance. The double-Cauchy hybrid kernel function in the MCC is analyzed, taking into account the parameters accessible to the system model. Considering the propagation of variance The kernel width is adaptively updated based on the Cauchy-Schwarz divergence of the noise distribution. For filtering, changing the kernel width alters the filter gain, thereby improving filtering performance for various types of non-Gaussian noise.

[0021] Because the kernel function in this application is not a traditional Gaussian kernel function, but a double Cauchy kernel function with a convex combination of certain coefficients. The kernel width is not a traditional fixed size, which is unsuitable for unstable or unknown noise, or time-varying noise. Instead, it is an adaptive kernel width obtained based on the system model and the shape parameters of the noise distribution. Therefore, an adaptive maximum correlation entropy KF filtering algorithm based on the Double-Cauchy hybrid kernel function is proposed, denoted as MDC_MCCKF.

[0022] Now combined Figure 1 The steps of this embodiment will be further explained as follows: Step 1: For the vehicle navigation problem with a linear measurement model, construct a discrete-time dynamic model of the linear system of vehicle states, including state equations and observation equations: (1) (2) In the formula, express Vehicle status at any given time. and They represent Vehicle position and speed at any given time; express Vehicle status at time -1 express The output of the observation at any given time; and They represent The state transition matrix at time t and The observation matrix at time points; and They represent The process noise and measurement noise at any given moment are unrelated. Assume the process noise follows a pattern with a mean of 0 and a covariance matrix of... The non-Gaussian distribution, the measurement noise follows a mean of 0 and a covariance matrix of... The non-Gaussian distribution satisfies , , Indicates the expectation operation, superscript T This indicates transpose, due to unknown outliers and interference. Process noise covariance at time step and Measurement noise covariance at time Neither of them are accurate. for n dimensional real number, for mdimensional real number, for 3D real matrix for A 3D real matrix.

[0023] For a basic vehicle navigation problem, let's take a four-dimensional state variable as an example. The first two state variables are the vehicle's position coordinates, and the last two state variables are the corresponding velocities. Therefore, and They are represented as follows: in, Indicates the sampling interval.

[0024] Based on the statistical characteristics of non-Gaussian noise in the discrete-time dynamic model of a linear system, any non-Gaussian distribution can be represented or approximated by the sum of a finite number of Gaussian distributions. Here, the process noise and measurement noise of the non-Gaussian distribution are represented by a convex combination of two Gaussian components, and the statistical model of non-Gaussian noise is constructed accordingly, as shown in equation (4): (4) For the weighting coefficients, satisfying ; ; The mean is covariance is It follows a normal distribution.

[0025] Step 2: Based on the non-Gaussian noise statistical model in Step 1, this embodiment adopts the adaptive kernel maximum correlation entropy Kalman filter method based on the hybrid kernel function to obtain better filtering performance.

[0026] To better understand this step, the traditional Kalman filtering process based on weighted least squares is explained as follows.

[0027] The traditional Kalman filtering process based on weighted least squares is as follows: For the vehicle navigation problem using a linear measurement model, a discrete-time dynamic model of the linear system of vehicle states is constructed, including state equations and observation equations: The state prediction phase includes one-step state prediction and one-step error covariance prediction, expressed as: (5) (6) In the formula, express Predicting vehicle status at any given moment. express Vehicle state estimation at any given time. This represents the prediction error covariance; express The state estimation error covariance at time t; Constructing the cost function , represented as: (15) Given the accurate process noise covariance and measurement noise covariance, by solving... The state update is represented as: (16) (17) (18) , (19) In the formula, express Vehicle state estimation at any given time. , They represent Filter gain at time and The state estimation error covariance at time t. express 3D identity matrix This is the prediction error term.

[0028] The filtering process in this embodiment also includes a state prediction stage and a state update stage.

[0029] Compared to traditional weighted least squares-based Kalman filtering, this embodiment uses the same state prediction stage as traditional weighted least squares-based Kalman filtering. That is, the state prediction stage in this embodiment is represented as follows: (5) (6) In this embodiment, the state update stage uses a correlation entropy cost function based on higher-order moments instead of the cost function based on weighted least squares algebra in traditional weighted least squares-based Kalman filtering.

[0030] Specifically: the cost function is the weighted sum of the kernel function for the residual term of the observation equation and the kernel function for the prediction error term of the state equation; whereby the residual term of the observation equation is expressed as: The prediction error term of the state equation is expressed as: ; The cost function, which is a convex combination of the double Cauchy mixed kernel functions for the residual term of the observation equation and the prediction error term of the state equation, is expressed as: (7) In the formula express The observation matrix at time, express Vehicle state estimation at any given time. express The state transition matrix at time t, express The nominal covariance of the state estimation error at time t. The nominal covariance represents the prediction error; This represents the nominal covariance of the measured noise. (Note: Here, "nominal covariance" refers to the covariance of a Gaussian distribution without outliers. The filtering accuracy obtained using the nominal covariance is actually somewhat biased because the non-Gaussian noise we study is usually characterized by a certain probability (e.g., α) of the Gaussian distribution without outliers plus a (1-α) probability of the distribution affected by large outliers. Since interference noise is diverse, the covariance under non-Gaussian noise is often difficult to obtain accurately. For simplicity, we use the nominal covariance instead of the actual non-Gaussian noise covariance. The filtering accuracy may be slightly worse than using the actual non-Gaussian noise covariance.) The superscript "-1" represents the inverse. Definition , These are the Cauchy kernel function terms of the observation equation, , It is the Cauchy kernel function term of the prediction error of the state equation; The kernel width is The Cauchy kernel function, The kernel width is The Cauchy kernel function, It is called the heavy-tailed regulation factor. Let e ​​be the error. For cases with covariance... vector , , This represents the 2-norm. This represents the mixing coefficient.

[0031] The kernel widths of the kernel function for the residual term of the observation equation and the kernel function for the prediction error term of the state equation are adaptively updated based on the residual term of the observation equation, and are expressed as follows: (8) In the formula, express Vehicle state estimation at any given time. It is the Cauchy-Schwarz divergence that characterizes the relationship between two probability distributions in observation noise.

[0032] The closed-form solution of the Cauchy-Schwarz divergence between two multivariate Gaussian distributions is: (9) if , These are positive definite square elements. ,and The diagonal matrix formed by these two multivariate Gaussian distributions can be expressed as the closed-form solution of the Cauchy-Schwarz divergence. (10) By maximizing the cost function, we obtain the vehicle state estimate at the current time and the covariance of the state estimate error at the current time, specifically: Regarding the cost function Vehicle state estimation at time 1 Find the derivative and set it to 0. This can be expressed as: (11) Approximation in kernel function ; The state update stage of this embodiment is as follows: (12) (13) (14) In the formula, express 3D identity matrix Indicates the filter gain. express The state estimation error covariance at time t. Indicates intermediate variables. .in , and make approximations ,have to , .

[0033] This embodiment employs the Maximum Correlation Entropy (MCC) cost function based on the higher-order moments of the error vector, which can better extract information from the error vector, achieving better filtering accuracy and robustness. Based on the characteristics of non-Gaussian noise distribution, a kernel function based on dual Cauchy mixing is used to achieve better filtering performance. Furthermore, this embodiment comprehensively considers the parameters accessible to the system model. Considering the propagation of variance The kernel width is adaptively updated based on the Cauchy-Schwarz divergence of the noise distribution. Compared to a fixed kernel width, this better reflects the characteristics of noise uncertainty in the state equation, achieving better filtering performance. To verify the filtering performance of the method in this embodiment, it is compared with other filtering algorithms.

[0034] The MDC_MCCKF proposed in this embodiment is applied to the basic navigation problem of linear uniform linear motion. The adaptive kernel maximum correlation entropy Kalman filter method based on a hybrid kernel function (a combination of two Cauchy kernel functions with certain coefficients) is compared with the maximum correlation entropy Kalman filter based on a single Gaussian kernel function (D_MCCKF), the hybrid maximum correlation entropy Kalman filter based on two Gaussian kernel functions (MDG_MCCKF), and the maximum correlation entropy Kalman filter based on a mixture of Gaussian and Cauchy kernel functions (MGC_MCCKF) under the adaptive kernel width criterion. The simulation results obtained using Matlab R2018a are as follows: D_MCCKF is typically the maximum correlation entropy Kalman filter using a single Gaussian kernel function, and its derivation will not be repeated here. The derivation of MDG_MCCKF and MGC_MCCKF is similar to that of MDC_MCCKF in this paper, and detailed derivations are omitted here. The derivation conclusions are as follows: (Note: For kernel width is Gaussian kernel function, It is a Gaussian kernel function with a heavy-tailed adjustment factor g; For kernel width is The Cauchy kernel function, This is the Cauchy kernel function with a heavy-tailed adjustment factor c.

[0035] Table 1. Cost functions and measurement update stages under different algorithms For the basic navigation problem applied to linear uniform linear motion, taking four state variables as an example, the first two state variables are the vehicle's position coordinates, and the last two state variables are the corresponding velocities.

[0036] (1) Add noise as Figure 2The shot non-Gaussian noise shown is illustrated. Under shot non-Gaussian noise, the root mean square error (RMSE) of variables 1 and 3 in the four states under the four filtering algorithms is as follows: Figure 3 and 4 As shown in Table 2, the estimation accuracy of the four filtering methods under shot non-Gaussian noise is illustrated.

[0037] Table 2. Estimation accuracy of four filtering methods under shot non-Gaussian noise. In the table, x1 represents the first state variable, x2 represents the second state variable, x3 represents the third state variable, and x4 represents the fourth state variable.

[0038] Therefore, it can be seen that under shot non-Gaussian noise, compared with the other three filtering algorithms, the filtering method of this embodiment greatly improves the filtering performance, verifying the superiority of the filtering method of this embodiment (under shot non-Gaussian noise, the ARMSE of the traditional KF algorithm is very large, which is not listed in the table, also indicating that the traditional KF is not suitable for filtering non-Gaussian noise).

[0039] (2) Add as Figure 5 The pulse-tailed non-Gaussian noise shown is illustrated. Under pulse-tailed non-Gaussian noise, the root mean square error (RMSE) of variables 1 and 3 in the four states under the four filtering algorithms is as follows: Figure 6 and 7 As shown in Table 3, the estimation accuracy of the four filtering methods is as follows: under impulse non-Gaussian noise.

[0040] Table 3. Estimation accuracy of three filtering methods under impulse non-Gaussian noise. Therefore, it can be seen that under pulse-tailed non-Gaussian noise, compared with the other three filtering algorithms, the filtering method of this embodiment greatly improves the filtering performance, verifying the superiority of the filtering method of this embodiment (under pulse-tailed non-Gaussian noise, the ARMSE of the traditional KF algorithm is very large, which is not listed in the table, also indicating that the traditional KF is not suitable for filtering non-Gaussian noise).

[0041] (3) Filtering performance of the proposed MDC_MCCKF algorithm under adaptive kernel width and fixed kernel width with heavy-tailed non-Gaussian noise in this embodiment.

[0042] Table 4. Estimation accuracy of the filtering algorithm under adaptive kernel width and fixed kernel width in the case of pulse-tailed non-Gaussian noise. As shown in Table 4, under impulse non-Gaussian noise, the filtering performance under adaptive kernel width, while not necessarily the best, is very close to that of the optimal kernel width. Furthermore, we can also see that the MDC_MCCKF proposed in this invention is insensitive to the kernel width relative to the Gaussian kernel function; the filtering performance remains relatively consistent regardless of changes in kernel width.

[0043] This application discloses a vehicle state estimation system based on hybrid kernel functions, including: Linear module: A linear system that constructs vehicle states based on state equations and observation equations; the vehicle states include vehicle position and vehicle speed; Noise module: Based on a non-Gaussian noise statistical model, it describes the process noise in the state equation and the measurement noise in the observation equation of a linear system; Estimation Module: For linear systems based on vehicle state, an adaptive kernel maximum correlation entropy Kalman filter method based on a hybrid kernel function is used to estimate the vehicle state, thereby obtaining the estimated vehicle state value. The adaptive kernel maximum correlation entropy Kalman filter method based on a hybrid kernel function includes: using a convex combination of two Cauchy hybrid kernel functions for the residual term of the observation equation and the prediction error term of the state equation as a cost function; maximizing the cost function to obtain the vehicle state estimate at the current time and the state estimate error covariance at the current time. Kernel width self-updating module: Adaptively updates the kernel widths of the kernel function for the residual term of the observation equation and the kernel function for the prediction error term of the state equation based on the residual term of the observation equation and the Cauchy-Schwarz divergence.

[0044] This application also discloses a terminal device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the computer program, it employs any of the vehicle state estimation methods based on hybrid kernel functions described in the above embodiments.

[0045] The terminal device can be a computer device such as a desktop computer, a laptop computer, or a cloud server. The terminal device includes, but is not limited to, a processor and a memory. For example, the terminal device may also include input / output devices, network access devices, and buses.

[0046] The processor can be a central processing unit (CPU). Of course, depending on the actual use, it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (SICs based on hybrid kernel functions for vehicle state estimation), off-the-shelf programmable gate arrays (FPGs based on hybrid kernel functions for vehicle state estimation), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc., and this application does not impose any restrictions on this.

[0047] The memory can be an internal storage unit of the terminal device, such as a hard disk or RAM of the terminal device, or an external storage device of the terminal device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD), or flash memory card (FC) equipped on the terminal device. Furthermore, the memory can be a combination of internal storage units and external storage devices of the terminal device. The memory is used to store computer programs and other programs and data required by the terminal device. The memory can also be used to temporarily store data that has been output or will be output. This application does not limit this.

[0048] In this terminal device, any of the vehicle state estimation methods based on hybrid kernel functions in the above embodiments can be stored in the memory of the terminal device and loaded and executed on the processor of the terminal device for convenient use. This application also discloses a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it employs any of the vehicle state estimation methods based on hybrid kernel functions described in the above embodiments.

[0049] The computer program can be stored in a computer-readable medium. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or certain middleware. The computer-readable medium includes any entity or device capable of carrying computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (R based on hybrid kernel function vehicle state estimation M), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the computer-readable medium includes, but is not limited to, the above-mentioned components.

[0050] In this computer-readable storage medium, any of the vehicle state estimation methods based on hybrid kernel functions in the above embodiments can be stored in the computer-readable storage medium and loaded and executed on the processor to facilitate the storage and application of the above methods.

[0051] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A vehicle state estimation method based on hybrid kernel functions, characterized in that, include: For linear systems based on vehicle state, an adaptive kernel maximum correlation entropy Kalman filter method based on hybrid kernel functions is used to estimate the vehicle state, thereby obtaining the state estimate of the vehicle state. The adaptive kernel maximum correlation entropy Kalman filter method based on hybrid kernel functions includes: The cost function is a convex combination of the double Cauchy mixture kernel function targeting the residual term of the observation equation and the prediction error term of the state equation. The cost function is maximized to obtain the vehicle state estimate and the state estimate error covariance at the current time. Among them, the kernel widths of the kernel function for the residual term of the observation equation and the kernel function for the prediction error term of the state equation are adaptively updated based on the residual term of the observation equation and the Cauchy-Schwarz divergence. The method of using a convex combination of two Cauchy mixed kernel functions as the cost function, targeting the residual term of the observation equation and the prediction error term of the state equation, includes the following steps: The residual term of the observation equation is expressed as: ; The prediction error term of the state equation is expressed as: ; in express The output of the observation at any moment, express The observation matrix at time, express Vehicle status at any given time; express The state transition matrix at time t; express Vehicle state estimation at any given time; The cost function is expressed as: (7) in express Vehicle state estimation at any given time. express Vehicle state estimation at any given time; express The nominal covariance of the state estimation error at time t. The nominal covariance represents the prediction error; Represents the nominal covariance of the measurement noise; the superscript "-1" indicates the inverse; , These are the Cauchy kernel function terms of the observation equation; , It is the Cauchy kernel function term of the prediction error of the state equation; The kernel width is The Cauchy kernel function, The kernel width is The Cauchy kernel function, This is called the heavy-tailed adjustment factor; e represents the error; for cases with covariance vector , , Represents the 2-norm; Indicates the mixing coefficient; The kernel widths of the kernel function for the residual term of the observation equation and the kernel function for the prediction error term of the state equation are adaptively updated based on the residual term of the observation equation and the Cauchy-Schwarz divergence, and are expressed as: (8) In the formula, express Vehicle state estimation at any given time; It is the Cauchy-Schwarz divergence that characterizes the two probability distributions in the observation noise; The closed-form solution of the Cauchy-Schwarz divergence is: like , These are positive definite square elements. ,and The closed-form solution of the Cauchy-Schwarz divergence can be expressed as: (The diagonal matrix formed by these two elements is not directly related to the main text and can be omitted.) 2. The vehicle state estimation method based on hybrid kernel functions according to claim 1, characterized in that, according to Vehicle state estimation at time and The state estimation error covariance at time t is obtained by performing a one-step prediction according to the following prediction equation. Predicted vehicle state estimates and prediction error covariance at time: (5) (6) In the formula, express Predicting vehicle status at any given moment. express Vehicle state estimation at any given time. This represents the prediction error covariance; express The state estimation error covariance at time t. express The state transition matrix at time t; express The nominal covariance matrix of the noise at each time step; superscript T This indicates transpose.

3. The vehicle state estimation method based on hybrid kernel functions according to claim 2, characterized in that, The process of maximizing the cost function to obtain the vehicle state estimate and the state estimate error covariance at the current time includes the following steps: Regarding the cost function Vehicle state estimation at time 1 Find the derivative and set it to 0. This can be expressed as: (11) Approximation in kernel function ,get: (12) (13) (14) In the formula, express 3D identity matrix Indicates the filter gain. express The state estimation error covariance at time t. Indicates intermediate variables. .

4. The vehicle state estimation method based on hybrid kernel functions according to claim 3, characterized in that, The linear system state equation and observation equation for the vehicle state are constructed; the vehicle state includes vehicle position and vehicle speed; the noise in the system is described using a non-Gaussian noise statistical model.

5. The vehicle state estimation method based on hybrid kernel functions according to claim 4, characterized in that, The construction of the linear system includes the following steps: Let the previous moment be... -1 time, current time is time; The state equation is expressed as: (1) In the formula, express Vehicle status at any given time. express The state transition matrix at time t, express Vehicle status at time -1 express The process noise at time t; the process noise follows a pattern with a mean of 0 and a covariance matrix of... is a non-Gaussian distribution, where , Indicates the expectation operation, superscript T Indicates transpose; The observation equation is expressed as: (2) In the formula, express The output of the observation at any moment, express The observation matrix at time, express Measurement noise at time; the measurement noise follows a pattern with a mean of 0 and a covariance matrix of... is a non-Gaussian distribution, where .

6. The vehicle state estimation method based on hybrid kernel functions according to claim 5, characterized in that, The noise in the system is described using a non-Gaussian noise statistical model, including the following steps: The mixed distribution consisting of L components can be represented as: (3) For the weighting coefficients, satisfying ; ; The mean is The covariance is The normal distribution; The heavy-tailed noise is characterized by a mixture model of two Gaussian density functions, expressed as: (4)。 7. A vehicle state estimation system based on hybrid kernel functions, characterized in that, include; Linear module: A linear system that constructs vehicle states based on state equations and observation equations; the vehicle states include vehicle position and vehicle speed; Noise module: Based on a non-Gaussian noise statistical model, it describes the process noise in the state equation and the measurement noise in the observation equation of a linear system; Estimation Module: For linear systems based on vehicle state, an adaptive kernel maximum correlation entropy Kalman filter method based on a hybrid kernel function is used to estimate the vehicle state, thereby obtaining the estimated vehicle state value. The adaptive kernel maximum correlation entropy Kalman filter method based on a hybrid kernel function includes: using a convex combination of two Cauchy hybrid kernel functions for the residual term of the observation equation and the prediction error term of the state equation as a cost function; maximizing the cost function to obtain the vehicle state estimate at the current time and the state estimate error covariance at the current time. Kernel width self-updating module: Adaptively updates the kernel widths of the kernel function for the residual term of the observation equation and the kernel function for the prediction error term of the state equation based on the residual term of the observation equation and the Cauchy-Schwarz divergence. The method of using a convex combination of two Cauchy mixed kernel functions as the cost function, targeting the residual term of the observation equation and the prediction error term of the state equation, includes the following steps: The residual term of the observation equation is expressed as: ; The prediction error term of the state equation is expressed as: ; in express The output of the observation at any moment, express The observation matrix at time, express Vehicle status at any given time; express The state transition matrix at time t; express Vehicle state estimation at any given time; The cost function is expressed as: (7) in express Vehicle state estimation at any given time. express Vehicle state estimation at any given time; express The nominal covariance of the state estimation error at time t. The nominal covariance represents the prediction error; Represents the nominal covariance of the measurement noise; the superscript "-1" indicates the inverse; , These are the Cauchy kernel function terms of the observation equation; , It is the Cauchy kernel function term of the prediction error of the state equation; The kernel width is The Cauchy kernel function, The kernel width is The Cauchy kernel function, This is called the heavy-tailed adjustment factor; e represents the error; for cases with covariance vector , , Represents the 2-norm; Indicates the mixing coefficient; The kernel widths of the kernel function for the residual term of the observation equation and the kernel function for the prediction error term of the state equation are adaptively updated based on the residual term of the observation equation and the Cauchy-Schwarz divergence, and are expressed as: (8) In the formula, express Vehicle state estimation at any given time; It is the Cauchy-Schwarz divergence that characterizes the two probability distributions in the observation noise; The closed-form solution of the Cauchy-Schwarz divergence is: like , These are positive definite square elements. ,and The closed-form solution of the Cauchy-Schwarz divergence can be expressed as: (The diagonal matrix formed by these two elements is not directly related to the main text and can be omitted.) 8. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, The memory stores a computer program that can run on a processor. When the processor loads and executes the computer program, it employs a vehicle state estimation method based on a hybrid kernel function as described in any one of claims 1 to 6.