Vehicle stability evaluation and early warning method and device, computer equipment and medium

Abstract

The invention relates to a vehicle stability evaluation and early warning method and device, computer equipment and a medium. The method comprises the following steps: constructing a high-dimensional feature space fusing vehicle lateral, yaw, heeling and tire dynamic states; obtaining a vehicle dynamic state estimation value in the high-dimensional feature space based on the multi-source sensor data; determining a local stable manifold boundary representing the stable motion limit of the vehicle in the high-dimensional feature space according to the current vehicle driving condition; extracting stability risk characteristics of multiple dimensions in parallel according to a geometrical relationship and a dynamic evolution trend between a vehicle dynamic state estimation value and a local stability manifold boundary; generating a comprehensive risk level through a probability fusion model, and identifying a dominant instability mode; and when the comprehensive risk level exceeds a preset threshold value, triggering a corresponding predictive cooperative chassis control strategy according to the dominant instability mode. By adopting the method, multi-dimensional states can be fused in real time, and different instability modes can be recognized and distinguished in advance.

Description

Technical Field

[0001] This application relates to the field of vehicle performance testing technology, and in particular to a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for vehicle stability assessment and early warning. Background Technology

[0002] Vehicle Stability Control (VSC) is the core of active automotive safety, with Electronic Stability Program (ESP) being the mainstream technology. This system uses sensors to collect signals such as yaw rate and lateral acceleration in real time and compares them with ideal values ​​calculated based on a linear two-degree-of-freedom model. When the deviation exceeds a preset threshold, the system applies braking to specific wheels to generate corrective torque. Despite its mature technology, ESP is essentially a "reactive" control system, requiring intervention only after sensors have clearly detected instability characteristics, resulting in a warning lag. Furthermore, its reliance on a linear model and fixed thresholds makes it difficult to accurately adapt to complex conditions such as changes in vehicle load, tire nonlinearity, and sudden changes in road surface adhesion, leading to a decline in control performance under extreme conditions.

[0003] To enhance the adaptability and foresight of systems, existing research has introduced more advanced dynamic analysis methods. For example, some patents design controllers by constructing a vehicle dynamic stability domain and combining it with model predictive control; however, these stability domains are often based on simplified models, making it difficult to accurately describe the strongly nonlinear dynamics under large operating conditions. Other approaches focus on improving predictive capabilities, such as using parameterized prototype models for prediction, but these suffer from high complexity and limited ability to distinguish multi-dimensional coupled instability modes. Still other technologies attempt to dynamically generate yaw rate thresholds to improve adaptability. While these methods touch upon the concepts of prediction and boundary analysis, they still face bottlenecks in nonlinear description, state dimensions, and early identification.

[0004] Therefore, there is an urgent need for a vehicle stability assessment and early warning method, device, computer equipment, computer-readable storage medium, and computer program product that can integrate multi-dimensional states in real time and identify and distinguish different instability modes in advance. Summary of the Invention

[0005] Therefore, it is necessary to provide a vehicle stability assessment and early warning method, device, computer equipment, computer-readable storage medium, and computer program product that can integrate multi-dimensional states in real time and identify and distinguish different instability modes in advance, in order to address the above-mentioned technical problems.

[0006] Firstly, this application provides a method for vehicle stability assessment and early warning, including:

[0007] Construct a high-dimensional feature space that integrates vehicle lateral, yaw, roll, and tire dynamic states;

[0008] Based on multi-source sensor data, the estimated value of the vehicle dynamic state in the high-dimensional feature space is obtained;

[0009] For the current vehicle driving conditions, determine the local stable manifold boundary in the high-dimensional feature space that characterizes the vehicle's stable motion limit;

[0010] Based on the geometric relationship and dynamic evolution trend between the estimated vehicle dynamic state and the local stable manifold boundary, stability risk features of multiple dimensions are extracted in parallel.

[0011] Based on the stability risk characteristics of the multiple dimensions, a comprehensive risk level is generated through a probabilistic fusion model, and the dominant instability mode is identified.

[0012] When the overall risk level exceeds a preset threshold, a corresponding predictive collaborative chassis control strategy is triggered based on the dominant instability mode to adjust the vehicle state.

[0013] In one embodiment, constructing a high-dimensional feature space that integrates vehicle lateral, yaw, roll, and tire dynamic states includes:

[0014] A vehicle coupled dynamics model is established, which includes the vehicle body roll degree of freedom. The vehicle coupled dynamics model is used to describe the coupling relationship between the vehicle's planar motion and roll motion, as well as the dynamic characteristics of each tire.

[0015] Based on the coupled dynamics model, a dimensionless feature state vector is defined, which includes the vehicle's center of gravity sideslip angle, the rate of change of the center of gravity sideslip angle, the yaw rate error, the difference in the average sideslip angle between the front and rear axles representing the understeer or oversteer tendency, and the difference in the sideslip angle between the left and right wheels of the front and rear axles representing the difference in tire grip utilization caused by load transfer.

[0016] The space containing the reduced-dimensional feature state vector is defined as the high-dimensional feature space.

[0017] In one embodiment, obtaining the vehicle dynamic state estimate in the high-dimensional feature space based on multi-source sensor data includes:

[0018] The coupled dynamics model is discretized to establish nonlinear state transition equations and observation equations that include vehicle state and tire-road adhesion parameters;

[0019] The Kalman filter algorithm is used to process the state transition equation and the observation equation. The state prediction covariance matrix is ​​modified by introducing a time-varying fading factor to enhance the tracking ability of system state and parameter abrupt changes.

[0020] Based on the multi-source sensor data, the vehicle dynamic state estimate is output in the high-dimensional feature space through recursive estimation using the Kalman filter algorithm.

[0021] In one embodiment, determining the local stable manifold boundary characterizing the vehicle's stable motion limit in the high-dimensional feature space for the current vehicle driving condition includes:

[0022] Based on the current vehicle speed, road surface adhesion estimation, and steering input parameters, retrieve the set of sample points in the adjacent state from the pre-stored full-condition historical state sample database;

[0023] The local linear embedding algorithm is used to reduce the dimensionality of the neighboring state sample point set to obtain the coordinates and corresponding projection transformation relationships in the low-dimensional local embedding space.

[0024] In the low-dimensional local embedding space, based on the stable sample points in the set of neighboring state sample points, the minimum bounding hypersphere is calculated using the support vector domain description method, and the spherical surface of the minimum bounding hypersphere is constructed as an approximate expression of the local stable manifold boundary under the current working condition.

[0025] Based on the projection transformation relationship, the real-time vehicle dynamic state estimate is mapped to the low-dimensional local embedding space. The Euclidean distance from the mapping point to the center of the minimum enclosing hypersphere is calculated, and the distance value is used as a geometric metric to quantify the deviation of the real-time vehicle state from the boundary of the local stable manifold.

[0026] In one embodiment, the stability risk characteristics include manifold geometric risk characteristics, tire force saturation risk characteristics, and roll motion risk characteristics; the generation of a comprehensive risk level based on the stability risk characteristics across multiple dimensions using a probabilistic fusion model, and the identification of the dominant instability mode, includes:

[0027] The manifold geometric risk features, the tire force saturation risk features, and the roll motion risk features are all input into a Bayesian neural network, which outputs the posterior probability distribution data of each preset mode, including the vehicle being in a stable state, sideslip mode, fishtailing mode, and rollover risk mode.

[0028] Based on the posterior probability distribution data and the severity weights of each preset instability mode, the comprehensive risk level is calculated, and the mode with the highest posterior probability is selected as the dominant instability mode.

[0029] In one embodiment, the step of triggering a corresponding predictive collaborative chassis control strategy based on the dominant instability mode when the overall risk level exceeds a preset threshold includes:

[0030] When the dominant instability mode is sideslip mode, a first additional yaw moment is generated with the goal of correcting the sideslip angle and the rate of change of the sideslip angle, and the drive system, braking system and steering system are coordinated to achieve the moment through differentiated torque distribution and angle compensation;

[0031] When the dominant instability mode is the tail-slip mode, a second additional yaw moment is generated with the goal of suppressing the yaw rate error and its divergence trend. In the moment calculation, a feedforward compensation term based on the maximum Lyapunov exponent is introduced. The second additional yaw moment is applied together by coordinating the directional differential braking function of the braking system and the reverse compensation function of the steering system.

[0032] When the dominant instability mode is the rollover risk mode, the suspension system damping is maximized to dissipate roll energy, and longitudinal uniform braking is controlled to reduce vehicle speed.

[0033] Secondly, this application also provides a vehicle stability assessment and early warning device, including:

[0034] The building module is used to construct a high-dimensional feature space that integrates the vehicle's lateral, yaw, roll, and tire dynamic states;

[0035] The processing module is used to obtain the vehicle dynamic state estimate in the high-dimensional feature space based on multi-source sensor data;

[0036] The processing module is also used to determine the local stable manifold boundary in the high-dimensional feature space that characterizes the vehicle's stable motion limit for the current vehicle driving conditions.

[0037] The extraction module is used to extract stability risk features of multiple dimensions in parallel based on the geometric relationship and dynamic evolution trend between the vehicle dynamic state estimate and the local stable manifold boundary.

[0038] The processing module is also used to generate a comprehensive risk level based on the stability risk characteristics of the multiple dimensions through a probability fusion model, and to identify the dominant instability mode;

[0039] The early warning module is used to trigger a corresponding predictive collaborative chassis control strategy based on the dominant instability mode when the overall risk level exceeds a preset threshold, so as to adjust the vehicle state.

[0040] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0041] Construct a high-dimensional feature space that integrates vehicle lateral, yaw, roll, and tire dynamic states;

[0042] Based on multi-source sensor data, the estimated value of the vehicle dynamic state in the high-dimensional feature space is obtained;

[0043] For the current vehicle driving conditions, determine the local stable manifold boundary in the high-dimensional feature space that characterizes the vehicle's stable motion limit;

[0044] Based on the geometric relationship and dynamic evolution trend between the estimated vehicle dynamic state and the local stable manifold boundary, stability risk features of multiple dimensions are extracted in parallel.

[0045] Based on the stability risk characteristics of the multiple dimensions, a comprehensive risk level is generated through a probabilistic fusion model, and the dominant instability mode is identified.

[0046] When the overall risk level exceeds a preset threshold, a corresponding predictive collaborative chassis control strategy is triggered based on the dominant instability mode to adjust the vehicle state.

[0047] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0048] Construct a high-dimensional feature space that integrates vehicle lateral, yaw, roll, and tire dynamic states;

[0049] Based on multi-source sensor data, the estimated value of the vehicle dynamic state in the high-dimensional feature space is obtained;

[0050] For the current vehicle driving conditions, determine the local stable manifold boundary in the high-dimensional feature space that characterizes the vehicle's stable motion limit;

[0051] Based on the geometric relationship and dynamic evolution trend between the estimated vehicle dynamic state and the local stable manifold boundary, stability risk features of multiple dimensions are extracted in parallel.

[0052] Based on the stability risk characteristics of the multiple dimensions, a comprehensive risk level is generated through a probabilistic fusion model, and the dominant instability mode is identified.

[0053] When the overall risk level exceeds a preset threshold, a corresponding predictive collaborative chassis control strategy is triggered based on the dominant instability mode to adjust the vehicle state.

[0054] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0055] Construct a high-dimensional feature space that integrates vehicle lateral, yaw, roll, and tire dynamic states;

[0056] Based on multi-source sensor data, the estimated value of the vehicle dynamic state in the high-dimensional feature space is obtained;

[0057] For the current vehicle driving conditions, determine the local stable manifold boundary in the high-dimensional feature space that characterizes the vehicle's stable motion limit;

[0058] Based on the geometric relationship and dynamic evolution trend between the estimated vehicle dynamic state and the local stable manifold boundary, stability risk features of multiple dimensions are extracted in parallel.

[0059] Based on the stability risk characteristics of the multiple dimensions, a comprehensive risk level is generated through a probabilistic fusion model, and the dominant instability mode is identified.

[0060] When the overall risk level exceeds a preset threshold, a corresponding predictive collaborative chassis control strategy is triggered based on the dominant instability mode to adjust the vehicle state.

[0061] The aforementioned vehicle stability assessment and early warning methods, devices, computer equipment, computer-readable storage media, and computer program products overcome the limitations of traditional low-dimensional or linear models in characterizing complex instability modes by establishing a unified high-dimensional feature space that integrates lateral, yaw, roll, and tire dynamics. By acquiring dynamic state estimates in this high-dimensional space in real time based on multi-source sensor data and determining the local stable manifold boundary representing the stability limit online for the current operating conditions, the system can perform advanced risk assessment based on dynamic boundaries without relying on fixed thresholds, effectively overcoming the shortcomings of traditional methods such as delayed early warning and poor adaptability to operating conditions. By extracting multi-dimensional geometric and dynamic risk features of state points relative to this dynamic boundary in parallel and using probabilistic models for fusion and pattern recognition, early and precise differentiation of instability modes with different physical mechanisms such as sideslip, fishtailing, and rollover is achieved, solving the problem of insufficient instability mode identification capability in existing technologies. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1 This is a flowchart illustrating a vehicle stability assessment and early warning method in one embodiment;

[0064] Figure 2 This is a flowchart illustrating the vehicle stability assessment and early warning method in yet another embodiment;

[0065] Figure 3 This is a flowchart illustrating the vehicle stability assessment and early warning method in another embodiment;

[0066] Figure 4 This is a flowchart illustrating the vehicle stability assessment and early warning method in another embodiment;

[0067] Figure 5 This is a flowchart illustrating the vehicle stability assessment and early warning method in the most detailed embodiment;

[0068] Figure 6 This is a structural block diagram of a vehicle stability assessment and early warning device in one embodiment;

[0069] Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0071] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0072] In one exemplary embodiment, such as Figure 1 As shown, a vehicle stability assessment and early warning method is provided, which is applied to... Figure 1 The server in the process is described, including the following steps S102 to S112. Wherein:

[0073] Step S102: Construct a high-dimensional feature space that integrates the vehicle's lateral, yaw, roll, and tire dynamic states.

[0074] Specifically, firstly, at the modeling level, an extended coupled vehicle dynamics model that surpasses the traditional two-degree-of-freedom model is established. This model not only includes the longitudinal, lateral, and yaw dynamic equations describing the vehicle's planar motion, but also specifically introduces the dynamic equations for the vehicle's roll degree of freedom, thus explicitly characterizing the coupling effect between roll motion and planar motion. Simultaneously, the model independently models the mechanical behavior of the four tires, employing the nonlinear Pacejka magic formula to describe the longitudinal and lateral forces of each tire as functions of tire slip angle, slip ratio, dynamic vertical load, and the real-time estimated road adhesion coefficient. This accurately characterizes the tire's properties in the nonlinear region, constructing a high-dimensional feature space that integrates the vehicle's lateral, yaw, roll, and tire dynamic states. Secondly, at the state definition level, based on the aforementioned complete dynamic model, a feature state vector for stability analysis is defined and derived. This vector not only includes the core states of the vehicle's motion, such as the center of mass slip angle and yaw rate, but also incorporates derived features reflecting the system's internal coupling mechanisms and instability tendencies. For example, the difference in average slip angle between the front and rear axles can be introduced to quantify understeer / oversteer trends, and the difference in slip angle between the left and right wheels of the front and rear axles can be introduced to reflect uneven tire grip utilization caused by load transfer.

[0075] Step S104: Based on multi-source sensor data, obtain the vehicle dynamic state estimate in the high-dimensional feature space.

[0076] Specifically, the multi-source sensors installed on the vehicle (e.g., inertial measurement units, wheel speed sensors, steering angle sensors, etc.) can only directly provide limited and typically low-dimensional measurement signals, such as vehicle yaw rate, longitudinal / lateral acceleration, and wheel speed. However, many key state variables defined in the aforementioned high-dimensional feature space, such as the centroid sideslip angle, the real-time sideslip angle of each tire, dynamic vertical load, and the crucial tire-road adhesion coefficient, cannot be directly measured by sensors. Therefore, a nonlinear state observer is established with an extended coupled vehicle dynamics model as its core. This observer receives the direct measurements from the aforementioned multi-source sensors as input and incorporates the high-dimensional state vector to be estimated (including vehicle motion state, attitude angles, and tire-related parameters) and the road adhesion coefficient into its estimation framework.

[0077] In a preferred embodiment, the Kalman Filter (STCKF) algorithm is employed. This algorithm operates through two recursive steps: time update (prediction) and measurement update (correction). First, the state at the next moment is predicted based on the dynamic model; then, the predicted value is compared with the actual sensor measurement at the current moment, and the predicted state is optimally corrected based on the Kalman gain.

[0078] Step S106: For the current vehicle driving conditions, determine the local stable manifold boundary in the high-dimensional feature space that represents the vehicle's stable motion limit.

[0079] Specifically, the locally stable manifold boundary essentially describes the boundary of the attractive domain of the system's stable equilibrium point (or stable motion mode). In a high-dimensional feature space, this boundary constitutes a complex, possibly curved, hypersurface. When the vehicle's state point lies on one side of this surface, its trajectory will converge to the stable region; once it crosses the boundary to the other side, the system will diverge and enter an unstable state. Therefore, this manifold boundary is the most accurate mathematical description of the vehicle's stable motion limit.

[0080] Based on real-time estimated key parameters of the current vehicle driving conditions (such as vehicle speed, road surface adhesion coefficient, and steering input), a dynamically changing local stable manifold boundary closely related to the current driving conditions (vehicle speed, road surface, and steering input) is calculated in real time using a data-driven method.

[0081] Step S108: Based on the geometric relationship and dynamic evolution trend between the vehicle dynamic state estimate and the local stable manifold boundary, multiple dimensions of stability risk features are extracted in parallel.

[0082] Specifically, geometric relationship refers to the spatial correlation between the vehicle dynamic state (coordinate points representing the instantaneous kinematics and dynamics of the vehicle) estimated in real time within a high-dimensional feature space spanned by the vehicle's sideslip, yaw, roll, and tire dynamics, and the locally stable manifold boundary (a topological surface representing the vehicle's stable motion limit under this condition) approximated online using a data-driven method under the current operating condition. Its core quantitative indicator is the geodesic distance from the state point to the manifold boundary or its projected distance in the reduced-dimensional embedding space. This distance intuitively reflects the static safety margin between the current vehicle state and the instability threshold.

[0083] The dynamic evolution trend focuses on the time-varying characteristics of the aforementioned geometric relationships and the inherent stability of the system. It is primarily manifested in the first and even higher-order time derivatives of the aforementioned safety margin distance, i.e., the rates and accelerations as the state point approaches or moves away from the stability boundary. At a deeper level, the evolution trend encompasses the analysis of the local dynamic stability of the system at the current state point. By performing Jacobi linearization on the nonlinear vehicle dynamics system at this point and calculating its dominant Lyapunov exponent or the real part of its eigenvalues, the local behavior of the system under small perturbations can be determined: negative values ​​indicate that the perturbation decays and the system is locally asymptotically stable; positive values ​​mean that the perturbation diverges exponentially, revealing that the system is already in an inherently unstable state, exhibiting a dynamic tendency towards spontaneous instability even if it does not immediately reach the geometric boundary.

[0084] Parallel extraction of stability risk features from multiple dimensions refers to the system adopting a synchronous computing architecture to independently and simultaneously generate a series of quantitative indicators from different physical and mathematical perspectives, which together constitute a three-dimensional and analytical assessment system for vehicle instability risk.

[0085] Step S110: Based on the stability risk characteristics of multiple dimensions, a comprehensive risk level is generated through a probability fusion model, and the dominant instability mode is identified.

[0086] Specifically, a probabilistic fusion model (preferably a Bayesian neural network or a similar probabilistic graphical model) is introduced. During the training phase, this model establishes a probabilistic association between features and outcomes by learning the complex mapping relationship between multidimensional risk features and the final actual stability outcome (stability, sideslip, tail-flip, roll, etc.) from a large amount of historical or simulation data. During online runtime, the model receives real-time multidimensional feature vectors as input, and its internal operations are not simple weighted summations but rather probabilistic inferences.

[0087] The model calculates a comprehensive risk level. This level is not a linear combination of the feature values, but rather a posterior probability or related continuous value reflecting the overall instability of the vehicle, derived from the joint inference of all features based on a probabilistic model. It comprehensively considers the contributions of all features and their coupling relationships, and includes a measure of the model's own uncertainty, making it more robust and reliable than any single indicator or fixed rule.

[0088] Simultaneously, the model outputs the probability distribution of the vehicle's most likely instability mode (e.g., 70% probability of sideslip, 25% probability of stability, and 5% probability of drift). The mode with the highest probability is identified as the dominant instability mode. This pattern recognition capability stems directly from the richness of information in the multi-dimensional feature vectors. For example, sideslip may be strongly correlated with tire adhesion saturation characteristics, drift may be strongly correlated with dynamic divergence characteristics, and rollover may be strongly correlated with roll energy characteristics.

[0089] Step S112: When the overall risk level exceeds the preset threshold, the corresponding predictive collaborative chassis control strategy is triggered according to the dominant instability mode to adjust the vehicle state.

[0090] Specifically, the overall risk level is a single indicator (e.g., a value between 0 and 1, or a "low, medium, high" level) generated through probability fusion, quantifying the likelihood of global instability. The preset threshold is a safety limit pre-defined based on safety standards and vehicle characteristics. Only when this overall risk level exceeds the threshold does the system determine that the risk has reached a point requiring active intervention. This ensures the necessity of control, avoids unnecessary interference during low-risk or normal driving, and guarantees driving comfort and system acceptability.

[0091] The dominant instability mode indicates the most likely type of instability (such as sideslip, fishtailing, and rollover risk). For sideslip: it may primarily coordinate drive torque vectoring and slight counter-steering compensation to generate corrective yaw moment and control the center of gravity sideslip angle. For fishtailing: it may primarily use active braking on the outer front wheel in the direction of rotation, combined with counter-steering compensation, to quickly suppress the divergence of yaw rate. For rollover risk: it may simultaneously instruct the suspension system to enter maximum damping mode to dissipate roll energy, and instruct the powertrain to reduce torque and the braking system to apply uniform braking force to reduce vehicle speed, thereby reducing lateral forces at the source.

[0092] The ultimate goal of all control actions is to pull the vehicle's high-dimensional dynamic state point back from the dangerous region approaching or exceeding the local stable manifold boundary and stabilize it within the safe domain defined by that boundary. This is not merely correcting a single yaw rate or sideslip angle, but rather comprehensively adjusting the overall coupled motion state of the vehicle to restore it to a stable and controllable motion mode.

[0093] The aforementioned vehicle stability assessment and early warning method overcomes the limitations of traditional low-dimensional or linear models in characterizing complex instability modes by establishing a unified high-dimensional feature space that integrates lateral, yaw, roll, and tire dynamics. By acquiring dynamic state estimates in this high-dimensional space in real time based on multi-source sensor data and determining the local stable manifold boundary representing the stability limit online for the current operating condition, the system can perform advanced risk assessment based on the dynamic boundary without relying on fixed thresholds. This effectively overcomes the shortcomings of traditional methods, such as delayed early warning and poor adaptability to operating conditions. Furthermore, by extracting multi-dimensional geometric and dynamic risk features of state points relative to this dynamic boundary in parallel and using probabilistic models for fusion and pattern recognition, early and precise differentiation of instability modes with different physical mechanisms, such as sideslip, fishtailing, and rollover, is achieved, solving the problem of insufficient instability mode identification capability in existing technologies.

[0094] In one exemplary embodiment, such as Figure 2 As shown, a high-dimensional feature space is constructed that integrates the vehicle's lateral, yaw, roll, and tire dynamic states, including:

[0095] Step S202: Establish a vehicle coupled dynamics model that includes the vehicle body roll degree of freedom. The vehicle coupled dynamics model is used to describe the coupling relationship between the vehicle's planar motion and roll motion, as well as the dynamic characteristics of each tire.

[0096] Specifically, a coupled dynamics model of the vehicle, including the roll degree of freedom, is established. The planar motion equations of the four-wheeled vehicle considering the roll degree of freedom are described by the following set of formulas:

[0097]

[0098]

[0099]

[0100]

[0101] Among them, subscript Represents four wheels; and These represent the longitudinal force and lateral force of each tire, respectively. The steering angle of each wheel; and The longitudinal and lateral distances from the tire contact patch center to the vehicle's center of gravity; For the overall vehicle weight; and These are the moments of inertia for yaw and tilt, respectively. This refers to the vehicle body roll angle; and Equivalent roll damping and stiffness; For the sprung mass; This is the vertical distance from the center of roll to the center of mass; Let be the acceleration due to gravity. The fourth equation above describes the coupling between roll dynamics and planar motion, where the lateral acceleration term... It is the main source of excitation for lateral tilting motion.

[0102] Step S204: Based on the coupled dynamics model, define a dimensionless feature state vector that includes the vehicle's center of gravity sideslip angle, the rate of change of the center of gravity sideslip angle, the yaw rate error, the difference in the average sideslip angles of the front and rear axles representing the understeer or oversteer trend, and the difference in the sideslip angles of the left and right wheels of the front and rear axles representing the difference in tire grip utilization caused by load transfer; define the space in which the dimensionless feature state vector is located as a high-dimensional feature space.

[0103] Specifically, accurate representation of tire forces is crucial to the reliability of the dynamic model. This includes the vertical load on each tire. It is dynamically changing, consisting of the superposition of static load and transferred load caused by inertial force:

[0104]

[0105] in, For static loads, and They are respectively lateral acceleration and longitudinal acceleration Calculated lateral and longitudinal load transfer amounts. Tire lateral force. The model is constructed using a coupled Pacejka magic formula model, which expresses tire force as the vehicle's center of gravity sideslip angle. slip ratio Dynamic vertical load and tire-road adhesion coefficient Functions:

[0106]

[0107] in, For shape parameters, and peak factor and and Proportional. Tire slip angle Calculated independently based on the velocity vector at the center of each wheel.

[0108] Based on the above model, the complete state vector upon which this scheme depends is defined. :

[0109]

[0110] This vector integrates vehicle motion, attitude, and tire interface information. To focus on stability analysis, a reduced-dimensional feature state vector is further derived. :

[0111]

[0112] in, It is the centroid sideslip angle; This represents the rate of change of the centroid sideslip angle. This refers to the yaw rate error. The difference between the average sideslip angles of the front and rear axles reflects the tendency for understeer or oversteer. and These represent the differences in side slip angles of the front and rear axles and the left and right wheels, respectively, reflecting the differences in grip utilization caused by load transfer. Feature Space This forms a complete coordinate system for subsequent stability manifold analysis and risk assessment.

[0113] In this embodiment, firstly, by establishing a coupled dynamics model that includes the vehicle body roll degree of freedom, the limitations of the traditional two-degree-of-freedom linear model are overcome. This model incorporates the roll motion, which characterizes rollover risk, and the planar motion, which leads to skidding / fishtailing, into a unified equation description. This allows the system to simultaneously characterize the dynamic coupling effects between different instability modes (such as rollover and skidding).

[0114] Secondly, the defined feature state vectors extract key state dimensions with clear physical meaning and mutual decoupling from the coupled model. For example, the center-of-gravity sideslip angle and the rate of change of the center-of-gravity sideslip angle directly characterize the vehicle's sideslip behavior; the yaw rate error reflects steering follow-up performance; the difference between the average sideslip angles of the front and rear axles quantifies the understeer / oversteer trend; and the difference between the sideslip angles of the left and right wheels of the front and rear axles dynamically reflects the imbalance in tire adhesion distribution caused by load transfer. These features together constitute a mathematical space that is both dimensionality-reduced and information-complete.

[0115] Ultimately, the construction of this high-dimensional feature space enables subsequent stability analysis to handle multiple factors such as lateral, yaw, roll, and tire force nonlinearity within a unified framework. This addresses the systemic shortcomings of traditional methods, such as poor adaptability to operating conditions, delayed early warning, and insufficient mode differentiation capabilities, resulting from model simplification or a single state dimension, in order to achieve data-driven online approximation of stable manifolds, extraction of multi-dimensional risk features, and accurate identification of instability modes.

[0116] In one exemplary embodiment, such as Figure 3 As shown, based on multi-source sensor data, the estimated value of the vehicle's dynamic state in the high-dimensional feature space is obtained, including:

[0117] Step S302: Discretize the coupled dynamics model and establish nonlinear state transition equations and observation equations that include vehicle state and tire-road adhesion parameters.

[0118] Specifically, by discretizing the continuous-time coupled dynamics model, the nonlinear state transition equations are obtained:

[0119]

[0120] in, This includes control inputs for steering wheel angle and master drive / brake commands. This is process noise.

[0121] Observation equations , where the observation vector Including directly measurable signals such as yaw rate, longitudinal / lateral acceleration, and wheel speed. To observe noise. and Assuming the noise is zero-mean Gaussian white noise, the covariance matrices are as follows: and .

[0122] In step S304, the Kalman filter algorithm is used to process the state transition equation and the observation equation. In this step, the state prediction covariance matrix is ​​modified by introducing a time-varying fading factor to enhance the tracking ability of system state and parameter abrupt changes.

[0123] Step S306: Based on multi-source sensor data, recursive estimation is performed using a strong tracking capacitive Kalman filter algorithm to output the estimated value of the vehicle's dynamic state in a high-dimensional feature space.

[0124] Specifically, the ability to track abrupt state changes is enhanced by introducing a time-varying fading factor to adjust the prediction covariance online. The core steps are as follows:

[0125] a) Initialization: Set initial values ​​for vehicle dynamic state estimation. and state prediction covariance matrix .

[0126] b) Time Update (Prediction):

[0127] 1. Calculate the volume point: , ,in Let be the state dimension.

[0128] 2. Propagation volume point: .

[0129] 3. Calculate the predicted state and covariance:

[0130]

[0131]

[0132] c) Calculation of fading factor and correction of state prediction covariance:

[0133] 1. Calculate the state prediction covariance matrix Estimation and intermediate matrix and .

[0134] 2. Calculate the time-varying fading factor: ,in Represents the locus of the matrix.

[0135] 3. Corrected state prediction covariance matrix: .

[0136] d) Measurement update (calibration):

[0137] 1. Based on Recalculate the volume point and propagate it to the observation space: .

[0138] 2. Calculate the Kalman gain And update the state estimate and covariance:

[0139]

[0140]

[0141] in, This represents the state prediction covariance matrix at time step k.

[0142] This algorithm can effectively handle conditions such as tire force saturation or sudden changes in road adhesion, and outputs high-precision vehicle dynamic state estimates in real time. .

[0143] In this embodiment, a strong tracking commensurate Kalman filter (STCKF) is used as the core estimation algorithm. This combines the high-precision approximation capability of commensurate Kalman filtering for general nonlinear systems with the rapid adaptive characteristics of strong tracking filtering for abrupt changes in conditions. By calculating online and introducing a time-varying fading factor to dynamically correct the prediction covariance, this algorithm effectively overcomes the estimation lag or divergence problems that occur in traditional filtering methods when encountering strongly nonlinear or parameter jump conditions such as tire force saturation or sudden changes in road adhesion. This ensures that even under extreme driving or low-adhesion road conditions, the system can still maintain stable and rapid tracking of key vehicle states (such as the center of gravity sideslip angle) and key parameters (such as the adhesion coefficients of each wheel).

[0144] In one exemplary embodiment, such as Figure 4 As shown, for the current vehicle driving conditions, the local stable manifold boundary representing the vehicle's stable motion limit in the high-dimensional feature space is determined, including:

[0145] Step S402: Based on the current vehicle speed, road surface adhesion estimation, and steering input parameters, retrieve the set of sample points in the neighboring state of the current working condition from the pre-stored full-condition historical state sample database.

[0146] Step S404: The local linear embedding algorithm is used to reduce the dimensionality of the neighboring state sample point set to obtain the coordinates and corresponding projection transformation relationships in the low-dimensional local embedding space.

[0147] Step S406: In the low-dimensional local embedding space, based on the stable sample points in the neighboring state sample point set, the minimum bounding hypersphere is calculated using the support vector domain description method, and the spherical surface of the minimum bounding hypersphere is constructed as an approximate expression of the local stable manifold boundary under the current working condition.

[0148] Step S408: Based on the projection transformation relationship, the real-time vehicle dynamic state estimate is mapped to a low-dimensional local embedding space, the Euclidean distance from the mapping point to the center of the minimum bounding hypersphere is calculated, and the distance value is used as a geometric metric to quantify the deviation of the real-time vehicle state from the boundary of the local stable manifold.

[0149] Specifically, after obtaining the high-dimensional feature state estimate, its stability needs to be evaluated within a unified framework.

[0150] For nonlinear dynamic systems The attraction domain boundary of its stable equilibrium point (or stable periodic solution) forms a low-dimensional differential manifold in the state space, i.e., a stable manifold. The system trajectory is in One side converges to a stable motion, while the other side diverges and becomes unstable. Therefore, It is the precise boundary of the stable domain in a high-dimensional space.

[0151] because Since the analytical form is difficult to obtain, a data-driven approach is used for modeling. Through high-fidelity simulations and experiments covering all operating conditions, a massive amount of state-label pairs are collected. , among which tags These correspond to the risks of stability, sideslip, drift, and rollover, respectively. A deep autoencoder is used to learn from a high-dimensional feature space. To low-dimensional potential space nonlinear mapping and its inverse mapping In potential space In this process, a classifier (such as a support vector machine or a deep neural network) is trained using labeled data to fit the boundaries between stable regions and unstable regions; these boundaries are called stable regions. Expression in latent space .

[0152] When running online, based on the current operating parameters ( , , The system retrieves neighboring sample points from the historical database. This database covers the vehicle's operating status under various possible speed, road surface adhesion coefficients (e.g., dry asphalt, wet, icy), and steering input combinations. Each data point not only contains the aforementioned operating condition parameters but, more importantly, includes high-dimensional state features (e.g., sideslip angle, yaw rate) collected or calculated under that condition, along with their corresponding stability labels (stable, sideslip, fishtailing, rollover risk, etc.). Using the current real-time operating condition feature vector as the query condition, the system retrieves a set of neighboring state sample points in the speed, adhesion, and steering input spaces of the historical database.

[0153] A fast local approximation is achieved by combining Local Linear Embedding (LLE) with Support Vector Domain Description (SVDD):

[0154] a) Local dimensionality reduction: For the retrieved set of state sample points... The low-dimensional local embedding is obtained by applying the LLE algorithm. and projection matrix.

[0155] b) Boundary fitting: In the local embedding space, SVDD is used to fit stable sample points. Compute a minimally bounding hypersphere whose surface forms the boundary of a locally stable manifold. .

[0156] c) State mapping and distance metric: mapping the current state... Obtained by mapping to the local embedding space using the same LLE projection. .calculate Euclidean distance to the center of the SVDD hypersphere And the signed distance to the sphere. The absolute value of this signed distance can be approximated as the distance from the state point to the higher-dimensional stable manifold. Local geodesic distance , which is a geometric metric value, with a sign indicating whether the state point is inside (negative) or outside (positive) the stability region.

[0157]

[0158]

[0159] when When, it means Located inside the hypersphere, the corresponding vehicle state is in a stable region;

[0160] when When, it means Located outside the hypersphere, the corresponding vehicle state has deviated from the stable region, posing a risk of instability.

[0161] when When, it means It falls precisely on the stability boundary.

[0162] In this embodiment, by retrieving neighboring sample points from the historical database based on the current operating conditions (vehicle speed, road surface adhesion, steering input), it is ensured that the constructed stable boundary can adapt to specific driving conditions, rather than relying on a globally fixed simplified model or threshold, thus solving the problem of inaccurate evaluation caused by model mismatch in traditional methods.

[0163] Secondly, by employing the Locally Linear Embedding (LLE) algorithm for intelligent dimensionality reduction, the system transforms the complex manifold boundary fitting problem into a low-dimensional space while preserving the local data structure (i.e., the proximity relationship between stable and unstable samples in high-dimensional space) to the greatest extent possible. This not only significantly reduces computational complexity, meeting the requirements of in-vehicle real-time systems, but also allows for the computation of a smooth, compact minimum-enclosing hypersphere using the Support Vector Domain Description (SVDD) method in low-dimensional space. This hypersphere effectively encloses stable sample points, and its surface naturally constitutes a good approximation of the stable region boundary, achieving an efficient mathematical representation of high-dimensional complex stable domains.

[0164] Finally, by mapping the real-time state estimates to the same low-dimensional space and calculating their Euclidean distance to the center of the hypersphere, the system obtains an intuitive, continuous, and physically meaningful quantitative stability index, namely the geometric metric.

[0165] In an exemplary embodiment, the stability risk characteristics include manifold geometric risk characteristics, tire force saturation risk characteristics, and roll motion risk characteristics. Based on the multi-dimensional stability risk characteristics, a comprehensive risk level is generated through a probabilistic fusion model, and the dominant instability mode is identified, including:

[0166] The manifold geometric risk features, the tire force saturation risk features, and the roll motion risk features are all input into the Bayesian neural network, which outputs the posterior probability distribution data of each preset mode, including the vehicle being in a stable state, sideslip mode, fishtailing mode, and rollover risk mode.

[0167] Based on the posterior probability distribution data and the severity weights of each preset instability mode, a comprehensive risk level is calculated, and the mode with the highest posterior probability is selected as the dominant instability mode.

[0168] Specifically, the following feature vectors are extracted from different physical perspectives. :

[0169] a) Geometric features of the manifold: (distance to manifold), (Rate of change of distance).

[0170] b) Local dynamic characteristics: In System Perform Jacobian matrix linearization and compute its maximum Lyapunov exponent. As a quantitative indicator of the local divergence of the trajectory, .

[0171] c) Tire force saturation characteristics: Calculate the coefficient of adhesion for each tire. Statistical features such as maximum values ​​and inter-axis differences are extracted to form a feature subset. .

[0172] d) Characteristics of lateral tilting motion: , , (Pan angle, angular velocity, and angular energy).

[0173] Define the system mode set Construct a lightweight Bayesian neural network (BNN) using the aforementioned feature vectors. The input is the mode. The output of this BNN is the modality. posterior probability distribution It also provides a measure of the uncertainty in the forecast. Ultimately, a comprehensive global risk level is determined. The following formula is used for fusion calculation:

[0174]

[0175] in, These are the normalized weighting coefficients. For distance scale parameters, To reflect the severity weights of each instability mode ( ).

[0176] Simultaneously output the dominant instability mode .

[0177] In this embodiment, firstly, by inputting four types of risk features derived from different physical mechanisms—manifold geometric features characterizing the relationship between the state and the stable boundary space, dynamic features reflecting the divergence trend of the system's local dynamics, tire force saturation features indicating the degree of tire adhesion depletion, and roll motion features quantifying the accumulation of vehicle roll energy—in parallel into a unified Bayesian neural network, an intelligent fusion framework capable of handling complex nonlinear correlations and uncertainties between features is established. This framework overcomes the limitations of traditional weighted averaging or rule-based judgment methods in accurately characterizing high-order coupling relationships between features, achieving deep integration of multidimensional heterogeneous information.

[0178] Secondly, the Bayesian neural network takes the aforementioned features as input and outputs not a single judgment result, but a complete posterior probability distribution of the vehicle in each preset mode (stable, sideslip, fishtailing, and rollover risk). This output format has a dual advantage: on the one hand, it quantifies the system's confidence in different judgment results, and its own output uncertainty measure (such as the entropy of the probability distribution) can serve as an additional indicator of risk reliability; on the other hand, the probabilistic output naturally possesses the ability to handle uncertainties such as sensor noise, model errors, and feature conflicts, significantly improving the robustness of pattern recognition in complex and ambiguous conditions.

[0179] Finally, based on the posterior probability distribution, the system generates a scalarized comprehensive risk level through fusion calculation (e.g., weighting the posterior probabilities of corresponding modes by weighting the severity of each instability mode). This level is no longer a reflection of a single physical quantity, but a global evaluation index that comprehensively considers all risk dimensions, different instability probabilities, and their severity. Simultaneously, by selecting the mode with the highest posterior probability as the dominant instability mode, the system achieves accurate identification of the specific manifestations of potential hazards.

[0180] In an exemplary embodiment, when the overall risk level exceeds a preset threshold, a corresponding predictive collaborative chassis control strategy is triggered based on the dominant instability mode, including:

[0181] When the dominant instability mode is sideslip, a first additional yaw moment is generated with the goal of correcting the sideslip angle and the rate of change of the sideslip angle, and the drive system, braking system and steering system are coordinated to achieve the moment through differentiated torque distribution and angle compensation.

[0182] When the dominant instability mode is the tail-slip mode, a second additional yaw moment is generated with the goal of suppressing the yaw rate error and its divergence trend. In the moment calculation, a feedforward compensation term based on the maximum Lyapunov exponent is introduced. The second additional yaw moment is applied together by coordinating the directional differential braking function of the braking system and the reverse compensation function of the steering system.

[0183] When the dominant instability mode is the rollover risk mode, the suspension system damping is maximized to dissipate roll energy, and longitudinal uniform braking is controlled to reduce vehicle speed.

[0184] Specifically, when the comprehensive risk level When the preset threshold is exceeded, the dominant instability mode is determined. The system triggers the corresponding predictive cooperative control strategy. For different identified instability modes, the system invokes the corresponding predictive cooperative control law, the core of which is to generate one or more generalized forces (torques) and apply this control quantity through the coordinated distribution of chassis actuators, thereby pulling the vehicle state back into a stable manifold.

[0185] For the sideslip-dominant mode ( The control objective is to generate a value used to correct the centroid sideslip angle. The first additional yaw moment and its changing trend Taking PID control as an example, the target torque is calculated using the following control law, which includes proportional and derivative feedback: ,in and To control the gain, and to achieve this torque requirement, the vehicle's torque vectoring system and steer-by-wire system work together to generate differentiated drive or braking torques on the front and rear axles, supplemented by small steering angle compensation, so that the total yaw torque increment contributed by each actuator accurately approximates the target torque. In addition, the system can pre-pressurize the front wheel brake cylinders on the outer side of the steering wheel in advance, preparing for possible emergency intervention.

[0186] When the system determines that the drift-dominant mode is ( When this happens, the control objective becomes rapidly suppressing the yaw rate error. And to counteract its divergent tendency. For this, a second additional yaw moment is required. The following control law is given: This formula not only includes feedback on the yaw rate error and its differential term (… , To provide the corresponding gain, a local maximum Lyapunov exponent was also introduced. Feedforward term (gain is This allows for predictive compensation of the system's divergent kinetic energy. This torque is primarily generated directly through the vehicle's torque vectoring system or Electronic Stability Program (ESP) applying active braking (coordinating the directional differential braking function of the braking system) to the outer front wheels in the direction of rotation; simultaneously, the Electric Power Steering (EPS) system adds a small counter-compensation steering angle. This helps stabilize the vehicle's posture.

[0187] Facing the rolling risk pattern ( The core of the control strategy is to rapidly dissipate the continuously accumulating roll energy and reduce vehicle speed to minimize the source of centrifugal force. This is achieved through coordinated intervention of the suspension system and longitudinal dynamics simultaneously. In terms of suspension control, the continuously adjustable damping shock absorber (CDC) is instantaneously switched to maximize its damping coefficient. This maximizes the power dissipation for roll oscillations. In terms of longitudinal force control, the drive system reduces torque to cut off power input, while ESP applies a force equal to the real-time estimated lateral acceleration to all wheels. Proportional uniform braking force ( (This is a proportionality coefficient), thereby achieving smooth and efficient deceleration and fundamentally reducing the lateral forces that cause roll.

[0188] In this embodiment, firstly, based on the dominant instability mode (sideslip, fishtailing, or rollover risk) identified by the Bayesian network, a precisely matched control law is invoked from a pre-set strategy library. For example, for sideslip caused by tire lateral force saturation, the focus is on correcting the vehicle's attitude through torque vector distribution; for fishtailing caused by yaw divergence, the focus is on directly suppressing rotational kinetic energy through differential braking; and for rollover risk caused by the accumulation of roll energy, the focus is on managing vertical and longitudinal dynamics through the coordinated use of suspension and braking. This fundamentally solves the problems of low correction efficiency or conflicting control effects under complex conditions caused by the single control objective and coarse intervention methods in traditional systems.

[0189] Secondly, based on the generated feedback torque for suppressing yaw error, a feedforward compensation term based on real-time calculation of the maximum Lyapunov exponent is introduced. This exponent directly quantifies the divergence rate of the system's local dynamics. Introducing it into the torque calculation allows the control system to inject active damping in advance that is equal in magnitude and opposite in direction to the system's spontaneous instability tendency, thereby achieving proactive cancellation of the instability tendency and significantly improving the intervention speed and stability in the initial stage of instability.

[0190] Ultimately, this solution achieves dynamic collaborative optimization of multiple chassis actuators. For different modes, the system does not independently drive individual actuators, but rather dynamically coordinates the drive, braking, steering, and suspension systems based on the principle of physical optimization. For example, when correcting sideslip, it simultaneously adjusts drive torque and steering compensation; when suppressing fishtailing, it coordinates differential braking and reverse steering; and when preventing rollover, it simultaneously maximizes suspension damping and applies uniform braking force. This collaboration not only leverages the unique advantages of each actuator, forming a combined control force, but more importantly, by comprehensively allocating control tasks, it avoids overloading individual systems (such as the braking system), ensuring smooth control and actuator lifespan, and achieving global optimization of vehicle dynamics safety.

[0191] The most detailed embodiment of this application is as follows:

[0192] like Figure 5 As shown, firstly, the system receives data input from multiple sensors located throughout the vehicle. This data includes, but is not limited to, longitudinal acceleration, lateral acceleration, and yaw rate collected by the inertial measurement unit; wheel speeds collected by the wheel speed sensors; steering wheel angle collected by the steering angle sensor; and auxiliary information provided by sensors such as the vehicle height sensor and roll angle sensor. These raw signals constitute the direct source for the system to perceive the vehicle's motion state.

[0193] Subsequently, the high-dimensional state joint estimation module begins operation. This module is centered on an extended coupled vehicle dynamics model that includes the vehicle's roll degree of freedom. This model details the vehicle's planar motion, roll motion, and the nonlinear mechanical characteristics of each tire. The module employs a strong tracking capacitive Kalman filter algorithm to deeply fuse the aforementioned sensor data with the dynamics model. By introducing a time-varying fading factor, this algorithm adaptively adjusts the prediction covariance online, thus maintaining robust tracking and accurate estimation of the system state and key parameters (such as the road adhesion coefficient of each wheel) even under highly nonlinear conditions such as sudden changes in tire force or road adhesion. This process outputs a high-dimensional state vector in real time, containing key states that cannot be directly measured, such as the vehicle's center of gravity sideslip angle, the real-time sideslip angle of each tire, and vertical loads, thereby reconstructing a complete dynamic picture of the vehicle's internal structure.

[0194] Next, the stability feature extraction module calculates and derives a dimensionality-reduced feature state vector specifically for stability analysis based on this high-dimensional state estimate. This vector is carefully designed and includes the following core features: vehicle center of gravity sideslip angle, rate of change of center of gravity sideslip angle, yaw rate tracking error, the difference between the average sideslip angles of the front and rear axles reflecting understeer / oversteer trends, and the difference between the sideslip angles of the left and right wheels of the front and rear axles characterizing uneven tire grip utilization caused by load transfer. These features together constitute a low-dimensional space with complete information and clear physical meaning, serving as the coordinate basis for subsequent in-depth analysis.

[0195] Building upon this foundation, the high-dimensional stable manifold online analysis module is initiated. This module does not solve for complex global stability boundaries but instead employs an efficient local data-driven approach. First, based on the current real-time vehicle speed, road adhesion estimation, and steering input, it quickly retrieves a set of nearest neighboring sample points from a pre-stored historical state sample database covering all operating conditions. Then, it applies a local linear embedding algorithm to intelligently reduce the dimensionality of this local sample point set, obtaining its coordinates and projection relationships in a low-dimensional local embedding space. In this low-dimensional space, the module uses the support vector domain description method to calculate a minimum bounding hypersphere for the retrieved stable sample points, and constructs the surface of this hypersphere as an approximate expression of the "stability domain" boundary under the current operating condition—that is, the local stable manifold. Finally, the real-time estimated vehicle characteristic state is mapped to this low-dimensional space through the same projection relationship, and its Euclidean distance to the center of the hypersphere is calculated. This distance value is the core geometric metric for quantifying the deviation of the vehicle state from the stability boundary.

[0196] Subsequently, the multi-dimensional risk feature calculation module operates in parallel, extracting risk indicators from different physical aspects. It simultaneously calculates: ① manifold geometric risk features, namely the Euclidean distance and its rate of change, quantifying the spatial location and motion trend of the state point; ② local dynamic risk features, assessing the system's local stability by performing Jacobian linearization on the system at real-time state points and calculating its maximum Lyapunov exponent; ③ tire force saturation risk features, calculating the adhesion coefficient of each tire using a tire model and extracting its maximum value, axle difference, and other statistics to reflect the degree of depletion of physical constraints; ④ roll motion risk features, including vehicle roll angle, roll rate, and roll energy, specifically used to assess rollover risk. These features are calculated in parallel and combined into a multi-dimensional risk feature vector.

[0197] Subsequently, the intelligent risk fusion and pattern recognition module receives this multi-dimensional feature vector. At the core of this module is a trained lightweight Bayesian neural network. This network takes the aforementioned multi-dimensional features as input, performs probabilistic inference, and outputs the posterior probability distribution of the vehicle's current state as a preset mode, including "stable state," "sideslip mode," "fishtailing mode," and "rollover risk mode." Based on this probability distribution and preset severity weights for each mode, the module calculates a scalarized comprehensive risk level and selects the mode with the highest posterior probability as the dominant instability mode.

[0198] The system then proceeds to the risk decision-making and pattern assessment phase. If the overall risk level is low and the vehicle remains in a stable state during continuous monitoring, the system will not intervene actively. If the risk level rises to the warning threshold but does not reach the control threshold, the system triggers a primary warning and provides a prompt to the driver through the human-machine interface. When the overall risk level exceeds the preset high-risk control threshold, the system enters a high-risk+ pattern recognition state and triggers the corresponding predictive collaborative control strategy based on the identified dominant instability pattern.

[0199] The specific collaborative control strategies vary highly depending on the mode: If the dominant mode is sideslip, a first additional yaw moment is generated to correct the sideslip angle and the rate of change of the sideslip angle. This is achieved collaboratively through the coordinated torque vector distribution of the drive system, the differential braking of the braking system, and the angle compensation of the steering system. If the dominant mode is drift, a second additional yaw moment is generated to suppress the yaw rate error and its divergence trend. This moment calculation incorporates a feedforward compensation term based on the aforementioned maximum Lyapunov exponent to predictively offset the divergent kinetic energy of the system. It is applied jointly through the coordinated directional differential braking of the braking system and the reverse compensation of the steering system. If the dominant mode is rollover risk, two core operations are executed simultaneously: first, the continuously adjustable damping shock absorber is switched to the maximum damping mode to quickly dissipate roll energy; second, the drive system is controlled to reduce torque and the electronic stability program is instructed to apply uniform braking force proportional to the real-time lateral acceleration to all wheels to reduce vehicle speed and reduce roll excitation at its source.

[0200] Ultimately, the chassis actuator coordination module executes the specific instructions generated by the aforementioned control strategy. Subsystems such as drive, braking, steering, and suspension coordinate their operations according to precise allocation logic, generating the required generalized forces and torques. These control quantities act on the vehicle, causing its dynamic response to change in the desired direction. The goal is to pull the vehicle's high-dimensional state point back from the danger zone and stabilize it within the safe domain defined by the local stable manifold, thereby achieving proactive suppression of instability trends and active preservation of vehicle stability, completing a full closed loop from state perception and intelligent assessment to predictive coordinated control.

[0201] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0202] Based on the same inventive concept, this application also provides a vehicle stability assessment and warning device for implementing the vehicle stability assessment and warning method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more vehicle stability assessment and warning device embodiments provided below can be found in the limitations of the vehicle stability assessment and warning method described above, and will not be repeated here.

[0203] In one exemplary embodiment, such as Figure 6 As shown, a vehicle stability assessment and early warning device is provided, comprising:

[0204] Module 602 is used to construct a high-dimensional feature space that integrates the vehicle's lateral, yaw, roll, and tire dynamic states.

[0205] Processing module 604 is used to obtain vehicle dynamic state estimates in a high-dimensional feature space based on multi-source sensor data;

[0206] The processing module 604 is also used to determine the local stable manifold boundary in the high-dimensional feature space that characterizes the vehicle's stable motion limit for the current vehicle driving conditions.

[0207] The extraction module 606 is used to extract stability risk features of multiple dimensions in parallel based on the geometric relationship and dynamic evolution trend between the vehicle dynamic state estimate and the local stable manifold boundary.

[0208] The processing module 604 is also used to generate a comprehensive risk level based on stability risk characteristics of multiple dimensions through a probability fusion model, and to identify the dominant instability mode;

[0209] The early warning module 608 is used to trigger a corresponding predictive collaborative chassis control strategy based on the dominant instability mode when the comprehensive risk level exceeds a preset threshold, so as to adjust the vehicle state.

[0210] In an exemplary embodiment, the construction module 602 is specifically used to establish a vehicle coupled dynamics model that includes the vehicle body roll degree of freedom. The vehicle coupled dynamics model is used to describe the coupling relationship between the vehicle's planar motion and roll motion, as well as the dynamic characteristics of each tire. Based on the coupled dynamics model, a dimensionless feature state vector is defined, which includes the vehicle's center of gravity sideslip angle, the rate of change of the center of gravity sideslip angle, the yaw rate error, the difference between the average sideslip angles of the front and rear axles representing the understeer or oversteer tendency, and the difference between the sideslip angles of the left and right wheels of the front and rear axles representing the difference in tire grip utilization caused by load transfer. The space in which the dimensionless feature state vector is located is defined as a high-dimensional feature space.

[0211] In an exemplary embodiment, the processing module 604 is specifically used to discretize the coupled dynamics model and establish nonlinear state transition equations and observation equations that include vehicle state and tire-road adhesion parameters; to process the state transition equations and observation equations using a Kalman filter algorithm, wherein a time-varying fading factor is introduced to correct the state prediction covariance matrix to enhance the tracking ability of system state and parameter abrupt changes; and to perform recursive estimation based on multi-source sensor data using the Kalman filter algorithm to output the vehicle dynamic state estimate in a high-dimensional feature space.

[0212] In an exemplary embodiment, the construction module 602 is specifically used to retrieve a set of neighboring state sample points from a pre-stored full-condition historical state sample database based on the current vehicle speed, road surface adhesion estimation, and steering input parameters; to perform dimensionality reduction processing on the neighboring state sample point set using a local linear embedding algorithm to obtain the coordinates and corresponding projection transformation relationships in the low-dimensional local embedding space; in the low-dimensional local embedding space, based on the stable sample points in the neighboring state sample point set, to calculate the minimum bounding hypersphere using the support vector domain description method, and to construct the surface of the minimum bounding hypersphere as an approximate expression of the local stable manifold boundary under the current condition; according to the projection transformation relationship, to map the real-time acquired vehicle dynamic state estimate to the low-dimensional local embedding space, to calculate the Euclidean distance from the mapping point to the center of the minimum bounding hypersphere, and to use the distance value as a geometric metric value to quantify the deviation of the real-time vehicle state from the local stable manifold boundary.

[0213] In an exemplary embodiment, the stability risk features include manifold geometric risk features, tire force saturation risk features, and roll motion risk features. These features are input into a Bayesian neural network, which outputs posterior probability distribution data for each preset mode, including stable state, sideslip mode, drift mode, and rollover risk mode. Based on the posterior probability distribution data and the severity weights of each preset instability mode, a comprehensive risk level is calculated, and the mode with the highest posterior probability is selected as the dominant instability mode.

[0214] In an exemplary embodiment, the warning module 608 is specifically configured to generate a first additional yaw moment aimed at correcting the sideslip angle and the rate of change of the sideslip angle when the dominant instability mode is sideslip mode, and coordinate the drive system, braking system, and steering system to achieve torque through differentiated torque distribution and angle compensation; when the dominant instability mode is tail-slip mode, generate a second additional yaw moment aimed at suppressing the yaw rate error and its divergence trend, introduce a feedforward compensation term based on the maximum Lyapunov exponent in the torque calculation, and jointly apply the second additional yaw moment by coordinating the directional differential braking function of the braking system and the reverse compensation function of the steering system; when the dominant instability mode is rollover risk mode, simultaneously control the suspension system to maximize damping to dissipate roll kinetic energy, and control longitudinal uniform braking to reduce vehicle speed.

[0215] The various modules in the aforementioned vehicle stability assessment and early warning device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0216] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 7As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores current multi-source sensor data and vehicle driving condition data. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a vehicle stability assessment and early warning method.

[0217] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0218] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0219] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0220] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the method described above.

[0221] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0222] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0223] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0224] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for assessing and warning about vehicle stability, characterized in that, The method includes: Construct a high-dimensional feature space that integrates vehicle lateral, yaw, roll, and tire dynamic states; Based on multi-source sensor data, the estimated value of the vehicle dynamic state in the high-dimensional feature space is obtained; For the current vehicle driving conditions, determine the local stable manifold boundary in the high-dimensional feature space that characterizes the vehicle's stable motion limit; Based on the geometric relationship and dynamic evolution trend between the estimated vehicle dynamic state and the local stable manifold boundary, stability risk features of multiple dimensions are extracted in parallel. Based on the stability risk characteristics of the multiple dimensions, a comprehensive risk level is generated through a probabilistic fusion model, and the dominant instability mode is identified. When the overall risk level exceeds a preset threshold, a corresponding predictive collaborative chassis control strategy is triggered based on the dominant instability mode to adjust the vehicle state.

2. The method according to claim 1, characterized in that, The construction of a high-dimensional feature space integrating vehicle lateral, yaw, roll, and tire dynamic states includes: A vehicle coupled dynamics model is established, which includes the vehicle body roll degree of freedom. The vehicle coupled dynamics model is used to describe the coupling relationship between the vehicle's planar motion and roll motion, as well as the dynamic characteristics of each tire. Based on the coupled dynamics model, a dimensionless feature state vector is defined, which includes the vehicle's center of gravity sideslip angle, the rate of change of the center of gravity sideslip angle, the yaw rate error, the difference in the average sideslip angle between the front and rear axles representing the understeer or oversteer tendency, and the difference in the sideslip angle between the left and right wheels of the front and rear axles representing the difference in tire grip utilization caused by load transfer. The space containing the reduced-dimensional feature state vector is defined as the high-dimensional feature space.

3. The method according to claim 2, characterized in that, The step of obtaining the vehicle dynamic state estimate in the high-dimensional feature space based on multi-source sensor data includes: The coupled dynamics model is discretized to establish nonlinear state transition equations and observation equations that include vehicle state and tire-road adhesion parameters; The Kalman filter algorithm is used to process the state transition equation and the observation equation. The state prediction covariance matrix is ​​modified by introducing a time-varying fading factor to enhance the tracking ability of system state and parameter abrupt changes. Based on the multi-source sensor data, the vehicle dynamic state estimate is output in the high-dimensional feature space through recursive estimation using the Kalman filter algorithm.

4. The method according to claim 3, characterized in that, The step of determining the local stable manifold boundary in the high-dimensional feature space that characterizes the vehicle's stable motion limit for the current vehicle driving condition includes: Based on the current vehicle speed, road surface adhesion estimation, and steering input parameters, retrieve the set of sample points in the adjacent state from the pre-stored full-condition historical state sample database; The local linear embedding algorithm is used to reduce the dimensionality of the neighboring state sample point set to obtain the coordinates and corresponding projection transformation relationships in the low-dimensional local embedding space. In the low-dimensional local embedding space, based on the stable sample points in the set of neighboring state sample points, the minimum bounding hypersphere is calculated using the support vector domain description method, and the spherical surface of the minimum bounding hypersphere is constructed as an approximate expression of the local stable manifold boundary under the current working condition. Based on the projection transformation relationship, the real-time vehicle dynamic state estimate is mapped to the low-dimensional local embedding space. The Euclidean distance from the mapping point to the center of the minimum enclosing hypersphere is calculated, and the distance value is used as a geometric metric to quantify the deviation of the real-time vehicle state from the boundary of the local stable manifold.

5. The method according to claim 1, characterized in that, The stability risk characteristics include manifold geometric risk characteristics, tire force saturation risk characteristics, and roll motion risk characteristics; based on these multiple dimensions of stability risk characteristics, a comprehensive risk level is generated through a probabilistic fusion model, and the dominant instability mode is identified, including: The manifold geometric risk features, the tire force saturation risk features, and the roll motion risk features are all input into a Bayesian neural network, which outputs the posterior probability distribution data of each preset mode, including the vehicle being in a stable state, sideslip mode, fishtailing mode, and rollover risk mode. Based on the posterior probability distribution data and the severity weights of each preset instability mode, the comprehensive risk level is calculated, and the mode with the highest posterior probability is selected as the dominant instability mode.

6. The method according to claim 5, characterized in that, When the overall risk level exceeds a preset threshold, a corresponding predictive collaborative chassis control strategy is triggered based on the dominant instability mode, including: When the dominant instability mode is sideslip mode, a first additional yaw moment is generated with the goal of correcting the sideslip angle and the rate of change of the sideslip angle, and the drive system, braking system and steering system are coordinated to achieve the moment through differentiated torque distribution and angle compensation; When the dominant instability mode is the tail-slip mode, a second additional yaw moment is generated with the goal of suppressing the yaw rate error and its divergence trend. In the moment calculation, a feedforward compensation term based on the maximum Lyapunov exponent is introduced. The second additional yaw moment is applied together by coordinating the directional differential braking function of the braking system and the reverse compensation function of the steering system. When the dominant instability mode is the rollover risk mode, the suspension system damping is maximized to dissipate roll energy, and longitudinal uniform braking is controlled to reduce vehicle speed.

7. A vehicle stability assessment and early warning device, characterized in that, The device includes: The building module is used to construct a high-dimensional feature space that integrates the vehicle's lateral, yaw, roll, and tire dynamic states; The processing module is used to obtain the vehicle dynamic state estimate in the high-dimensional feature space based on multi-source sensor data; The processing module is also used to determine the local stable manifold boundary in the high-dimensional feature space that characterizes the vehicle's stable motion limit for the current vehicle driving conditions. The extraction module is used to extract stability risk features of multiple dimensions in parallel based on the geometric relationship and dynamic evolution trend between the vehicle dynamic state estimate and the local stable manifold boundary. The processing module is also used to generate a comprehensive risk level based on the stability risk characteristics of the multiple dimensions through a probability fusion model, and to identify the dominant instability mode; The early warning module is used to trigger a corresponding predictive collaborative chassis control strategy based on the dominant instability mode when the overall risk level exceeds a preset threshold, so as to adjust the vehicle state.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images