Vehicle handling stability evaluation method based on working condition state distribution

Abstract

The invention belongs to the technical field of vehicle handling stability control, and particularly relates to a vehicle handling stability evaluation method based on working condition state distribution. According to the evaluation method, a comprehensive stability evaluation method is designed by combining a multi-dimensional Markov transition probability matrix. According to the evaluation method, the probability distribution and the transfer characteristic of each working condition state are fully considered, it is ensured that the high-probability state has greater influence on overall evaluation, and comprehensive and dynamic evaluation of the vehicle handling stability is achieved. Compared with a traditional fixed working condition evaluation method, the method more accurately reflects the stability performance of the vehicle in a complex and changeable actual driving environment, has high adaptability and practicability, and provides a scientific and reliable analysis method based on actual data for optimization design and dynamic control of systems such as vehicle control and suspension.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle handling stability control, and particularly relates to a method for evaluating vehicle handling stability based on the distribution of operating conditions. Background Art

[0002] Vehicle handling stability refers to the ability of a vehicle to maintain an expected driving trajectory under various driving conditions, which is directly related to driving safety and comfort. Traditional methods for evaluating handling stability usually rely on fixed test condition parameters (such as the double lane change test), and cannot fully reflect the performance of the vehicle in complex and changing actual driving environments. As the types and parameter dimensions of the operating conditions faced by vehicles continue to increase, traditional methods have obvious deficiencies in terms of comprehensiveness and accuracy.

[0003] Traditional methods for evaluating vehicle handling stability usually rely on preset standard test conditions, such as the double lane change test, the slalom test, etc. These methods simulate the handling performance of the vehicle by setting fixed experimental conditions. The basic idea of these methods is to obtain the stability indicators of the vehicle by simulating specific driving scenarios or handling actions, and then evaluate the stability of the vehicle based on these indicators. This results in their inability to fully reflect the performance of the vehicle in various complex and dynamic environments. These methods usually cannot take into account the diverse operating conditions and changes that may be encountered in actual driving.

[0004] In traditional methods, many state partitions or threshold settings are artificial and lack the ability to adapt according to actual data. Manually set thresholds may not match the actual driving environment, resulting in a decrease in the accuracy of the evaluation. Traditional methods usually can only evaluate the stability of the vehicle under specific handling conditions and lack a comprehensive consideration of multi-dimensional driving conditions. For example, only considering longitudinal and lateral stability while ignoring yaw stability or other dynamic characteristics, resulting in incomplete evaluation results. Summary of the Invention

[0005] The present invention proposes a method for evaluating vehicle handling stability based on the distribution of operating conditions, which solves the problems of the above-mentioned traditional fixed-condition test methods, such as being unable to evaluate according to the probability distribution and transfer characteristics of actual operating conditions, having poor effect in evaluating vehicle dynamic stability, and insufficient adaptability and comprehensiveness of the evaluation results.

[0006] The technical solution of the present invention is realized as follows:

[0007] A method for evaluating vehicle handling stability based on the distribution of operating conditions, comprising the following steps:

[0008] S1. Data collection of actual vehicle operating conditions;

[0009] S2. Data preprocessing;

[0010] S3. Modeling of operating condition state distribution: adaptively adjust the boundary of state division according to the probability density of data; specifically including the following content:

[0011] S31. Probability density estimation; for the denoised one-dimensional operating condition data Perform probability density estimation based on kernel density estimation;

[0012] S32. Perform quantile division; according to the estimated probability density, automatically determine the state boundary using quantiles to achieve equal-probability division;

[0013] S4. Stability analysis; based on the operating condition state distribution, determine the stable state of the vehicle and score the stable state.

[0014] Through the above technical solutions, probability density estimation and quantile division can automatically determine appropriate state boundaries according to the distribution of actual operating condition data, ensuring that the divided states can accurately reflect the stability of the actual operating conditions. These steps ensure that the division of different operating condition states is scientific and adaptive. After completing the modeling of the operating condition state distribution, step S4 further evaluates the stability performance of the vehicle under various operating conditions by analyzing the transition probabilities under different states and scores each stability state. This process is the core of the present invention, which can quantitatively evaluate the handling stability of the vehicle according to the distribution and transition of states, and finally obtain the comprehensive stability score (CSS).

[0015] Optionally, in step S1, collect multi-dimensional operating condition data in real time to form an operating condition data vector;

[0016] X(t) = [x1(t), x2(t), …, x n (t)];

[0017] where, x i (t) represents the i-th operating condition parameter collected at time point t, and i takes values from 1 to n, where n is the total dimension of the operating condition parameters.

[0018] Through the above technical solutions, the above operation steps form a dynamic data vector including all key operating condition parameters, providing a data basis for subsequent modeling.

[0019] Optionally, in step S2, assume that x i (t) is the operating condition state data of the vehicle, and perform multi-layer discrete wavelet transform on the signal x i (t) to decompose it into approximate coefficients and detail coefficients of different scales:

[0020]

[0021] where, J is the decomposition level, φ J,k (t) and ψj,k (t) are the scaling function and the wavelet basis function respectively, and a J,k is the approximation coefficient of the J-th layer, and d j,k is the detail coefficient of the j-th layer;

[0022] Perform threshold processing on the obtained detail coefficient d j,k ;

[0023] d′ j,k = sign(d j,k ) · max(|d j,k | - λ j , 0);

[0024] Among them, d′ j,k is the processed detail coefficient, and λ j is the threshold of the j-th layer, using the ViSuShrink method:

[0025]

[0026] Among them, σ j is the noise standard deviation of the j-th layer, which can be estimated by the median absolute deviation: l refers to the detail coefficient level; n is the number of data points; N is the total number of data points;

[0027]

[0028] Use the processed detail coefficient d′ j,k and the original approximation coefficient a J,k to perform inverse discrete wavelet transform to reconstruct the denoised working condition data

[0029]

[0030] Through the above technical solution, through multi-layer discrete wavelet transform, the working condition state data can be decomposed into approximation coefficients and detail coefficients of different scales. This process can effectively distinguish different frequency components of the signal, so that high-frequency noise and low-frequency actual signals can be processed separately, and the noise components in the data can be removed.

[0031] Optionally, in step S31, use the kernel density estimation method f(x) to perform probability density estimation on the one-dimensional working condition data ;

[0032]

[0033] Among them, N is the number of data points, h is the bandwidth parameter that determines the smoothness of the kernel function, K(·) is the kernel function, is the denoised working condition data;

[0034]

[0035] Through the above technical solution, by performing kernel density estimation on the denoised working condition data, the distribution of the data can be accurately described.

[0036] Optionally, in step S32, set the number of states m to be divided i , and calculate the equal-probability quantiles:

[0037]

[0038] where is the inverse function of the cumulative distribution function of x i (t); k represents the index of the probability quantile, and m i is the number of states to be divided, and the working condition data is divided into m i states;

[0039] Divide the value range of x i (t) into m i equal-probability states:

[0040]

[0041] According to the state space of each working condition parameter, construct a multi-dimensional state space S by calculating the Cartesian product, represents the state interval corresponding to the last interval after the division of the i-th working condition parameter x i (t), represents the boundary of the interval after the division of the i-th working condition parameter x i (t);

[0042] S = S1 × S2 × … × S n = {s1 × s2 × … × s n};

[0043] where each state s = (s1, s2, …, s n ) represents a combined state of the multi-dimensional working condition. For each moment t k , map the multi-dimensional working condition data X(t k ) = [x1(t k ), x2(t k ), …, x n (t k )] to the corresponding state s(t k ) ∈ S: S n refers to the set of the divided state spaces corresponding to the n-th working condition parameter in the multi-dimensional state space, that is, the set of all state spaces related to the n-th working condition parameter;

[0044]

[0045] Among them, is the state corresponding to x i (t k ).

[0046] Count the number of transitions N i ,s j ) of each pair of multi-dimensional states (s ij :

[0047]

[0048] Among them, is the indicator function, and s(t k ) represents the multi-dimensional state to which the working condition vector data is mapped at time t k . Here, s i and s j are certain states in the multi-dimensional state space, and s(t k + 1) represents the multi-dimensional state to which the working condition vector data is mapped at the time point t k + 1;

[0049] Calculate the transition probability based on the number of transitions:

[0050]

[0051] Among them, m is the total number of states in the multi-dimensional state space, and N ij represents the number of transitions from state s i to state s j , that is, the number of times the working condition transfers from s i to s j at all times, and N i is the total number of transitions of state s i , that is, the sum of the number of times of transferring from state s i to all other states;

[0052] Construct the transition probability matrix:

[0053]

[0054] Among them, a high transition probability can reflect common working condition switches, and a low transition probability can reflect rare or abnormal working condition switches. If the probability is close to 1, it indicates a high transition probability, and if the probability is close to 0, it indicates a low transition probability.

[0055] Through the above technical solutions, the above operation steps ensure that the divided states are adaptive and can accurately reflect the distribution characteristics of the data. Compared with the fixed interval division, the equal probability division can better adapt to the actual distribution of the data, avoid the deviation of the artificially divided boundary, and make the division of the working condition data more scientific and reasonable. The transition probability matrix P can reveal the transition relationship between each state, and thus provide an important basis for the dynamic stability evaluation.

[0056] Optionally, in step S4, define the stable and unstable states of the vehicle to obtain the stable state set

[0057] S stable ={s ∈ S | the reasonable operating condition range of the vehicle design}

[0058] Obtain the unstable state set

[0059] S unstable = S \ S stable .

[0060] Through the above technical solutions, the above operation steps make the stability analysis of the entire system have a clear boundary and make the stability classification of different states clear.

[0061] Optionally, in step S4, the stable state score C(s):

[0062]

[0063] where α is an adjustment factor (0 < α ≤ 1) used to artificially balance the contribution of the unstable state to the stable state; is the probability that state s transfers to the stable state s j ; for the stable state, the score is 1, indicating complete stability; for the unstable state, the score is the total probability of its transfer to the stable state, reflecting the ability to recover from the unstable state to the stable state;

[0064] Calculate the comprehensive stability score CSS: CSS is calculated based on the steady-state distribution and the index score C(s) as follows:

[0065]

[0066] where the first term represents the proportion of the system in the stable state under the steady state, and the second term represents the part of the system in the unstable state under the steady state that has a probability of transferring to the stable state.

[0067] Through the above technical solutions, a comprehensive and dynamic stability assessment of vehicle stability performance is achieved. The design of the stable state score C(s) can accurately reflect the overall stability performance of the vehicle during operation. Moreover, by introducing the adjustment factor α, the contribution of unstable states to stable states can be flexibly adjusted, enabling the system to adapt to different stability requirements in practical applications. Through this comprehensive scoring method, the system can provide accurate and real-time feedback on vehicle handling stability, and provide a scientific basis for dynamic control and optimization to improve vehicle safety and stability.

[0068] After adopting the above technical solutions, the beneficial effects of the present invention are as follows:

[0069] 1. The evaluation method in the present invention is based on the probability distribution of different working conditions during actual driving, and can truly reflect the handling stability of the vehicle in a changing environment. By independently dividing and combining the state space, the information of working condition parameters in each dimension is fully utilized to capture multi-dimensional dynamic characteristics. According to the actual working conditions of the target vehicle, the state transitions with high occurrence probability have a greater impact on the overall stability score, ensuring that the evaluation results are more representative and practical.

[0070] 2. The present invention reasonably controls the scale of the multi-dimensional state space through independent division and equiprobable combination, avoiding the "curse of dimensionality". Based on actual driving data for state division and transition probability statistics, it ensures that the evaluation results are of practical significance.

[0071] 3. The present invention realizes a completely data-driven state definition based on probability density and quantile coarsening of states, without the need for artificial threshold setting, improving the automation and adaptability of the method. It reduces subjectivity and improves the objectivity and consistency of evaluation.

[0072] 4. The present invention can quantify the overall handling stability of the vehicle and provide clear evaluation indicators. Moreover, this evaluation method is directly based on high-dimensional data without separate splitting, and can comprehensively consider multi-dimensional dynamic characteristics such as longitudinal, lateral, and angular momentum to provide a comprehensive evaluation of handling stability. Moreover, this evaluation method not only focuses on the proportion of stable states, but also evaluates the ability to recover from unstable states, comprehensively reflecting the dynamic stability of the vehicle, and can reflect the stability performance of the vehicle under different working conditions in real time, adapting to changes during driving. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for description in the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings without creative efforts based on these drawings.

[0074] Figure 1 It is the automated evaluation result of the CSS stability under working conditions;

[0075] Figure 2 It is the comparison between the CSS evaluation result and the manual annotation result. Specific implementation manners

[0076] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0077] The embodiment of the present application discloses a method for evaluating the vehicle handling stability based on the working condition state distribution.

[0078] Embodiment

[0079] According to Figures 1 to 2 As shown, a method for evaluating the vehicle handling stability based on the working condition state distribution includes the following contents:

[0080] I. Data collection of the actual vehicle working conditions

[0081] Install a variety of sensors (such as vehicle speed sensors, yaw rate sensors, acceleration sensors, etc.) on the vehicle. Real-time collect multi-dimensional working condition data to form a working condition data vector.

[0082] X(t) = [x1(t), x2(t), …, x n (t)];

[0083] Among them, x i (t) represents the i-th working condition parameter collected at the time point t, where i ranges from 1 to n, and n is the total dimension of the working condition parameters.

[0084] II. Data preprocessing

[0085] The actually collected working condition data is often disturbed by various noises, such as sensor measurement errors, electromagnetic interference, environmental changes, etc. These noises will affect the subsequent data analysis and stability evaluation results. Therefore, it is necessary to perform effective denoising processing on the collected working condition data to improve the data quality and ensure the accuracy of the evaluation results.

[0086] The present invention selects the wavelet transform denoising method. The wavelet transform is a time-frequency analysis tool that can decompose and reconstruct signals at different scales, and is particularly suitable for denoising non-stationary signals. By performing multi-scale decomposition on the signal, the wavelet transform can effectively separate the noise and useful information in the signal.

[0087] Assume x i (t) The operating condition state data of the vehicle. For the signal x i (t), perform multi-level discrete wavelet transform to decompose it into approximation coefficients and detail coefficients at different scales:

[0088]

[0089] where J is the decomposition level, φ J,k (t) and ψ j,k (t) are the scaling function and wavelet basis function respectively. After selecting a suitable wavelet basis function, a J,k is the approximation coefficient of the Jth level. The coefficient obtained by convolving the signal with the scaling function and then downsampling represents the low-frequency component of the signal. d j,k is the detail coefficient of the jth level. The coefficient obtained by convolving the signal with the scaling function and then downsampling represents the low-frequency component of the signal.

[0090] Perform threshold processing on the decomposed detail coefficient d j,k to suppress noise. The present invention adopts the soft threshold method.

[0091] d′ j,k = sign(d j,k )·max(|d j,k |-λ j ,0);

[0092] where: d′ j,k is the processed detail coefficient. λ j is the threshold of the jth level, usually determined according to the noise level. For example, using the ViSuShrink method:

[0093]

[0094] where σ j is the standard deviation of the noise of the jth level, which can be estimated by the median absolute deviation: l refers to the level of the detail coefficient; n is the number of data points; N is the total number of data points.

[0095]

[0096] Use the processed detail coefficient d′ j,k and the original approximation coefficient a J,k to perform inverse discrete wavelet transform to reconstruct the denoised operating condition data

[0097]

[0098] III. Modeling of Operating Condition State Distribution

[0099] Extract the state transition distribution of X(t) using Markov theory, and introduce an innovative adaptive partitioning method based on probability density during the state coarse-graining process to improve the accuracy and adaptability of state partitioning.

[0100] A Markov chain is a discrete-time, discrete-state-space stochastic process with memorylessness, i.e., the current state depends only on the previous state and is independent of earlier historical states.

[0101] The state space is represented as S = {s1, s2, …, s m}. The transition probability matrix is represented as P = [p ij , where p ij = P(s j |s i ) represents the probability of transitioning from state s i to state s j .

[0102] To effectively analyze the dynamic characteristics of multi-dimensional operating condition data using Markov theory, it is necessary to discretize the continuous operating condition data into a finite state space. Traditional state partitioning methods mostly use fixed interval partitioning, which has the problem that the partitioning granularity does not adapt to the data characteristics.

[0103] The present invention proposes an adaptive partitioning method based on probability density, which adaptively adjusts the state partitioning boundary according to the probability density of the data to improve the accuracy and adaptability of state definition.

[0104] First, perform probability density estimation. For the denoised one-dimensional operating condition data perform probability density estimation based on kernel density estimation. Use the kernel density estimation method f(x) to perform probability density estimation on the one-dimensional operating condition data :

[0105]

[0106] where N is the number of data points. h is the bandwidth parameter, which determines the smoothness of the kernel function. K(·) refers to the kernel function, and in this patent, the Gaussian kernel is used as the kernel function. The Gaussian kernel is a classic kernel function, and it is used for operating condition data for the first time. is the denoised operating condition data.

[0107]

[0108] Then perform quantile partitioning. According to the estimated probability density, automatically determine the state boundary using quantiles to achieve equal-probability partitioning. That is, each state contains the same probability mass to ensure the balance of transition probabilities between states. Set the number of states m to be partitionedi Calculate the equal - probability quantiles:

[0109]

[0110] where, is the inverse function of the cumulative distribution function of x i (t). k represents the index of the probability quantile. m i is the number of states to be divided, and the operating condition data is divided into m i states.

[0111] Obtain the state - division definition;

[0112] Divide the value range of x i (t) into m i equal - probability states:

[0113]

[0114] According to the state space of each operating - condition parameter, construct the multi - dimensional state space S by calculating the Cartesian product: represents the state interval corresponding to the last interval after the division of the i - th operating - condition parameter x i (t); represents the boundary of the interval after the division of the i - th operating - condition parameter x i (t).

[0115] S = S1×S2×…×S n = {s1×s2×…×s n};

[0116] where each state s=(s1, s2,…, s n ) represents a combined state of the multi - dimensional operating conditions. For each moment t k , map the multi - dimensional operating - condition data X(t k ) = [x1(t k ), x2(t k ),…, x n (t k )] to the corresponding state s(t k ) ∈ S: S n refers to the set of the divided state spaces corresponding to the n - th operating - condition parameter in the multi - dimensional state space, that is, the set of all state spaces related to the n - th operating - condition parameter. t k refers to a certain moment.

[0117]

[0118] where, is x i (t k)The corresponding state.

[0119] Count the number of transitions N of each pair of multi-dimensional states (s i , s j ): ij :

[0120]

[0121] Among them, is the indicator function. s(t k ) represents the multi-dimensional state to which the working condition vector data is mapped at time t k . Here, s i and s j are certain states in the multi-dimensional state space. s(t k + 1) represents the multi-dimensional state to which the working condition vector data is mapped at time point t k + 1.

[0122] Calculate the transition probability based on the number of transitions:

[0123]

[0124] Among them, m is the total number of states in the multi-dimensional state space. N ij represents the number of transitions from state s i to state s j , that is, at all times, the number of times the working condition transfers from s i to s j . N i is the total number of transitions of state s i , that is, the sum of the number of times of transferring from state s i to all other states.

[0125] Construct the transition probability matrix:

[0126]

[0127] Among them, a high transition probability can reflect common working condition switches. A low transition probability can reflect rare or abnormal working condition switches. A high transition probability means a relatively large probability of transferring from one state to another. If the probability is close to 1, it indicates a high transition probability. If the probability is close to 0, it indicates a low transition probability.

[0128] IV. Comprehensive stability score

[0129] In order to achieve a comprehensive stability evaluation based on the working condition state distribution, the following comprehensive stability scoring method is proposed.

[0130] First, define stable and unstable states:

[0131] Obtain the set of stable states

[0132] S stable = {s ∈ S | range of reasonable operating conditions for vehicle design}

[0133] Obtain the set of unstable states

[0134] S unstable = S \ S stable ;

[0135] Calculate the stable state score C(s):

[0136]

[0137] where α is an adjustment factor (0 < α ≤ 1) used to artificially balance the contribution of unstable states transitioning to stable states. is the probability that state s transitions to the stable state s j For stable states, the score is 1, indicating complete stability. For unstable states, the score is the total probability of transitioning to a stable state, reflecting the ability to recover from an unstable state to a stable state.

[0138] Calculate the comprehensive stability score CSS: CSS is calculated based on the steady-state distribution and the index score C(s) as follows:

[0139]

[0140] where the first term characterizes the proportion of the system in a stable state under steady state. The second term characterizes the part of the system that has a probability of transitioning to a stable state when it is in an unstable state under steady state.

[0141] The closer CSS is to 1, the more time the system is in a stable state, or even when it is in an unstable state, it has a high probability of recovering to a stable state, indicating better overall stability. The closer CSS is to 0, the more time the system is in an unstable state, or the lower the probability of recovering to a stable state when it is in an unstable state, indicating poorer overall stability.

[0142] Compare the comprehensive stability scoring (CSS) method proposed in the present invention with the stability results manually annotated to verify the effectiveness of the method in the present invention. The experiment used 100 randomly intercepted working condition data segments, and compared the stability states of these data segments with the manually annotated results. First, manually annotate the stability states of these data segments, and classify them into three categories: stable, relatively stable, and unstable, represented by different identifiers. Then, use the CSS method to perform stability scoring on the same 100 data segments.

[0143] Figure 1 The relationship between the CSS score (vertical axis) and the stability status manually marked (horizontal axis) was compared. Each point represents a data segment, and the shapes are distinguished according to the stability of the manual marking (circles, squares, triangles). The horizontal axis represents the working condition number, and the vertical axis shows the corresponding CSS score of that working condition. It can be seen from the scatter plot that the CSS score is highly consistent with the stability status manually marked. The CSS score of the stable state is higher, while the CSS score of the unstable state is lower. For the relatively stable state, the CSS score is between circles and triangles. Figure 2 The distribution of the CSS score in different stability categories (stable, relatively stable, unstable) is shown. The box plot shows the median of the CSS score for each category and its distribution range. The CSS score of the stable state is higher, the CSS score of the unstable state is lower, and the scores of the relatively stable state are distributed between the two. From these two charts, it can be concluded that the CSS method is highly consistent with the stability status manually marked, proving that this method can effectively evaluate the stability of the vehicle. The CSS score provides an automated, data-driven way of stability evaluation and highly matches the results of the experts' manual judgment.

[0144] In summary, the present invention is based on manually marking 100 randomly intercepted working condition data segments and comparing them with the results of the comprehensive stability score CSS score. As Figure 1 and Figure 2 shown, the results show that the comprehensive stability score (CSS) method proposed by the present invention can achieve the automated evaluation of the stability of vehicle operating conditions. The CSS score is highly matched with the stability status manually marked, verifying the effectiveness and reliability of this method.

[0145] The comprehensive vehicle handling stability evaluation method based on the working condition state distribution of the present invention uses multi-dimensional data (such as vehicle speed, yaw rate, acceleration, etc.) as inputs and analyzes the dynamic changes of these data through a Markov chain model, so as to be able to more realistically reflect the stability of the vehicle in a complex and changeable actual driving environment. Compared with the traditional fixed working condition test, the method of the present invention can dynamically evaluate the vehicle stability according to the probability distribution and transfer characteristics of the actual working conditions, ensuring that the evaluation results are more adaptable and comprehensive. The present invention can dynamically and comprehensively evaluate the stability of the vehicle under complex and changeable driving conditions: the traditional method relies on fixed test conditions and cannot adapt to complex driving environments. The present invention can fully consider the probability distribution of each working condition state and its transfer characteristics by combining a multi-dimensional Markov transition probability matrix, and realize the dynamic and comprehensive evaluation of the vehicle handling stability.

[0146] In addition, the present invention can process and analyze high-dimensional and complex working condition data: The performance of a vehicle under different working conditions is multi-dimensional (such as vehicle speed, yaw rate, acceleration, etc.), which is often overlooked by traditional methods. Through an adaptive partitioning method based on probability density, the present invention can effectively process high-dimensional data, avoid the "curse of dimensionality" caused by fixed partitioning spaces, and better reflect the impact of each working condition dimension on vehicle stability.

[0147] The present invention can improve the adaptability and accuracy of the evaluation method: Traditional methods often rely on thresholds set by humans, and these thresholds may not be suitable for working condition data under all distributions. The present invention uses kernel density estimation and quantile partitioning to adaptively adjust the state partitioning boundary according to actual data.

[0148] In terms of the evaluation of handling stability under working condition state transitions, traditional methods usually only focus on the vehicle performance under specific working conditions. The present invention, by comprehensively considering multi-dimensional data and the vehicle's recovery ability (the probability of recovering from an unstable state to a stable state), provides a more comprehensive stability evaluation index (CSS score), which can better reflect the overall handling stability of the vehicle under working condition state transitions.

[0149] In terms of reducing the subjectivity of the evaluation method, the present invention, through state partitioning and transition probability statistics completely based on data-driven, avoids manual intervention and improves the automation and consistency of the method. This makes the evaluation of vehicle handling stability more objective and reliable.

[0150] The present invention is funded by the Shandong Natural Science Foundation (ZR2023QE208), the China Postdoctoral Science Foundation (2024M751577), and the Key Research and Development Program of Shandong Province (2022CXGC020302).

[0151] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the technical scope of the present invention shall be included in the protection scope of the present invention.

Claims

1. A vehicle handling stability evaluation method based on the distribution of operating conditions, characterized in that It includes the following steps: S1. Data collection of the actual vehicle working conditions; S2. Data preprocessing; S3. Modeling of the working condition state distribution: adaptively adjust the boundary of state division according to the probability density of the data; specifically it includes the following content: S31. Probability density estimation; for the denoised one-dimensional operating condition data Perform probability density estimation based on kernel density estimation; S32. Conduct quantile division; automatically determine the state boundary using quantiles according to the estimated probability density to achieve equal-probability division; S4. Stability analysis; based on the working condition state distribution, determine the stable state of the vehicle and score the stable state.

2. The vehicle handling stability evaluation method based on the working condition state distribution according to claim 1, characterized in that In step S1, multi-dimensional working condition data is collected in real time to form a working condition data vector; X(t) = [x1(t), x2(t), …, x n (t)]; where x i (t) represents the i-th operating condition parameter collected at time point t, where i ranges from 1 to n, and n is the total dimension of the operating condition parameters.

3. A vehicle handling stability evaluation method based on the distribution of operating conditions according to claim 1, characterized in that In step S2, assume that x i (t) is the vehicle's operating condition data, and perform a multi-level discrete wavelet transform on the signal x i (t) to decompose it into approximation coefficients and detail coefficients at different scales: Among them, J is the decomposition level, φ J,k (t) and ψ j,k (t) are the scaling function and the wavelet basis function respectively, a J,k is the approximation coefficient of the J-th layer, d j,k is the detail coefficient of the j-th layer; Perform threshold processing on the decomposed detail coefficient d j,k ; d′ j,k = sign(d j,k )·max(|d j,k | - λ j , 0); Among them, d′ j,k is the processed detail coefficient, λ j is the threshold of the j-th layer, using the ViSuShrink method: where, σ j is the standard deviation of the noise of the j-th layer, which can be estimated by the median absolute deviation: l refers to the level of the detail coefficient; n is the number of data points; N is the total number of data points; Using the processed detail coefficient d' j,k and the original approximation coefficient a J,k perform an inverse discrete wavelet transform to reconstruct the denoised operating condition data 4. A vehicle handling stability evaluation method based on the distribution of operating conditions according to claim 1, characterized in that In step S31, the probability density estimation is performed on the one-dimensional working condition data by using the kernel density estimation method f(x). The probability density estimation is as follows: where N is the number of data points, h is the bandwidth parameter that determines the smoothness of the kernel function, K(·) is the kernel function, is the working condition data after denoising; 5. A vehicle handling stability evaluation method based on the distribution of operating conditions according to claim 4, characterized in that In step S32, set the number of states m to be partitioned i , and calculate the equal-probability quantiles: Among them, is the inverse function of the cumulative distribution function of x i (t), k represents the index of the probability quantile, and m i is the number of states to be divided, and the operating condition data is divided into m i number of states; Divide x i (t) into m i equally probable states: According to the state space of each working condition parameter, a multi-dimensional state space S is constructed by calculating the Cartesian product. Denote the state interval corresponding to the last interval after the division of the i-th working condition parameter x i (t). Denote the boundary of the interval after the division of the i-th working condition parameter x i (t). S = S1 × S2 × … × S n = {s1 × s2 × … × s n}; where each state s = (s1, s2, …, s n ) represents a combined state of multi-dimensional working conditions. For each moment t k , the multi-dimensional working condition data X(t k ) = [x1(t k ), x2(t k ), …, x n (t k )] is mapped to the corresponding state s(t k ) ∈ S: S n refers to the set of partitioned state spaces corresponding to the nth working condition parameter in the multi-dimensional state space, that is, the set of all state spaces related to the nth working condition parameter; Among them, is the state corresponding to x i (t k ). Count the number of transitions N i for each pair of multi-dimensional states (s j , s ij ): Among them, is an indicator function, and s(t k ) represents the multi-dimensional state to which the working condition vector data is mapped at time t k . Here, s i and s j are certain states in the multi-dimensional state space. s(t k + 1) represents the multi-dimensional state to which the working condition vector data is mapped at time point t k + 1; Calculate the transition probability according to the number of transitions: Among them, m is the total number of states in the multi-dimensional state space, N ij represents the number of transitions from state s i to state s j , that is, at all times, the operating condition transitions from s i to s j . N i is the total number of transitions of state s i , that is, the total number of times of transitioning from state s i to all other states; Construct a transition probability matrix: Among them, a high transition probability reflects common working condition switches, and a low transition probability reflects rare or abnormal working condition switches. If the probability is close to 1, it indicates a high transition probability, and if the probability is close to 0, it indicates a low transition probability.

6. The vehicle handling stability evaluation method based on the distribution of operating conditions according to claim 5, wherein In step S4, the stable and unstable states of the vehicle are defined to obtain the set of stable states S stable = {s ∈ S | range of reasonable operating conditions for vehicle design} Obtain the set of unstable states S unstable = S\S stable .

7. A vehicle handling stability evaluation method based on the distribution of operating conditions according to claim 6, characterized in that In step S4, the stable state score C(s): where α is an adjustment factor (0 < α ≤ 1) used to artificially balance the contribution of the transition from the unstable state to the stable state; is the probability of the state s transitioning to the stable state s j ; for the stable state, the score is 1, indicating complete stability; for the unstable state, the score is the total probability of its transition to the stable state, reflecting the ability to recover from the unstable state to the stable state. Calculate the comprehensive stability score CSS: CSS is based on the steady-state distribution and the index score C(s), and is calculated as follows: Among them, the first item represents the proportion of the system in a stable state under steady state, and the second item represents the part that has a probability of transitioning to a stable state when the system is in an unstable state under steady state.

Images