Unstructured road elevation estimation method based on electric vehicle suspension system response

Abstract

The present invention discloses an unstructured road elevation estimation method based on the response of an electric vehicle suspension system, the method comprising: constructing a sprung system dynamics model and a multi-scale tire model set considering changes in road adhesion; combining the sprung system dynamics model and the tire model set to form a vehicle suspension system dynamics model set covering a variety of driving conditions, selecting the suspension system dynamics model with the highest matching degree under the current condition according to the condition division; constructing a nonlinear fault-tolerant filter based on the suspension system dynamics model with the highest matching degree; developing a filter gain prediction model based on a long sequence recurrent neural network, and fusing it with the nonlinear fault-tolerant filter to construct a unified filter, so as to realize non-Gaussian time-varying noise road elevation estimation under complex conditions. Through the processing scheme disclosed in the present invention, the adaptability of road elevation estimation to different conditions is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy vehicle design and manufacturing, and in particular to a method for estimating the elevation of an unstructured road surface based on the response of an electric vehicle suspension system. Background Art

[0002] Active control of electric vehicle suspension system plays a vital role in improving vehicle driving performance and safety. Road elevation information is used as the input of suspension system, especially unstructured road surface, which can easily cause the vehicle to lose driving ability. However, due to the complex and changeable vehicle driving conditions and the time-varying and non-Gaussian noise of on-board sensors, the traditional single suspension system model is difficult to adapt to a variety of complex conditions, and the road elevation information estimation algorithm is difficult to adapt to non-Gaussian time-varying noise, resulting in the estimation accuracy being difficult to meet the requirements of active suspension system control.

[0003] It can be seen that how to create a new unstructured road elevation estimation method has become a goal that the industry urgently needs to improve. Summary of the invention

[0004] In view of this, an embodiment of the present disclosure provides an unstructured road elevation estimation method based on the response of an electric vehicle suspension system to solve the problems existing in the prior art.

[0005] In a first aspect, an embodiment of the present disclosure provides a method for estimating elevation of an unstructured road surface based on the response of an electric vehicle suspension system, the method comprising the following steps:

[0006] Construct a sprung system dynamics model and a multi-scale tire model set that considers road adhesion changes;

[0007] Combine the sprung system dynamics model and the tire model set to form a vehicle suspension system dynamics model set covering various driving conditions, and select the suspension system dynamics model with the highest matching degree under the current condition according to the working condition classification;

[0008] Building a nonlinear fault-tolerant filter based on the suspension system dynamics model with the highest matching degree;

[0009] A filter gain prediction model based on a long sequence recurrent neural network is developed, and it is fused with a nonlinear fault-tolerant filter to construct a unified filter to achieve non-Gaussian time-varying noise road elevation estimation under complex working conditions.

[0010] According to a specific implementation of the embodiment of the present disclosure, the sprung system dynamics model is constructed based on the following formula:

[0011]

[0012] Among them, m sis the sprung mass; z s is the vertical displacement of the sprung mass; k s is the spring stiffness; z w is the vertical displacement of the unsprung mass; c s is the damping coefficient of the suspension system; F z is the equivalent vertical force of the tire; m w is the wheel mass; k t is the stiffness of the tire; z r Input for road surface;

[0013] The multi-scale tire model set is constructed based on the following formula:

[0014]

[0015] Among them, F x is the tire longitudinal force; C σ is the tire longitudinal stiffness; σ is the tire slip rate; λ is the tire nonlinear factor; f(λ) is the nonlinear function of λ; F y is the tire lateral force; C α is the lateral stiffness of the tire; α is the tire slip angle;

[0016] The tire nonlinear factor λ is calculated based on the following formula:

[0017]

[0018]

[0019] Wherein, μ is the tire-road adhesion coefficient, and the value range of μ is [0,1]; F z is the equivalent vertical force of the tire; v x is the longitudinal velocity of the vehicle; v is the speed influencing factor; σ is the tire slip rate; α is the tire side slip angle; F x is the tire longitudinal force; F y is the lateral force of the tire.

[0020] According to a specific implementation of the embodiment of the present disclosure, the nonlinear fault-tolerant filter is constructed based on the suspension system dynamics model with the highest matching degree, including:

[0021] According to the dynamics model of the sprung system, the state space equation is obtained:

[0022]

[0023] The state variables are:

[0024] X τ =[z r ] T ;

[0025] The system observation vector is:

[0026]

[0027] The input variables are:

[0028] u τ =[F z ] T ;

[0029] Among them, X τ is the state variable of the system at time τ; f is the nonlinear state function; X τ-1 is the state vector of the system at time τ-1; u τ-1 is the input variable; W τ-1 is the process noise; Z τ is the observation vector of the system; η τ is the probability of data loss; h is the measurement function; u τ is the known control input variable; V τ is the measurement noise; r is the road surface input; T is the matrix transpose; z s is the vertical displacement of the sprung mass; z w is the vertical displacement of the unsprung mass;

[0030] Initial state X τ With W τ 、V τ Unrelated to each other; are g independent random variables, diag{·} represents a diagonal matrix, is the probability density function;

[0031] Iterate the nonlinear error-tolerant filter, including:

[0032] S1. Initialization:

[0033]

[0034] in, is the initial state prediction value; E is the mathematical expectation; X0 is the initial state value; P0 is the covariance matrix;

[0035] S2. Time update:

[0036] The state one-step forecast is calculated based on the following formula:

[0037]

[0038] in, is the posterior estimate of the state at time τ-1; X τ-1|τ-1is the state prior value at time τ-1; u τ-1|τ-1 is the input at time τ-1;

[0039] The covariance of the state prediction error is calculated based on the following formula:

[0040]

[0041] Among them, P τ|τ-1 A is the prior covariance at time τ-1; τ is the bias matrix; P τ-1|τ-1 is the posterior covariance at time τ-1; A τ T A τ The transpose of ; Q is the process noise matrix; In order to seek partial guidance;

[0042] S3, measurement update:

[0043] The one-step predicted value of the measurement is calculated based on the following formula

[0044]

[0045] The filter gain is calculated based on the following formula:

[0046]

[0047] Among them, K τ is the filter gain; is the measurement Jacobian matrix; is the data loss probability density, R τ is the measurement noise matrix;

[0048] Estimates updated in real time:

[0049]

[0050] in, is the posterior estimate of the state at time τ; is the new interest variance;

[0051] Further update the error covariance:

[0052]

[0053] Where I is the identity matrix; P τ-1|τ is the prior covariance at time τ;

[0054] The data of the measurement vector innovation variance, the measurement vector observation error, the state posterior update error and the filter gain matrix are recorded respectively, and a one-to-one correspondence is established to construct a data set.

[0055] According to a specific implementation of the embodiment of the present disclosure, the development of a filter gain prediction model based on a long sequence recurrent neural network includes:

[0056] Selecting an LSTM network to train the data set to obtain a real-time prediction model of filter gain;

[0057] LSTM first receives the input data of information variance and error serialization, and passes the information in time series through the gating mechanism and the state of the memory unit.

[0058] The filter gain prediction model is integrated with the nonlinear fault-tolerant filter to construct a unified filter to achieve non-Gaussian time-varying noise road elevation estimation under complex working conditions, including:

[0059] The nonlinear fault-tolerant filter is decomposed into a data flow-oriented framework structure according to its function, and the LSTM gain prediction model is embedded in it;

[0060] The LSTM model is used to directly predict the optimal Kalman gain required for iterative updates of the system state, and a priori estimates are obtained using the state value of the previous time step and the state update function. The predicted value of the latest measured variable is obtained using the measurement update function and the predicted measured variable at the previous moment. The final road surface estimate is obtained by iterative loops, taking into account the deviation between the actual measured value and the predicted measured value, as well as the predicted value of the filter gain prediction LSTM model.

[0061] The unstructured road elevation estimation method based on the response of the electric vehicle suspension system in the embodiment of the present invention estimates the road elevation parameters based on the active suspension system, and designs an adaptive estimator taking into account the influence of process noise and measurement noise uncertainty, which can fill the current technical gap; the present invention establishes a data-driven noise online update method, improves the adaptability of road elevation estimation to different working conditions, and increases the scope of application of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The above is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0063] Figure 1 A flow chart of a method for estimating the elevation of an unstructured road surface based on the response of an electric vehicle suspension system provided by an embodiment of the present disclosure;

[0064] Figure 2 A schematic diagram of driving condition area division provided in an embodiment of the present disclosure;

[0065] Figure 3A schematic diagram of a prediction filter gain model provided by an embodiment of the present disclosure;

[0066] Figure 4 A schematic diagram of a unified road elevation estimator provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0067] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0068] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.

[0069] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement a device and / or practice a method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.

[0070] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0071] The embodiment of the present invention provides an unstructured road elevation estimation method based on the response of the suspension system of an electric vehicle. First, a sprung system dynamics model and a multi-scale tire model set considering the road adhesion change are constructed based on the multi-body dynamics theory. The sprung system dynamics model and the tire model are combined to form a vehicle suspension system dynamics model set covering a variety of driving conditions. The optimal suspension system model under the current working condition is selected according to the working condition classification. Secondly, a nonlinear fault-tolerant filter is constructed based on the optimal suspension system model, and a filter gain prediction model based on a long sequence recurrent neural network is developed. It is then organically integrated with the nonlinear fault-tolerant filter to construct a unified filter, so as to achieve accurate estimation of non-Gaussian time-varying noise road elevation under complex working conditions.

[0072] Figure 1 A schematic diagram of the process of a method for estimating the elevation of an unstructured road surface based on the response of an electric vehicle suspension system provided in an embodiment of the present disclosure.

[0073] like Figure 1 As shown, at step S110, a sprung system dynamics model and a multi-scale tire model set considering road adhesion changes are constructed;

[0074] More specifically, a sprung system dynamics model and a multi-scale tire model set considering the road adhesion variation are constructed based on multi-body dynamics theory.

[0075] In an embodiment of the present invention, the construction of the sprung system dynamics model and the multi-scale tire model set considering the road adhesion change includes:

[0076] The sprung system dynamics model is constructed based on the following formula:

[0077]

[0078] Among them, m s is the sprung mass; z s is the vertical displacement of the sprung mass; k s is the spring stiffness; z w is the vertical displacement of the unsprung mass; c s is the damping coefficient of the suspension system; F z is the equivalent vertical force of the tire; m w is the wheel mass; k t is the stiffness of the tire; z r Input for road surface;

[0079] The multi-scale tire model set is constructed based on the following formula:

[0080]

[0081] Among them, F x is the longitudinal force of the tire; Cσ is the tire longitudinal stiffness; σ is the tire slip rate; λ is the tire nonlinear factor, and the nonlinear characteristics of the tire force caused by tire slip are determined according to the value of this variable; f(λ) is the nonlinear function of λ; F y is the tire lateral force; C α is the lateral stiffness of the tire; α is the tire slip angle.

[0082] The tire nonlinear factor λ is calculated based on the following formula:

[0083]

[0084] f(λ) is calculated based on the following formula:

[0085]

[0086]

[0087] Wherein, μ is the tire-road adhesion coefficient, and the value range of μ is [0,1]. For example, if the tire model is divided into equal intervals of 0.1, it can correspond to different road surfaces (such as asphalt road surface, ice and snow road surface, etc.). Then, the above tire model can be composed of 10 tire models corresponding to different road surfaces to form a tire model set; v x is the longitudinal velocity of the vehicle; v is the speed influencing factor; σ is the tire slip rate; α is the tire side slip angle; F z is the equivalent vertical force of the tire; F x is the tire longitudinal force; F y is the lateral force of the tire.

[0088] More specifically, the process proceeds to step S120.

[0089] In step S120, the sprung system dynamics model and the tire model set are combined to form a vehicle suspension system dynamics model set covering various driving conditions, and the suspension system dynamics model with the highest matching degree under the current condition is selected according to the condition classification.

[0090] More specifically, the tire model set and the sprung system dynamics model are combined based on Fz to obtain the vehicle suspension system dynamics model set. For example, based on the common variable Fz of the tire model set and the sprung system dynamics model, the sprung system dynamics model is combined with each tire model in the tire model set to obtain the vehicle suspension system dynamics model set.

[0091] According to the different road adhesion coefficients, the friction ellipse is constructed, and the vehicle driving conditions are divided into two categories: normal conditions and extreme conditions. The longitudinal and lateral accelerations obtained by the vehicle acceleration sensor correspond to Figure 2area in order to select the model that best matches the current working conditions.

[0092] Next, go to step S130.

[0093] In step S130, a nonlinear fault-tolerant filter is constructed based on the suspension system dynamics model with the highest matching degree.

[0094] More specifically, a nonlinear fault-tolerant filter is constructed, and colored noise sequences obeying various distributions are randomly generated to construct a data set including the innovation variance, the measurement vector observation error, the state posterior update error, and the filter gain matrix;

[0095] In an embodiment of the present invention, the step of constructing a nonlinear fault-tolerant filter based on the suspension system dynamics model with the highest matching degree includes:

[0096] According to the sprung system dynamics model (Formula (1), (2)), the state space equation is obtained:

[0097]

[0098] The state variables are:

[0099] X τ =[z r ] T ……(9)

[0100] The system observation vector is:

[0101]

[0102] The input variables are:

[0103] u τ =[F z ] T ……(11)

[0104] Among them, X τ is the state variable of the system at time τ; f is the nonlinear state function; X τ-1 is the state vector of the system at time τ-1; u τ-1 is the input variable; W τ-1 is the process noise; Z τ is the observation vector of the system; η τ is the probability of data loss; h is the measurement function; u τ is the known control input variable; V τ is the measurement noise; r is the road surface input; T is the matrix transpose; z s is the vertical displacement of the sprung mass; z w is the vertical displacement of the unsprung mass;

[0105] Initial state X τ With W τ 、V τ Unrelated, are g independent random variables, diag{·} represents a diagonal matrix, is the probability density function.

[0106] The nonlinear error-tolerant filter is iterated, including the following steps:

[0107] S1. Initialization:

[0108]

[0109] in, is the initial state prediction value; E is the mathematical expectation; X0 is the initial state value; P0 is the covariance matrix;

[0110] S2. Time update:

[0111] The state one-step forecast is calculated based on the following formula:

[0112]

[0113] in, is the posterior estimate of the state at time τ-1; X τ-1|τ-1 is the state prior value at time τ-1; u τ-1|τ-1 is the input at time τ-1;

[0114] The covariance of the state prediction error is calculated based on the following formula:

[0115]

[0116] Among them, P τ|τ1 A is the prior covariance at time τ-1; τ is the bias matrix; P τ-1|τ-1 is the posterior covariance at time τ-1; A τ T A τ The transpose of ; Q is the process noise matrix; In order to seek partial guidance;

[0117] S3, measurement update:

[0118] The one-step predicted value of the measurement is calculated based on the following formula

[0119]

[0120] The filter gain is calculated based on the following formula:

[0121]

[0122] Among them, K τ is the filter gain; is the measurement Jacobian matrix; is the data loss probability density, R τ is the measurement noise matrix;

[0123] Estimates updated in real time:

[0124]

[0125] in, is the posterior estimate of the state at time τ; is the new interest variance;

[0126] Further update the error covariance:

[0127]

[0128] Where I is the identity matrix; P τ-1|τ The prior covariance at time τ is recorded respectively, and the data of the measurement vector innovation variance, the measurement vector observation error, the state posterior update error and the filter gain matrix are established, and a one-to-one correspondence is established to construct a data set.

[0129] More specifically, based on the established nonlinear fault-tolerant filter, the filter is run by setting different driving conditions, different distributed noise interferences and different colored noise influences, and the data of the measurement vector innovation variance, measurement vector observation error, state posterior update error and filter gain matrix are recorded respectively, and a one-to-one correspondence is established. Finally, a data set is constructed through a large number of parameter settings.

[0130] Next, go to step S140.

[0131] In step S140, a filter gain prediction model based on a long sequence recurrent neural network is developed, and it is fused with a nonlinear fault-tolerant filter to construct a unified filter to achieve non-Gaussian time-varying noise road elevation estimation under complex working conditions.

[0132] More specifically, a filter gain prediction model based on a long sequence recurrent neural network is developed, and it is organically integrated with a nonlinear fault-tolerant filter to construct a unified filter to achieve road elevation estimation.

[0133] In an embodiment of the present invention, the development of a filter gain prediction model based on a long sequence recurrent neural network includes: selecting an LSTM network to train the data set to obtain a real-time prediction model for filter gain; the LSTM first receives serialized input data of information variance and error, and the information is transmitted and updated in a time series through a gating mechanism and state transfer of memory units.

[0134] The filter gain prediction model is integrated with the nonlinear fault-tolerant filter to construct a unified filter to realize the non-Gaussian time-varying noise road elevation estimation under complex working conditions, including: deconstructing the nonlinear fault-tolerant filter into a data flow-oriented framework structure according to its function, and embedding the LSTM gain prediction model into it; using the LSTM model to directly predict the optimal Kalman gain required for the iterative update of the system state, and using the state value and state update function of the previous time step to obtain a priori estimate; using the measurement update function and the predicted measurement variables at the previous moment to obtain the predicted value of the latest measurement variable; through an iterative cycle, considering the deviation between the actual measurement value and the predicted measurement value, as well as the predicted value of the filter gain prediction LSTM model to obtain the final road surface estimation value.

[0135] More specifically, the unified filter design is as follows:

[0136] Based on the data set, an LSTM network is selected to train the data set to obtain a real-time prediction model of the filter gain, and a prediction model structure is constructed as follows: Figure 3 shown.

[0137] Figure 3 The memory cell in LSTM is used to store historical information. The update and use of historical information are controlled by three gates: input gate, forget gate, and output gate. h is the output of the LSTM cell, and x is the input data. LSTM first receives serialized input data such as information variance and error, and passes the information in the time series through the gating mechanism and the state of the memory cell. This enables LSTM to effectively predict the filter gain while preventing the problem of gradient vanishing or gradient exploding.

[0138] The fault-tolerant estimator is deconstructed into a data flow-oriented framework structure according to its function, and the LSTM gain prediction model is embedded in it, as shown in Figure 4 As shown in Figure 2, accurate estimation of road elevation is achieved. Figure 4 As shown in the figure, the original equation responsible for calculating the error covariance is removed. This part uses the LSTM model to directly predict the optimal Kalman gain required for iterative updates of the system state. In the system state update module, a priori estimation is obtained using the state value and state update function of the previous time step; then, the predicted value of the latest measured variable is obtained using the measurement update function and the predicted measured variable of the previous moment; finally, the final road surface estimation value is obtained by iterative looping, considering the deviation between the actual measured value and the predicted measured value, and the predicted value of the filter gain prediction LSTM model.

[0139] The unstructured road elevation estimation method based on the response of the electric vehicle suspension system proposed in the present invention has the following beneficial effects compared with the prior art:

[0140] 1. The present invention estimates the road elevation parameters based on the active suspension system, considers the influence of process noise and measurement noise uncertainty, and designs an adaptive estimator, which can fill the current technical gap;

[0141] 2. The present invention establishes a data-driven online noise updating method, which improves the adaptability of road elevation estimation to different working conditions and increases the scope of application of the present invention.

[0142] The above is only a specific implementation of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present disclosure should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims

1. A method for estimating the elevation of an unstructured road surface based on the response of an electric vehicle suspension system, characterized in that: The method comprises the following steps: Construct a sprung system dynamics model and a multi-scale tire model set that considers road adhesion changes; Combine the sprung system dynamics model and the tire model set to form a vehicle suspension system dynamics model set covering various driving conditions, and select the suspension system dynamics model with the highest matching degree under the current condition according to the working condition classification; Building a nonlinear fault-tolerant filter based on the suspension system dynamics model with the highest matching degree; Develop a filter gain prediction model based on a long sequence recurrent neural network and integrate it with a nonlinear fault-tolerant filter to construct a unified filter to achieve non-Gaussian time-varying noise road elevation estimation under complex working conditions; The method of constructing a nonlinear fault-tolerant filter based on the suspension system dynamics model with the highest matching degree includes: According to the dynamics model of the sprung system, the state space equation is obtained: The state variables are: X τ =[z r ] T ; The system observation vector is: The input variables are: u τ =[F z ] T ; Among them, X τ is the state variable of the system at time τ; f is the nonlinear state function; X τ-1 is the state vector of the system at time τ-1; u τ-1 is the input variable; W τ-1 is the process noise; Z τ is the observation vector of the system; η τ is the probability of data loss; h is the measurement function; u τ is the known control input variable; V τ is the measurement noise; r is the road surface input; T is the matrix transpose; z s is the vertical displacement of the sprung mass; z w is the vertical displacement of the unsprung mass; Initial state X τ With W τ 、V τ Unrelated to each other; are g independent random variables, diag{·} represents a diagonal matrix, is the probability density function; Iterate the nonlinear error-tolerant filter, including: S1. Initialization: in, is the initial state prediction value; E is the mathematical expectation; X0 is the initial state value; P0 is the covariance matrix; S2. Time update: The state one-step forecast is calculated based on the following formula: in, is the posterior estimate of the state at time τ-1; X τ-1|τ-1 is the state prior value at time τ-1; u τ-1|τ-1 is the input at time τ-1; The covariance of the state prediction error is calculated based on the following formula: Among them, P τ|τ-1 A is the prior covariance at time τ-1; τ is the bias matrix; P τ-1|τ-1 is the posterior covariance at time τ-1; A τ T A τ The transpose of ; Q is the process noise matrix; In order to seek partial guidance; S3, measurement update: The one-step predicted value of the measurement is calculated based on the following formula The filter gain is calculated based on the following formula: Among them, K τ is the filter gain; is the measurement Jacobian matrix; is the data loss probability density, R τ is the measurement noise matrix; Estimates updated in real time: in, is the posterior estimate of the state at time τ; is the new interest variance; Further update the error covariance: Among them, P τ|τ is the error covariance; I is the unit matrix; P τ-1|τ is the prior covariance at time τ; The data of the measurement vector innovation variance, the measurement vector observation error, the state posterior update error and the filter gain matrix are recorded respectively, and a one-to-one correspondence is established to construct a data set.

2. The unstructured road elevation estimation method based on the electric vehicle suspension system response according to claim 1 is characterized in that: The sprung system dynamics model is constructed based on the following formula: Among them, m s is the sprung mass; z s is the vertical displacement of the sprung mass; k s is the spring stiffness; z w is the vertical displacement of the unsprung mass; c s is the damping coefficient of the suspension system; F z is the equivalent vertical force of the tire; m w is the wheel mass; k t is the stiffness of the tire; z r Input for road surface; The multi-scale tire model set is constructed based on the following formula: Among them, F x is the tire longitudinal force; C σ is the tire longitudinal stiffness; σ is the tire slip rate; λ is the tire nonlinear factor; f(λ) is the nonlinear function of λ; F y is the tire lateral force; C α is the lateral stiffness of the tire; α is the tire slip angle; The tire nonlinear factor λ is calculated based on the following formula: Wherein, μ is the tire-road adhesion coefficient, and the value range of μ is [0,1]; v x is the longitudinal velocity of the vehicle; v is the speed influencing factor; σ is the tire slip rate; α is the tire side slip angle; F z is the equivalent vertical force of the tire; F x is the tire longitudinal force; F y is the lateral force of the tire.

3. The unstructured road elevation estimation method based on the electric vehicle suspension system response according to claim 1 is characterized in that: The development of a filter gain prediction model based on a long sequence recurrent neural network includes: Selecting an LSTM network to train the data set to obtain a real-time prediction model of filter gain; LSTM first receives the serialized input data of the new information variance and error, and transmits the information in the time series through the gating mechanism and the state of the memory unit; The filter gain prediction model is integrated with the nonlinear fault-tolerant filter to construct a unified filter to achieve non-Gaussian time-varying noise road elevation estimation under complex working conditions, including: The nonlinear fault-tolerant filter is decomposed into a data flow-oriented framework structure according to its function, and the LSTM gain prediction model is embedded in it; The LSTM model is used to directly predict the optimal Kalman gain required for iterative updates of the system state, and a priori estimates are obtained using the state value of the previous time step and the state update function. The predicted value of the latest measured variable is obtained using the measurement update function and the predicted measured variable at the previous moment. The final road surface estimate is obtained by iterative loops, taking into account the deviation between the actual measured value and the predicted measured value, as well as the predicted value of the filter gain prediction LSTM model.

Images