Vehicle key driving state estimation method based on low-orbit satellite and vehicle-mounted sensor multi-source information fusion

Abstract

The application discloses a vehicle key driving state estimation method based on low-orbit satellite and vehicle-mounted sensing multi-source information fusion, which comprises the following steps: acquiring real-time longitudinal speed of a vehicle based on a low-orbit satellite measurement signal and a Doppler frequency shift formula; acquiring real-time lateral force of a tire based on an intelligent tire measurement signal and a Gaussian process regression algorithm; calculating a tire side slip angle based on a vehicle-mounted sensing signal, establishing a tire side slip stiffness estimation algorithm combined with prior information of the lateral force, and acquiring a real-time side slip stiffness adaptive value; acquiring vehicle parameters and real-time sensing signals, and establishing a vehicle dynamics model combined with the side slip stiffness adaptive value; taking the lateral force estimation value provided by the intelligent tire system, the longitudinal speed value measured by the low-orbit satellite, and the yaw angular velocity and transverse acceleration sensing measurement values as measurement signals, combining a innovation adaptive extended Kalman filter, designing a vehicle key driving state estimation algorithm, and acquiring real-time center of mass side slip angle, yaw angular velocity and longitudinal speed of the vehicle.

Description

Technical Field

[0001] This invention belongs to the field of vehicle control technology and relates to a method for estimating key vehicle driving states, particularly a method for estimating key vehicle driving states based on the fusion of multi-source information from low-orbit satellites and on-board sensors. Background Technology

[0002] Key vehicle driving states (including longitudinal speed, sideslip angle, and yaw rate) are important parameters describing vehicle motion characteristics, among which longitudinal speed (v... x The longitudinal control of autonomous vehicles is used to regulate the acceleration, braking and cruise control of the vehicle. The sideslip angle (β) describes the angle between the actual driving direction of the vehicle and the longitudinal axis of the vehicle body. It is an important parameter for lateral stability analysis. The yaw rate (γ) describes the angular velocity of the vehicle about its center of mass. It is an important control variable for trajectory tracking and path planning.

[0003] Traditional state estimation methods typically rely on acceleration and angular velocity data provided by onboard sensors (such as inertial measurement units, IMUs) to calculate key vehicle driving states using vehicle dynamics models. However, these methods often fall short of accuracy requirements under specific driving conditions (such as slippery surfaces, high-speed curves, or emergency obstacle avoidance), primarily because achieving an accurate and concise representation of tire forces is challenging. In other words, while simplified tire models can accurately describe tire forces under most driving conditions, they often fail to accurately represent the actual tire motion behavior when the vehicle is in special driving situations. Although employing more complex tire models can improve the accuracy of tire force representation under specific driving conditions, this also increases the complexity of the vehicle driving state estimation system, making it difficult to meet the real-time requirements of the vehicle control system.

[0004] Smart tire technology involves installing sensors inside the tire to directly measure dynamic parameters during tire movement and combining this with machine learning or theoretical models to estimate key tire states. Compared to traditional tire state estimation methods, this technology provides more accurate and reliable tire state data. Therefore, if tire force prediction can be achieved based on smart tire technology, it can simplify the process of traditional tire force prediction methods while improving the accuracy and reliability of tire force estimation, especially under special driving conditions, thereby enhancing the overall performance of the vehicle driving state estimation system. Although many scholars have proposed tire force prediction methods based on smart tire technology, there are still areas for optimization, mainly including improving the signal acquisition quality of the smart tire system and optimizing the mapping relationship between signal features and tire force. Regarding signal acquisition, the current field of smart tire force prediction lacks sufficient theoretical support for the scientific selection of sensor installation locations, making it difficult to guarantee continuous and high-quality acquisition of tire dynamic features during tire operation. In terms of mapping relationship construction, incorrect or inappropriate selection of mapping methods can also lead to poor performance of the final tire force prediction model.

[0005] While smart tires can directly measure lateral forces and slip ratios, enhancing vehicle lateral dynamics estimation, their performance can be limited by road conditions and tire contact patch. Against the backdrop of rapid advancements in satellite technology, low Earth orbit (LEO) satellites have become a disruptive innovation. These satellites, located at altitudes of approximately 100 to 500 miles above the Earth's surface, have fundamentally changed our communication methods, data collection techniques, and Earth monitoring models. Simultaneously, the rapid development of LEO technology has made it possible to acquire high-precision information on vehicle longitudinal speed and position. As longitudinal vehicle speed is crucial for estimation, accurate measurement can clearly further improve the accuracy and reliability of vehicle driving state estimation.

[0006] However, neither academia nor industry has yet proposed a scheme for estimating key vehicle driving states based on the fusion of multi-source information from low-orbit satellites (for longitudinal vehicle speed measurement) and on-board sensors (traditional on-board sensors + intelligent tire sensors). Summary of the Invention

[0007] To address the shortcomings of the existing technologies, the present invention aims to provide a method for estimating key vehicle driving states based on the fusion of multi-source information from low-orbit satellites and on-board sensors. This method uses intelligent tire lateral force estimates, low-orbit satellite longitudinal vehicle speed measurements, and yaw rate and lateral acceleration measurements as measurement signals. A vehicle key driving state estimation model is designed using a novel adaptive extended Kalman filter to obtain the vehicle's real-time centroid sideslip angle, yaw rate, and longitudinal speed. This multi-source information fusion method achieves vehicle key driving state estimation, improving the accuracy and reliability of vehicle driving state estimation.

[0008] The technical solution adopted in this invention is as follows:

[0009] A method for estimating key vehicle driving states based on the fusion of multi-source information from low-orbit satellites and vehicle-mounted sensors is characterized by the following steps:

[0010] S1. Based on low-orbit satellite measurement signals and combined with the Doppler frequency shift formula, obtain the real-time longitudinal velocity of the vehicle;

[0011] S2. Based on the intelligent tire measurement signal and combined with the Gaussian process regression algorithm, obtain the real-time lateral force of the tire;

[0012] S3. Based on the real-time lateral force of the tire and combined with the vehicle sensor information, design a tire lateral stiffness adaptive algorithm to obtain the real-time lateral stiffness adaptive value.

[0013] S4. Acquire vehicle parameters and real-time sensor signals, and combine them with real-time adaptive lateral stiffness values ​​to establish a vehicle dynamics model.

[0014] S5. Based on the vehicle dynamics model and the new adaptive extended Kalman filter, establish a key vehicle driving state estimation algorithm to obtain real-time vehicle longitudinal velocity, centroid sideslip angle and yaw rate.

[0015] Furthermore, S1 includes the following steps:

[0016] S1-1. Based on the low-orbit satellite ephemeris data received by the vehicle-mounted receiver, calculate the position of each satellite in the geodetic coordinate system;

[0017] S1-2. Based on the position of each low-Earth orbit satellite, determine the visible range of each low-Earth orbit satellite, obtain the intersection of the visible ranges, and divide the grid to select candidate initial value points.

[0018] S1-3. Establish the Doppler frequency shift equation using the relative motion between the vehicle and the satellite, linearize the Doppler frequency shift equation, and substitute the candidate initial points; use the least squares method to solve the real-time position of the vehicle and calculate the real-time longitudinal velocity of the vehicle.

[0019] Furthermore, the constructed Doppler frequency shift equation is expressed as:

[0020]

[0021] Where, Δf i Let f0 be the i-th frequency shift observation, and P be the satellite's signal transmission frequency. u P represents the location to be determined for the vehicle. s Let v be the satellite's position vector. u The desired speed of the vehicle, v sτ represents the satellite's velocity, c represents the signal propagation speed, and τ represents the t-axis. i This represents the frequency shift measurement error;

[0022] By linearizing the Doppler frequency shift equation and substituting the candidate initial points, the Doppler observation equation is approximated as:

[0023]

[0024] Where x is the variable to be determined; Δx is the vehicle position and speed correction to be solved, denoted as Δx = (Δx u ,Δy u ,Δz u ,Δv x , u ,Δv y,u ,Δv z,u ), Δx u Δy u Δz u Δv x,u Δv y,u Δv z,u These are the vehicle's longitudinal position correction, lateral position correction, vertical position correction, longitudinal speed correction, lateral speed correction, and vertical speed correction. The Jacobian matrix representing the variables to be determined in the observation equation; v u (0) Let be the initial velocity vector of the ground vehicle.

[0025] Furthermore, taking the nadir point of each low-Earth orbit satellite as the center, its visible radius r on the ground is calculated. v :

[0026]

[0027] The visible projection range of each low-Earth orbit satellite in the latitude and longitude directions is:

[0028]

[0029] Take the visible range of all low-Earth orbit satellites, calculate their intersection area, and represent it as a rectangle:

[0030]

[0031] Divide the intersection matrix into an N×N grid, and use the center point of each grid as a candidate initial value point:

[0032]

[0033] Where R is the Earth's radius, h s It is the altitude of satellite s, L lonL represents the longitudinal boundary of the satellite's visible projection range. lat The rectangular area represents the lateral boundary of the satellite's visible projection range, where ΔΦ is the span of this rectangular area in the latitudinal direction. min Φ represents the southernmost boundary of the intersection region, i.e., the lowest latitude value within the rectangular area. max Let λ be the northernmost boundary of the intersection region, i.e., the highest latitude value within the rectangular area, and let Δλ be the span of this rectangular area in the longitude direction. min λ represents the westernmost boundary of the intersection region, i.e., the lowest longitude value within the rectangular area. max Φ represents the easternmost boundary of the intersection region, i.e., the highest longitude value within the rectangular area. i Let λ be the latitude value of the i-th row of the grid, N be the number of grid divisions, and λ be the latitude value of the grid. j Let be the longitude value of the j-th grid.

[0034] Furthermore, S2 includes the following steps:

[0035] S2-1. Establish a PVDF-type intelligent tire finite element model, verify the accuracy of the modeling method through real vehicle tests, and build an array-type PVDF-type intelligent tire finite element model based on this model.

[0036] S2-2. Determine the influencing factors and their value range of the output voltage of the intelligent tire PVDF sensor array based on daily driving conditions, and generate the sample matrix S by combining the Sobol sequence method;

[0037] The sample matrix S is decomposed into 2+D sub-sample matrices, where D represents the number of factors affecting the output voltage of the array-type piezoelectric thin film sensor.

[0038] S2-3. Quantize the z installation positions of the circumferential sensor array in the inner liner by the tire lateral tread pattern, and install z PVDF sensor arrays to form a new smart tire finite element model. On this basis, use the new smart tire finite element model to calculate 2+D sub-sample matrices, and extract the output voltage of the z sensor arrays and their corresponding tire lateral forces to form a voltage dataset Data_VT1 and a lateral force dataset Data_Fy.

[0039] S2-4. The average value of the first-order main effect index of the output voltage values ​​of all sensors in the g-th PVDF sensor array is taken as the first-order main effect index of the g-th installation location, denoted as... The g first-order main effect indices are sorted in descending order, and the top t installation positions are selected as the optimal installation positions for the PVDF sensor array.

[0040] S2-5. Based on the determined installation positions of the t sensor arrays, process the data of the g-th PVDF sensor array, Data_VT1. gFurther extraction is performed, and l specific voltage values ​​are selected as input features for each sensor array voltage data. The selection logic is as follows: taking the maximum voltage value in the sensor array under each operating condition as a reference, c voltage value points are selected along the direction of voltage decrease to both sides of that point, and finally l voltage value points are selected as input features. The processed data is denoted as Data_VT2. g ;

[0041] Extract the datasets Data_Fy and Data_VT2 g The dataset is divided into a training set and a test set, with the training set denoted as Data_Fy. x and Data_VT2 g_x The test sets are denoted as Data_Fy. c and Data_VT2g c ;

[0042] S2-6, Based on Data_Fy and Data_VT2 g Based on the data characteristics, the types of the mean function m(x) and covariance function k(x,x') are determined by combining empirical methods;

[0043] The voltage data and tire lateral force in the training dataset are expressed as a Gaussian process distribution, i.e., the lateral force Data_Fy. x The function f(Data_VT2) g_x ) follows a mean of m(Data_VT2) g_x ), covariance is k(Data_VT2) g_x Data_VT2 g_x A Gaussian process is used; based on this, the initial values ​​of the hyperparameters and the number of iterations in the Gaussian process regression model are determined, and the hyperparameters σ and F in the covariance function are determined by minimizing the negative logarithm of the marginal likelihood function using the gradient descent method. The negative logarithm of the marginal likelihood function is specifically expressed as:

[0044]

[0045] Where -logp(f|x) represents the negative logarithm of the marginal likelihood function, f is the Gaussian process regression function, and x is the voltage input characteristic. Used to fit training data, n is the dimension of the training dataset, k n Let be an n-dimensional covariance function. Let I be an n-dimensional variance vector. n It is an n-dimensional unit vector. It is a penalty item. It is a normalization constant;

[0046] Based on the prior definition, and combining the training and test sets collected in S2-1, the joint prior distribution can be derived as follows:

[0047]

[0048] Where, f(Data_VT2) g_x f(Data_VT2) represents the lateral force of the tires corresponding to the training set. g_c ) represents the tire lateral force corresponding to the test number, and N represents the normal distribution;

[0049] S2-7, Calculate the estimated value f(Data_VT2) g_c The posterior distribution of ) is obtained and substituted into the hyperparameter solution of tire real-time lateral force obtained in S2-6.

[0050] Furthermore, the expression for the first-order main effect index is:

[0051]

[0052] in, The first-order principal effect index represents the effect of the Δth influencing factor on the voltage output of the gth sensor array, where 1 ≤ Δ ≤ D. When Δ = 5, the influencing factor is the tire slip angle, and D is the number of influencing factors; Mean(Data_VT1(A) i Data_VT1(B) i ) represents the average value of the numerical vector formed by concatenating the i-th column of the sensor array output voltage matrix corresponding to matrix A and the i-th column of the sensor array output voltage matrix corresponding to matrix B. This represents the j-th value in the i-th column of the sensor array output voltage matrix corresponding to matrix B. Representative matrix AB Δ The corresponding value of the j-th column of the sensor array output voltage matrix. This represents the j-th value in the i-th column of the sensor array output voltage matrix corresponding to matrix A.

[0053] Furthermore, step S3 includes the following steps:

[0054] S3-1. Obtain the front and rear wheelbases of the vehicle. f and l r Calculate the vertical loads on the four wheels of the vehicle when it is stationary, including the static vertical loads Ld of the left front, right front, left rear, and right rear tires. fl , Ld fr , Ld rl and Ld rr ;

[0055] The tire load was Ld respectively. fl and Ld rlSide slip characteristic tests under certain conditions were conducted to obtain side slip angle-lateral force data. The fixed values ​​K of the front and rear wheel side slip stiffness were then obtained by combining the magic tire formula for lateral force and curve fitting methods. f and K r ;

[0056] S3-2. Define the steering angles of the two front wheels as equal to δ, and calculate δ based on the signals from the on-board sensors; define the slip angles of the two front axle wheels as equal to α. f The two rear axle wheels have the same slip angle, which is equal to α. r ;

[0057] S3-3. The tire's lateral stiffness is adaptively adjusted using the difference between the linear tire force and the intelligent tire lateral force estimate; the adaptive values ​​of the front and rear tire lateral stiffness can be expressed as:

[0058]

[0059] in, and These are the adaptive values ​​for the lateral stiffness of the front and rear wheels, respectively. and These are the sum of the estimated lateral forces of the two front wheels and the two rear wheels, respectively; F yf and F yr These are the sum of the linear tire forces of the two front wheels and the two rear wheels, respectively.

[0060] Furthermore, the fixed values ​​of the lateral stiffness of the front and rear tires can be expressed as:

[0061]

[0062] Among them, B f C f and D f For load Ld fl or Ld fr The parameters B obtained by fitting the corresponding lateral characteristic data r C r and D r For load Ld rl or Ld rr The parameters are obtained by fitting the corresponding lateral deviation characteristic data.

[0063] Furthermore, S4 specifically includes the following steps:

[0064] S4-1. Acquire vehicle parameters, including vehicle mass, vehicle yaw moment of inertia, distance from the vehicle's center of gravity to the front axle and distance to the rear axle; acquire real-time signals from onboard sensors, including vehicle longitudinal and lateral acceleration, yaw rate and steering wheel angle.

[0065] S4-2. A vehicle dynamics model is constructed based on the real-time adaptive values ​​of the front and rear wheel lateral stiffness, vehicle parameters, and real-time signals from onboard sensors. Specifically, it is expressed as follows:

[0066]

[0067] Among them, a x For the longitudinal acceleration of the vehicle, I z Let the vehicle's yaw moment of inertia be... For longitudinal acceleration, v x γ is the longitudinal velocity of the vehicle, and γ is the yaw rate. Let β be the yaw acceleration, β be the sideslip angle of the vehicle's center of gravity, and m be the mass of the vehicle. and These are the adaptive values ​​of the lateral stiffness of the front and rear wheels, l f and l r These are the front and rear wheelbases of the vehicle, I z Let be the moment of inertia of the vehicle about the z-axis.

[0068] Furthermore, step S5 includes the following steps:

[0069] S5-1. Define the state vector J, input vector U, and observation vector Z of the vehicle critical driving state estimation system as follows:

[0070]

[0071] Among them, a x Let a be the longitudinal acceleration of the vehicle. y Let v be the lateral acceleration of the vehicle, γ be the yaw rate, and v be the lateral acceleration. x β is the vehicle's longitudinal velocity, β is the vehicle's sideslip angle, and γ is the yaw rate.

[0072] S5-2. Define the prior state estimation equation P(·) and the observation estimation equation H(·), as follows:

[0073]

[0074] Where P1, P2, and P3 represent the prior estimates of the vehicle's longitudinal velocity, sideslip angle, and yaw rate, respectively. x0 β0 and γ0 represent the initial values ​​of the vehicle's longitudinal velocity, sideslip angle, and yaw rate, respectively. s The system sampling time; and These are the sum of the estimated lateral forces of the two front wheels and the two rear wheels, respectively.

[0075] S5-3. Establish an adaptive adjustment strategy for the noise covariance matrix based on sliding window self-adjustment, and obtain the adaptive noise covariance matrix. First, define the covariance matrix of the innovation as:

[0076] C k =E[v k v k T ]

[0077] Among them, v k Let be the error between the actual and predicted values ​​of the measured variable at step k, where k > 0; E is the expected value.

[0078] The actual covariance matrix of the innovation is defined as follows:

[0079]

[0080] Where M represents the length of the sliding window, expressed as:

[0081]

[0082] Where int(·) is the floor function. It is a constant;

[0083] Noise covariance adaptive matrix The expression is written as:

[0084]

[0085] Where trace(·) is the trace function, and R is the noise covariance matrix;

[0086] S5-4. Based on the principle of the extended Kalman filter and the first-order Taylor series expansion theorem, the calculation process calculates the parameters, including the Jacobian matrices Pm and Hm, as well as the prior estimation error covariance matrix Pf and the Kalman gain Km, which are expressed as follows:

[0087]

[0088] Pf k =Pm k ·Pl k-1 ·Pm k T +Q

[0089]

[0090] Among them, Pm k Let Hm be the Jacobian matrix of the state transition equation at step k. k Let Pl be the Jacobian matrix of the k-th step measurement equation. k-1Let Q be the error covariance matrix at step k-1, and let Q be the process noise covariance matrix. The parameters of Q can be manually calibrated by trial and error.

[0091] S5-5. Estimate the key driving states of the vehicle based on prior state estimation information, measurement estimation information, and sensor measurement information, and update the error covariance matrix Pl. The specific formula is as follows:

[0092]

[0093] Pl k =(I3-Km) k ·Hm k )·Pl k

[0094] Where I is the identity matrix, J k This represents the estimated critical driving state of the vehicle at step k, including the vehicle's longitudinal velocity, sideslip angle, and yaw rate; Km k Let K be the Kalman gain at step k.

[0095] Beneficial effects:

[0096] (1) The present invention establishes an array-type PVDF intelligent tire finite element model based on finite element software, which greatly reduces the time and financial costs required for the scientific selection of the PVDF sensor array installation position in the tire and the extraction of training data for the tire lateral force estimation algorithm.

[0097] (2) For the established array-type PVDF smart tire, the present invention also measures the sensitivity of the lateral force change during tire operation to the output voltage of the PVDF sensor array based on the Sobol sensitivity analysis method, and determines the optimal installation position of the sensor array based on this, which can improve the signal acquisition capability of the smart tire system.

[0098] (3) Based on array-type PVDF intelligent tire technology, this invention combines Gaussian process regression algorithm to propose a tire lateral force estimation method with low cost, high accuracy, good stability and strong generalization.

[0099] (4) Based on the prior information of tire lateral force provided by the array-type PVDF intelligent tire system, this invention combines vehicle-mounted sensing information and vehicle dynamics methods to design a tire lateral stiffness adaptive algorithm, which is beneficial to improving the accuracy of the vehicle model built on this basis.

[0100] (5) Based on vehicle state information measured by low-orbit satellites and vehicle-mounted sensors, and vehicle dynamics methods, this invention combines the principle of the novel adaptive extended Kalman filter to design a high-precision and high-reliability method for estimating key vehicle driving states.

[0101] (6) Based on the tire lateral force estimation value provided by the intelligent tire system and the tire lateral force calculation value calculated by the vehicle dynamics method, combined with the Kalman filter innovation adaptive principle, a sliding window self-adjusting noise covariance matrix adaptive adjustment strategy is designed, which is beneficial to improve the estimation accuracy of key vehicle driving states. Attached Figure Description

[0102] Figure 1 This is a schematic flowchart of an embodiment of the method for estimating key vehicle driving states based on the fusion of multi-source information from low-orbit satellites and vehicle-mounted sensors in this invention patent.

[0103] Figure 2 This is the modeling process for the finite element model of the piezoelectric intelligent tire in this invention patent.

[0104] Figure 3 This is a schematic diagram of the finite element model of the array-type PVDF intelligent tire in this invention patent.

[0105] Figure 4 This diagram illustrates all potential installation locations of the PVDF sensor array in the tire liner.

[0106] Figure 5 The results show the calculation of the first-order main effect indexes for different PVDF sensor array installation positions under varying lateral forces.

[0107] Figure 6 This is a schematic diagram of a three-free vehicle dynamics model.

[0108] Figure 7 The results show the estimated lateral force of the left front wheel intelligent tire when a vehicle travels at high speed through a double lane change test road on a high-adhesion road surface.

[0109] Figure 8 This is the estimated vehicle speed when a vehicle on a high-adhesion road passes through a double lane change test road at high speed.

[0110] Figure 9 This is the estimated center of gravity sideslip angle when a vehicle travels at high speed through a double lane change test road under a high-adhesion surface.

[0111] Figure 10 This is the estimated yaw rate of a vehicle traveling at high speed through a double lane change test road on a high-adhesion surface.

[0112] Figure 11 The results show the estimated lateral force of the left front wheel intelligent tire when a vehicle travels at a medium speed through a double lane change test road on a low-friction surface.

[0113] Figure 12 This is the estimated vehicle speed when a vehicle travels at a medium speed through a double lane change test road on a low-friction surface.

[0114] Figure 13This is the estimated center of gravity sideslip angle when a vehicle travels at a medium speed through a double lane change test road on a low-adhesion surface.

[0115] Figure 14 This is the estimated yaw rate of a vehicle traveling at a medium speed through a double lane change test road on a low-friction surface. Detailed Implementation

[0116] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0117] Combined with appendix Figure 1-14 As shown, this invention proposes a method for estimating key vehicle driving states based on the fusion of multi-source information from low-orbit satellites and vehicle-mounted sensors, comprising the following steps:

[0118] S1. Based on low-orbit satellite measurement signals and combined with the Doppler frequency shift formula, obtain the real-time longitudinal velocity of the vehicle;

[0119] First, based on the low-Earth orbit satellite ephemeris data received by the vehicle-mounted receiver, the position of each satellite in the geodetic coordinate system is calculated. The acquired ephemeris data includes the three-dimensional coordinates P of each satellite. s =(X s ,Y s Z s ), velocity v s and the transmit signal rate f0, where X s Y s Z s The coordinates of satellite s are X, Y, and Z, respectively.

[0120] Furthermore, the three-dimensional coordinates of the low-orbit satellite are converted to the geodetic coordinate system, which can be specifically expressed as:

[0121]

[0122] Where, Φ s , λ s and h s These are the latitude, longitude, and altitude of satellite s, respectively, and R is the Earth's radius.

[0123] After obtaining the position of each low-Earth orbit satellite, it is necessary to determine the visible range of each low-Earth orbit satellite, calculate the intersection of the visible ranges, and divide the grid to select candidate initial value points in order to narrow down the solution range.

[0124] First, using the nadir point of each low-Earth orbit satellite as the center, calculate its visible radius r on the ground.v :

[0125]

[0126] Furthermore, the visible projection range of each low-orbit satellite in the latitude and longitude directions is as follows:

[0127]

[0128] Among them, L lon L represents the longitudinal boundary of the satellite's visible projection range. lat This represents the lateral boundary of the satellite's visible projection range.

[0129] Furthermore, taking the visible range of all low-Earth orbit satellites, we calculate the intersection region of these visible ranges, representing it as an intersection rectangle:

[0130]

[0131] Where ΔΦ is the span of the rectangular region in the latitudinal direction, Φ min Φ represents the southernmost boundary of the intersection region, i.e., the lowest latitude value within the rectangular area. max Let λ be the northernmost boundary of the intersection region, i.e., the highest latitude value within the rectangular area, and let Δλ be the span of this rectangular area in the longitude direction. min λ represents the westernmost boundary of the intersection region, i.e., the lowest longitude value within the rectangular area. max This represents the easternmost boundary of the intersection region, i.e., the highest longitude value within the rectangular area.

[0132] Furthermore, the intersection matrix is ​​divided into a 200×200 grid, and the center point of each grid is used as a candidate initial value point:

[0133]

[0134] Where, Φ i Let λ be the latitude value of the i-th row of the grid, N be the number of grid divisions, and λ be the latitude value of the grid. j Let be the longitude value of the j-th grid.

[0135] The Doppler frequency shift equation is established using the relative motion between the vehicle and the satellite, and the real-time position of the vehicle is calculated using the least squares method. The constructed Doppler frequency shift equation is as follows:

[0136]

[0137] Where, Δf i Let f0 be the i-th frequency shift observation, and P be the satellite's signal transmission frequency. u P represents the location to be determined for the vehicle. s Let v be the satellite's position vector. uThe desired speed of the vehicle is represented by τ, where c represents the signal propagation speed, and τ represents the speed of signal propagation. i This represents the frequency shift measurement error. Based on this, the Doppler frequency shift equation is linearized, and candidate initial points are substituted. At this point, the Doppler observation equation can be approximated as:

[0138]

[0139] Where x is the variable to be determined; Δx is the vehicle position and speed correction to be solved, denoted as Δx = (Δx u ,Δy u ,Δz u ,Δv x , u ,Δv y,u ,Δv z,u ), Δx u Δy u Δz u Δv x,u Δv y,u Δv z,u These are the vehicle's longitudinal position correction, lateral position correction, vertical position correction, longitudinal speed correction, lateral speed correction, and vertical speed correction. The Jacobian matrix represents the variables to be determined in the observation equation.

[0140] Furthermore, based on the vehicle position and velocity correction Δx to be solved, the real-time longitudinal velocity of the vehicle is solved using the least squares method. First, for observations from multiple low-orbit satellites, an overdetermined system of equations is constructed:

[0141] d=H·Δx+τ (8)

[0142] Among them, d=[Δf1,Δf2,...,Δf n ] T , represents the frequency shift observation vector, H represents the observation matrix, which contains the unit direction vectors of the satellite and the vehicle and the satellite velocity, Δx is the vehicle position and velocity correction to be solved, and τ is the observation noise.

[0143] Furthermore, the solution is obtained using the least squares method:

[0144] Δx=(H T H) -1 H T d (9)

[0145] Furthermore, the vehicle's position and speed are updated iteratively in real time:

[0146]

[0147] Finally, once the residual ||d|| meets the set threshold (0.001 in this embodiment), v is extracted. u The velocity in the x-direction is taken as the real-time longitudinal velocity of the vehicle.

[0148] S2. Based on intelligent tire measurement signals and combined with Gaussian process regression algorithm, obtain the real-time lateral force of the tire.

[0149] First, based on Figure 2 The modeling process shown establishes a finite element model of a smart tire with a piezoelectric thin-film sensor installed in the inner liner, and verifies the accuracy of the modeling method through a real vehicle test platform. Based on this, as... Figure 3 As shown, an array-type piezoelectric thin film smart tire finite element model is built by using a circumferential array of individual piezoelectric thin film sensors, where a single circumferential sensing array contains a total of 35 piezoelectric thin film sensors.

[0150] Furthermore, to ensure the selected PVDF sensor array installation location has good adaptability to various operating conditions, the influence of output voltage should be considered as many factors as possible when designing the simulation scheme for calculating the first-order main effect index. Given that the purpose of this invention is to construct a tire lateral force estimation model, the influencing factors considered include tire slip ratio, sideslip angle, vertical load, road adhesion coefficient, vehicle speed, and tire pressure, and the value ranges shown in Table 1 are determined based on daily vehicle driving conditions.

[0151] Table 1: Value range of each influencing factor

[0152]

[0153]

[0154] Furthermore, by combining sequence generation methods, a 200×12 dimensional sample matrix S is generated, written as:

[0155]

[0156] Among them, S ij These are the calculated values ​​of the upper and lower limits of the Sobol sequence generation algorithm combined with the values ​​of six influencing factors.

[0157] Furthermore, the sample matrix S is decomposed into 8 sub-sample matrices, written as:

[0158]

[0159] Where, change(B i A i This represents replacing the i-th column of matrix B with the i-th column of matrix A to form a new matrix AB. iA total of 6 matrix replacement operations were performed to form 6 subsample matrices, where each row of each subsample matrix represents an additional set of test conditions.

[0160] In fact, the position of the PVDF sensor array in the tire circumferential direction does not need to be determined, because it is always symmetrical with respect to the tire's longitudinal central symmetry plane. Therefore, as... Figure 4 As shown, this invention quantifies and names the potential installation locations of the sensor arrays on the lateral tread pattern of a single tire pitch using a finite element tire model, and installs 62 PVDF sensor arrays to form a new intelligent tire finite element model. Based on this, the new model is used to calculate eight sub-sample matrices, and the output voltages of the 62 sensor arrays and their corresponding tire lateral forces are extracted, forming a voltage dataset Data_VT1 and a lateral force dataset Data_Fy, written as:

[0161]

[0162] Among them, Data_VT1 g Represents the g-th PVDF sensor array, fy represents the output voltage of the j-th sensor on the g-th PVDF sensor array under the i-th test condition. i This represents the tire lateral force under the i-th test condition. The average of the first-order principal effect index (FMEI) of the output voltage values ​​of all sensors in the g-th PVDF sensor array is taken as the first-order principal effect index for the g-th installation location, denoted as MFMEI, which can be expressed as:

[0163]

[0164] in, Mean(Data_VT1(A)) represents the first-order principal effect index of the voltage output of the g-th sensor array on the Δ-th influencing factor (1≤Δ≤6, where Δ=1 represents tire pressure, Δ=2 represents vertical load, Δ=3 represents vehicle speed, Δ=4 represents adhesion coefficient, Δ=5 represents sideslip angle, and Δ=6 represents slip ratio). i Data_VT1(B) i ) represents the average value of the numerical vector formed by concatenating the i-th column of the sensor array output voltage matrix corresponding to matrix A and the i-th column of the sensor array output voltage matrix corresponding to matrix B. This represents the j-th value in the i-th column of the sensor array output voltage matrix corresponding to matrix B. Representative matrix AB Δ The corresponding value of the j-th column of the sensor array output voltage matrix. This represents the j-th value in the i-th column of the sensor array output voltage matrix corresponding to matrix A.

[0165] Furthermore, considering that tire force cannot be directly controlled, this invention indirectly controls tire lateral force by controlling the tire slip angle. Figure 5 The results show the top 10 MFMIE rankings in descending order as the sideslip angle changes. Considering the limitations of in-tire installation space in practical applications, this invention selects the top 4 positions, namely 24, 37, 44, and 17, as the optimal installation positions for the PVDF sensor array. Based on this, the finite element model of the array-type PVDF smart tire is reconstructed to obtain the smart tire finite element model with the optimal in-tire sensor configuration.

[0166] After determining the installation locations of the four sensor arrays, the obtained voltage data Data_VT1 was processed. g Further extraction was performed, and five specific voltage values ​​were selected as input features for each sensor array voltage data point. The specific selection logic was as follows: taking the maximum voltage value in the sensor array under each operating condition as a reference, four voltage value points were selected along the direction of voltage decrease to both sides of that point, and finally five voltage value points were selected as input features. The reselected data was denoted as Data_VT2. g Specifically, it can be written as:

[0167]

[0168] Furthermore, extract the obtained datasets Data_Fy and Data_VT2. g 80% of it is used as the training set, denoted as Data_Fy x and Data_VT2 g_x 20% was used as the test set, denoted as Data_Fy c and Data_VT2g c Based on this, according to Data_Fy and Data_VT2 g Based on the data characteristics, and combined with empirical methods, the types of the mean function m(x) and covariance function k(x,x') are determined, written as:

[0169] m(x)=E[f(x)] (16)

[0170]

[0171] Where f(x) is the latent function, representing Data_VT2 g_x enter The corresponding tire lateral force fy, E is the mathematical expectation, σ 2 Let F be the signal variance. i Let x represent the variance scale of the i-th dimension. i Given the known data in dimension i, x i 'for x iThe transpose of x, where d is the dimension of variable x.

[0172] Furthermore, the voltage data and tire lateral force in the training dataset are expressed as a Gaussian process distribution, i.e., the lateral force Data_Fy is represented by this distribution. x The function f(Data_VT2) g_x ) follows a mean of m(Data_VT2) g_x ), covariance is k(Data_VT2) g_x Data_VT2 g_x A Gaussian process is used. Based on this, the initial values ​​of the hyperparameters and the number of iterations in the Gaussian process regression model are determined, and the hyperparameters σ and F in the covariance function are determined by minimizing the negative logarithm of the marginal likelihood function using the gradient descent method. The negative logarithm of the marginal likelihood function can be specifically expressed as:

[0173]

[0174] Where -logp(f|x) represents the negative logarithm of the marginal likelihood function, f is the Gaussian process regression function, and x is the voltage input characteristic. Used to fit training data, n is the dimension of the training dataset, k n Let be an n-dimensional covariance function. Let I be an n-dimensional variance vector. n It is an n-dimensional unit vector. It is a penalty item. It is a normalization constant.

[0175] Furthermore, based on the prior definition and combining the collected training and test sets, the joint prior distribution can be derived as follows:

[0176]

[0177] Where, f(Data_VT2) g_x f(Data_VT2) represents the lateral force of the tires corresponding to the training set. g_c ) represents the tire lateral force corresponding to the test number, and N represents the normal distribution.

[0178] Furthermore, the estimated value f(Data_VT2) is calculated. g_c The posterior distribution of ) is written as:

[0179]

[0180] in, The mean value represents the estimated value. Substituting the obtained hyperparameters into the equation yields the real-time tire lateral force f(Data_VT2g_c). cov(f(Data_VT2) g_x)) represents the variance corresponding to f(Data_VT2g_x).

[0181] S3. Based on S2, obtain the real-time lateral force of the tire, and combine it with the vehicle sensor information to design a tire lateral stiffness adaptive algorithm to obtain the real-time lateral stiffness adaptive value.

[0182] Obtain the front and rear wheelbase of the vehicle. f (In this embodiment, it is 1.015m) and l r (In this embodiment, the value is 1.895m). The vertical load on the four wheels of the vehicle when it is stationary can be calculated as follows:

[0183]

[0184] Among them, Ld fl , Ld fr , Ld rl and Ld rr These represent the static vertical loads on the left front, right front, left rear, and right rear tires of the vehicle, respectively. m represents the total mass of the vehicle body (1412 kg in this example), and g represents the acceleration due to gravity (9.81 m / s² in this example). 2 );l f The distance from the front axle to the center of gravity of the vehicle, l r This is the distance from the rear axle of the vehicle to its center of gravity.

[0185] Furthermore, lateral characteristic tests were conducted under tire loads of 4510.1 N and 2415.7 N to obtain lateral angle-lateral force data. The fixed values ​​K of the front and rear wheel lateral stiffness were then obtained by combining the lateral force magic tire formula and curve fitting method. f and K r The formula for the lateral force magic tire is as follows:

[0186] Y=Dsin(Carctan(Bx-E(Bx-arctan(Bx)))) (22)

[0187] Where Y is the tire lateral force output term, x is the tire slip angle input term, and B, C, D, and E are the parameters to be fitted.

[0188] Furthermore, the fixed values ​​of the front and rear tire lateral stiffness can be expressed as:

[0189]

[0190] Among them, B f C f and D f For load Ld fl (or Ld) fr The parameters obtained by fitting the lateral deviation characteristic data, B r Cr and D r For load Ld rl (or Ld) rr The parameters are obtained by fitting the corresponding sideslip characteristic data. Define the steering angles of the two front wheels as being the same and equal to δ. Based on this, δ is calculated using signals from the onboard sensors, which can be expressed as:

[0191]

[0192] Where, δ sw Represents the steering wheel angle, i w The angular ratio of the vehicle's steering system is 17.6 in this embodiment. The slip angles of the two front axle wheels are defined to be the same and equal to α. f The two rear axle wheels have the same slip angle, which is equal to α. r At this point, the slip angles of the front and rear wheels can be expressed as:

[0193]

[0194] Among them, β, γ and v x These are the vehicle's center of gravity sideslip angle, yaw rate, and longitudinal velocity, respectively, which can be obtained from the vehicle's key driving state estimator. The tire's lateral stiffness is adaptively adjusted using the difference between the linear tire force and the intelligent tire lateral force estimate. At this point, the adaptive values ​​of the front and rear tire lateral stiffness can be expressed as:

[0195]

[0196] in, and These are the adaptive values ​​for the lateral stiffness of the front and rear wheels, respectively. and F represents the sum of the estimated lateral forces of the two front wheels and the two rear wheels, respectively, which can be obtained from the established tire lateral force estimation model. yf and F yr These are the sum of the linear tire forces of the two front wheels and the two rear wheels, respectively, which can be obtained by multiplying the obtained fixed value of lateral stiffness by the tire lateral slip angle.

[0197] S4. Acquire vehicle parameters and real-time sensor signals, and combine them with real-time adaptive lateral stiffness values ​​to establish a vehicle dynamics model.

[0198] Obtain vehicle parameters, including vehicle mass (1412 kg in this embodiment) and vehicle yaw moment of inertia (1536.7 kg / m in this embodiment). 2 The distance from the vehicle's center of gravity to the front axle (1.015m in this embodiment) and the distance to the rear axle (1.895m in this embodiment) are measured; real-time signals from onboard sensors are acquired, including vehicle longitudinal and lateral acceleration, yaw rate, and steering wheel angle.

[0199] Establish such as Figure 6 The three-degree-of-freedom vehicle dynamics model is shown, and the following assumptions are made:

[0200] (1) Ignoring car aerodynamics;

[0201] (2) Ignore the car's roll, pitch and vertical motion, and only consider the longitudinal, lateral and yaw motion;

[0202] (3) The road surface slope angle and inclination angle are zero;

[0203] (4) The car's center of mass is fixed and coincides with the origin of the car's coordinate system;

[0204] (5) The front wheel turning angle of a car is linearly related to the steering wheel turning angle, while the rear wheel turning angle remains constant.

[0205] Furthermore, the dynamic equations of the vehicle in the longitudinal, lateral, and yaw directions can be obtained:

[0206]

[0207] Among them, a x For the longitudinal acceleration of the vehicle, I z Let the vehicle's yaw moment of inertia be... For longitudinal acceleration, ...

[0208] S5. Based on the vehicle dynamics model and the new adaptive extended Kalman filter, establish a key vehicle driving state estimation algorithm to obtain the real-time vehicle longitudinal velocity, centroid sideslip angle and yaw rate.

[0209] First, the state vector J, input vector U, and observation vector Z of the vehicle key driving state estimation system are defined as follows:

[0210] J = (v x , β, γ) T =(J1, J2, J3) T (28)

[0211] U=(a x , δ) T =(U1, U2) T (29)

[0212] Z = (a y , v x , γ) T =(Z1, Z2, Z3) T (30)

[0213] Among them, a x Let a be the longitudinal acceleration of the vehicle. yThe lateral acceleration of the vehicle can be obtained from onboard sensors. Similarly, the yaw rate γ can also be easily obtained from onboard sensors. The longitudinal velocity v of the vehicle... x It can be obtained by low-orbit satellite measurement. The vehicle's center of gravity sideslip angle β is in a state to be estimated, and δ is the front wheel steering angle.

[0214] Then, the state prior estimation equation P(·) and the observation estimation equation H(·) are defined as follows:

[0215]

[0216] Where P1, P2, and P3 represent the prior estimates of the vehicle's longitudinal velocity, sideslip angle, and yaw rate, respectively. x0 β0 and γ0 represent the initial values ​​of the vehicle's longitudinal velocity, sideslip angle, and yaw rate, respectively, and are all set to 0 in this embodiment. s The system sampling time is set to 0.001s in this embodiment. This is the adaptive value for the front wheel lateral stiffness. denoted as the adaptive value for rear wheel lateral stiffness, where m is the vehicle mass.

[0217] Furthermore, an adaptive adjustment strategy for the noise covariance matrix based on sliding window self-adjustment is established to obtain the adaptive noise covariance matrix. First, the covariance matrix of the innovation is defined as:

[0218] C k =E[v k v k T (33)

[0219] Among them, v k Let E be the error between the actual and predicted values ​​of the measured variable at step k (k>0), and E be the expected value.

[0220] Furthermore, the actual covariance matrix of the innovation is defined as:

[0221]

[0222] Where M represents the length of the sliding window, it can be further expressed as:

[0223]

[0224] Where int(·) is the floor function. The constant is 10 in this embodiment; at this time, the noise covariance adaptive matrix is... The expression can be written as:

[0225]

[0226] Where trace(·) is the trace function, and R is the noise covariance matrix, which in this embodiment is:

[0227]

[0228] Furthermore, based on the principle of the extended Kalman filter and the first-order Taylor series expansion theorem, the calculation process computes parameters, including the Jacobian matrices Pm and Hm, as well as the prior estimation error covariance matrix Pf and the Kalman gain Km. Their expressions are as follows:

[0229]

[0230] Pf k =Pm k ·Pl k-1 ·Pm k T +Q (40)

[0231]

[0232] Among them, Pm k Let Hm be the Jacobian matrix of the state transition equation at step k. k Let Pl be the Jacobian matrix of the k-th step measurement equation. k-1 Let Pl0 be the error covariance matrix at step k-1. Pl0 refers to the value of pl at step k=1, which is the initial value of pl, obtained manually by trial and error. Q is the process noise covariance matrix, whose parameters can be manually calibrated by trial and error. In this embodiment, they are as follows:

[0233]

[0234] Furthermore, based on the prior state estimation information, measurement estimation information, and sensor measurement information, the key driving states of the vehicle are estimated, and the error covariance matrix Pl is updated. Their formulas can be specifically expressed as follows:

[0235]

[0236] Pl k =(I3-Km) k ·Hm k )·Pl k (45)

[0237] Where I is the identity matrix, J k The key driving state estimates for the vehicle at step k include the vehicle's longitudinal velocity, sideslip angle, and yaw rate.

[0238] Figures 7 to 10The key vehicle driving state estimation method proposed in this invention is tested at 80 km / h under double lane change conditions on a high-adhesion (adhesion coefficient of 0.8) road surface. C0 represents the measured quantities as sideslip angle and yaw rate, while C1 represents the measured quantities as longitudinal vehicle speed provided by a low-orbit satellite, voltage acquired by the in-tire PVDF array sensor, sideslip angle, and yaw rate. The results show that the estimation accuracy of the left front wheel lateral force is high, with an absolute error within ±40 N. Under condition C0, the maximum absolute errors of the estimated vehicle longitudinal speed, sideslip angle, and yaw rate are 0.0265 m / s, 0.004 rad, and 0.0235 rad / s, respectively. Under condition C1, the maximum absolute errors of the estimated vehicle longitudinal speed, sideslip angle, and yaw rate are 0.0011 m / s, 0.0017 rad, and 0.0006 rad / s, respectively. Compared to using only vehicle yaw rate and lateral acceleration as measurement quantities, the addition of low-orbit satellite and tire-mounted sensor information reduced the maximum absolute error of vehicle longitudinal velocity, center of gravity sideslip angle, and yaw rate estimation by 95.84%, 57.54%, and 97.45%, respectively.

[0239] Figures 11 to 14 The results of the vehicle key driving state estimation method proposed in this invention are presented in a double lane change test at 50 km / h on a high-adhesion (adhesion coefficient of 0.3) road surface. The results show that the estimated curve of the left front wheel lateral force has a very high degree of overlap with the actual curve, with an absolute estimation error within ±24 N. Under condition C0, the maximum absolute errors of the estimated vehicle longitudinal velocity, sideslip angle, and yaw rate are 0.0178 m / s, 0.0039 rad, and 0.0114 rad / s, respectively. Under condition C1, the maximum absolute errors of the estimated vehicle longitudinal velocity, sideslip angle, and yaw rate are 0.004 m / s, 0.0029 rad, and 0.0022 rad / s, respectively. Compared to using only vehicle yaw rate and lateral acceleration as measurements, the addition of low-orbit satellite and in-tire sensor information reduces the maximum absolute errors of the estimated vehicle longitudinal velocity, sideslip angle, and yaw rate by 77.52%, 25.64%, and 80.7%, respectively. In summary, compared with traditional vehicle driving state estimation methods that use onboard sensor information as the measurement quantity, the vehicle key driving state estimation method based on the fusion of multi-source information from low-orbit satellites and onboard sensors proposed in this invention exhibits superior estimation performance.

[0240] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for estimating key vehicle driving states based on the fusion of multi-source information from low-orbit satellites and vehicle-mounted sensors, characterized in that, Includes the following steps: S1. Based on low-orbit satellite measurement signals and combined with the Doppler frequency shift formula, obtain the real-time longitudinal velocity of the vehicle; S2. Based on the intelligent tire measurement signal and combined with the Gaussian process regression algorithm, obtain the real-time lateral force of the tire; S3. Based on the real-time lateral force of the tire and combined with the vehicle sensor information, design a tire lateral stiffness adaptive algorithm to obtain the real-time lateral stiffness adaptive value. S4. Acquire vehicle parameters and real-time sensor signals, and combine them with real-time adaptive lateral stiffness values ​​to establish a vehicle dynamics model. S5. Based on the vehicle dynamics model and the new adaptive extended Kalman filter, establish a key vehicle driving state estimation algorithm to obtain real-time vehicle longitudinal velocity, centroid sideslip angle and yaw rate. Step S5 includes the following steps: S5-1. Define the state vector J, input vector U, and observation vector Z of the vehicle critical driving state estimation system as follows: ； ； ； Among them, a x Let a be the longitudinal acceleration of the vehicle. y Let v be the lateral acceleration of the vehicle, γ be the yaw rate, and v be the lateral acceleration. x β is the vehicle's longitudinal velocity, β is the vehicle's sideslip angle, and γ is the yaw rate. S5-2. Define the prior state estimation equation P(·) and the observation estimation equation H(·), as follows: ； ； Where P1, P2, and P3 represent the prior estimates of the vehicle's longitudinal velocity, sideslip angle, and yaw rate, respectively. x0 β0 and γ0 represent the initial values ​​of the vehicle's longitudinal velocity, sideslip angle, and yaw rate, respectively. s The system sampling time; and These are the sum of the estimated lateral forces of the two front wheels and the two rear wheels, respectively. S5-3. Establish an adaptive adjustment strategy for the noise covariance matrix based on sliding window self-adjustment, and obtain the adaptive noise covariance matrix. First, define the covariance matrix of the innovation as: ； Among them, v k Let $\mathbf{k}$ be the error between the actual and predicted values ​​of the measured variable at step k, where $k > 0. For mathematical expectation; The actual covariance matrix of the innovation is defined as follows: ； Where M represents the length of the sliding window, expressed as: ； Where int(·) is the floor function and φ is a constant; Noise covariance adaptive matrix The expression is written as: ； Where trace(·) is the trace function, and R is the noise covariance matrix; S5-4. Based on the principle of the extended Kalman filter and the first-order Taylor series expansion theorem, the calculation process calculates the parameters, including the Jacobian matrices Pm and Hm, as well as the prior estimation error covariance matrix Pf and the Kalman gain Km, which are expressed as follows: ； ； ； ； in, Let be the Jacobian matrix of the state transition equation at step k. Let Pl be the Jacobian matrix of the k-th step measurement equation. k-1 Let Q be the error covariance matrix at step k-1, and let Q be the process noise covariance matrix. The parameters of Q can be manually calibrated by trial and error. S5-5. Estimate the key driving states of the vehicle based on prior state estimation information, measurement estimation information, and sensor measurement information, and update the error covariance matrix Pl. The specific formula is as follows: ； ； Where I is the identity matrix, J k The key driving state estimates for the vehicle at step k include the vehicle's longitudinal velocity, sideslip angle, and yaw rate. Let K be the Kalman gain at step k.

2. The method for estimating key vehicle driving states based on the fusion of multi-source information from low-orbit satellites and vehicle-mounted sensors according to claim 1, characterized in that, S1 includes the following steps: S1-1. Based on the low-orbit satellite ephemeris data received by the vehicle-mounted receiver, calculate the position of each satellite in the geodetic coordinate system; S1-2. Based on the position of each low-Earth orbit satellite, determine the visible range of each low-Earth orbit satellite, obtain the intersection of the visible ranges, and divide the grid to select candidate initial value points. S1-3. Establish the Doppler frequency shift equation using the relative motion between the vehicle and the satellite, linearize the Doppler frequency shift equation, and substitute the candidate initial points; use the least squares method to solve the real-time position of the vehicle and calculate the real-time longitudinal velocity of the vehicle.

3. The method for estimating key vehicle driving states based on the fusion of multi-source information from low-orbit satellites and onboard sensors according to claim 1, characterized in that, The constructed Doppler frequency shift equation is expressed as: ； in, For the i-th frequency shift observation, P is the satellite's signal transmission frequency. u This represents the location of the vehicle in question. Let v be the satellite's position vector. u The desired speed of the vehicle, v s τ represents the satellite's velocity, c represents the signal propagation speed, and τ represents the t-axis. i This represents the frequency shift measurement error; By linearizing the Doppler frequency shift equation and substituting the candidate initial points, the Doppler observation equation is approximated as: ； in, Let be the variable to be solved; Δx is the vehicle position and speed correction amount to be solved, denoted as Δx = (Δx u , Δy u ,Δz u , Δv x, u , Δv y, u , Δv z, u ), Δx u Δy u Δz u Δv x, u Δv y, u Δv z, u These are the vehicle's longitudinal position correction, lateral position correction, vertical position correction, longitudinal speed correction, lateral speed correction, and vertical speed correction. The Jacobian matrix represents the variables to be determined in the observation equation; Let be the initial velocity vector of the ground vehicle.

4. The method for estimating key vehicle driving states based on the fusion of multi-source information from low-orbit satellites and on-board sensors according to claim 1, characterized in that, Calculate the visible radius r of each low-Earth orbit satellite on the ground, centered on its nadir point. v : ； The visible projection range of each low-Earth orbit satellite in the latitude and longitude directions is: ； Take the visible range of all low-Earth orbit satellites, calculate their intersection area, and represent it as a rectangle: ； Divide the intersection matrix into an N×N grid, and use the center point of each grid as a candidate initial value point: ； Where R is the Earth's radius, h s It is the altitude of satellite 's'. The longitudinal boundary of the satellite's visible projection range. The horizontal boundary of the satellite's visible projection range. This represents the span of the rectangular region in the latitudinal direction. This represents the southernmost boundary of the intersection region, i.e., the lowest latitude value within the rectangular area. This represents the northernmost boundary of the intersection region, i.e., the highest latitude value within the rectangular area. This represents the span of the rectangular region along the longitude direction. This represents the westernmost boundary of the intersection region, i.e., the lowest longitude value within the rectangular area. This represents the easternmost boundary of the intersection region, i.e., the highest longitude value within the rectangular area. Let be the latitude value of the i-th row of the grid. Number of grid divisions, Let be the longitude value of the j-th grid.

5. The method for estimating key vehicle driving states based on the fusion of multi-source information from low-orbit satellites and on-board sensors according to claim 1, characterized in that, S2 includes the following steps: S2-1. Establish a PVDF-type intelligent tire finite element model, verify the accuracy of the modeling method through real vehicle tests, and build an array-type PVDF-type intelligent tire finite element model based on this model. S2-2. Determine the influencing factors and their value range of the output voltage of the intelligent tire PVDF sensor array based on daily driving conditions, and generate the sample matrix S by combining the Sobol sequence method; The sample matrix S is decomposed into 2+D sub-sample matrices, where D represents the number of factors affecting the output voltage of the array-type piezoelectric thin film sensor. S2-3. Quantize the z installation positions of the circumferential sensor array in the inner liner by the tire lateral tread pattern, and install z PVDF sensor arrays to form a new smart tire finite element model. On this basis, use the new smart tire finite element model to calculate 2+D sub-sample matrices, and extract the output voltage of the z sensor arrays and their corresponding tire lateral forces to form a voltage dataset Data_VT1 and a lateral force dataset Data_Fy. S2-4. The average value of the first-order main effect index of the output voltage values ​​of all sensors in the g-th PVDF sensor array is taken as the first-order main effect index of the g-th installation location, denoted as... The g first-order main effect indices are sorted in descending order, and the first t installation positions are selected as the optimal installation positions for the PVDF sensor array. S2-5. Based on the determined installation positions of the t sensor arrays, process the data of the g-th PVDF sensor array, Data_VT1. g Further extraction is performed, and l specific voltage values ​​are selected as input features for each sensor array voltage data. The selection logic is as follows: taking the maximum voltage value in the sensor array under each operating condition as a reference, c voltage value points are selected along the direction of voltage decrease to both sides of that point, and finally l voltage value points are selected as input features. The processed data is denoted as Data_VT2. g ; Extract the datasets Data_Fy and Data_VT2 g The dataset is divided into a training set and a test set, with the training set denoted as Data_Fy. x and Data_VT2 g_x The test sets are denoted as Data_Fy. c and Data_VT2g c ; S2-6, Based on Data_Fy and Data_VT2 g Based on the data characteristics, the types of the mean function m(x) and covariance function k(x, x') are determined by combining empirical methods; The voltage data and tire lateral force in the training dataset are expressed as a Gaussian process distribution, i.e., the lateral force Data_Fy. x The function f(Data_VT2) g_x ) follows a mean of m(Data_VT2) g_x ), covariance is k(Data_VT2) g_x Data_VT2 g_x ’ A Gaussian process is used; based on this, the initial values ​​of the hyperparameters and the number of iterations in the Gaussian process regression model are determined, and the hyperparameters σ and F in the covariance function are determined by minimizing the negative logarithm of the marginal likelihood function using the gradient descent method. The negative logarithm of the marginal likelihood function is specifically expressed as: ； Where -logp(f|x) represents the negative logarithm of the marginal likelihood function. Let Gaussian process regression function, For voltage input characteristics, Used to fit training data, n is the dimension of the training dataset, k n Let be an n-dimensional covariance function. Let I be an n-dimensional variance vector. n It is an n-dimensional unit vector. It is a penalty item. It is a normalization constant; Based on the prior definition, and combining the training and test sets collected in S2-1, the joint prior distribution can be derived as follows: ； Where, f(Data_VT2) g_x f(Data_VT2) represents the lateral force of the tires corresponding to the training set. g_c ) represents the tire lateral force corresponding to the test number, and N represents the normal distribution; S2-7, Calculate the estimated value f(Data_VT2) g_c The posterior distribution of ) is obtained and substituted into the hyperparameter solution of tire real-time lateral force obtained in S2-6.

6. The method for estimating key vehicle driving states based on the fusion of multi-source information from low-orbit satellites and vehicle-mounted sensors according to claim 5, characterized in that, The expression for the first-order main effect index is: ； in, The first-order principal effect index represents the effect of the Δth influencing factor on the voltage output of the gth sensor array, where 1 ≤ Δ ≤ D. When Δ = 5, the influencing factor is the tire slip angle, and D is the number of influencing factors. Mean(Data_VT1(A)) i Data_VT1(B) i ) represents the average value of the numerical vector formed by concatenating the i-th column of the sensor array output voltage matrix corresponding to matrix A and the i-th column of the sensor array output voltage matrix corresponding to matrix B. This represents the j-th value in the i-th column of the sensor array output voltage matrix corresponding to matrix B. Representative matrix AB Δ The corresponding value of the j-th column of the sensor array output voltage matrix. This represents the j-th value in the i-th column of the sensor array output voltage matrix corresponding to matrix A.

7. The method for estimating key vehicle driving states based on the fusion of multi-source information from low-orbit satellites and vehicle-mounted sensors according to claim 1, characterized in that, Step S3 includes the following steps: S3-1. Obtain the front and rear wheelbases of the vehicle. f and l r Calculate the vertical loads on the four wheels of the vehicle when it is stationary, including the static vertical loads Ld of the left front, right front, left rear, and right rear tires. fl , Ld fr , Ld rl and Ld rr ; The tire load was Ld respectively. fl and Ld rl Side slip characteristic tests under certain conditions were conducted to obtain side slip angle-lateral force data. The fixed values ​​K of the front and rear wheel side slip stiffness were then obtained by combining the magic tire formula for lateral force and curve fitting methods. f and K r ; S3-2. Define the steering angles of the two front wheels as equal to δ, and calculate δ based on the signals from the on-board sensors; define the slip angles of the two front axle wheels as equal to α. f The two rear axle wheels have the same slip angle, which is equal to α. r ; S3-3. The tire's lateral stiffness is adaptively adjusted using the difference between the linear tire force and the intelligent tire lateral force estimate; the adaptive values ​​of the front and rear tire lateral stiffness can be expressed as: ； in, and These are the adaptive values ​​for the lateral stiffness of the front and rear wheels, respectively. and These are the sum of the estimated lateral forces of the two front wheels and the two rear wheels, respectively; F yf and F yr These are the sum of the linear tire forces of the two front wheels and the two rear wheels, respectively.

8. The method for estimating key vehicle driving states based on the fusion of multi-source information from low-orbit satellites and vehicle-mounted sensors according to claim 7, characterized in that, The fixed values ​​of front and rear tire lateral stiffness can be expressed as: ； Among them, B f C f and D f For load Ld fl or Ld fr The parameters B obtained by fitting the corresponding lateral characteristic data r C r and D r For load Ld rl or Ld rr The parameters are obtained by fitting the corresponding lateral deviation characteristic data.

9. The method for estimating key vehicle driving states based on the fusion of multi-source information from low-orbit satellites and vehicle-mounted sensors according to claim 1, characterized in that, S4 specifically includes the following steps: S4-1. Acquire vehicle parameters, including vehicle mass, vehicle yaw moment of inertia, distance from the vehicle's center of gravity to the front axle and distance to the rear axle; acquire real-time signals from onboard sensors, including vehicle longitudinal and lateral acceleration, yaw rate and steering wheel angle. S4-2. A vehicle dynamics model is constructed based on the real-time adaptive values ​​of the front and rear wheel lateral stiffness, vehicle parameters, and real-time signals from onboard sensors. Specifically, it is expressed as follows: ； Among them, a x For the longitudinal acceleration of the vehicle, I z Let the vehicle's yaw moment of inertia be... For longitudinal acceleration, v x γ is the longitudinal velocity of the vehicle, and γ is the yaw rate. Let β be the yaw acceleration and β be the sideslip angle of the vehicle's center of gravity. For the quality of the car, and These are the adaptive values ​​of the lateral stiffness of the front and rear wheels, l f and l r These are the front and rear wheelbases of the vehicle, respectively. Let be the moment of inertia of the vehicle about the z-axis.

Images