CN112758097B - State prediction and estimation method for unmanned vehicle - Google Patents

State prediction and estimation method for unmanned vehicle Download PDF

Info

Publication number
CN112758097B
CN112758097B CN202011631422.0A CN202011631422A CN112758097B CN 112758097 B CN112758097 B CN 112758097B CN 202011631422 A CN202011631422 A CN 202011631422A CN 112758097 B CN112758097 B CN 112758097B
Authority
CN
China
Prior art keywords
vehicle
estimation
state
acceleration
time
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202011631422.0A
Other languages
Chinese (zh)
Other versions
CN112758097A (en
Inventor
陈勇
任宏斌
陈思忠
高泽鹏
赵玉壮
齐志权
李长隆
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Institute of Technology BIT
Original Assignee
Beijing Institute of Technology BIT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Institute of Technology BIT filed Critical Beijing Institute of Technology BIT
Priority to CN202011631422.0A priority Critical patent/CN112758097B/en
Publication of CN112758097A publication Critical patent/CN112758097A/en
Application granted granted Critical
Publication of CN112758097B publication Critical patent/CN112758097B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W40/00—Estimation or calculation of non-directly measurable driving parameters for road vehicle drive control systems not related to the control of a particular sub unit, e.g. by using mathematical models
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W40/00—Estimation or calculation of non-directly measurable driving parameters for road vehicle drive control systems not related to the control of a particular sub unit, e.g. by using mathematical models
    • B60W40/10—Estimation or calculation of non-directly measurable driving parameters for road vehicle drive control systems not related to the control of a particular sub unit, e.g. by using mathematical models related to vehicle motion
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • B60W2050/0001—Details of the control system
    • B60W2050/0043—Signal treatments, identification of variables or parameters, parameter estimation or state estimation
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation

Landscapes

  • Engineering & Computer Science (AREA)
  • Automation & Control Theory (AREA)
  • Transportation (AREA)
  • Mechanical Engineering (AREA)
  • Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Human Computer Interaction (AREA)
  • Control Of Driving Devices And Active Controlling Of Vehicle (AREA)

Abstract

The invention discloses a state prediction and estimation method for an unmanned vehicle, which comprises the steps of utilizing a proposed speed prediction formula with vehicle torque correction, determining optimal parameters of the speed prediction formula through characteristics such as acceleration, acceleration derivative and the like, training by utilizing a genetic algorithm neural network, introducing historical data of the current running working condition of the vehicle to update and optimize the acceleration and the acceleration derivative in time, estimating the road slope angle by utilizing a kinematic method and a dynamic method in a fusion manner, estimating the vehicle mass by utilizing a least square method, and improving the estimation precision of the vehicle mass and the sprung mass; estimating the vertical force of the wheel by using an information fusion method, introducing a road adhesion coefficient, and performing joint estimation on the lateral force and the state of the vehicle by using a nonlinear estimation method; finally, a vehicle state rolling estimation method introducing vehicle variable parameters and variable working conditions is provided, and the accuracy of vehicle speed prediction under the conditions of the vehicle variable parameters and the variable working conditions is improved.

Description

State prediction and estimation method for unmanned vehicle
Technical Field
The invention belongs to the technical field of state estimation of unmanned vehicles, and particularly relates to a state prediction and estimation method for an unmanned vehicle.
Background
Real-time estimation of vehicle states is the basis for vehicle dynamics control. From the physical model, there are generally classified into a kinetic method and a kinematic method. The kinematics method is generally based on a kinematics method, and a sensor is used for observing the quantity to be estimated, and the kinematics method is based on a kinematics method, and limited sensor configuration is used for realizing vehicle state estimation observation. The estimation algorithm can be classified into a Kalman filtering algorithm, a Longbeige algorithm, a robust algorithm, a sliding mode algorithm and a nonlinear observation algorithm. The kalman filter algorithm is currently the most common method for estimating the state of a vehicle.
The Kalman filtering is based on that system noise and measurement noise both accord with Gaussian white noise distribution, and the estimation process utilizes time updating and measurement updating to complete estimation and measurement in one period. The focus of designing a kalman filter is to determine the feedback gain. The kalman filter feedback gain represents the proportional weight of the model predicted value and the measured value in the estimation measurement. The kalman feedback gain is inversely proportional to the covariance of the measurement noise, and when the measurement value of the sensor is not accurate, that is, the larger the covariance of the measurement noise, the smaller the kalman feedback gain, the greater the reliability of the estimated value predicted by the model. If the prediction output of the model is accurate, the covariance of the prior error is almost 0, the Kalman gain also tends to be 0, and the estimation value of the vehicle state mainly depends on the model.
However, Kalman filtering is generally only suitable for linear systems, most application systems are nonlinear, the condition is difficult to meet, and extended Kalman filtering is used for linearizing the nonlinear systems by using Taylor expansion, determining Jacobian matrixes and determining system matrixes and observation matrixes. In order to overcome the defects of extended Kalman filtering, tasteless Kalman filtering directly utilizes a nonlinear equation to observe the vehicle state, so that the influence of a nonlinear system in a linearization process is avoided, and a Jacobian matrix does not need to be calculated. The kalman filter assumes that both the state noise and the measurement noise conform to a gaussian distribution. In order to release the limit of the Gaussian hypothesis and better accord with the practical application process, the Gaussian hypothesis is cancelled by particle filtering, and the high-order nonlinear state observation problem is effectively processed. However, particle filtering has a great disadvantage, and as the number of filtering iterations increases, particle exhaustion easily occurs, and most of the smaller-weight particles become few or none.
In order to adapt to state observation of a strong nonlinear vehicle system and quickly estimate the vehicle state, a new method is needed for estimating the vehicle state, and the state observation needs to consider parameter change of a distributed driving unmanned vehicle and complex and changeable running conditions. The method utilizes the measured value or the error value of the state time in the past estimation time domain, releases the Markov hypothesis, solves the nonlinear problem of the optimal target function in the rolling time domain without the linearization error, and realizes real-time iterative update and rapid accurate estimation of the vehicle state.
Control of a conventional manned vehicle is typically based on a vehicle stability determination to decide whether to apply a control command, which has some lag in control effectiveness. For a highly complex unmanned vehicle, the state (speed) of the vehicle needs to be predicted in advance, and the stability of the vehicle needs to be predicted in advance, so that the effectiveness and the real-time performance of the control of the unmanned vehicle are ensured. Therefore, the speed prediction of the unmanned vehicle is very critical for the control of the unmanned vehicle.
Disclosure of Invention
In view of the above, the present invention provides a method for predicting and estimating a state of an unmanned vehicle, which is suitable for estimating a vehicle state of a vehicle under variable parameters and variable conditions, and has high estimation accuracy and calculation efficiency.
The technical scheme for realizing the invention is as follows:
a state prediction and estimation method for an unmanned vehicle, comprising the steps of:
step 1, collecting the wheel speed, wheel torque, longitudinal acceleration, lateral acceleration and vertical acceleration of a vehicle body of each wheel, and the roll angle speed, pitch angle speed and yaw angle speed of the vehicle;
2, randomly selecting 20 working conditions from the standard running working conditions based on the vehicle state information obtained in the step 1, then randomly extracting 400 working condition blocks with different time lengths from the selected 20 working conditions, extracting 10 characteristics of each working condition block from the selected working condition blocks, performing characteristic calculation on the extracted 400 working condition blocks, performing speed prediction, correcting the speed prediction by using the acceleration and acceleration derivative parameters, optimizing parameter values by using an algorithm, and training the speed prediction by using a genetic algorithm neural network; then, in order to improve the adaptability of the vehicle speed prediction under any working condition, the historical state of the vehicle under the current running working condition is learned by using a rolling updating method, so that the problem of vehicle speed prediction precision caused by the limitation of the vehicle running working condition is avoided;
step 3, building a vehicle model based on the vehicle state information obtained in real time in the step 1, wherein the model comprises a vehicle body kinematics model, a vehicle body dynamics model, a tire model and a suspension model, and a non-linear model and a measurement model of noise-free state propagation are given;
step 4, integrating road slope angle estimation by using a kinematic method and a dynamic method, and improving the static deviation of low-frequency noise and the model dependency of high-frequency noise; estimating the vehicle mass by using a least square method, and improving the estimation precision of the vehicle mass and the sprung mass; estimating the vertical force of the wheel by using an information fusion method, and ensuring the estimation precision of the vertical force in the vehicle attitude change; introducing a road adhesion coefficient, and performing joint estimation on the lateral force and the vehicle state of the vehicle by adopting a nonlinear estimation method;
and 5, based on the vehicle parameters and the working condition information obtained in the step 4, performing vehicle state rolling estimation on the vehicle variable parameters and the variable working conditions, and improving the accuracy of vehicle speed prediction under the conditions of the vehicle variable parameters and the variable working conditions.
Further, the vehicle speed prediction in step 2 specifically includes:
from the standard driving conditions WVUSSUB, WVUITER, VAIL2NRE, US06_ HWY, US06, UNIF01, UKBUS6, UDDSHDV, UDDS, SC03, REP05, NYCRUCK, NYCOMP, NYCC, NurembergR36, NREL2VAIL, NewYorkBus, MEASURED _ MASS1, MANHATTAN, LA92, INDIA _ URBAN _ SAMPLE, INDIA _ HWY _ SAMPLE, IM240, HWFET, HL07, EUDC _ LOW, EUDC, ECE, CSHVR _ VeRandomly selecting 20 working conditions from HICLE, COMMENT, CBDTRUCK, CBDBUS, CBD14, BUSRTE, ARTERIAL, ARB02 and 1015_6PRIUS, then randomly extracting 400 working condition blocks with different time lengths from the selected 20 working conditions, and extracting 10 characteristics of each working condition block from the selected working condition blocks, wherein each characteristic corresponds to a parameter, and the 10 characteristic parameters specifically comprise the highest vehicle speed
Figure BDA0002876635860000031
Average vehicle speed
Figure BDA0002876635860000032
Maximum acceleration AmaxAverage acceleration AavMaximum acceleration derivative dAmaxMean acceleration dAavMaximum deceleration BmaxAverage deceleration BavDerivative of maximum deceleration dBmaxAverage acceleration derivative dBav(ii) a The specific calculation formula for the 10 parameters is represented by the following formula:
Figure BDA0002876635860000033
wherein, N is the time length of each working condition block, and N1, N2, N3 and N4 are the acceleration time length, deceleration time length, acceleration derivative time length and deceleration derivative time length of each working condition block respectively;
the speed prediction in the future time domain [ k +1, k + n ] is specifically as follows:
Figure BDA0002876635860000034
wherein the content of the first and second substances,
Figure BDA0002876635860000035
is the speed at the present moment in time,
Figure BDA0002876635860000036
to predict the velocity in the time domain at time k + n, ReffIs a vehicleEffective rolling radius of the wheel, TmIs the driving and braking torque of the vehicle, tdIs the exponential type variable attenuation coefficient of the motor torque, m is the whole vehicle mass, f is the rolling resistance coefficient of the vehicle, rhoa,CdAir density and windward resistance coefficient;
during acceleration of the vehicle, the vehicle drive torque TmThe correction value is expressed as:
Figure BDA0002876635860000041
by introducing an acceleration parameter k1,k2Preventing the vehicle speed from dropping too fast in a high-speed steady state, and accelerating the attenuation of the torque required by the vehicle when the vehicle speed is high; acceleration derivative parameter k3,k4When the acceleration of the vehicle is increased sharply, a larger required torque is kept, and the expected vehicle speed is increased in time in a prediction time domain; a. theav,dAavThe speed of the historical time period of the vehicle is obtained through a segmented acceleration algorithm and a segmented acceleration derivative algorithm, in order to reduce the complexity of algorithm calculation and ensure good real-time performance of a controller, the segmented time is selected to be consistent, 5s is selected, the speed of the historical time period is updated in a rolling mode in real time according to the time state of the vehicle, namely new historical speed data is added, and historical speed data with earlier time is discarded; the acceleration parameters and the acceleration derivative parameters are subjected to particle swarm optimization to obtain optimal parameter values under 400 typical working conditions respectively, and the ranges of the acceleration parameters and the acceleration derivative parameters are given in table 1;
TABLE 1 range of acceleration parameters and acceleration derivative parameters
Figure BDA0002876635860000042
Training 400 groups of different acceleration parameters and acceleration derivative parameters obtained based on a genetic algorithm neural network in real time according to historical data of the current running condition of the vehicle, and utilizing the error square sum EiMeasuring the adaptive value;
Ei＝∑∑(d0-y0)2
wherein d is0,y0Actual output and desired output of the network, respectively;
according to the historical vehicle speed, recording the vehicle speed data of 10s in the past historical time period, extracting 10 parameters, taking the parameters as input, training the genetic algorithm neural network, and performing online optimization updating on the parameters, thereby realizing the rolling prediction of the vehicle speed.
Further, the step 4 specifically includes:
the basic idea of road slope estimation is to retain the high frequency part of the kinematic estimation and filter out the low frequency part, and retain the low frequency part of the kinematic estimation and filter out the high frequency part;
Figure BDA0002876635860000043
wherein the content of the first and second substances,
Figure BDA0002876635860000044
is the road slope estimated by the fusion method,
Figure BDA0002876635860000045
a dynamically estimated road slope and a kinematically estimated road slope, respectively;
the dynamics estimation method relies on a vehicle dynamics model, which is expressed as:
Figure BDA0002876635860000051
wherein, FxIs the longitudinal force of the vehicle, is directly obtained through the torque vector control of the distributed front-driving unmanned vehicle,
Figure BDA0002876635860000052
acceleration in the longitudinal direction of the vehicle, p, CdA is the air density and wind resistance of the vehicle respectivelyThe coefficient and the windward area, theta and f are the road surface gradient and the road surface rolling resistance coefficient respectively;
the model-based dynamics estimation road slope method is represented as follows:
Figure BDA0002876635860000053
the observed value of the acceleration of the inertial navigation in the longitudinal direction is related to the derivative of the speed of the vehicle as follows:
Figure BDA0002876635860000054
a kinematic estimation method of road gradient can be obtained by the above equation:
Figure BDA0002876635860000055
estimation and estimation of vehicle mass based on recursive least square method
Figure BDA0002876635860000056
Figure BDA0002876635860000057
Wherein k is the current sampling moment, and k-1 is the last sampling moment; l, K are the least squares gain and error covariance update, respectively;
estimation of sprung mass
Figure BDA0002876635860000058
The unsprung mass m may be subtracted from the estimated vehicle massusObtaining:
Figure BDA0002876635860000059
in the observation of the vertical forces of the tires, the movements of the vehicle, in particular the longitudinal, transverse and rolling movements, are taken into account;
the load transfer of the front/rear axle caused by longitudinal and lateral motion is estimated by the following formula:
Figure BDA00028766358600000510
the roll motion of the vehicle also results in vertical load transfer; estimating the change of the axial force in the vehicle rolling process according to a calculation formula of the vehicle rolling motion:
Figure BDA00028766358600000511
vertical load per tire FzijThe method can be based on the estimation:
Figure BDA00028766358600000512
estimating the vehicle state and the lateral force by a nonlinear estimation method, estimating the vehicle state and the tire lateral force by the nonlinear method, wherein the number of Sigma points is related to the number of state points, the number of n state points generates 2n +1 Sigma points, the first one is defined as the average value of the random current state, and the Sigma point χ is the value of the random current statekCan be calculated by the following formula:
Figure BDA0002876635860000061
wherein, PxIs the covariance of the state x at random,
Figure BDA0002876635860000062
is the square root of the matrix, λ is a scaling parameter, and the calculation formula is as follows:
λ＝α2(n+κ)-n,(1≤α≤10-4)
wherein the constant alpha determines the distribution of sigma points around the average value of the current state, and if the alpha value is larger, the sigma points are away from the state
Figure BDA0002876635860000063
Average value of (2)
Figure BDA0002876635860000064
The farther away, the closer the sigma point is when the alpha value is smaller
Figure BDA0002876635860000065
Predicting sigma point χ at point k +1k+1Updating can be carried out based on a nonlinear function according to the current sigma point;
χk+1＝χk+f(χk,uk+1,wk+1)·ΔT
wherein u isk+1For control input of the system, wk+1Is process noise, Δ T is the sampling time;
the non-linearity measurement of the system can be obtained by the non-linearity measurement of the system:
Yk+1＝h(xk+1,uk+1,vk+1)
wherein v isk+1To measure noise, it is assumed that the noise has no nonlinear effect on the system;
Figure BDA0002876635860000066
thereby obtaining a state vector xk+1|kAnd the measured value yk+1Average value of (d):
Figure BDA0002876635860000067
wherein the content of the first and second substances,
Figure BDA0002876635860000068
is a weight matrix;
the covariance of the state vector and the measurement vector may be updated by:
Figure BDA0002876635860000069
wherein the content of the first and second substances,
Figure BDA00028766358600000610
is a weight matrix; q is the covariance of the system process noise, R is the measurement noise; q and R are both Gaussian white noise; the measurement update equation is:
Figure BDA0002876635860000071
wherein z isk+1Is a measurement signal from a sensor;
the upper and lower boundaries of the slip angle can be expressed by the following relation:
Figure BDA0002876635860000072
based on the estimate of the mid-lateral dynamics, the peak self-aligning moment is sensitive to the tire road friction coefficient when the slip angle varies in the middle zone (up and down); estimating a coefficient of friction from the SAT using a brush tire model; the brush tire model well simulates the alignment torque behavior of the tire,
τ＝μFzlθy tan(α){1-3|θytan(α)|+3|θy tan(α)|2-|θy tan(α)|3}
when the peak value is 27/256 mu FzAt 4tan (. alpha.) thetayWhen 1, the road friction coefficient can be estimated
Figure BDA0002876635860000073
Figure BDA0002876635860000074
τmax＝max(τ(t)),t∈[t-Δt,t]
In the formula, τmaxEstimating the maximum value of the aligning moment in the past window delta t;
based on the estimation of large dynamics, according to the friction circle theory, when the slip angle is larger than the upper boundary, the tire force is in the saturation region; the tire longitudinal and lateral forces follow non-linear constraints:
Fxi_sat 2+Fyi_sat 2≤(μFz)2
the road friction is estimated using the following equation:
Figure BDA0002876635860000075
non-linear estimation under other conditions, assuming that the tire-road friction coefficient is constant during the sampling time Δ t; the estimated state function can thus be used:
Figure BDA0002876635860000076
the measured output in the non-linear estimator is defined as:
Figure BDA0002876635860000077
wherein the content of the first and second substances,
Figure BDA0002876635860000078
different friction estimation algorithms are designed according to different excitation levels; these are integrated into a road friction estimator.
Further, the step 5 of performing the rolling estimation of the vehicle state of the vehicle variable parameter and the variable working condition specifically includes:
discretizing the vehicle noiseless state propagation model and the measurement model in the general form obtained in the step 3 to obtain a discretized vehicle propagation model and a discretized measurement model:
Figure BDA0002876635860000081
performing feature extraction on the attitude influencing the vehicle state estimation, wherein the attitude comprises longitudinal motion, lateral motion, vertical motion, pitching motion, rolling motion and yawing motion of the vehicle, and correspondingly extracting corresponding variables comprises the following steps:
xa|t＝[u v w p q r]T
the vehicle state rolling estimation method is based on an optimization objective function of J (x, u, p); wherein x, u and p are respectively the state of a vehicle system, the input of the system and the parameter variable of the vehicle;
the rolling time domain optimization method is that in each step state estimation time domain, vehicle measurement output of a vehicle at the current moment and the vehicle state at the current moment are increased, the vehicle state at the moment of q-N +1 is discarded, and the steps are circulated in sequence at a new moment; at the time of k-1, the optimal objective function and the constraint conditions thereof are as follows:
J(xk-1,uk-1,pk-1)
Figure BDA0002876635860000082
Figure BDA0002876635860000083
Figure BDA0002876635860000084
Figure BDA0002876635860000085
the least squares objective function J, can be expressed as:
Figure BDA0002876635860000086
k-1 represents the last time instant,
Figure BDA0002876635860000087
and
Figure BDA0002876635860000088
representing the system output and system input actually measured at time k-1;
Figure BDA0002876635860000089
the representation has a diagonal positive definite weighting matrix Wk-1And Vk-1The euclidean norm of (d); using a symmetrical positive definite matrix PLEstimating parameters for the oldest state and a few priors in a rolling time domain window
Figure BDA00028766358600000810
Is corrected with a weight coefficient of the symmetric positive array PL(ii) a Furthermore, all estimates in the least squares method may be limited by the upper and lower bounds of the system physical limitations; at a cost function
Figure BDA00028766358600000811
In the dynamic optimization problem, a control quantity is also included to illustrate the actual control of the system and the control calculated by the controller
Figure BDA00028766358600000812
The deviation therebetween;
at time k, the optimal objective function and its constraint conditions are:
J(xk,uk,pk)
Figure BDA00028766358600000813
Figure BDA00028766358600000814
Figure BDA00028766358600000815
Figure BDA00028766358600000816
least square method objective function
Figure BDA00028766358600000817
It can be expressed as:
Figure BDA0002876635860000091
k represents the current time;
Figure BDA0002876635860000092
and
Figure BDA0002876635860000093
representing the system output and system input actually measured at the current time.
At the time k +1, the optimal objective function and its constraint conditions are:
J(xk+1,uk+1,pk+1)
Figure BDA0002876635860000094
Figure BDA0002876635860000095
Figure BDA0002876635860000096
Figure BDA0002876635860000097
least square method objective function
Figure BDA0002876635860000098
It can be expressed as:
Figure BDA0002876635860000099
k +1 represents the current time;
Figure BDA00028766358600000910
and
Figure BDA00028766358600000911
system outputs and system inputs representing actual measurements at the current time;
in the least square method objective function, the weighting matrix coefficient PL,Vk,WkThe value selection of (a) affects the accuracy of the vehicle state estimation; for proper selection of the weighting matrix, the output of the measured values is assumed to have a gaussian distribution; assume an initial value xLAnd the parameter p is a random variable with a normal distribution, whose covariance matrices are respectively
Figure BDA00028766358600000912
The mean values thereof are respectively
Figure BDA00028766358600000913
Assuming measured values of the system
Figure BDA00028766358600000914
And unknown control inputs
Figure BDA00028766358600000915
Are each yk,ukThe covariance matrices are respectively
Figure BDA00028766358600000916
(ii) a gaussian distribution of; pLOf two block diagonal matrices of
Figure BDA00028766358600000917
Vk,WkAre respectively
Figure BDA00028766358600000918
The least square method target function accords with the maximum likelihood estimation of the current window; expression P of the weighting matrixL,Vk,WkRespectively as follows:
Figure BDA00028766358600000919
the first term in the objective function, representing all the information collected by the measurement before time t; reference estimator
Figure BDA00028766358600000920
Selecting a value to estimate a solution for the instantaneous rolling time domain optimization; selecting the arrival cost matrix P in different waysLIt may be a constant zero matrix, or a so-called smooth extended kalman filter update of the sensitivity information obtained when solving the previous rolling time domain optimization problem; classical extended kalman filtering is equivalent to updating and rolling time domain optimization estimation with a rolling time domain of 1 using a smooth kalman filter.
Has the advantages that:
the method of the invention utilizes the measured value or the error value of the state time in the past estimation time domain, releases the Markov assumption, solves the nonlinear problem of the optimal objective function in the rolling time domain without linearization error, carries out real-time iterative update and quickly realizes the accurate estimation of the vehicle state.
Drawings
Fig. 1 is a schematic diagram of a vehicle speed prediction structure provided by the present invention.
FIG. 2 is a schematic diagram of a vehicle parameter and state estimation architecture provided by the present invention.
Fig. 3 is a mixed road surface adhesion coefficient estimation flowchart.
FIG. 4 lateral force and SAT model under different friction conditions. (mu-0.2, 0.4,0.6,0.8, 1.0; SAT for thick lines; lateral force for thin dotted lines)
Detailed Description
The invention is described in detail below by way of example with reference to the accompanying drawings.
The invention provides a state estimation method for a distributed drive unmanned vehicle. As shown in fig. 1, first, a new speed prediction method suitable for an unmanned vehicle is proposed. The method comprises the steps of randomly extracting 400 working condition blocks with different time lengths under the known 20 working conditions, extracting 10 characteristics corresponding to each working condition block, utilizing a proposed vehicle torque correction speed prediction formula, determining optimal parameters of the speed prediction formula through characteristics of acceleration, acceleration derivative and the like, training by utilizing a genetic algorithm neural network, introducing historical data of the current running working condition of the vehicle to timely update and optimize the parameters of the acceleration, the acceleration derivative and the like, improving the prediction accuracy of the vehicle, and having learning capacity. Then, a vehicle parameter, condition and lateral force estimation method is proposed. The method comprises the following steps of utilizing a kinematic method and a dynamic method to fuse road slope angle estimation, and improving static deviation of low-frequency noise and model dependency of high-frequency noise; estimating the vehicle mass by using a least square method, and improving the estimation precision of the vehicle mass and the sprung mass; estimating the vertical force of the wheel by using an information fusion method, and ensuring the estimation precision of the vertical force in the vehicle attitude change; and introducing a road adhesion coefficient, and performing joint estimation on the vehicle lateral force and the vehicle state by adopting a nonlinear estimation method. Finally, a vehicle state rolling estimation method introducing vehicle variable parameters and variable working conditions is provided, and the accuracy of vehicle speed prediction under the conditions of the vehicle variable parameters and the variable working conditions is improved.
As shown in fig. 2, the method specifically includes the following steps:
step 1, collecting wheel speed, wheel torque, longitudinal acceleration, lateral acceleration and vertical acceleration of a vehicle body of each wheel, and parameters such as vehicle roll angle speed, pitch angle speed and yaw angle speed.
And 2, randomly selecting 20 working conditions from the standard running working conditions based on the observation information of the vehicle longitudinal acceleration, the wheel torque and the like obtained in the step 1, then randomly extracting 400 working condition blocks with different time lengths from the selected 20 working conditions, extracting 10 characteristics of each working condition block from the selected working condition blocks, performing characteristic calculation on the extracted 400 working condition blocks, providing a new speed prediction method suitable for the unmanned vehicle, correcting the speed prediction by using the acceleration and acceleration derivative parameters, optimizing parameter values by using an algorithm, and training the speed prediction by using a genetic algorithm neural network. Then, in order to improve the adaptability of the vehicle speed prediction under any working condition, the historical state of the vehicle under the current running working condition is learned by using a rolling updating method, and the problem of vehicle speed prediction precision caused by the limitation of the vehicle running working condition is avoided.
And 3, establishing a vehicle model based on the vehicle state acquired in real time in the step 1, wherein the model comprises a vehicle body kinematics model, a vehicle body dynamics model, a tire model and a suspension model, and providing a non-linear model and a measurement model of noise-free state propagation.
Step 4, providing a vehicle parameter, working condition and lateral force estimation method, fusing road slope angle estimation by using a kinematic method and a dynamic method, and improving static deviation of low-frequency noise and model dependency of high-frequency noise; estimating the vehicle mass by using a least square method, and improving the estimation precision of the vehicle mass and the sprung mass; estimating the vertical force of the wheel by using an information fusion method, and ensuring the estimation precision of the vertical force in the vehicle attitude change; and introducing a road adhesion coefficient, and performing joint estimation on the vehicle lateral force and the vehicle state by adopting a nonlinear estimation method.
And step 5, providing a vehicle state rolling estimation method introducing vehicle variable parameters and variable working conditions based on the vehicle parameters and the working condition information obtained in the step 4, and improving the accuracy of vehicle speed prediction under the conditions of the vehicle variable parameters and the variable working conditions.
The vehicle speed prediction in step 2 specifically includes:
the method includes selecting 20 conditions at random from among the standard driving conditions WVUSSUB, WVULTER, VAIL2NRE, US06_ HWY, US06, UNIF01, UKBUS6, UDDSHDV, UDDS, SC03, REP05, NYCRUCK, NYCOMP, NYCC, NurembergR36, NREL2VAIL, NewYorkBus, MEASURED _ MASS1, MANHATTAN, LA92, INDIA _ URBAN _ SAMPLE, INDIA _ HWY _ SAMPLE, IM240, HWFET, HL07, EUDC _ LOW, EUDC, ECE, CSHVR _ Vehicle, COMMUTER, CBDTRUCK, DBUS, CBD14, BUS, IARTE, ERARL, ARB02, ARB 1015_6PRIUS, extracting 400 from the selected 20 conditions, 400, 10, and 10 time segments, and extracting a specific characteristic from each condition block including a selected characteristic, wherein the characteristic is selected and the characteristic of the Vehicle speed segment
Figure BDA0002876635860000111
Average vehicle speed
Figure BDA0002876635860000112
Maximum acceleration AmaxAverage acceleration AavMaximum acceleration derivative dAmaxMean acceleration dAavMaximum deceleration BmaxAverage deceleration BavDerivative of maximum deceleration dBmaxAverage acceleration derivative dBav. The specific calculation formula for the 10 parameters is represented by the following formula:
Figure BDA0002876635860000121
where N is the time length of each condition block, and N1, N2, N3, and N4 are the acceleration time length, deceleration time length, acceleration derivative time length, and deceleration derivative time length of each condition block, respectively.
The conventional vehicle speed prediction methods mainly comprise three types, namely a Markov chain-based speed prediction method, an exponential decay-based speed prediction method and a neural network-based speed prediction method. In order to improve the accuracy of vehicle speed prediction and achieve a better active safety control effect of a vehicle, a new method for predicting the speed of an unmanned vehicle in a future time domain [ k +1, k + n ] is provided:
Figure BDA0002876635860000122
wherein the content of the first and second substances,
Figure BDA0002876635860000123
is the speed at the present moment in time,
Figure BDA0002876635860000124
to predict the velocity in the time domain at time k + n, ReffIs the effective rolling radius, T, of the wheelmIs the driving and braking torque of the vehicle, tdIs the exponential type variable attenuation coefficient of the motor torque, m is the whole vehicle mass, f is the rolling resistance coefficient of the vehicle, rhoa,CdAir density and windward resistance coefficient, respectively.
From the above equation, it can be found that the driving and braking torque T of the vehiclemThe vehicle speed prediction accuracy in a future time domain is directly influenced, and in order to further realize the vehicle prediction accuracy based on a speed prediction model, the driving and braking torque of the vehicle is corrected, so that the vehicle has good adaptability of prediction accuracy under different running conditions. When the vehicle is in a sharp acceleration (deceleration) working condition, the predicted vehicle speed can be rapidly increased (reduced) according to the torque demand, so that the torque can be rapidly responded according to the vehicle demand in a predicted time domain; when the vehicle is in a high-speed steady state, the required torque of the vehicle can be maintained in a certain range value, and the unreasonable reduction caused by the rapid attenuation of the torque is avoided, so that the stability of vehicle speed prediction is influenced. Taking the acceleration process of the vehicle as an example, the vehicle drive torque TmThe correction value is expressed as:
Figure BDA0002876635860000125
by introducing an acceleration parameter k1,k2To prevent the vehicle from being stable at high speedThe vehicle speed drops excessively fast in this state, and the torque required to accelerate the vehicle decays when the vehicle speed is high. Acceleration derivative parameter k3,k4When the acceleration of the vehicle is increased sharply, the expected vehicle speed is increased in time in the prediction time domain while a large required torque is maintained. A. theav,dAavThe speed of the historical time period of the vehicle is obtained through a segmented acceleration algorithm and a segmented acceleration derivative algorithm, in order to reduce the complexity of algorithm calculation and ensure good real-time performance of a controller, the segmented time is selected to be consistent, 5s is selected, and the speed of the historical time period is updated in a rolling mode in real time according to the time state of the vehicle, namely new historical speed data is added, and historical vehicle speed data with earlier time are discarded. The acceleration parameters and the acceleration derivative parameters are subjected to particle swarm optimization to obtain optimal parameter values under 400 typical working conditions respectively, and the ranges of the acceleration parameters and the acceleration derivative parameters are given in table 1.
TABLE 1 range of acceleration parameters and acceleration derivative parameters
Figure BDA0002876635860000131
400 groups of different optimal values of the acceleration parameters and the acceleration derivative parameters pass through 400 working condition blocks, but the accuracy of prediction is difficult to guarantee when form conditions with complicated and changeable working conditions are predicted. In order to ensure the adaptability and the accuracy of the predicted vehicle speed under any other working conditions, a design method based on parameter online optimization is provided, the acceleration parameter and the acceleration derivative parameter are updated and optimized in time through historical data of the current running working condition of the vehicle, and the problem of vehicle speed prediction accuracy caused by the limitation of the running working condition of the vehicle is avoided. Training 400 groups of different acceleration parameters and acceleration derivative parameters obtained based on a genetic algorithm neural network in real time according to historical data of the current running condition of the vehicle, and utilizing the error square sum EiAnd measuring the adaptation value.
Ei＝∑∑(d0-y0)2
Wherein d is0,y0Respectively the actual output of the networkOut and desired output.
According to the historical vehicle speed, recording the vehicle speed data of 10s in the past historical time period, extracting 10 parameters, taking the parameters as input, training the genetic algorithm neural network, and performing online optimization updating on the parameters, thereby realizing the rolling prediction of the vehicle speed.
The noise-free state propagation model and the measurement model in the general form of the vehicle described in step 3 specifically include:
the vehicle chassis is modeled as a rigid body, the absolute position and the heading of the vehicle are represented by an X-Y coordinate system, and the relative speed of the vehicle is represented by a local coordinate system X-Y-z. The four wheels are modeled as independent entities, having only rotational inertia. Roll, pitch, and vertical motions of the vehicle are taken into account, and these motions affect the vehicle load transfer. The distributed drive unmanned vehicle is provided with a front wheel steering system and a four-wheel independent drive system, and the control input is the steering wheel angle and the wheel drive braking moment.
The vehicle kinematics equation is expressed as:
Figure BDA0002876635860000132
the vehicle dynamic chassis system is represented as:
Figure BDA0002876635860000141
wherein u, v, w respectively represent the longitudinal vehicle speed, the lateral vehicle speed and the vertical vehicle speed of the vehicle body. p, q, r are the roll angular velocity, pitch angular velocity and yaw angular velocity of the vehicle body, respectively. m, mbRespectively representing the vehicle mass and the vehicle body mass. Ca,CaAre the longitudinal and lateral aerodynamic factors. I isx,Iy,IzThe moment of inertia of the vehicle about the x, y, z axes, respectively. I isxs,Iys,IzsThe moment of inertia of the body about the x, y, z axes, respectively. FX,FY,FZ,MX,MY,MZRespectively representing the tire forces generated by four tiresOr suspension forces, respectively, in total force and moment along or about the x, y, z axes, respectively, are expressed as:
Figure BDA0002876635860000142
wherein, Fxij,FyijAnd ij epsilon { fl, fr, rl, rr } represents the longitudinal force and the lateral force of the tire, and the longitudinal force and the lateral force are determined by the road adhesion coefficient, the vertical load of the vehicle, the lateral deflection angle of the tire, the slip ratio of the tire and the like. FzsijRepresenting the suspension force generated by the suspension. The corner marks ij ═ fl, fr, rl, rr denote the left front wheel, the right front wheel, the left rear wheel, and the right rear wheel, respectively.
The mechanical motion of the vehicle is slower than the electro-magnetic dynamic response of the motor, so, ignoring the dynamic response of the motor drive and the in-wheel or wheel-side motor, the motor controller and the in-wheel motor unit can be controlled by the control gain k if considered as one unit for the in-wheel motor and the motor controllerijThe following steps are described:
Figure BDA0002876635860000143
wherein, TijIs the output torque of the in-wheel motor uijIs the control signal to the drive brake motor and the control gain of the motor can be obtained from experimental data.
The wheel rotation dynamics may be represented by the following equation:
Figure BDA0002876635860000151
wherein, Iw,RweThe moment of inertia and the effective rolling radius of the wheel, respectively.
Tire model
Vehicle longitudinal acceleration, lateral acceleration and vertical acceleration, as well as vehicle pitch, roll and yaw rates, can all be effectively measured by global positioning systems and inertial navigators. The tire slip angle is defined as the angle between the speed direction of the vehicle wheel center and the wheel orientation:
Figure BDA0002876635860000152
the tire slip ratio, i.e., the degree of slip of the tire, represents the relative difference between the tire peripheral speed and the tire center speed:
Figure BDA0002876635860000153
the speed at the wheel center can be represented by:
Figure BDA0002876635860000154
the magic formula tire model has the capability of fitting a curve between the vehicle tire force and the tire slip rate and the tire slip angle, and the magic formula is represented by the following formula:
Figure BDA0002876635860000155
where y (h) represents a tire longitudinal force, a tire lateral force, or a tire aligning moment, and h is a tire slip ratio and a tire slip angle. Sv,ShIndicating vertical and horizontal offsets, respectively. Z, W, BtAnd E represents a crest factor, a shape factor, a stiffness factor and a curve factor, respectively.
A suspension model. The suspension vertical force refers to the upward force which is perpendicular to the plane of the vehicle and acts on the wheel center of a suspension system, and mainly comprises pre-pressure of the suspension in the static state of the vehicle, spring force and damping force generated by the vertical motion of the suspension.
The vehicle pre-pressure refers to the vehicle body acting force applied to the wheel center when the vehicle is in a static state or runs at a constant speed, and is directly related to the weight of the vehicle body:
Figure BDA0002876635860000161
the spring is stretched and compressed to generate a suspension force, and the magnitude of the spring force is related to the stiffness and deflection of the spring.
Fzskij＝ksdij
Wherein k iss,dijRespectively representing the spring rate and the amount of deformation of the spring.
The spring force calculated above is converted into a spring force acting at the wheel center,
Fzs2ij＝KsFzskij
wherein, KsIndicating the lever ratio of the spring.
The shock absorber generates damping force by the dynamic deflection speed of the suspension,
Fzscij＝cdDij
wherein, cd,DijRespectively representing the damper damping and the dynamic deflection speed.
The conversion of the damping force into a force at the wheel center can be expressed as:
Fzs3ij＝CdFzscij
wherein, CdRepresenting the damper lever ratio.
Neglecting the anti-roll and anti-pitch forces due to vehicle roll and pitch, the total suspension vertical force:
Fzsij＝Fzs1ij+Fzs2ij+Fzs3ij
the vertical force of the wheel can be expressed as
Fzwij＝Fzsij+mwijg
Wherein m iswijIndicating the unsprung wheel weight. The above wheel vertical force ignores the elastic deformation force caused by the elastic deformation of the tire.
An optimization strategy for vehicle state estimation based on non-linear rolling time domain estimation is proposed. The vehicle noise-free state propagation model and the measurement model are written in the general form:
Figure BDA0002876635860000162
wherein the state variables and control inputs of the system are respectively
Figure BDA0002876635860000163
And
Figure BDA0002876635860000164
the output of the measurement model is
Figure BDA0002876635860000165
Figure BDA0002876635860000166
The vehicle parameter, working condition and lateral force estimation method in the step 4 specifically comprises the following steps:
the estimation of the road gradient is generally performed by a dynamic estimation method and a kinematic estimation method. The two different methods have the respective defects and influence the estimation precision, the kinematic estimation method has the defects that the static deviation of low-frequency noise has large influence on the observed value of the acceleration sensor, and the dynamic estimation method has the defect that high-frequency noise has large influence on the vehicle model parameters. Therefore, in order to ensure the accuracy and precision of the road surface gradient estimation, an estimation method combining a kinematics method and a dynamics method is adopted, and the basic idea is to reserve a high-frequency part of the kinematics estimation and filter a low-frequency part, reserve a low-frequency part of the dynamics estimation and filter a high-frequency part.
Figure BDA0002876635860000171
Wherein the content of the first and second substances,
Figure BDA0002876635860000172
is a fusion methodThe estimated gradient of the road is used,
Figure BDA0002876635860000173
respectively a kinematically estimated road gradient and a kinematically estimated road gradient.
The dynamics estimation method relies on a vehicle dynamics model, which is expressed as:
Figure BDA0002876635860000174
wherein, FxThe longitudinal force of the vehicle can be directly obtained through the torque vector control of the distributed front-drive unmanned vehicle,
Figure BDA0002876635860000175
acceleration in the longitudinal direction of the vehicle, p, CdA is the air density, the wind resistance coefficient and the windward area of the vehicle, and theta and f are the road gradient and the road rolling resistance coefficient.
The model-based dynamics estimation road slope method may be represented as follows:
Figure BDA0002876635860000176
the observed value directly measured by the vehicle acceleration sensor is not only related to the running state of the vehicle, but also has close relation with the gradient of the road. The observed value of the acceleration of the inertial navigation in the longitudinal direction is related to the derivative of the speed of the vehicle as follows:
Figure BDA0002876635860000177
a kinematic estimation method of road gradient can be obtained by the above equation:
Figure BDA0002876635860000178
the mass of the vehicle has a large effect on the dynamics of the vehicle, so that the vehicle's empty and full load changes the sprung mass of the vehicle, thereby affecting the dynamics of the vehicle. The vehicle mass parameters then have no way of directly obtaining the observation to be estimated by measurement, which is very important for improving the control of the vehicle. Estimation and estimation of vehicle mass based on recursive least square method
Figure BDA0002876635860000179
Figure BDA00028766358600001710
Wherein k is the current sampling time, and k-1 is the last sampling time. L, K are the least squares gain and error covariance updates, respectively.
The sprung mass is calculated using a very fast calculation method, and in general the chassis of a distributed drive unmanned vehicle, including the frame and wheel masses, is substantially constant, so that the estimate of the sprung mass is
Figure BDA00028766358600001711
The unsprung mass m may be subtracted from the estimated vehicle massusObtaining:
Figure BDA00028766358600001712
for vertical loads, the sensor cannot directly measure and acquire. The estimation of the tire force is also made more complicated if observed by the suspension force. A vertical load observation method based on vertical load transmission is provided. In the observation of the vertical forces of the tires, the movements of the vehicle are taken into account, in particular the longitudinal, lateral and roll movements.
The load transfer of the front/rear axle caused by longitudinal and lateral motion can be estimated by the following formula
Figure BDA0002876635860000181
As shown in fig. 3, roll motion of the vehicle also results in vertical load transfer. According to the calculation formula of the vehicle rolling motion, the change of the axle force during the vehicle rolling process can be estimated:
Figure BDA0002876635860000182
vertical load per tire FzijThe method can be based on the estimation:
Figure BDA0002876635860000183
and estimating the vehicle state and the lateral force by adopting a nonlinear estimation method. The vehicle state and the tire lateral force and the like are estimated using a nonlinear method. The number of Sigma points is related to the number of status points. The number of n state points will generate 2n +1 Sigma points, the first of which is defined as the average of the random current state. sigma point xkCan be calculated by the following formula:
Figure BDA0002876635860000184
wherein, PxIs the covariance of state x at random.
Figure BDA0002876635860000185
Is the square root of the matrix. λ is a scaling parameter. The calculation formula is as follows:
λ＝α2(n+κ)-n,(1≤α≤10-4)
where the constant alpha determines the distribution of sigma points around the mean of the current state. If the value of alpha is large, the sigma point is off state
Figure BDA0002876635860000186
Average value of (2)
Figure BDA0002876635860000187
The further away. When the value of alpha is smaller, the sigma point is closer
Figure BDA0002876635860000188
Predicting sigma point χ at point k +1k+1The update may be based on a non-linear function according to the current sigma point.
χk+1＝χk+f(χk,uk+1,wk+1)·ΔT
Wherein u isk+1Is a control input to the system. w is ak+1Is process noise. Δ T is the sampling time.
The non-linearity measurement of the system can be obtained by the non-linearity measurement of the system:
Yk+1＝h(xk+1,uk+1,vk+1)
wherein v isk+1To measure noise. It is assumed that the noise has no non-linear effect on the system.
Figure BDA0002876635860000191
Thereby obtaining a state vector xk+1|kAnd the measured value yk+1Average value of (d):
Figure BDA0002876635860000192
wherein the content of the first and second substances,
Figure BDA0002876635860000193
is a weight matrix.
The covariance of the state vector and the measurement vector can be updated by:
Figure BDA0002876635860000194
wherein the content of the first and second substances,
Figure BDA0002876635860000195
is a weight matrix. Q is the covariance of the system process noise and R is the measurement noise. Both Q and R are Gaussian white noise. The measurement update equation is:
Figure BDA0002876635860000196
wherein z isk+1Is the measurement signal from the sensor.
In order to improve the robustness of the unmanned vehicle to external interference, a hybrid road adhesion coefficient estimator is provided, and the flow of the estimator is shown in the figure. FIG. 4 plots tire lateral force and self-aligning torque under different road friction. The tire cornering moment is linearly related to the slip angle, and increases to a peak value and then decreases to zero as the slip angle increases. It can be noted that when the slip angle is in the media zone (between the lower and upper boundaries in fig. 4), the coefficient of friction is easily distinguished compared to lateral dynamics based methods.
The upper and lower boundaries of the slip angle can be represented by the following relationship:
Figure BDA0002876635860000197
based on the estimate of mid-lateral dynamics, the peak self-aligning torque is sensitive to the tire road surface friction coefficient when the slip angle varies in the middle region (up and down). Herein, we estimate the coefficient of friction from the SAT using the brush tire model; the brush tire model well simulates the alignment torque behavior of the tire,
τ＝μFzlθy tan(α){1-3|θy tan(α)|+3|θy tan(α)|2-|θy tan(α)|3}
when the peak value is 27/256 mu FzAt 4tan (. alpha.) thetayWhen 1, the road friction coefficient can be estimated
Figure BDA0002876635860000198
Figure BDA0002876635860000201
τmax＝max(τ(t)),t∈[t-Δt,t]
In the formula, τmaxThe maximum value of the alignment moment is estimated for the past window deltat.
Based on the estimation of large dynamics, according to the friction circle theory, when the slip angle is larger than the upper boundary, the tire force is in the saturation region; the tire longitudinal and lateral forces follow non-linear constraints:
Fxi_sat 2+Fyi_sat 2≤(μFz)2
the road friction is estimated using the following equation:
Figure BDA0002876635860000202
non-linear estimation under other conditions, it can be assumed that the tire-road friction coefficient is constant during the sampling time Δ t; the estimated state function can thus be used:
Figure BDA0002876635860000203
the measured output in the non-linear estimator is defined as:
Figure BDA0002876635860000204
wherein the content of the first and second substances,
Figure BDA0002876635860000205
as previously mentioned, different friction estimation algorithms are designed according to different excitation levels. To improve its robustness and extend its range of application, these estimators are implemented as an integrated road friction estimator, as shown in fig. 3.
The vehicle state rolling estimation method introducing the vehicle variable parameters and the variable working conditions in the step 5 specifically comprises the following steps:
discretizing the vehicle noiseless state propagation model and the measurement model in the general form obtained in the step 3 to obtain a discretized vehicle propagation model and a discretized measurement model:
Figure BDA0002876635860000206
extracting characteristics of postures influencing vehicle state estimation, wherein the postures mainly comprise longitudinal motion, lateral motion, vertical motion, pitching motion, rolling motion and yawing motion of a vehicle, and correspondingly extracting corresponding variables comprises the following steps:
xa|t＝[u v w p q r]T
in order to improve the precision of vehicle state estimation and improve the adaptability of vehicle parameters and working conditions, a vehicle state rolling estimation method with self-adaptive vehicle parameters and working conditions is provided. The vehicle state roll estimation method is based on an optimized objective function of J (x, u, p). Where x, u, p are the state of the vehicle system, the inputs to the system and the parametric variables of the vehicle, respectively.
The rolling time domain optimization method is that in each step state estimation time domain, vehicle measurement output of a vehicle at the current moment and the vehicle state of the current moment are increased, the vehicle state at the moment of q-N +1 is discarded, and the steps are circulated in sequence at a new moment. At the time of k-1, the optimal objective function and the constraint conditions thereof are as follows:
J(xk-1,uk-1,pk-1)
Figure BDA0002876635860000211
Figure BDA0002876635860000212
Figure BDA0002876635860000213
Figure BDA0002876635860000214
the least squares objective function J, can be expressed as:
Figure BDA0002876635860000215
k-1 represents the last time instant,
Figure BDA0002876635860000216
and
Figure BDA0002876635860000217
representing the system output and system input actually measured at time k-1;
Figure BDA0002876635860000218
the representation has a diagonal positive definite weighting matrix Wk-1And Vk-1The euclidean norm of (a). Using a symmetrical positive definite matrix PLEstimating parameters for the oldest state and a few priors in a rolling time domain window
Figure BDA0002876635860000219
Is corrected with a weight coefficient of the symmetric positive array PL. Furthermore, all estimates in the least squares method may be limited by the upper and lower bounds of the system's physical limitations. In the function of cost
Figure BDA00028766358600002110
In the dynamic optimization problem, a control quantity is also included to illustrate the actual control of the system and the control calculated by the controller
Figure BDA00028766358600002111
And such variations may be caused by actuator noise or inaccuracies.
At time k, the optimal objective function and its constraint conditions are:
J(xk,uk,pk)
Figure BDA00028766358600002112
Figure BDA00028766358600002113
Figure BDA00028766358600002114
Figure BDA00028766358600002115
least square method objective function
Figure BDA00028766358600002116
It can be expressed as:
Figure BDA00028766358600002117
k represents the current time;
Figure BDA00028766358600002118
and
Figure BDA00028766358600002119
representing the system output and system input actually measured at the current time.
At the time k +1, the optimal objective function and its constraint conditions are:
J(xk+1,uk+1,pk+1)
Figure BDA00028766358600002120
Figure BDA00028766358600002121
Figure BDA00028766358600002122
Figure BDA00028766358600002123
least square method objective function
Figure BDA0002876635860000221
It can be expressed as:
Figure BDA0002876635860000222
k +1 represents the current time;
Figure BDA0002876635860000223
and
Figure BDA0002876635860000224
representing the system output and system input actually measured at the current time.
In the least square method objective function, the weighting matrix coefficient PL,Vk,WkThe value selection of (c) affects the accuracy of the vehicle state estimation. For a proper selection of the weighting matrix, it is assumed that the output of the measured values has a gaussian distribution characteristic. Assume an initial value xLAnd the parameter p is a random variable with a normal distribution, whose covariance matrices are respectively
Figure BDA0002876635860000225
The mean values thereof are respectively
Figure BDA0002876635860000226
Assuming measured values of the system
Figure BDA0002876635860000227
And unknown control inputs
Figure BDA0002876635860000228
Are each yk,ukThe covariance matrices are respectively
Figure BDA0002876635860000229
A gaussian distribution of (a). PLOf two block diagonal matrices of
Figure BDA00028766358600002210
Vk,WkAre respectively
Figure BDA00028766358600002211
The least squares objective function conforms to the maximum likelihood estimate of the current window. Expression P of the weighting matrixL,Vk,WkRespectively as follows:
Figure BDA00028766358600002212
the first term in the objective function represents all the information collected by the measurement before time t. Reference estimator
Figure BDA00028766358600002213
Selection of values the solution of the rolling temporal optimization at the current instant of estimation. Selecting the arrival cost matrix P in different waysLIt may be a constant zero matrix or a so-called smooth extended kalman filter update of the sensitivity information obtained when solving the previous rolling time domain optimization problem. Classic extended Kalman Filter is equivalent to updating and rolling using a smooth Kalman FilterAnd (4) performing rolling time domain optimization estimation with the moving time domain of 1.
The method is a direct method for obtaining the objective function through the multiple target practice discrete rolling time domain optimization problem, the finite dimension least square nonlinear programming is solved by using a generalized Gauss-Newton method, and the method solves the quadratic programming problem once during each iteration. This allows multiple active sets to vary, ensuring that the performance of the non-linear rolling time domain optimization algorithm is no worse than a rolling time domain estimator based on a differential equation model.
In summary, the above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (5)

1. A state prediction and estimation method for an unmanned vehicle, comprising the steps of:
step 1, collecting the wheel speed, wheel torque, longitudinal acceleration, lateral acceleration and vertical acceleration of a vehicle body of each wheel, and the roll angle speed, pitch angle speed and yaw angle speed of the vehicle;
2, randomly selecting 20 working conditions from the standard running working conditions based on the vehicle state information obtained in the step 1, then randomly extracting 400 working condition blocks with different time lengths from the selected 20 working conditions, extracting 10 characteristics of each working condition block from the selected working condition blocks, performing characteristic calculation on the extracted 400 working condition blocks, performing speed prediction, correcting the speed prediction by using the acceleration and acceleration derivative parameters, optimizing parameter values by using an algorithm, and training the speed prediction by using a genetic algorithm neural network; then, in order to improve the adaptability of the vehicle speed prediction under any working condition, the historical state of the vehicle under the current running working condition is learned by using a rolling updating method, so that the problem of vehicle speed prediction precision caused by the limitation of the vehicle running working condition is avoided;
step 3, building a vehicle model based on the vehicle state information obtained in real time in the step 1, wherein the model comprises a vehicle body kinematics model, a vehicle body dynamics model, a tire model and a suspension model, and a non-linear model and a measurement model of noise-free state propagation are given;
step 4, integrating road slope angle estimation by using a kinematic method and a dynamic method, and improving the static deviation of low-frequency noise and the model dependency of high-frequency noise; estimating the vehicle mass by using a least square method, and improving the estimation precision of the vehicle mass and the sprung mass; estimating the vertical force of the wheel by using an information fusion method, and ensuring the estimation precision of the vertical force in the attitude change of the vehicle; introducing a road adhesion coefficient, and performing joint estimation on the lateral force and the vehicle state of the vehicle by adopting a nonlinear estimation method;
and 5, based on the vehicle parameters and the working condition information obtained in the step 4, performing vehicle state rolling estimation on the vehicle variable parameters and the variable working conditions, and improving the accuracy of vehicle speed prediction under the conditions of the vehicle variable parameters and the variable working conditions.
2. A state prediction and estimation method for unmanned vehicles according to claim 1, wherein the speed prediction in step 2 specifically comprises:
randomly selecting 20 working conditions from standard running working conditions, randomly extracting 400 working condition blocks with different time lengths from the selected 20 working conditions, and extracting 10 characteristics of each working condition block from the selected working condition blocks, wherein each characteristic corresponds to a parameter, and the 10 characteristic parameters specifically comprise the highest vehicle speed
Figure FDA0003576746160000011
Average vehicle speed
Figure FDA0003576746160000012
Maximum acceleration AmaxAverage acceleration AavMaximum acceleration derivative dAmaxMean acceleration dAavMaximum deceleration BmaxAverage deceleration BavDerivative of maximum deceleration dBmaxAverage acceleration derivative dBav(ii) a The specific calculation formula of 10 parameters is represented by the following formula：
Figure FDA0003576746160000021
Wherein, N is the time length of each working condition block, and N1, N2, N3 and N4 are the acceleration time length, deceleration time length, acceleration derivative time length and deceleration derivative time length of each working condition block respectively;
the speed prediction in the future time domain [ k +1, k + n ] is specifically as follows:
Figure FDA0003576746160000022
wherein the content of the first and second substances,
Figure FDA0003576746160000023
is the speed at the present moment in time,
Figure FDA0003576746160000024
to predict the velocity in the time domain at time k + n, ReffIs the effective rolling radius of the wheel, TmIs the driving and braking torque of the vehicle, tdIs the exponential type variable attenuation coefficient of the motor torque, m is the whole vehicle mass, f is the rolling resistance coefficient of the vehicle, rhoa,CdAir density and windward resistance coefficient;
during acceleration of the vehicle, the vehicle drive torque TmThe correction value is expressed as:
Figure FDA0003576746160000025
by introducing an acceleration parameter k1,k2Preventing the vehicle speed from being reduced too fast in a high-speed steady state, and accelerating the attenuation of the torque required by the vehicle when the vehicle speed is higher; acceleration derivative parameter k3,k4When the acceleration of the vehicle is sharply increased, a large required torque is maintained atThe expected speed is increased in time in the prediction time domain; a. theav,dAavThe speed of the historical time period of the vehicle is obtained through a segmented acceleration algorithm and a segmented acceleration derivative algorithm, in order to reduce the complexity of algorithm calculation and ensure good real-time performance of a controller, the segmented time is selected to be consistent, 5s is selected, the speed of the historical time period is updated in a rolling mode in real time according to the time state of the vehicle, namely new historical speed data is added, and historical speed data with earlier time is discarded; the acceleration parameter and the acceleration derivative parameter are subjected to particle swarm optimization to obtain optimal parameter values under 400 typical working conditions respectively, and the ranges of the acceleration parameter and the acceleration derivative parameter are given in table 1;
TABLE 1 range of acceleration parameter and acceleration derivative parameter
Figure FDA0003576746160000026
Figure FDA0003576746160000031
Training 400 groups of different acceleration parameters and acceleration derivative parameters obtained based on a genetic algorithm neural network in real time according to historical data of the current running condition of the vehicle, and utilizing the error square sum EiMeasuring the adaptive value;
Ei＝∑∑(d0-y0)2
wherein d is0,y0Actual output and desired output of the network, respectively;
according to the historical vehicle speed, recording the vehicle speed data of 10s in the past historical time period, extracting 10 parameters, taking the parameters as input, training the genetic algorithm neural network, and performing online optimization updating on the parameters, thereby realizing the rolling prediction of the vehicle speed.
3. A state prediction and estimation method for unmanned vehicles according to claim 2 wherein said standard driving conditions include WVUSUB, WVUINTER, VAIL2NRE, US06_ HWY, US06, UNIF01, UKBUS6, UDDSHDV, UDDS, SC03, REP05, NYCTRUCK, NYCCOMP, NYCC, numbergr 36, NREL2VAIL, NewYorkBus, MEASURED _ MASS1, MANHATTAN, LA92, INDIA _ URBAN _ ply, INDIA _ HWY _ same, IM240, HWFET, HL07, EUDC _ LOW, EUDC, ECE, hvr _ Vehicle, commulter, cbdtru, CBD14, rte, art 1015, ARB02, pri _ 866.
4. A state prediction and estimation method for unmanned vehicles according to claim 1, characterized in that step 4 is specifically:
the basic idea of road slope estimation is to retain the high frequency part of the kinematic estimation and filter out the low frequency part, and retain the low frequency part of the kinematic estimation and filter out the high frequency part;
Figure FDA0003576746160000032
wherein the content of the first and second substances,
Figure FDA0003576746160000033
is the road slope estimated by the fusion method,
Figure FDA0003576746160000034
a dynamically estimated road slope and a kinematically estimated road slope, respectively;
the dynamics estimation method relies on a vehicle dynamics model, which is expressed as:
Figure FDA0003576746160000035
wherein, FxIs the longitudinal force of the vehicle, is directly obtained through the torque vector control of the distributed front-driving unmanned vehicle,
Figure FDA0003576746160000036
acceleration in the longitudinal direction of the vehicle, p, CdA is vehicle air density, wind resistance coefficient and windward area, and theta and f are road gradient and road rolling resistance coefficient;
the model-based dynamics estimation road slope method is represented as follows:
Figure FDA0003576746160000037
the observed value of the acceleration of the inertial navigation in the longitudinal direction is related to the derivative of the speed of the vehicle as follows:
Figure FDA0003576746160000041
a kinematic estimation method of road gradient can be obtained by the above equation:
Figure FDA0003576746160000042
estimation and estimation of vehicle mass based on recursive least square method
Figure FDA0003576746160000043
Figure FDA0003576746160000044
Wherein k is the current sampling moment, and k-1 is the last sampling moment; l, K are the least squares gain and error covariance update, respectively;
estimation of sprung mass
Figure FDA0003576746160000045
The unsprung mass m may be subtracted from the estimated vehicle massusObtaining:
Figure FDA0003576746160000046
in the observation of the vertical force of the tire, the longitudinal movement, the lateral movement and the rolling movement of the vehicle are considered;
the load transfer of the front/rear axle caused by longitudinal and transverse motion is estimated by the following formula:
Figure FDA0003576746160000047
the roll motion of the vehicle also results in vertical load transfer; estimating the change of the axial force in the vehicle rolling process according to a calculation formula of the vehicle rolling motion:
Figure FDA0003576746160000048
vertical load F of each tirezijThe method can be based on the estimation:
Figure FDA0003576746160000049
estimating the vehicle state and the lateral force by a nonlinear estimation method, estimating the vehicle state and the tire lateral force by the nonlinear method, wherein the number of Sigma points is related to the number of state points, and the number of n state points generates 2n +1 Sigma points, wherein the first one is defined as the average value of the random current state and the Sigma points
Figure FDA00035767461600000410
Can be calculated by the following formula:
Figure FDA00035767461600000411
wherein, PxIs the covariance of the state x at random,
Figure FDA00035767461600000412
is the square root of the matrix, λ is a scaling parameter, and the calculation formula is as follows:
λ＝α2(n+κ)-n,(1≤α≤10-4)
wherein the constant alpha determines the distribution of sigma points around the average value of the current state, and if the alpha value is larger, the sigma points are away from the state
Figure FDA0003576746160000051
Average value of (2)
Figure FDA0003576746160000052
The farther away, the closer the sigma point is when the alpha value is smaller
Figure FDA0003576746160000053
Predicting sigma point χ at point k +1k+1Updating based on a nonlinear function according to the current sigma point;
χk+1＝χk+f(χk,uk+1,wk+1)·ΔT
wherein u isk+1As a control input to the system, wk+1Is process noise, Δ t is the sampling time;
the nonlinear measurement of the system can be obtained by the nonlinear measurement of the system:
Yk+1＝h(xk+1,uk+1,vk+1)
wherein v isk+1To measure noise, it is assumed that the noise has no nonlinear effect on the system;
Figure FDA0003576746160000054
thereby obtaining a state vector xk+1|kAnd measuringValue yk+1Average value of (d):
Figure FDA0003576746160000055
wherein the content of the first and second substances,
Figure FDA0003576746160000056
is a weight matrix;
the covariance of the state vector and the measurement vector can be updated by:
Figure FDA0003576746160000057
wherein the content of the first and second substances,
Figure FDA0003576746160000058
is a weight matrix; q is the covariance of the system process noise, R is the measurement noise; q and R are both Gaussian white noise; the measurement update equation is:
Figure FDA0003576746160000059
wherein z isk+1Is a measurement signal from a sensor;
the upper and lower boundaries of the slip angle can be represented by the following relationship:
Figure FDA00035767461600000510
based on the estimate of the mid-lateral dynamics, the peak self-aligning torque is sensitive to the tire road friction coefficient when the slip angle varies in the middle zone (up and down); estimating a coefficient of friction from the SAT using a brush tire model; the brush tire model well simulates the alignment torque behavior of the tire,
τ＝μFzlθytan(α){1-3|θytan(α)|+3|θytan(α)|2-|θytan(α)|3}
when the peak value is 27/256 mu FzAt 4tan (. alpha.) thetayWhen 1, the road friction coefficient can be estimated
Figure FDA0003576746160000061
Figure FDA0003576746160000062
τmax＝max(τ(t)),t∈[t-Δt,t]
In the formula, τmaxEstimating the maximum value of the aligning moment in the past window delta t;
based on the estimation of large dynamics, according to the friction circle theory, when the slip angle is larger than the upper boundary, the tire force is in the saturation region; the tire longitudinal and lateral forces follow non-linear constraints:
Fxi_sat 2+Fyi_sat 2≤(μFz)2
the road friction is estimated using the following equation:
Figure FDA0003576746160000063
non-linear estimation under other conditions, assuming that the tire-road friction coefficient is constant during the sampling time Δ t; the estimated state function is thus expressed as:
Figure FDA0003576746160000064
the measured output in the non-linear estimator is defined as:
Figure FDA0003576746160000065
wherein the content of the first and second substances,
Figure FDA0003576746160000066
different friction estimation algorithms are designed according to different excitation levels; these estimators are integrated into a road friction estimator.
5. The method as claimed in claim 1, wherein the step 5 of performing rolling estimation of the vehicle state of the vehicle variable parameters and the variable working conditions specifically comprises:
discretizing the vehicle noiseless state propagation model and the measurement model in the general form obtained in the step 3 to obtain a discretized vehicle propagation model and a discretized measurement model:
Figure FDA0003576746160000067
performing feature extraction on the attitude influencing the vehicle state estimation, wherein the attitude comprises longitudinal motion, lateral motion, vertical motion, pitching motion, rolling motion and yawing motion of the vehicle, and correspondingly extracting corresponding variables comprises the following steps:
xa|t＝[u v w p q r]T
the vehicle state rolling estimation method is based on an optimization objective function of J (x, u, p); wherein x, u and p are respectively the state of a vehicle system, the input of the system and the parameter variable of the vehicle;
the rolling time domain optimization method is that in each step state estimation time domain, vehicle measurement output of a vehicle at the current moment and the vehicle state at the current moment are increased, the vehicle state at the moment of q-N +1 is discarded, and the steps are circulated in sequence at a new moment; at the time of k-1, the optimal objective function and the constraint conditions thereof are as follows:
J(xk-1,uk-1,pk-1)
s.t.xk＝F(xk-1,uk-1)
Figure FDA0003576746160000071
Figure FDA0003576746160000072
Figure FDA0003576746160000073
Figure FDA0003576746160000074
the least squares objective function J, can be expressed as:
Figure FDA0003576746160000075
k-1 represents the last time instant,
Figure FDA0003576746160000076
and
Figure FDA0003576746160000077
representing the system output and system input actually measured at time k-1;
Figure FDA0003576746160000078
the representation has a diagonal positive definite weighting matrix Wk-1And Vk-1The euclidean norm of (d); using a symmetrical positive definite matrix PLEstimating parameters for the oldest state and a few priors in a rolling time domain window
Figure FDA0003576746160000079
Is corrected with a weight coefficient of a symmetrical positive array PL(ii) a In addition, all the estimators in the least squares method may beCan be limited by the upper and lower bounds of the physical limitations of the system; in the function of cost
Figure FDA00035767461600000710
In the dynamic optimization problem, a control quantity is also included to illustrate actual control of the system and control calculated by the controller
Figure FDA00035767461600000711
The deviation therebetween;
at time k, the optimal objective function and its constraint conditions are:
J(xk,uk,pk)
s.t.xk+1＝F(xk,uk)
Figure FDA00035767461600000712
Figure FDA00035767461600000713
Figure FDA00035767461600000714
Figure FDA00035767461600000715
least square method objective function
Figure FDA00035767461600000716
It can be expressed as:
Figure FDA00035767461600000717
k represents the current time;
Figure FDA00035767461600000718
and
Figure FDA00035767461600000719
system outputs and system inputs representing actual measurements at the current time;
at the time k +1, the optimal objective function and its constraint conditions are:
J(xk+1,uk+1,pk+1)
s.t.xk+1＝F(xk,uk)
Figure FDA0003576746160000081
Figure FDA0003576746160000082
Figure FDA0003576746160000083
Figure FDA0003576746160000084
least square method objective function
Figure FDA0003576746160000085
It can be expressed as:
Figure FDA0003576746160000086
k +1 represents the current time;
Figure FDA0003576746160000087
and
Figure FDA0003576746160000088
system outputs and system inputs representing actual measurements at the current time;
in the least square method objective function, the weighting matrix coefficient PL,Vk,WkThe value selection of (a) affects the accuracy of the vehicle state estimation; for proper selection of the weighting matrix, the output of the measured values is assumed to have a gaussian distribution; assume an initial value xLAnd the parameter p is a random variable with a normal distribution, whose covariance matrices are respectively
Figure FDA0003576746160000089
The mean values thereof are respectively
Figure FDA00035767461600000810
Assuming measurements of the system
Figure FDA00035767461600000811
And unknown control inputs
Figure FDA00035767461600000812
Are each yk,ukThe covariance matrices are respectively
Figure FDA00035767461600000813
(ii) a gaussian distribution of; pLOf two block diagonal matrices of
Figure FDA00035767461600000814
Vk,WkAre respectively
Figure FDA00035767461600000815
The least square method target function accords with the maximum likelihood estimation of the current window; expression P of the weighting matrixL,Vk,WkRespectively as follows:
Figure FDA00035767461600000816
the first term in the objective function, representing all the information collected by the measurement before time t; reference estimator
Figure FDA00035767461600000817
Selecting a value to estimate a solution for the instantaneous rolling time domain optimization; selecting the arrival cost matrix P in different waysLIt may be a constant zero matrix, or a so-called smooth extended kalman filter update of the sensitivity information obtained when solving the previous rolling time domain optimization problem; classical extended kalman filtering is equivalent to using a smooth kalman filter update and a rolling time domain optimization estimate with a rolling time domain of 1.
CN202011631422.0A 2020-12-30 2020-12-30 State prediction and estimation method for unmanned vehicle Active CN112758097B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011631422.0A CN112758097B (en) 2020-12-30 2020-12-30 State prediction and estimation method for unmanned vehicle

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011631422.0A CN112758097B (en) 2020-12-30 2020-12-30 State prediction and estimation method for unmanned vehicle

Publications (2)

Publication Number Publication Date
CN112758097A CN112758097A (en) 2021-05-07
CN112758097B true CN112758097B (en) 2022-06-03

Family

ID=75699598

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011631422.0A Active CN112758097B (en) 2020-12-30 2020-12-30 State prediction and estimation method for unmanned vehicle

Country Status (1)

Country Link
CN (1) CN112758097B (en)

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113392518B (en) * 2021-06-08 2024-05-17 阿波罗智联(北京)科技有限公司 Method and apparatus for estimating vehicle weight
CN113408047B (en) * 2021-08-05 2024-03-22 江苏大学 Vehicle dynamics prediction model based on time-lag feedback neural network, training data acquisition method and training method
CN113335294B (en) * 2021-08-09 2021-10-29 天津所托瑞安汽车科技有限公司 Method and device for estimating longitudinal gradient of road surface, electronic device and storage medium
CN114043986B (en) * 2021-08-20 2024-04-26 东南大学 Multi-model fusion estimation method for tire road surface adhesion coefficient by considering quality mismatch
CN114291093A (en) * 2021-11-30 2022-04-08 安徽海博智能科技有限责任公司 Layered correction control method and system for cooperative automatic driving of vehicle and road
CN113978470B (en) * 2021-12-13 2024-01-12 郑州轻工业大学 On-line quick estimation method for friction force between tire and road surface
CN113895446B (en) * 2021-12-13 2022-02-22 智道网联科技(北京)有限公司 Method, device, equipment and storage medium for determining speed of automatic driving vehicle
CN114312811B (en) * 2022-01-27 2023-11-07 清华大学 Method, device and equipment for approximately optimally estimating self-driving state of automatic driving automobile
US20240092397A1 (en) * 2022-08-26 2024-03-21 Zoox, Inc. Interpretable kalman filter comprising neural network component(s) for autonomous vehicles
CN116985825B (en) * 2023-09-26 2024-03-19 安徽蔚来智驾科技有限公司 Method, system and storage medium for determining vehicle state
CN117251818B (en) * 2023-11-17 2024-01-26 上海伯镭智能科技有限公司 Data management method for safe operation of unmanned mine car

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102514560A (en) * 2011-12-09 2012-06-27 北京理工大学 Method for acquiring longitudinal running speed information of vehicle in anti-skid control system
CN107813830A (en) * 2016-08-31 2018-03-20 法乐第(北京)网络科技有限公司 A kind of method and device for aiding in vehicle drive
EP3353494A1 (en) * 2016-12-14 2018-08-01 Baidu USA LLC Speed control parameter estimation method for autonomous driving vehicles
CN109766745A (en) * 2018-11-22 2019-05-17 四川大学 The combination of intensified learning tri-state long Memory Neural Networks system and training and prediction technique in short-term
CN110497916A (en) * 2019-08-15 2019-11-26 太原科技大学 Vehicle driving state estimation method based on BP neural network
EP3696734A1 (en) * 2019-02-15 2020-08-19 Siemens Aktiengesellschaft Tensor-train recurrent neural network supported autonomous and assisted driving
DE102019211738B3 (en) * 2019-08-05 2020-11-12 Zf Friedrichshafen Ag Checking a driver takeover based on image data

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109102073A (en) * 2017-06-21 2018-12-28 上海寒武纪信息科技有限公司 A kind of sparse training method

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102514560A (en) * 2011-12-09 2012-06-27 北京理工大学 Method for acquiring longitudinal running speed information of vehicle in anti-skid control system
CN107813830A (en) * 2016-08-31 2018-03-20 法乐第(北京)网络科技有限公司 A kind of method and device for aiding in vehicle drive
EP3353494A1 (en) * 2016-12-14 2018-08-01 Baidu USA LLC Speed control parameter estimation method for autonomous driving vehicles
CN109766745A (en) * 2018-11-22 2019-05-17 四川大学 The combination of intensified learning tri-state long Memory Neural Networks system and training and prediction technique in short-term
EP3696734A1 (en) * 2019-02-15 2020-08-19 Siemens Aktiengesellschaft Tensor-train recurrent neural network supported autonomous and assisted driving
DE102019211738B3 (en) * 2019-08-05 2020-11-12 Zf Friedrichshafen Ag Checking a driver takeover based on image data
CN110497916A (en) * 2019-08-15 2019-11-26 太原科技大学 Vehicle driving state estimation method based on BP neural network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
Longitudinal Dynamic Control under Complex Driving Conditions via Fuzzy Logic Sliding-mode Control;赵玉壮 等;《IEEE International Conference on Mechatronics and Automation》;20190807;第1161-1166页 *

Also Published As

Publication number Publication date
CN112758097A (en) 2021-05-07

Similar Documents

Publication Publication Date Title
CN112758097B (en) State prediction and estimation method for unmanned vehicle
CN108594652B (en) Observer information iteration-based vehicle state fusion estimation method
Acosta et al. Virtual tyre force sensors: An overview of tyre model-based and tyre model-less state estimation techniques
US7571039B2 (en) Vehicle yaw/roll stability control with semi-active suspension
US20090177346A1 (en) Dynamic estimation of vehicle inertial parameters and tire forces from tire sensors
CN110588657B (en) Joint estimation method for vehicle motion state and road gradient
CN111645699B (en) Model self-adaptive lateral speed estimation method based on multi-sensor information fusion
JP2020504049A (en) System and method for calibrating vehicle tires
CN114379583A (en) Automatic driving vehicle trajectory tracking system and method based on neural network dynamics model
CN113830088A (en) Intelligent semi-trailer tractor trajectory tracking prediction control method and vehicle
CN116061934A (en) Architecture for managing chassis and driveline actuators and model-based predictive control method
CN116560223A (en) Intelligent vehicle dynamics model, ILQR control algorithm and track tracking controller based on physical information neural network
CN116680873A (en) Vehicle state estimation method based on improved self-adaptive extended Kalman filtering
CN112613125B (en) Method for identifying and dynamically correcting automobile roll center under condition of road surface excitation
US10940734B2 (en) Suspension control system and vehicle
CN112287289A (en) Vehicle nonlinear state fusion estimation method for cloud control intelligent chassis
CN114435371B (en) Road gradient estimation method and device
CN113650621B (en) Distributed driving electric vehicle state parameter estimation method facing complex working conditions
CN113341997B (en) Transverse control method and system based on multi-state parameter collaborative estimation
CN113071282B (en) Vehicle adjusting method and device
CN117465465B (en) Vehicle sprung mass and road transverse gradient joint estimation method
CN115946679B (en) Vehicle stability judging method and system
Biglarbegian et al. Design of a novel fuzzy controller to enhance stability of vehicles
Das Vehicle Dynamics Modelling: Lateral And Longitudinal
CN117807703B (en) Method for estimating key parameters of scooter chassis vehicle with mutually corrected dynamic and static parameters

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant