CN109885883A - A kind of control method of the unmanned vehicle transverse movement based on GK clustering algorithm model prediction - Google Patents
A kind of control method of the unmanned vehicle transverse movement based on GK clustering algorithm model prediction Download PDFInfo
- Publication number
- CN109885883A CN109885883A CN201910052382.5A CN201910052382A CN109885883A CN 109885883 A CN109885883 A CN 109885883A CN 201910052382 A CN201910052382 A CN 201910052382A CN 109885883 A CN109885883 A CN 109885883A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- model
- control
- tire
- clustering algorithm
- 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.)
- Granted
Links
Classifications
-
- 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
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Control Of Position, Course, Altitude, Or Attitude Of Moving Bodies (AREA)
- Steering Control In Accordance With Driving Conditions (AREA)
Abstract
The invention discloses a kind of control method of unmanned vehicle transverse movement based on GK clustering algorithm model prediction, 1, vehicle's current condition is obtained in real time;2, vehicle-periphery is acquired, plans expected path in real time;3, single track whole vehicle model is established using GK clustering algorithm;4, the single track whole vehicle model that step 3 obtains is converted into the state space equation of linearity error model, and sliding-model control;5, utilize the Model Predictive Control Algorithm of linear time-varying, establish the model predictive controller of linear time-varying, using under vehicle axis system systemic velocity, course angle, yaw velocity, vehicle location as model predictive controller input, front wheel angle is exported as controller, according to current state and target trajectory, calculates the control sequence in the tracing point and control time domain in prediction time domain, problem is converted into quadratic programming problem and seeks optimal solution, updates vehicle-state;6, the control amount obtained according to model predictive controller successively controls the steering of target vehicle.
Description
Technical field
The invention belongs to intelligent automobile control technology fields, and in particular to a kind of nothing based on GK clustering algorithm model prediction
The control method of people's vehicle transverse movement.
Background technique
In prior art research dynamics of vehicle crosswise joint, done it is many it is assumed that especially in terms of tire characteristics,
All it is usually that only considered linear region of the side drift angle less than 5 °, the precision of model is enormously simplified, to also reduce control
Accuracy, and in actual vehicle movement, especially under high-speed working condition, side drift angle can easily exceed linear region, reach non-
Cause the unstability of vehicle in linear region, it is easy to lead to vehicle rollover, therefore fully consider the cornering behavior of tire, for vehicle
The research of transverse movement control is of great significance to.
Summary of the invention
The present invention is based on a kind of improved G-K clustering algorithms by the nonlinear portion of relationship between side force of tire and side drift angle
Divide piece-wise linearization, the crosswise joint of vehicle is then carried out by linear time-varying model predictive control algorithm.The specific skill used
Art scheme is as follows:
A kind of control method of the unmanned vehicle transverse movement based on GK clustering algorithm model prediction is that control object vehicle is real
When steering wheel angle is provided, thus realize to control target transverse movement control, comprising the following steps:
Step 1, vehicle's current condition, such as systemic velocity, course angle, yaw angle speed are obtained in real time using vehicle sensors
Degree, vehicle changing coordinates and slip angle of tire and vehicle speed information.
Step 2, environment surrounding automobile information is acquired using industrial camera and millimetre-wave radar, determines travelable region, from
And expected path is planned in real time.
Step 3, single track whole vehicle model is established: will be between tire cornering power and side drift angle in relationship using GK clustering algorithm
Non-linear partial piece-wise linearization obtains the piecewise function of affine rear relationship between tire cornering power and slip angle of tire, utilizes
This piecewise function, which is brought into vehicle dynamic model, obtains single track whole vehicle model expression formula.
Step 4, this non-linear single track whole vehicle model is converted into the state space of linearity error model using Taylor's formula
Equation, wherein quantity of state takes the vehicle under systemic velocity under vehicle axis system, course angle, yaw velocity, earth coordinates
Position, control measure front wheel angle, and by its sliding-model control.
Step 5, using the Model Predictive Control Algorithm of linear time-varying, the model predictive controller of linear time-varying is established, it will
The vehicle location under systemic velocity, course angle, yaw velocity, earth coordinates under vehicle axis system is as model prediction control
Prediction time domain is calculated according to current state and target trajectory in the input of device processed, output of the front wheel angle as controller
Control sequence in interior tracing point and control time domain, control sequence, establishes objective function, and problem is converted into order to obtain
Quadratic programming problem seeks optimal solution, and using first element of control sequence as the control amount of practical control target.It updates
The state of vehicle repeats the rolling optimization function of above step implementation model prediction.
Step 6, the control amount obtained according to model predictive controller successively controls the steering of target vehicle.
Further, in institute's step 1:
Obtaining the current status information of vehicle in real time is acquired in real time using inertial navigation, and vehicle speed information is by wheel speed sensors
Acquisition obtains in real time.
Further, in institute's step 2:
The method of planning expected path is to acquire environment surrounding automobile information using industrial camera and millimetre-wave radar in real time,
Travelable region is determined, to plan expected path in real time.
Further, in institute's step 3:
Single track whole vehicle model is established, detailed process is as follows:
Using the method for GK cluster by the non-linear partial piecewise affine in tire cornering power F and slip angle of tire α relationship
At several sections of inearized models, identification of Model Parameters includes the segmentation of data subspace, the estimation of each subspace PARAMETERS IN THE LINEAR MODEL and cuts
Change the estimation of face equation coefficient.The test number of slip angle of tire α and lateral force F under all kinds of operating conditions are acquired using multi-dimension force sensor
According to carrying out parameter identification to the piecewise affine models of tire cornering characteristics and illustrate the technical solution of this method by taking front-wheel as an example.
Based on the experimental data that sensor obtains, regression equation: y (k)=f (x (k))+e (k) is constructed, wherein x (k) is regression vector,
It outputs and inputs vector by the history of system to constitute, x (k)=[y (k-1) ... y (k-na),u(k-1)…u(k-na)];E (k) is
The additive noise of known probability Density Distribution, y (k) are actual measurement output signal, and u (k) is actual measurement input signal.
1) piecewise affine data Subspace partition
For input/output signal sequence Z=(x (k), y (k)), k=1 ... T, T > 0, defining mode function: μ (k):
{ 1 ... T } → { 1 ... s }, μ (k) :=i, whenever x (k) ∈ χiInvention, using improved G-K clustering method implementation pattern
The solution of function mu, so that cluster result meets following objective function:
Wherein, m ∈ [1, ∞) characterization clustering fuzzy degree adjustable parameter, represent the overlapping degree between of all categories, take m
=2.d(zj,vi) indicate sample zjWith cluster centre viThe distance between, it determines the shape of cluster, defines this algorithm here and adopt
With adaptive distance metric method:
d2(zj-vi)=(zj-vi)TMi(zj-vi) (2)
Wherein, MiIt is positive definite matrix,
Wherein, FiIt is cluster covariance matrix,
Data-oriented collection Z, clustering algorithm key step as shown in Fig. 2,
Step 1: the covariance matrix of experiment with computing data set Z, it is initial cluster that farthest two samples are selected from Z
Center calculates subordinated-degree matrixEnable l=1
Step 2: cluster covariance matrix F is calculatedi, characteristic value and feature vector are extracted, covariance adjustment is carried out
Step 3: M is calculatediWith squared-distance d2
Step 4: subordinated-degree matrix is recalculated
Step 5: judgement | | Ul-Ul-1| | whether > ε is true, if it is return step two, if not then continuing to walk
Rapid six
Step 6: retain this cluster centreWith subordinated-degree matrixClustering validation is carried out, that is, judges c
≤cmax, if it is leaping to step 9, if not then continuing step 7
Step 7: c=c+1, according to subordinated-degree matrixFind out a sample z dissimilar with each subsetk
Step 8: according to new cluster initial center, corresponding new initial subordinated-degree matrix is calculated
Step 9: determine that best cluster number carries out data set division according to the degree of membership of every group of affiliated subset of sample.
2) piecewise affine submodel parameter identification
After obtaining data Subspace partition based on clustering method, submodel parameter identification problem reduction is asked for linear optimization
Topic, the present invention are calculated using weighted least-squares method.According to subclass data set ZiInputoutput data, with practical system
Output bias quadratic sum between system and submodel constitutes criterion function as follows, and the minimum of the criterion corresponds to submodel ginseng
Number vector θiEstimated value.
Wherein, yiIt (j) is output signal, xiIt (j) is input signal.
3) diverter surface equation coefficient is estimated
Due to each scopeConvex polyhedron there is no lap, diverter surface equation coefficient in addition to public boundary
Identification can be converted into the linear partition problem of cluster data.The present invention carries out hyperplane side using the method based on support vector machines
The solution of journey coefficient, for the convex polyhedron section of submodel scope:Switching
Face equation can be described as: hji={ wix(k)+bi=0 }, wi、biIt is coefficient vector respectively, compromise considers maximum class interval and most
Few error sample establishes Generalized optimal classifying face, and the linear kernel function that selection meets Mercer condition constitutes supporting vector, thus
Obtain system hyperplane equation coefficient h.
Since the subspace number c of Piecewise affine systems is determined by the extreme value of optimizing index, estimate it is possible that crossing
The case where, the method that proposed adoption of the present invention verifies afterwards judges the similitude of each submodel, realizes similar submodel
Merging and the identification again of model parameter.
It is as follows that the cluster affine relation that affine tire cornering power F and side drift angle α are obtained is clustered by GK clustering algorithm:
Wherein,For the parameter vector of affine submodel,For the effect of affine submodel
Domain convex body section, and assume that each convex body does not have overlapping interval in addition to public boundary, i.e.,It switches when the state of system reaches borderline region.The lateral deviation in control process
Angle α can be calculated according to following relational expression:
WhereinWithBeing is longitudinal velocity under vehicle axis system, side velocity and yaw rate respectively,
This tittle is all that the quantity of state of auto model can be calculated in real time in control process, δfIt is the front wheel angle of vehicle, is
The control amount of auto model is calculated in real time in control process, and a, b are distance of the vehicle centroid to antero posterior axis respectively.
It brings lateral force into single track whole vehicle model, obtains mathematical model:
Wherein, Fcf,FcrLateral force suffered by the forward and backward tire of vehicle respectively is represented via the segmentation of above-mentioned clustering algorithm
Come;Flf, FlrRespectively longitudinal force suffered by the forward and backward tire of vehicle, it is related with the longitudinal rigidity of tire, slip rate;δfBefore vehicle
Take turns corner;For Vehicular yaw angle;A is distance of the front axle to mass center, and b is distance of the rear axle to mass center, IZIt is vehicle around z-axis
Rotary inertia.
Further, in the step 4:
It is nonlinear model for single-track vehicle kinetic model obtained above, it is therefore desirable to which linearisation could use state
Space equation expression, wherein quantity of state takes systemic velocity, course angle, yaw velocity, earth coordinates under vehicle axis system
Under vehicle location, control measure front wheel angle.The specific method is as follows:
Single-track vehicle kinetic model is write as:
Wherein, ξ is quantity of state, and μ is control amount, by it in arbitrary point (ξr,ur) at Taylor expansion, only retain single order item, suddenly
Slightly higher order term, obtains:
In addition also have at this arbitrary point:
Formula (9) and (10) are subtracted each other to obtain:
In formula,
Sliding-model control is carried out with the method for single order difference coefficient to this state space equation, obtains Discrete Linear model tormulation
Formula:
Wherein, A (k)=I+TA (t), B (k)=TB (t), T are the sampling times.
Further, in institute's step 5:
Auto model is converted to linearity error model and brings Model Predictive Control by the characteristics of according to Model Predictive Control
One group of control sequence is calculated by the Optimization Solution function inside model predictive controller in device, and by the of control sequence
One amount acts on controlled device, then repeating scrolling above step, is specifically divided into the description, excellent of state variable and output variable
Change and solves, feeds back calculating.Detailed process is as follows:
1) description of state variable and output variable
In order to solve convenience, above formula (15) are merged and are write as:
Obtain new state-space expression:
Wherein,
It is the output predicted in time domain.
Choosing prediction time domain is 10, and control time domain is 3, and the output quantity in system prediction time domain is expressed with a matrix type
Are as follows:Control time domain in output beThe output in system future can be write as: Y
(k+10 | k)=Sxx(k)+SuΔ U (k), wherein
C is state sky
Between output matrix.
2) Optimization Solution:
Cost function is set are as follows:
Wherein first item reflects the tracking ability to reference locus, and Section 2 reflects the steady change of control process,
Q, R is this two weight matrix, ρ ε respectively2Be in order to avoid in optimization process occur without solution situation and it is increased relaxation because
Son.Be converted to the quadratic programming problem of standard:
3) feedback updates:
The solution of the quadratic programming problem is completed, the control amount in available control time domain:
Further, in institute's step 6:
According to first element of the control sequence that linear time-varying model predictive controller (MPC) obtains, vehicle is acted on,
Since the model is established based on linearity error, obtained control amount also needs the control amount plus last moment,
Obtain the practical control amount of subsequent time.Namely:
Beneficial effects of the present invention:
1, the present invention relative to tradition assume side drift angle less than 5 ° to only consider the range of linearity in it is assumed that this method can
To improve the scope of application of controller, so that control precision is improved, high speed, intelligence for increasingly developed intelligent vehicle
With positive facilitation.
2 and present invention employs Model Predictive Control Algorithm, relative to traditional PID control, there is preferably control essence
Degree, can improve the stability of vehicle in galloping.
3, present invention employs the Model Predictive Controls based on linear time-varying, for Nonlinear Model Predictive Control
With higher real-time, the especially nowadays raising of computer hardware performance, so that it is quicker to calculate the speed solved.
Detailed description of the invention
Fig. 1 is vehicle single track model schematic;
Fig. 2 is based on the Subspace partition schematic diagram for improving G-K clustering algorithm;
Fig. 3 is control flow chart.
Specific embodiment
It is the present invention is based on a kind of improved G-K clustering algorithm, relationship between side force of tire and side drift angle is nonlinear
Partial segments linearisation, then carries out the crosswise joint of vehicle by linear time-varying model predictive control algorithm.Control flow is such as
Shown in Fig. 3, the specific technical solution of use is as follows:
A kind of control method of the unmanned vehicle transverse movement based on GK clustering algorithm model prediction is that control object vehicle is real
When steering wheel angle is provided, thus realize to control target transverse movement control, comprising the following steps:
Step 1, vehicle's current condition, such as systemic velocity, course angle, yaw angle speed are obtained in real time using vehicle sensors
Degree, vehicle changing coordinates and slip angle of tire and vehicle speed information.
Step 2, environment surrounding automobile information is acquired using industrial camera and millimetre-wave radar, determines travelable region, from
And expected path is planned in real time.
Step 3, single track whole vehicle model is established: will be between tire cornering power and side drift angle in relationship using GK clustering algorithm
Non-linear partial piece-wise linearization obtains the piecewise function of affine rear relationship between tire cornering power and slip angle of tire, utilizes
This piecewise function, which is brought into vehicle dynamic model, obtains single track whole vehicle model expression formula.
Step 4, using Taylor's formula by the state space of this non-linear linear error model of single-track vehicle model conversion
Equation, wherein quantity of state takes the vehicle under systemic velocity under vehicle axis system, course angle, yaw velocity, earth coordinates
Position, control measure front wheel angle, and by its sliding-model control.
Step 5, using the Model Predictive Control Algorithm of linear time-varying, the model predictive controller of linear time-varying is established, it will
The vehicle location under systemic velocity, course angle, yaw velocity, earth coordinates under vehicle axis system is as model prediction control
Prediction time domain is calculated according to current state and target trajectory in the input of device processed, output of the front wheel angle as controller
Control sequence in interior tracing point and control time domain, control sequence, establishes objective function, and problem is converted into order to obtain
Quadratic programming problem seeks optimal solution, and using first element of control sequence as the control amount of practical control target.It updates
The state of vehicle repeats the rolling optimization function of above step implementation model prediction.
Step 6, the control amount obtained according to model predictive controller successively controls the steering of target vehicle.
Further, in institute's step 1:
Obtaining the current status information of vehicle in real time is acquired in real time using inertial navigation, and vehicle speed information is by wheel speed sensors
Acquisition obtains in real time.
Further, in institute's step 2:
The method of planning expected path is to acquire environment surrounding automobile information using industrial camera and millimetre-wave radar in real time,
Travelable region is determined, to plan expected path in real time.
Further, in institute's step 3:
Single-track vehicle kinetic model is established, detailed process is as follows:
Using the method for GK cluster by the non-linear partial piecewise affine in tire cornering power F and slip angle of tire α relationship
At several sections of inearized models, identification of Model Parameters includes the segmentation of data subspace, the estimation of each subspace PARAMETERS IN THE LINEAR MODEL and cuts
Change the estimation of face equation coefficient.The test number of slip angle of tire α and lateral force F under all kinds of operating conditions are acquired using multi-dimension force sensor
According to carrying out parameter identification to the piecewise affine models of tire cornering characteristics and illustrate the technical solution of this method by taking front-wheel as an example.
Based on the experimental data that sensor obtains, regression equation: y (k)=f (x (k))+e (k) is constructed, wherein x (k) is regression vector,
It outputs and inputs vector by the history of system to constitute, x (k)=[y (k-1) ... y (k-na),u(k-1)…u(k-na)];E (k) is
The additive noise of known probability Density Distribution, y (k) are actual measurement output signal, and u (k) is actual measurement input signal.
1) piecewise affine data Subspace partition
For input/output signal sequence Z=(x (k), y (k)), k=1 ... T, T > 0, defining mode function: μ (k):
{ 1 ... T } → { 1 ... s }, μ (k) :=i, whenever x (k) ∈ χiInvention, using improved G-K clustering method implementation pattern
The solution of function mu, so that cluster result meets following objective function:
Wherein, m ∈ [1, ∞) characterization clustering fuzzy degree adjustable parameter, represent the overlapping degree between of all categories, take m
=2.d(zj,vi) indicate sample zjWith cluster centre viThe distance between, it determines the shape of cluster, defines this algorithm here and adopt
With adaptive distance metric method:
d2(zj-vi)=(zj-vi)TMi(zj-vi) (2)
Wherein, MiIt is positive definite matrix, by cluster covariance matrix FiIt determines
Data-oriented collection Z, clustering algorithm key step as shown in Fig. 2,
Step 1: the covariance matrix of experiment with computing data set Z, it is initial cluster that farthest two samples are selected from Z
Center calculates subordinated-degree matrixEnable l=1
Step 2: cluster covariance matrix F is calculatedi, characteristic value and feature vector are extracted, covariance adjustment is carried out
Step 3: M is calculatediWith squared-distance d2
Step 4: subordinated-degree matrix is recalculated
Step 5: judgement | | Ul-Ul-1| | > ε, if it is return step two, if not then continuing step 6
Step 6: retain this cluster centreWith subordinated-degree matrixClustering validation is carried out, that is, judges c
≤cmax, if it is leaping to step 9, if not then continuing step 7
Step 7: c=c+1, according to subordinated-degree matrixFind out a sample zk dissimilar with each subset
Step 8: according to new cluster initial center, corresponding new initial subordinated-degree matrix is calculated
Step 9: determine that best cluster number carries out data set division according to the degree of membership of every group of affiliated subset of sample.
2) piecewise affine submodel parameter identification
After obtaining data Subspace partition based on clustering method, submodel parameter identification problem reduction is asked for linear optimization
Topic, the present invention are calculated using weighted least-squares method.According to subclass data set ZiInputoutput data, with practical system
Output bias quadratic sum between system and submodel constitutes criterion function as follows, and the minimum of the criterion corresponds to submodel ginseng
Number vector θiEstimated value.
3) diverter surface equation coefficient is estimated
Due to each scopeConvex polyhedron there is no lap in addition to public boundary, diverter surface equation coefficient is distinguished
Know the linear partition problem that can be converted into cluster data.The present invention carries out hyperplane equation using the method based on support vector machines
The solution of coefficient, for the convex polyhedron section of submodel scope:Diverter surface equation
It can be described as: hji={ wix(k)+bi=0 }, compromise considers maximum class interval and minimum error sample, establishes Generalized optimal point
Class face, the linear kernel function that selection meets Mercer condition constitutes supporting vector, to obtain system hyperplane equation coefficient h.
Since the subspace number c of Piecewise affine systems is determined by the extreme value of optimizing index, estimate it is possible that crossing
The case where, the method that proposed adoption of the present invention verifies afterwards judges the similitude of each submodel, realizes similar submodel
Merging and the identification again of model parameter.
It is as follows that the cluster affine relation that affine tire cornering power and side drift angle obtain is clustered by GK clustering algorithm:
Wherein,For the parameter vector of affine submodel,It is convex for affine submodel scope
Face body section, and assume that each convex body does not have overlapping interval in addition to public boundary, i.e.,Work as system
State reach borderline region when switch.Side drift angle α can be calculated according to following relational expression in control process:
WhereinWithBeing is longitudinal velocity under vehicle axis system, side velocity and yaw rate respectively,
This tittle is all that the quantity of state of auto model can be calculated in real time in control process, δfIt is the front wheel angle of vehicle, is
The control amount of auto model is calculated in real time in control process, and a, b are distance of the vehicle centroid to antero posterior axis respectively.
It brings lateral force into single-track vehicle model (such as Fig. 1), obtains mathematical model:
Wherein, Fcf,FcrLateral force suffered by the forward and backward tire of vehicle respectively is represented via the segmentation of above-mentioned clustering algorithm
Come;Flf, FlrRespectively longitudinal force suffered by the forward and backward tire of vehicle, it is related with the longitudinal rigidity of tire, slip rate;δfBefore vehicle
Take turns corner;For Vehicular yaw angle;A is distance of the front axle to mass center, and b is distance of the rear axle to mass center, IZIt is vehicle around z-axis
Rotary inertia.
Further, in institute's step 4:
It is nonlinear model for single-track vehicle kinetic model obtained above, it is therefore desirable to which linearisation could use state
Space equation expression, wherein quantity of state takes systemic velocity, course angle, yaw velocity, earth coordinates under vehicle axis system
Under vehicle location, control measure front wheel angle.The specific method is as follows:
Single-track vehicle kinetic model is write as:
Wherein, ξ is quantity of state, and μ is control amount, and by it, Taylor expansion, reservation single order item ignore high-order at any point
, it obtains:
In addition also have at this arbitrary point:
Formula (9) and (10) are subtracted each other to obtain:
In formula,
Sliding-model control is carried out with the method for single order difference coefficient to this state space equation, obtains Discrete Linear model tormulation
Formula:
Wherein, A (k)=I+TA (t), B (k)=TB (t).
Further, in institute's step 5:
Auto model is converted to linearity error model and brings Model Predictive Control by the characteristics of according to Model Predictive Control
One group of control sequence is calculated by the Optimization Solution function inside model predictive controller in device, and by the of control sequence
One amount acts on controlled device, then repeating scrolling above step, is specifically divided into the description, excellent of state variable and output variable
Change and solves, feeds back calculating.Detailed process is as follows:
1) description of state variable and output variable
In order to solve convenience, above formula (15) are merged and are write as:
Obtain new state-space expression:
Wherein,
η (k | t) it is the output predicted in time domain.
Choosing prediction time domain is 10, and control time domain is 3, and the output quantity in system prediction time domain is expressed with a matrix type
Are as follows:Control time domain in output beThe output in system future can be write as: Y
(k+10 | k)=Sxx(k)+SuΔ U (k), wherein
2) Optimization Solution:
Cost function is set are as follows:
Wherein first item reflects the tracking ability to reference locus, and Section 2 reflects the steady change of control process,
Q, R is this two weight matrix, ρ ε respectively2Be in order to avoid in optimization process occur without solution situation and it is increased relaxation because
Son.Be converted to the quadratic programming problem of standard:
3) feedback updates:
The solution of the quadratic programming problem is completed, the control amount in available control time domain:
Further, in institute's step 6:
According to first element of the control sequence that linear time-varying model predictive controller (MPC) obtains, vehicle is acted on,
Since the model is established based on linearity error, obtained control amount also needs the control amount plus last moment,
Obtain the practical control amount of subsequent time.Namely:
The series of detailed descriptions listed above only for feasible embodiment of the invention specifically
Protection scope bright, that they are not intended to limit the invention, it is all without departing from equivalent implementations made by technical spirit of the present invention
Or change should all be included in the protection scope of the present invention.
Claims (8)
1. a kind of control method of the unmanned vehicle transverse movement based on GK clustering algorithm model prediction, which is characterized in that including such as
Lower step:
Step 1, vehicle's current condition, such as systemic velocity, course angle, yaw velocity, vehicle are obtained in real time using vehicle sensors
Changing coordinates and slip angle of tire and vehicle speed information;
Step 2, environment surrounding automobile information is acquired using industrial camera and millimetre-wave radar, determines travelable region, advises in real time
Draw expected path;
Step 3, single track whole vehicle model is established: will be non-thread in relationship between tire cornering power and side drift angle using GK clustering algorithm
Property partial segments linearisation, obtain between tire cornering power and slip angle of tire it is affine after relationship piecewise function, utilize this point
Section function, which is brought into vehicle dynamic model, obtains single track whole vehicle model;
Step 4, linearity error mould is converted into using the non-linear Three Degree Of Freedom single track whole vehicle model that Taylor's formula obtains step 3
The state space equation of type, wherein quantity of state takes systemic velocity, course angle, yaw velocity, the earth seat under vehicle axis system
The lower vehicle location of mark system, control measurement front wheel angle, and by its sliding-model control;
Step 5, using the Model Predictive Control Algorithm of linear time-varying, the model predictive controller of linear time-varying is established, by vehicle
The vehicle location under systemic velocity, course angle, yaw velocity, earth coordinates under coordinate system is as model predictive controller
Input, output of the front wheel angle as controller is calculated in prediction time domain according to current state and target trajectory
Control sequence in tracing point and control time domain, control sequence, establishes objective function in order to obtain, and problem is converted into secondary
Planning problem seeks optimal solution, and using first element of control sequence as the control amount of practical control target.More new vehicle
State, repeat above step implementation model prediction rolling optimization function;
Step 6, the control amount obtained according to model predictive controller successively controls the steering of target vehicle.
2. a kind of controlling party of unmanned vehicle transverse movement based on GK clustering algorithm model prediction according to claim 1
Method, which is characterized in that in the step 1, obtaining the current status information of vehicle in real time is acquired in real time using inertial navigation, vehicle
Fast information is acquired acquisition by wheel speed sensors in real time.
3. a kind of controlling party of unmanned vehicle transverse movement based on GK clustering algorithm model prediction according to claim 1
Method, which is characterized in that the single track whole vehicle model expression formula of the step 3 are as follows:
Wherein, Fcf,FcrLateral force suffered by the forward and backward tire of vehicle respectively shows via the segmentation of above-mentioned clustering algorithm;
Flf, FlrRespectively longitudinal force suffered by the forward and backward tire of vehicle, it is related with the longitudinal rigidity of tire, slip rate;δfTo be rotated before vehicle
Angle;For Vehicular yaw angle;A is distance of the front axle to mass center, and b is distance of the rear axle to mass center, IZIt is vehicle around the rotation of z-axis
Inertia.
4. a kind of controlling party of unmanned vehicle transverse movement based on GK clustering algorithm model prediction according to claim 3
Method, which is characterized in that the method for building up of the single track whole vehicle model is as follows:
Using GK cluster method by tire cornering power F with the non-linear partial piecewise affine in slip angle of tire α relationship at several
Section inearized model, identification of Model Parameters include the segmentation of data subspace, the estimation of each subspace PARAMETERS IN THE LINEAR MODEL and diverter surface
Equation coefficient estimation;The test data for acquiring slip angle of tire α and lateral force F under all kinds of operating conditions, divides tire cornering characteristics
Section affine model carries out parameter identification, based on the experimental data that sensor obtains, constructs regression equation: y (k)=f (x (k))+e
(k), wherein x (k) is regression vector, outputs and inputs vector by the history of system and constitutes, x (k)=[y (k-1) ... y (k-
na),u(k-1)…u(k-na)];E (k) is the additive noise of known probability Density Distribution, and y (k) is actual measurement output signal, u (k)
To survey input signal;
1) piecewise affine data Subspace partition
For input/output signal sequence Z=(x (k), y (k)), k=1 ... T, T > 0, defining mode function: μ (k): { 1 ...
T } → { 1 ... s }, μ (k) :=i, whenever x (k) ∈ χiInvention, using improved G-K clustering method implementation pattern function mu
Solution so that cluster result meets following objective function:
Wherein, m ∈ [1, ∞) characterization clustering fuzzy degree adjustable parameter, represent the overlapping degree between of all categories, take m=2.
d(zj,vi) indicate sample zjWith cluster centre viThe distance between, the shape of cluster is determined, using adaptive distance metric
Method:
d2(zj-vi)=(zj-vi)TMi(zj-vi)
Wherein, MiIt is positive definite matrix, by cluster covariance matrix FiIt determines
2) piecewise affine submodel parameter identification
After obtaining data Subspace partition based on clustering method, submodel parameter identification problem reduction is linear optimization problem,
It is calculated using weighted least-squares method;According to subclass data set ZiInputoutput data, with real system and submodule
Output bias quadratic sum between type constitutes criterion function as follows, and the minimum of the criterion corresponds to submodel parameter vector θi
Estimated value;
3) diverter surface equation coefficient is estimated
Due to each scopeConvex polyhedron there is no lap in addition to public boundary, the identification of diverter surface equation coefficient can
It is converted into the linear partition problem of cluster data;Asking for hyperplane equation coefficient is carried out using the method based on support vector machines
Solution, for the convex polyhedron section of submodel scope:Diverter surface equation can be retouched
It states are as follows: hji={ wix(k)+bi=0 }, compromise considers maximum class interval and minimum error sample, establishes Generalized optimal classification
Face, the linear kernel function that selection meets Mercer condition constitutes supporting vector, to obtain system hyperplane equation coefficient h;
The similitude of each submodel is judged using the method verified afterwards again, realizes merging and the mould of similar submodel
The identification again of shape parameter;
It is as follows that the cluster affine relation that affine tire cornering power and side drift angle obtain is clustered by GK clustering algorithm:
Wherein,For the parameter vector of affine submodel,It is convex for affine submodel scope
Face body section, and assume that each convex body does not have overlapping interval in addition to public boundary, i.e.,
It switches when the state of system reaches borderline region.Side drift angle α can be calculated according to following relational expression in control process
It obtains:
WhereinWithBeing is longitudinal velocity under vehicle axis system, side velocity and yaw rate respectively, these
Amount is all that the quantity of state of auto model can be calculated in real time in control process, δfIt is the front wheel angle of vehicle, is vehicle
The control amount of model is calculated in real time in control process, and a, b are distance of the vehicle centroid to antero posterior axis respectively.By side
Single track whole vehicle model is brought into power, obtains mathematical model:
Wherein, Fcf,FcrLateral force suffered by the forward and backward tire of vehicle respectively shows via the segmentation of above-mentioned clustering algorithm;
Flf, FlrRespectively longitudinal force suffered by the forward and backward tire of vehicle, it is related with the longitudinal rigidity of tire, slip rate;δfTo be rotated before vehicle
Angle;For Vehicular yaw angle;A is distance of the front axle to mass center, and b is distance of the rear axle to mass center, IZIt is vehicle around the rotation of z-axis
Inertia.
5. a kind of controlling party of unmanned vehicle transverse movement based on GK clustering algorithm model prediction according to claim 4
Method, which is characterized in that the step of clustering algorithm is as follows:
Step 1: the covariance matrix of experiment with computing data set Z, it is initial cluster centre that farthest two samples are selected from Z,
Calculate subordinated-degree matrixEnable l=1;
Step 2: cluster covariance matrix F is calculatedi, characteristic value and feature vector are extracted, covariance adjustment is carried out;
Step 3: M is calculatediWith squared-distance d2;
Step 4: subordinated-degree matrix is recalculated
Step 5: judgement | | Ul-Ul-1| | whether > ε is true, if it is return step two, if not then continuing step 6;
Step 6: retain this cluster centreWith subordinated-degree matrixCarry out Clustering validation, that is, judge c≤
cmax, if it is leaping to step 9, if not then continuing step 7;
Step 7: variable c=c+1 is calculated, according to subordinated-degree matrixFind out a sample z dissimilar with each subsetk;
Step 8: according to new cluster initial center, corresponding new initial subordinated-degree matrix is calculated;
Step 9: determine that best cluster number carries out data set division according to the degree of membership of every group of affiliated subset of sample.
6. a kind of controlling party of unmanned vehicle transverse movement based on GK clustering algorithm model prediction according to claim 1
Method, which is characterized in that the specific implementation of the step 4 includes the following:
Single track whole vehicle model is write as:
Wherein, ξ is quantity of state, and μ is control amount, and by it, Taylor expansion, reservation single order item ignore higher order term at any point,
It obtains:
In addition also have at this arbitrary point:
By formulaWithSubtract each other to obtain:
In formula,
Sliding-model control is carried out with the method for single order difference coefficient to this state space equation, obtains Discrete Linear model expression:
Wherein, A (k)=I+TA (t), B (k)=TB (t).
7. a kind of controlling party of unmanned vehicle transverse movement based on GK clustering algorithm model prediction according to claim 6
Method, which is characterized in that the specific implementation of the step 5 includes the following:
1) description of state variable and output variable
It, will in order to solve convenientlyMerging is write as:
Obtain new state-space expression:
Wherein,
It is the output predicted in time domain;
Choosing prediction time domain is 10, and control time domain is 3, and the output quantity in system prediction time domain is expressed with a matrix type are as follows:Control time domain in output beThe output in system future can be write as: Y (k+
10 | k)=Sxx(k)+SuΔ U (k), wherein
2) Optimization Solution:
Cost function is set are as follows:
Wherein first item reflects the tracking ability to reference locus on the right of equation, and Section 2 reflects stablizing for control process and becomes
Change, Q, R are this two weight matrix, ρ ε respectively2It is in order to avoid occurring without solution situation and increased relaxation in optimization process
The factor;
Be converted to the quadratic programming problem of standard:
3) feedback updates:
The solution for completing the quadratic programming problem obtains the control amount in control time domain:
8. a kind of controlling party of unmanned vehicle transverse movement based on GK clustering algorithm model prediction according to claim 7
Method, which is characterized in that in the step 6, control the practical control amount of the steering of target vehicle are as follows: obtained by model predictive controller
To control amount obtained plus the control amount of last moment, i.e., are as follows:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910052382.5A CN109885883B (en) | 2019-01-21 | 2019-01-21 | Unmanned vehicle transverse motion control method based on GK clustering algorithm model prediction |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910052382.5A CN109885883B (en) | 2019-01-21 | 2019-01-21 | Unmanned vehicle transverse motion control method based on GK clustering algorithm model prediction |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109885883A true CN109885883A (en) | 2019-06-14 |
CN109885883B CN109885883B (en) | 2023-04-18 |
Family
ID=66926363
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910052382.5A Active CN109885883B (en) | 2019-01-21 | 2019-01-21 | Unmanned vehicle transverse motion control method based on GK clustering algorithm model prediction |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109885883B (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110321605A (en) * | 2019-06-19 | 2019-10-11 | 中汽研(天津)汽车工程研究院有限公司 | A kind of human-computer interaction coordination control strategy based on Multiple Velocity Model PREDICTIVE CONTROL |
CN110321938A (en) * | 2019-06-20 | 2019-10-11 | 西北工业大学 | A kind of state space construction method and device of Intelligent unattended cluster |
CN110459272A (en) * | 2019-08-19 | 2019-11-15 | 中南大学 | One kind being based on power matched aluminium electroloysis molten salt system field of force approximating method |
CN111216738A (en) * | 2020-02-19 | 2020-06-02 | 北京百度网讯科技有限公司 | Vehicle control method and device, electronic equipment and vehicle |
CN111338346A (en) * | 2020-03-05 | 2020-06-26 | 中国第一汽车股份有限公司 | Automatic driving control method and device, vehicle and storage medium |
CN111443611A (en) * | 2020-05-25 | 2020-07-24 | 上海大学 | Multi-model switching-based high-speed unmanned ship speed controller determination method and system |
CN111580391A (en) * | 2020-05-29 | 2020-08-25 | 中车青岛四方车辆研究所有限公司 | Motor train unit traction torque control method based on model prediction |
CN111679575A (en) * | 2020-05-14 | 2020-09-18 | 江苏大学 | Intelligent automobile trajectory tracking controller based on robust model predictive control and construction method thereof |
CN112182892A (en) * | 2020-09-30 | 2021-01-05 | 西安空间无线电技术研究所 | Power amplifier behavior modeling method and device based on piecewise affine |
CN112937587A (en) * | 2021-02-26 | 2021-06-11 | 南京经纬达汽车科技有限公司 | Road feel simulation method based on K-Medoids and classification regression tree |
CN113110045A (en) * | 2021-03-31 | 2021-07-13 | 同济大学 | Model prediction control real-time optimization parallel computing method based on computational graph |
CN113296515A (en) * | 2021-05-25 | 2021-08-24 | 北京理工大学 | Explicit model prediction path tracking method for double-independent electrically-driven vehicle |
CN113341994A (en) * | 2021-06-23 | 2021-09-03 | 江苏大学 | Intelligent automobile path tracking control method based on optimal control of piecewise affine system |
CN113468698A (en) * | 2021-06-30 | 2021-10-01 | 上海理工大学 | Traffic flow model parameter calibration method facing intersection traffic trajectory discreteness |
CN114506388A (en) * | 2021-12-27 | 2022-05-17 | 广东机电职业技术学院 | Vehicle steering control method and device, computer equipment and storage medium |
CN115534925A (en) * | 2022-12-01 | 2022-12-30 | 广汽埃安新能源汽车股份有限公司 | Vehicle control method, device, equipment and computer readable medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107561942A (en) * | 2017-09-12 | 2018-01-09 | 重庆邮电大学 | Intelligent vehicle track following model predictive control method based on model compensation |
CN108597058A (en) * | 2017-12-11 | 2018-09-28 | 江苏大学 | Distributed-driving electric automobile state based on pseudo- measurement information cascades method of estimation |
CN108646756A (en) * | 2018-07-05 | 2018-10-12 | 合肥工业大学 | Intelligent automobile crosswise joint method and system based on piecewise affine fuzzy sliding mode |
CN108717268A (en) * | 2018-06-22 | 2018-10-30 | 南京航空航天大学 | Automatic Pilot minimum time maneuver control system and its control method based on optimum control and safe distance |
CN109017778A (en) * | 2018-07-31 | 2018-12-18 | 大连民族大学 | The expected path active steering control method of four motorized wheels vehicle |
-
2019
- 2019-01-21 CN CN201910052382.5A patent/CN109885883B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107561942A (en) * | 2017-09-12 | 2018-01-09 | 重庆邮电大学 | Intelligent vehicle track following model predictive control method based on model compensation |
CN108597058A (en) * | 2017-12-11 | 2018-09-28 | 江苏大学 | Distributed-driving electric automobile state based on pseudo- measurement information cascades method of estimation |
CN108717268A (en) * | 2018-06-22 | 2018-10-30 | 南京航空航天大学 | Automatic Pilot minimum time maneuver control system and its control method based on optimum control and safe distance |
CN108646756A (en) * | 2018-07-05 | 2018-10-12 | 合肥工业大学 | Intelligent automobile crosswise joint method and system based on piecewise affine fuzzy sliding mode |
CN109017778A (en) * | 2018-07-31 | 2018-12-18 | 大连民族大学 | The expected path active steering control method of four motorized wheels vehicle |
Non-Patent Citations (2)
Title |
---|
王艺等: "基于模型预测控制的智能车辆路径跟踪控制器设计", 《汽车技术》 * |
邓涛等: "基于MPC与EKF算法的电动汽车状态估计与横向稳定性控制", 《汽车安全与节能学报》 * |
Cited By (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110321605A (en) * | 2019-06-19 | 2019-10-11 | 中汽研(天津)汽车工程研究院有限公司 | A kind of human-computer interaction coordination control strategy based on Multiple Velocity Model PREDICTIVE CONTROL |
CN110321938A (en) * | 2019-06-20 | 2019-10-11 | 西北工业大学 | A kind of state space construction method and device of Intelligent unattended cluster |
CN110321938B (en) * | 2019-06-20 | 2022-10-11 | 西北工业大学 | State space construction method and device of intelligent unmanned cluster |
CN110459272A (en) * | 2019-08-19 | 2019-11-15 | 中南大学 | One kind being based on power matched aluminium electroloysis molten salt system field of force approximating method |
CN110459272B (en) * | 2019-08-19 | 2023-03-31 | 中南大学 | Force matching-based force field fitting method for aluminum electrolysis molten salt system |
CN111216738B (en) * | 2020-02-19 | 2022-02-01 | 阿波罗智能技术(北京)有限公司 | Control method and device for vehicle in automatic driving, electronic equipment and vehicle |
CN111216738A (en) * | 2020-02-19 | 2020-06-02 | 北京百度网讯科技有限公司 | Vehicle control method and device, electronic equipment and vehicle |
CN111338346A (en) * | 2020-03-05 | 2020-06-26 | 中国第一汽车股份有限公司 | Automatic driving control method and device, vehicle and storage medium |
CN111679575A (en) * | 2020-05-14 | 2020-09-18 | 江苏大学 | Intelligent automobile trajectory tracking controller based on robust model predictive control and construction method thereof |
CN111443611A (en) * | 2020-05-25 | 2020-07-24 | 上海大学 | Multi-model switching-based high-speed unmanned ship speed controller determination method and system |
CN111443611B (en) * | 2020-05-25 | 2021-05-04 | 上海大学 | Multi-model switching-based high-speed unmanned ship speed controller determination method and system |
CN111580391A (en) * | 2020-05-29 | 2020-08-25 | 中车青岛四方车辆研究所有限公司 | Motor train unit traction torque control method based on model prediction |
CN112182892A (en) * | 2020-09-30 | 2021-01-05 | 西安空间无线电技术研究所 | Power amplifier behavior modeling method and device based on piecewise affine |
CN112182892B (en) * | 2020-09-30 | 2024-01-02 | 西安空间无线电技术研究所 | Power amplifier behavior modeling method and device based on piecewise affine |
CN112937587A (en) * | 2021-02-26 | 2021-06-11 | 南京经纬达汽车科技有限公司 | Road feel simulation method based on K-Medoids and classification regression tree |
CN113110045A (en) * | 2021-03-31 | 2021-07-13 | 同济大学 | Model prediction control real-time optimization parallel computing method based on computational graph |
CN113110045B (en) * | 2021-03-31 | 2022-10-25 | 同济大学 | Model prediction control real-time optimization parallel computing method based on computation graph |
CN113296515A (en) * | 2021-05-25 | 2021-08-24 | 北京理工大学 | Explicit model prediction path tracking method for double-independent electrically-driven vehicle |
CN113341994A (en) * | 2021-06-23 | 2021-09-03 | 江苏大学 | Intelligent automobile path tracking control method based on optimal control of piecewise affine system |
CN113468698A (en) * | 2021-06-30 | 2021-10-01 | 上海理工大学 | Traffic flow model parameter calibration method facing intersection traffic trajectory discreteness |
CN114506388A (en) * | 2021-12-27 | 2022-05-17 | 广东机电职业技术学院 | Vehicle steering control method and device, computer equipment and storage medium |
CN115534925A (en) * | 2022-12-01 | 2022-12-30 | 广汽埃安新能源汽车股份有限公司 | Vehicle control method, device, equipment and computer readable medium |
Also Published As
Publication number | Publication date |
---|---|
CN109885883B (en) | 2023-04-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109885883A (en) | A kind of control method of the unmanned vehicle transverse movement based on GK clustering algorithm model prediction | |
CN112356830B (en) | Intelligent parking method based on model reinforcement learning | |
CN108698595B (en) | For controlling the method for vehicle movement and the control system of vehicle | |
Bouton et al. | Belief state planning for autonomously navigating urban intersections | |
CN109597864B (en) | Method and system for real-time positioning and map construction of ellipsoid boundary Kalman filtering | |
CN109415089A (en) | The controller and method and non-transitory computer-readable memory of control vehicle | |
CN107357168B (en) | Unmanned vehicle obstacle avoidance method based on opportunity constraint model predictive control | |
Qin et al. | An improved real-time slip model identification method for autonomous tracked vehicles using forward trajectory prediction compensation | |
Kapania | Trajectory planning and control for an autonomous race vehicle | |
Wu et al. | Route planning and tracking control of an intelligent automatic unmanned transportation system based on dynamic nonlinear model predictive control | |
Akhtar et al. | The interacting multiple model smooth variable structure filter for trajectory prediction | |
CN117222915A (en) | System and method for tracking an expanded state of a moving object using a composite measurement model | |
Dou et al. | Path planning and tracking for autonomous mining articulated vehicles | |
CN116560223A (en) | Intelligent vehicle dynamics model, ILQR control algorithm and track tracking controller based on physical information neural network | |
Zhu et al. | A gaussian process model for opponent prediction in autonomous racing | |
Xiong et al. | Surrounding vehicle trajectory prediction and dynamic speed planning for autonomous vehicle in cut-in scenarios | |
Fan et al. | Spatiotemporal path tracking via deep reinforcement learning of robot for manufacturing internal logistics | |
Wang et al. | Extraction of preview elevation information based on terrain mapping and trajectory prediction in real-time | |
CN112829744A (en) | Vehicle long time domain track prediction method based on longitudinal and transverse coupling | |
Xu et al. | Actor–critic reinforcement learning for autonomous control of unmanned ground vehicles | |
CN115343950A (en) | Vehicle path tracking control method and control system suitable for complex road surface | |
Chen et al. | An Intelligent Navigation Control Approach for Autonomous Unmanned Vehicles via Deep Learning-Enhanced Visual SLAM Framework | |
CN113296515A (en) | Explicit model prediction path tracking method for double-independent electrically-driven vehicle | |
Yang et al. | SLAM self-cruise vehicle based on ROS platform | |
Li et al. | Navigation Simulation of Autonomous Mobile Robot Based on TEB Path Planner |
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 |