CN115027499A - Vehicle automatic driving prediction control method based on dynamic neural network Hammerstein model - Google Patents
Vehicle automatic driving prediction control method based on dynamic neural network Hammerstein model Download PDFInfo
- Publication number
- CN115027499A CN115027499A CN202210722735.XA CN202210722735A CN115027499A CN 115027499 A CN115027499 A CN 115027499A CN 202210722735 A CN202210722735 A CN 202210722735A CN 115027499 A CN115027499 A CN 115027499A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- hammerstein model
- neural network
- state quantity
- model
- 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.)
- Pending
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
- B60W60/00—Drive control systems specially adapted for autonomous road vehicles
- B60W60/001—Planning or execution of driving tasks
-
- 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
-
- 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
Abstract
The invention belongs to the technical field of vehicle automatic driving prediction control, and discloses a vehicle automatic driving prediction control method based on a dynamic neural network Hammerstein model, which comprises the following steps: s1, constructing a Hammerstein model based on a dynamic neural network, wherein the Hammerstein model comprises a static nonlinear module and a dynamic linear module which are connected in series, and the static nonlinear module comprises a multilayer feedforward neural network and a Map table for reflecting a relation curve of a tire cornering angle and a tire lateral force; and S2, designing a prediction controller based on a Hammerstein model, calculating a control quantity of the vehicle according to the expected state quantity and an actual state quantity fed back and output by a vehicle system in real time by the prediction controller, and inputting the control quantity into a vehicle control system. In conclusion, the Hammerstein model has the advantages of high modeling accuracy and rapidity, so that the nonlinear modeling accuracy of the vehicle is effectively guaranteed, and the modeling complexity is reduced.
Description
Technical Field
The invention belongs to the technical field of vehicle automatic driving prediction control, and particularly relates to a vehicle automatic driving prediction control method based on a dynamic neural network Hammerstein model.
Background
The vehicle dynamics model has important significance for realizing the vehicle motion planning based on model predictive control. Tires and steering systems have highly nonlinear characteristics as important subsystems of vehicle systems, and in order to improve vehicle control accuracy, it is necessary to sufficiently consider nonlinear dynamics of vehicles.
At present, vehicle dynamics modeling comprises two types of linear models and nonlinear models in the field of vehicle automatic driving controller design. For specific working conditions, most researches adopt classical linear models such as 2DOF, 3DOF and 5DOF to design a controller. The two-degree-of-freedom vehicle model considers the yaw degree and the lateral degree of freedom, only considers the linear characteristic of a tire, and simplifies the whole vehicle into a single vehicle model. The three-degree-of-freedom vehicle model increases the longitudinal degree of freedom, the mass center slip angle and the yaw angular velocity can change along with the longitudinal velocity, and the three-degree-of-freedom vehicle model is more consistent with an actual vehicle system. In addition, the vehicle model further considers a suspension and a steering system, and simultaneously, the analysis of the tire model is expanded from a linear region to a non-linear region, such as 12DOFs, 17DOFs, 19DOFs and the like, and compared with the linear model, the non-linear model can analyze the motion state of the vehicle under the combined working condition.
In summary, vehicle modeling accuracy requirements are increasing for vehicle autonomous driving control. The linear model is difficult to adapt to the design requirement of the controller under the complex working condition, meanwhile, the complexity of the nonlinear vehicle coupling model provides higher challenge for the design of the controller, and the transverse and longitudinal coupling characteristics of the vehicle are difficult to be comprehensively and accurately analyzed. Therefore, aiming at the problem that the transverse and longitudinal coupling complex nonlinear system of the vehicle automatic driving controller is difficult to model, the applicant provides a vehicle automatic driving control method which is based on a dynamic neural network, constructs a Hammerstein transverse and longitudinal coupling model and carries out predictive control on the basis of the transverse and longitudinal coupling model.
Disclosure of Invention
In view of the above, the present invention aims to provide a vehicle automatic driving prediction control method based on a dynamic neural network Hammerstein model, and in particular, the present invention describes the tire cornering nonlinearity through a Map table, so as to reduce the tire cornering nonlinearity modeling difficulty and effectively reduce the computation complexity.
In order to achieve the above object, the present invention provides the following technical solutions
A vehicle automatic driving prediction control method based on a dynamic neural network Hammerstein model comprises the following steps:
s1, constructing a Hammerstein model based on a dynamic neural network, wherein the Hammerstein model comprises a static nonlinear module and a dynamic linear module which are connected in series, and the static nonlinear module comprises a multilayer feedforward neural network and a Map table for reflecting a relation curve of a tire side deflection angle and a tire side force;
and S2, designing a prediction controller based on a Hammerstein model, calculating a control quantity of the vehicle according to the expected state quantity and an actual state quantity fed back and output by a vehicle system in real time by the prediction controller, and inputting the control quantity into a vehicle control system.
Preferably, the dynamic linear module is a linear state space equation, and the equation expression is:
in the formula (I), the compound is shown in the specification,for the intermediate variables output by the static nonlinear module, A, B is the constant matrix to be identified, v x (k)、v y (k)、Is the state quantity at time k, v x (k+1)、v y (k+1)、Is the state quantity at the time k + 1.
Preferably, the intermediate variable output by the static nonlinear moduleThe method comprises the following steps:
Obtaining an initial variable u through the multi-layer feedforward neural network 1 (k) And transition variablesCorresponding intermediate variable
Wherein the initial variable u 1 (k)、u 2 (k) Resultant force F of longitudinal force of vehicle on X axis at k time x (k) Angle delta with front wheel f (k) Transition variableIs the slip angle alpha to the tire in Map table f Corresponding tire side force F y 。
Preferably, the initial variable u is obtained through the Map table 2 (k) Corresponding transition variableThe method comprises the following steps:
the front wheel steering angle and tire slip angle are converted as follows:l f obtaining the value of the tire slip angle for the distance from the mass center of the vehicle to the front axle through the current vehicle state information and the front wheel steering angle;
setting a range [ -a, a ] of a tire slip angle;
dividing the range of the tire slip angle into N intervals, numbering each interval in sequence, and enabling the tire slip angle alpha to be within the range of the slip angle [ -a, a [ ]]Internal time, interval numberIs composed ofStoring the corresponding tire lateral force F y 。
Preferably, the Hammerstein model is combined with a soft constraint objective function in the predictive controller, the soft constraint objective function is used for correspondingly constraining the expected state quantity, the actual state quantity and the controlled quantity increment, and an expression of the soft constraint objective function is as follows:
where k denotes the current time of the model, k + i denotes the i-th predicted time of the model, and x j (. h) is the j-th predicted state quantity, r, predicted by the Hammerstein model j (. is) the jth reference state quantity, Δ u j (. cndot.) is the jth controlled variable,. DELTA.u (n + i) ═ u (n + i) -u (n + i-1), and the controlled variable u 1 、u 2 Resultant forces F being longitudinal forces on the X-axis, respectively x And front wheel angle delta f ,N p To predict the time domain, N c For controlling the time domain, Q is a prediction control output weight coefficient matrix, and S is a control increment weight coefficient matrix.
Preferably, the state quantities in the desired state quantity and the actual state quantity each include a longitudinal speed v x Transverse velocity v y And yaw rate
Preferably, the control amount includes a front wheel turning angle δ f And a longitudinal force F x 。
Compared with the prior art, the invention has the following beneficial effects:
(1) high accuracy: the neural network has nonlinear mapping capability and richly-variable network structures, and the multilayer feedforward neural network and the Map table are combined into a static nonlinear module in the Hammerstein model, so that the nonlinear coupling characteristic of a vehicle system can be completely reflected.
(2) And (3) reducing modeling complexity: in the invention, the static nonlinear module and the dynamic linear module of the Hammerstein model are separately constructed, so that compared with the traditional nonlinear dynamic neural network, the modeling complexity of the model can be effectively reduced.
Drawings
FIG. 1 is a schematic diagram of a predictive control system of the present invention, in which MPC is a predictive controller;
FIG. 2 is a structural diagram of a Hammerstein model based on a dynamic neural network in the present invention;
FIG. 3 is a Map table showing a relationship between a tire slip angle and a tire lateral force according to the present invention;
FIG. 4 is a Hammerstein model topology of the dynamic neural network of the present invention;
FIG. 5 is a flow chart of the genetic algorithm of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1 to 5, in the method, a Hammerstein model based on a dynamic neural network is first constructed, then a prediction controller based on the Hammerstein model is designed, and finally the prediction controller is utilized to realize the prediction control of the automatic driving of the vehicle.
S1, a Hammerstein model based on a dynamic neural network:
specifically, the Hammerstein model designed according to fig. 2 includes a static nonlinear module and a dynamic linear module connected in series. Wherein:
a) the static nonlinear module comprises a multilayer feedforward neural network and a Map table for reflecting a relation curve of a tire slip angle and a tire lateral force;
in the Map table, data is read through a hash table look-up method, and the method comprises the following steps:
the front wheel steering angle and tire slip angle are converted as follows:l f obtaining the value of the tire slip angle for the distance from the mass center of the vehicle to the front axle through the current vehicle state information and the front wheel steering angle;
setting a range [ -a, a ] of a tire slip angle;
dividing the range of the tire slip angle into N intervals, numbering each interval in sequence, and enabling the tire slip angle alpha to be within the range of the slip angle [ -a, a [ ]]When inside, the interval number is corresponding toThe data correspondingly stored in the interval is the lateral force F of the tire y 。
b) The dynamic linear module is a linear state space equation:
in the formula (I), the compound is shown in the specification,for the intermediate variables output by the static nonlinear module, A, B is the constant matrix to be identified, v x (k)、v y (k)、Is the state quantity at time k, v x (k+1)、v y (k+1)、Is the state quantity at the time k + 1.
Aiming at the designed Hammerstein model, the Hammerstein model is subjected to parameter training before application, and specifically during training:
providing an initial variable u 1 (k)、u 2 (k) And the state quantity v at the time k x (k)、v y (k)、And with respect to the initial variable u 1 (k)、u 2 (k) And the state quantity v at the time k x (k)、v y (k)、All the parameters can be acquired by an actual vehicle system or automobile dynamics simulation software;
Obtaining an initial variable u through the multi-layer feedforward neural network 1 (k) And transition variablesCorresponding intermediate variable
And acquiring the actual output state quantity of the vehicle at the moment of k +1, and verifying whether the error between the generated state quantity and the actual output state quantity is within an acceptable error range, if so, finishing the training of the Hammerstein model.
Above, the initial variable u 1 (k)、u 2 (k) Respectively, the resultant delta of the longitudinal force of the vehicle on the X axis at the time k f (k) And front wheel corner F x (k) Transition variableAs Map tableNeutral and tire slip angle α f Corresponding tire side force F y . In addition, in the training, the updating of the weight values and the threshold values in the multilayer feedforward neural network and the linear model is realized through a real-time loop learning algorithm.
In fig. 4, the dynamic neural network Hammerstein model topology is composed of Map and 5-layer neural network layers.
Intermediate layer neuronal output Z (3) =ω 2 Z (2) +b 2 ,ω 2 As a weight, b 2 Is a threshold value;
neuronal output Z of hidden layer 2 (4) =ω 3 Z (3) +νx(k),ω 3 V is the weight;
output layer neuron output Z (5) =ω 4 Z (4) ,ω 4 Are the weights.
And (3) real-time cyclic learning algorithm: the layers in the recurrent neural network have temporal connections, so the gradient updates the parameters in the network along the temporal path. Based on the optimization goal as follows,wherein x (k) is the state quantity expected to be output by the vehicle system and the state quantity output by the neural network Hammerstein model so as to update the ith row and j column elements v in the weight v ij For example, the update rules are as follows:
Besides the real-time cyclic learning algorithm, parameters can be updated through a back propagation algorithm, a particle swarm algorithm and the like along with time.
S2, designing a prediction controller based on a Hammerstein model:
and (3) based on the trained Hammerstein model, combining a soft constraint objective function to correspondingly constrain the expected state quantity, the actual state quantity and the controlled quantity increment. Wherein the expression of the soft constraint objective function is:
where k denotes the current time of the model, k + i denotes the i-th predicted time of the model, and x j (. h) is the j-th predicted state quantity, r, predicted by the Hammerstein model j (. h) is the jth reference state quantity, Δ u j (. cndot.) is the jth controlled variable variation, Δ u (n + i) ═ u (n + i) -u (n + i-1), and the controlled variable u 1 、u 2 Resultant forces F being longitudinal forces on the X-axis, respectively x Angle delta with front wheel f ,N p To predict the time domain, N c For controlling the time domain, Q is a prediction control output weight coefficient matrix, and S is a control increment weight coefficient matrix.
S3, the predictive controller is utilized to realize the predictive control of the automatic driving of the vehicle:
the state quantity includes a longitudinal speed v x Transverse velocity v y And yaw rateThe control quantity including the front wheel turning angle delta f And front wheel longitudinal force F x 。
To sum up, with reference to fig. 1, the optimization calculation process of the prediction controller based on the Hammerstein model designed by the present invention is as follows:
acquiring actual state information of a vehicle system by a sensor;
solving the optimal control quantity of the vehicle based on a genetic algorithm;
judging whether the optimal solution is feasible or not; if the vehicle system is feasible, determining the vehicle system as a control quantity and acting on the vehicle system; if not, re-solving.
Specifically, with reference to the flow chart of the genetic algorithm in fig. 5, the solving process is as follows:
initializing a population, wherein the number of the population is N;
evaluating the population, and taking a constraint target function as the fitness of the individual;
the evolutionary computation comprises selection, crossing and variation. Wherein the selection operation is a roulette method, and the probability that the individual i is selected is P i :
Wherein, F i The fitness of an individual i in the population is shown, and k is a coefficient;
performing crossover operations, e.g. using the real number crossover method, chromosome c m And c n The interleaving operation at the j-th bit is as follows:
c mj =c mj (1-d)+c nj d
c nj =c nj (1-d)+c mj d
d is random number in [0,1 ];
mutation operation with gene c ij For example, the following steps are carried out:
wherein, c max And c min Respectively the upper and lower bounds of the gene, G the number of iterations, GEN the maximum number of iterations, r is [0,1]]Internal random number r 1 Is [0,1]]An internal random number;
judging whether the maximum iteration times are met, if so, outputting an optimal scheme; if not, returning to iteration.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (7)
1. The vehicle automatic driving prediction control method based on the dynamic neural network Hammerstein model is characterized by comprising the following steps:
s1, constructing a Hammerstein model based on a dynamic neural network, wherein the Hammerstein model comprises a static nonlinear module and a dynamic linear module which are connected in series, and the static nonlinear module comprises a multilayer feedforward neural network and a Map table for reflecting a relation curve of a tire side deflection angle and a tire side force;
and S2, designing a prediction controller based on a Hammerstein model, calculating a control quantity of the vehicle according to the expected state quantity and an actual state quantity fed back and output by a vehicle system in real time by the prediction controller, and inputting the control quantity into a vehicle control system.
2. The automatic driving prediction control method for the vehicle based on the Hammerstein model of the dynamic neural network as claimed in claim 1, characterized in that: the dynamic linear module is a linear state space equation, and the equation expression is as follows:
3. The method as claimed in claim 2, wherein the intermediate variable output by the static nonlinear module is an intermediate variable output by the Hammerstein modelThe method comprises the following steps:
Obtaining an initial variable u through the multi-layer feedforward neural network 1 (k) And transition variablesCorresponding intermediate variable
4. The method as claimed in claim 3, wherein the initial variable u is obtained from the Map table 2 (k) Corresponding transition variableThen, a Hash search method is adopted, and the method comprises the following steps:
the front wheel steering angle and tire slip angle are converted as follows:l f obtaining the value of the tire slip angle for the distance from the mass center of the vehicle to the front axle through the current vehicle state information and the front wheel steering angle;
setting a range [ -a, a ] of a tire slip angle;
dividing the range of the tire slip angle into N intervals, numbering each interval in sequence, and enabling the tire slip angle alpha to be within the range of the slip angle [ -a, a [ ]]When inside, the interval is numbered asThe tire longitudinal force F processed by the Hash function and stored by the output interval number y 。
5. The automatic vehicle driving prediction control method based on the Hammerstein model of the dynamic neural network as claimed in claim 4, wherein: combining the Hammerstein model with a soft constraint objective function in the predictive controller, the soft constraint objective function being used for correspondingly constraining the expected state quantity, the actual state quantity and the controlled quantity increment, and the expression of the soft constraint objective function being:
where k denotes the current time of the model, k + i denotes the i-th predicted time of the model, and x j (. h) is the j-th predicted state quantity, r, predicted by the Hammerstein model j (. h) is the jth reference state quantity, Δ u j (. cndot.) is the jth controlled variable,. DELTA.u (n + i) ═ u (n + i) -u (n + i-1), and the controlled variable u 1 、u 2 Resultant forces F being longitudinal forces on the X-axis, respectively x Angle delta with front wheel f ,N p To predict the time domain, N c For controlling the time domain, Q is a prediction control output weight coefficient matrix, and S is a control increment weight coefficient matrix.
7. The method for predictive control of vehicle automatic driving based on the Hammerstein model of claim 6, wherein: the control quantity comprises a longitudinal force F x Angle delta with front wheel f 。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210722735.XA CN115027499A (en) | 2022-06-24 | 2022-06-24 | Vehicle automatic driving prediction control method based on dynamic neural network Hammerstein model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210722735.XA CN115027499A (en) | 2022-06-24 | 2022-06-24 | Vehicle automatic driving prediction control method based on dynamic neural network Hammerstein model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115027499A true CN115027499A (en) | 2022-09-09 |
Family
ID=83127788
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210722735.XA Pending CN115027499A (en) | 2022-06-24 | 2022-06-24 | Vehicle automatic driving prediction control method based on dynamic neural network Hammerstein model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115027499A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117291230A (en) * | 2023-11-23 | 2023-12-26 | 湘江实验室 | Hammerstein nonlinear system hybrid identification method with closed state |
-
2022
- 2022-06-24 CN CN202210722735.XA patent/CN115027499A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117291230A (en) * | 2023-11-23 | 2023-12-26 | 湘江实验室 | Hammerstein nonlinear system hybrid identification method with closed state |
CN117291230B (en) * | 2023-11-23 | 2024-03-08 | 湘江实验室 | Hammerstein nonlinear system hybrid identification method with closed state |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112162555B (en) | Vehicle control method based on reinforcement learning control strategy in hybrid vehicle fleet | |
US6721718B2 (en) | System for intelligent control based on soft computing | |
CN112356830B (en) | Intelligent parking method based on model reinforcement learning | |
US6216083B1 (en) | System for intelligent control of an engine based on soft computing | |
Villagra et al. | Flatness-based vehicle steering control strategy with SDRE feedback gains tuned via a sensitivity approach | |
US6463371B1 (en) | System for intelligent control of a vehicle suspension based on soft computing | |
Liu et al. | Multi-kernel online reinforcement learning for path tracking control of intelligent vehicles | |
CN114967676A (en) | Model prediction control trajectory tracking control system and method based on reinforcement learning | |
JP2022544474A (en) | Improved machine learning for technical systems | |
CN111880412B (en) | Reconfigurable robot zero and neural optimal control method based on single evaluation network | |
CN114170789A (en) | Intelligent network connected vehicle lane change decision modeling method based on space-time diagram neural network | |
Zhu et al. | An adaptive path tracking method for autonomous land vehicle based on neural dynamic programming | |
CN115027499A (en) | Vehicle automatic driving prediction control method based on dynamic neural network Hammerstein model | |
CN113777918A (en) | Automobile intelligent drive-by-wire chassis control method of digital twin architecture | |
CN116460860A (en) | Model-based robot offline reinforcement learning control method | |
CN113687654A (en) | Neural network training method and path planning method based on evolutionary algorithm | |
CN114819068A (en) | Hybrid target track prediction method and system | |
CN116560223A (en) | Intelligent vehicle dynamics model, ILQR control algorithm and track tracking controller based on physical information neural network | |
CN113341696A (en) | Intelligent setting method for attitude control parameters of carrier rocket | |
Picotti et al. | Data-driven Tuning of a NMPC Controller for a Virtual Motorcycle through Genetic Algorithm | |
CN108319146B (en) | Discrete particle swarm training-based method for radial basis function neural network | |
CN114967472A (en) | Unmanned aerial vehicle trajectory tracking state compensation depth certainty strategy gradient control method | |
CN115373287A (en) | Adaptive parameter model prediction path tracking control method for articulated tractor | |
CN116300850A (en) | Automatic driving man-like safety self-evolution method and system based on data mechanism fusion | |
Al-Dabooni et al. | Mobile robot control based on hybrid neuro-fuzzy value gradient reinforcement learning |
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 |