CN103057436B - Yawing moment control method of individual driven electromobile based on multi-agent - Google Patents

Abstract

The invention discloses a yawing moment control method of an individual driven electromobile based on a multi-agent. The yawing moment control method includes enabling a yawing moment electric control unit to ba a regulating controller of every motor of the individual driven electromobile to obtain a turning intention and a driving state of a driver in real time through a state observation and measurement module; obtaining torque adjusting motion sequence which is needed by four-wheel of the individual electromobile through a reinforced learning Q-algorithm which is based on a multi-agent; sending the motion sequence to every driven motor controller of the individual driven electromobile through the muli-agent, sending instructions to every driven motor to carry out received motion sequence, generating needed yawing moment and rectifying the posture of the electromobile. The yawing moment control method of the individual driven electromobile based on the multi-agent has the advantages of avoiding energy consumption during actuating process by utilizing driving force control, disposing non-linear elements existing in the system and guaranteeing the stability of turning.

Description

A kind of individual drive electronlmobil yaw moment control method based on multiple agent

Technical field

The invention belongs to electronlmobil yaw moment control technical field, relate to a kind of individual drive electronlmobil yaw moment control method based on multiple agent.

Background technology

Individual drive electronlmobil is as a kind of electronlmobil research direction of novelty, four-wheel all can individual drive by configuration wheel hub motor or hub reduction motor, if the rotating speed when Turning travel just simply by kinematic relation configuration four-wheel makes outside differential in vehicle run, change due to the parameter in kinematics model can not ensure the steering stability of vehicle, must coordinate respectively to take turns the handling stability that torque could improve vehicle by dynamics Controlling.The electronic stability program (ESP) having dropped into application in conventional fuel oil car, by judging Vehicular turn state, applies corresponding braking force when deficiency or oversteer in respective wheel and produces yaw moment.The electronlmobil of four motorized wheels can reach by control four drive motor the loss that same object reduces braking procedure simultaneously.

The object of yaw moment control makes the yaw velocity of vehicle and side slip angle follow the tracks of the yaw velocity and side slip angle expected. in case of emergency, vehicle yaw stability control system is by turning to, differential braking or the input of differential polling power controlling produce necessary correction yaw moment and compensate chaufeur, help chaufeur to stablize vehicle movement.

Traditional PID control, the method for optimal control is mostly adopted at present in direct yaw moment control method, basic ideas are all calculate required yaw torque modification deviation by observation vehicle centroid sideslip angle and yaw velocity deviometer, then configure each propulsive effort of taking turns or braking force.These method Problems existing are excessive to the dependence of system model parameter, and the Correction and Control of vehicle yaw moment is the nonlinear time_varying system being difficult to set up accurate model, and therefore its effect is all undesirable.

Summary of the invention

The problem that the present invention solves is to provide a kind of individual drive electronlmobil yaw moment control method based on multiple agent, the high and strong robustness of the road-holding property of the method.

The present invention is achieved through the following technical solutions:

Based on an individual drive electronlmobil yaw moment control method for multiple agent, comprise following operation:

1) each motor coordination controller using yaw moment electronic control unit as individual drive electronlmobil, yaw moment electronic control unit is turned to by state observation and measurement module Real-time Obtaining chaufeur and is intended to and motoring condition;

2) signal that yaw moment electronic control unit analysis collection is next, obtains vehicle running state, calculates the error of the yaw velocity of reality and demand and the error delta β of the vehicle centroid sideslip angle of reality and demand, then use the intensified learning Q algorithm based on multiple agent to obtain the adjustment of the torque needed for the four-wheel action sequence of individual drive electronlmobil;

3) action sequence is sent to each drive motor controller of individual drive electronlmobil by multiple agent, and in conjunction with the rate request of vehicle, sends instruction and performs the action sequence obtained to each drive motor, the yaw moment needed for generation, the attitude of correction vehicle.

Described state observation and measurement module, between yaw moment electronic control unit and electric machine controller, all transmit the signal needed for control process by CAN.

Described driver intention input comprises:

δ is wheel steering angle, and to turn left for just, signal is from corner displacement sensor;

U is vehicular longitudinal velocity, signal from longitudinal speed sensor,

for yaw velocity is from yaw-rate sensor;

β is side slip angle, estimates that vehicular longitudinal velocity u and vehicle side obtain to speed v by auto model, $\mathrm{\β}=\arctan \frac{v}{u};$

When side slip angle and yaw velocity computing value exceed maxim, get respective maxim as its estimated valve, the maxim of side slip angle and yaw velocity is got as follows:

A _ymaxfor lateral acceleration maxim.

Producing with working direction the yaw velocity turned left just is, to overlook conter clockwise from vehicle up direction for moment positive dirction, the unstable attitude of vehicle judges as follows:

1) δ >0, turn left, revolver is interior wheel, oversteer, the yaw moment that demand is negative;

2) δ >0, turn left, revolver is interior wheel, understeering, the yaw moment that demand is positive;

3) δ <0, turn right, right wheel is taken turns for interior, oversteer, the yaw moment that demand is positive;

4) δ <0, turn right, right wheel is taken turns for interior, understeering, the yaw moment that demand is negative;

Wherein, δ is deflection angle, for yaw velocity deviation, for yaw velocity, for expecting yaw velocity;

When understeering, increase the propulsive effort that outer front-wheel reduces inner rear wheel simultaneously, when oversteer, reduce the propulsive effort that outer front-wheel increases inner rear wheel simultaneously, maximum to the contribution of replying steady-state quantities.

Described expectation side slip angle is β _d, expect that yaw velocity is wherein, function for wheel steering angle δ and speed of a motor vehicle u:

${\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}_d\&le;\min \left\{\right|\frac{u}{L(1+{\mathrm{Ku}}^2)}\mathrm{\&delta;}|,|\frac{0.8{\mathrm{\&mu;}}_s}{v}\left|\right\}\mathrm{sign}\left(\mathrm{\&delta;}\right)$

In formula, K is the understeering coefficient of vehicle m is vehicle mass, and a is that vehicle centroid arrives front axle distance, and b is that vehicle centroid arrives rear axle distance, and L=a+b is wheel base, k ₁for front tyre cornering stiffness, k ₂for rear tire cornering stiffness, μ _sfor vehicle side is to adhesion value;

Expect that side slip angle controls, in scope little as far as possible, to get β _d=0.

Described in order to produce yaw moment, the propulsive effort of left and right sides needs Differential Driving, the change note Δ F of four wheel drive power _xi, then the propulsive effort after Four wheel alignment needed for reality is F _xi ^*=F _xi+ Δ F _xi;

Wherein i=1,2,3,4 represent the near front wheel, off front wheel, left rear wheel, off hind wheel respectively, are just with vehicle forward direction;

After yaw moment electronic control unit obtains information of vehicles, Δ F required under adopting the intensified learning Q algorithm based on multiple agent to calculate different conditions _xi, by arranging Δ F under different conditions _xicost function use Q computing to obtain restraining minimum action sequence:

Will in span becoming different conditions with combination of two after Δ β decile, by selecting different conditions, calculating the Q value Q (s of the different actions under this state, a), then by selecting next action and state updating Q value, its principle upgraded is Q (s, a)=g+rminQ (s ', a ') wherein, (s, a) is the Q value of current action and state to Q, Q (s ', a ') be the Q value of a upper action and state, g is cost immediately, and r is discount factor; Selecting to restrain to obtain the minimum action sequence of one group of Q value by repeatedly circulating, reaching dbjective state namely trend towards 0 with Δ β, obtain the action sequence of Least-cost, meet control overflow.

The described intensified learning Q algorithm based on multiple agent, comprises the following steps:

1) initialization Q shows;

2) judge interior foreign steamer according to corner value, the action of left and right wheels is changed into the action of interior foreign steamer;

3) content of action and state to be determined before calculation cost, calculate following value most according to auto model:

$\left\{\begin{array}{c}\\ \max \mathrm{\&Delta;\&beta;}={\mathrm{\&beta;}}_{\max }\\ \max \mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}={\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}_{\max }-{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}_d\\ \max \mathrm{\&Delta;}{F}_{\mathrm{xi}}=\frac{k_{\mathrm{\&beta;}}\max \mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\max \mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4}\end{array}\right.$

Max Δ β is side slip angle error maximum permissible value, for yaw-rate error maximum permissible value, max Δ F _xithe maximum permissible value of each wheel drive force change;

4) define cost function, cost function setting principle is that the cost of inner rear wheel and outer front-wheel is less, by this two-wheeled mean allocation demand yaw torque, is the operation point of Least-cost in 1/2 of demand yaw torque,

5) discount factor r is set;

6) plot number is set;

7) state is selected at random, by side slip angle error delta β decile between ± max Δ β; Yaw-rate error ? between decile, then by Δ β and combination of two becomes final state, each state comprise different Δ β and

Dbjective state is $\left\{\begin{array}{c}\left|\mathrm{\&Delta;\&beta;}\right|\&le;\min \left|\mathrm{\&Delta;\&beta;}\right|\\ \left|\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}\right|\&le;\min \left|\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}\right|\end{array}\right.$

represent that actual side slip angle and yaw velocity trend towards expectation value and reach within error allowed band;

8) define Q (s, a)=g+rminQ (s ', a '), wherein (s, a) is the Q value of current state and action to Q, the Q value that Q (s ', a ') is a upper action and state, and g is cost immediately;

Random select a motor and perform an action obtain cost g immediately, determine next state according to current state and performed action; It is different that different states performs identical action cost, is determined by cost function, then initialization Q matrix, and obtains state action cost matrix;

9) constantly repeatedly access each state, show to make Q to be converged in Inner eycle; Constantly perform an action, obtain cost, find next state, upgrade the Q value that current state performs current action, arrive new state, again perform an action and upgrade Q value, arrive dbjective state always and just exit Inner eycle; From the Q matrix finally obtained, find optimum coordination strategy, make accumulative Least-cost.

Described cost function is:

$\left\{\begin{array}{c}\mathrm{if\&delta;}>0\\ g\left(\mathrm{\&Delta;}{F}_{x1}\right)=a_1{({\mathrm{\&Delta;F}}_{x1}+\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_1\\ g\left(\mathrm{\&Delta;}{F}_{x4}\right)=a_1{(\mathrm{\&Delta;}{F}_{x4}-\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_1\\ g\left(\mathrm{\&Delta;}{F}_{x2}\right)=a_2{(\mathrm{\&Delta;}{F}_{x2}-\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_2\\ g\left(\mathrm{\&Delta;}{F}_{x3}\right)=a_2{({\mathrm{\&Delta;F}}_{x3}+\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_2\\ \mathrm{if\&delta;}<0\\ g\left(\mathrm{\&Delta;}{F}_{x1}\right)=a_2{(\mathrm{\&Delta;}{F}_{x1}+\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_2\\ g\left(\mathrm{\&Delta;}{F}_{x4}\right)=a_2{(\mathrm{\&Delta;}{F}_{x4}-\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_2\\ g\left(\mathrm{\&Delta;}{F}_{x2}\right)=a_1{(\mathrm{\&Delta;}{F}_{x2}-\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_1\\ g\left(\mathrm{\&Delta;}{F}_{x3}\right)=a_1{(\mathrm{\&Delta;}{F}_{x3}+\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_1\end{array}\right.$

Wherein a ₁, a ₂, b ₁, b ₂for setting constant and a ₁>a ₂, b ₁>b ₂;

β is side slip angle, F _xifor the propulsive effort of four-wheel, Δ F _xifor the changing value of the propulsive effort of four-wheel, g (Δ F _xi) be each cost immediately of taking turns action, β _dfor expecting side slip angle, Δ β=β-β _dfor side slip angle deviation, k _β, for proportionality coefficient, d is width between two-wheeled, and δ is wheel steering angle, for producing the propulsive effort changing value needed for 1/2 demand yaw torque.

Described change cost function can meet non-all-wheel powered automobile:

For forward drive vehicle, only the propulsive effort of front turbine or electrical motor is controlled, i.e. Δ F _x1, Δ F _x2just can bear, brake-power control only be done to trailing wheel namely Δ F is set _x3<0, Δ F _x4<0;

For rear drive vehicle, then make contrary setting.

Compared with prior art, the individual drive electronlmobil yaw moment control method based on multiple agent provided by the invention has following useful technique effect:

1) a set of failure-free moment variations scheme and action sequence can be made according to vehicle-state automatically, the contribution margin of four-wheel in turning to can be quantized by arranging cost function, thus there is good comformability, make transition pulsation-free meet control accuracy by the propulsive effort controlling four-wheel simultaneously simultaneously;

2) energy ezpenditure of braking procedure is avoided in being controlled by propulsive effort;

3) owing to being be applicable to high-speed cruising situation based on the control process of dynamics analysis, processed the non-linear factor existed in system preferably simultaneously, ensured the stability turned to.

4) applicable equally for non-all-wheel powered automobile, only need change cost function, as only controlled the propulsive effort of front turbine or electrical motor for forward drive vehicle, brake-power control only be done to trailing wheel namely Δ F is set _x3<0, Δ F _x4<0 meets the demands equally, and the alerting ability that the present invention applies at all-wheel powered electronlmobil is stronger.

Accompanying drawing explanation

Fig. 1 is system architecture schematic block diagram of the present invention;

Fig. 2 is control system schematic block diagram of the present invention;

Fig. 3 is Q operational flowchart;

The curve synoptic diagram of cost function when Fig. 4 is left-hand rotation;

The curve synoptic diagram of cost function when Fig. 5 is left-hand rotation.

Detailed description of the invention

Below in conjunction with specific embodiment, the present invention is described in further detail, and the explanation of the invention is not limited.

Based on the four motorized wheels electronlmobil yaw moment control method of multiple agent, mainly comprise:

1) electronlmobil of four motorized wheels, state observation and measurement module, yaw moment electronic control unit (ECU), electric machine controller and CAN are set;

2) state observation and measurement module are by four wheel speed sensors, corner displacement sensor, pedal force sensor, yaw-rate sensor Real-time Collection wheel speed signal, wheel steering angle, treadle effort, yaw velocity;

3) yaw moment electronic control unit (ECU) uses the intensified learning Q algorithm based on multiple agent to obtain the action sequence of four-wheel by analyzing the signal gathering;

4) drive motor controller of each wheel is lower floor's actr, performs the action sequence that step 3) calculates;

5) state observation and measurement module, transmit the signal needed for control process by CAN communication between yaw moment electronic control unit (ECU) and electric machine controller.

Concrete, as shown in Figure 1, one electronlmobil 1 comprising four motorized wheels, state observation and measurement module 2, yaw moment electronic control unit (ECU) 3, electric machine controller 4, electric machine controller 5, electric machine controller 6, electric machine controller 7 and CAN 8 are set, ECU is by accepting from after the signal analysis of measurement module, use the torque instruction that Q computing obtains needed for four-wheel electric machine controller, allly independently all transmit data by CAN between controller and sensor.

Will identify that before implementing yaw moment chaufeur turns to the differentiation of intention and motoring condition, driver intention is by receiving the signal acquisition from steering angle sensor and wheel speed sensor:

Vehicle-state receives yaw-rate sensor signal acquisition, and producing with working direction the yaw velocity turned left just is, to overlook conter clockwise (left-hand rotation) from vehicle up direction for moment positive dirction, the unstable attitude judgment rule of vehicle is as follows:

1) δ >0, turn left, revolver is interior wheel, oversteer, the yaw moment that demand is negative;

2) δ >0, turn left, revolver is interior wheel, understeering, the yaw moment that demand is positive;

3) δ <0, turn right, right wheel is taken turns for interior, oversteer, the yaw moment that demand is positive;

4) δ <0, turn right, right wheel is taken turns for interior, understeering, the yaw moment that demand is negative;

Wherein, δ is deflection angle, for yaw velocity deviation, for yaw velocity, for expecting yaw velocity;

By knowing conditional electronic stability program (ESP) analysis: when turning to, the additional yaw moment that outer front-wheel is caused by braking force is with reduced the additional yaw moment direction caused by side force identical, all contrary with steering direction, when oversteer, this take turns apply braking force to correction oversteer the most effective; In like manner, inner rear wheel applies braking force and the most effectively understeering is corrected for correction understeering.

Therefore in four motorized wheels electronlmobil, arrange the effect that can increase or reduce inner rear wheel and outer front-wheel when many wheels are coordinated to produce yaw moment according to steering state comparatively obvious, the effect of other two-wheeleds is comparatively gentle.Increase outer front-wheel during understeering and reduce the propulsive effort of inner rear wheel simultaneously, reducing outer front-wheel during oversteer, to increase the contribution of propulsive effort to reply steady-state quantities of inner rear wheel maximum simultaneously.

Concrete, control method of the present invention adopts the Q computing based on multiple agent when calculating yaw moment, as shown in Figure 2, driver intention input wherein comprises control system block diagram:

Method, is characterized in that, described driver intention input comprises:

δ is wheel steering angle, and to turn left for just, signal is from corner displacement sensor;

U is vehicular longitudinal velocity, signal from longitudinal speed sensor,

for yaw velocity is from yaw-rate sensor;

β is side slip angle, estimates that vehicular longitudinal velocity u and vehicle side obtain to speed v by auto model, $\mathrm{\&beta;}=\arctan \frac{v}{u};$

When side slip angle and yaw velocity computing value exceed maxim, directly get respective maxim as its estimated valve, the maxim of side slip angle and yaw velocity is got as follows

According to 2DOF reference model, function for wheel steering angle δ and speed of a motor vehicle u:

${\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}_d\&le;\min \left\{\right|\frac{u}{L(1+{\mathrm{Ku}}^2)}\mathrm{\&delta;}|,|\frac{0.8{\mathrm{\&mu;}}_s}{v}\left|\right\}\mathrm{sign}\left(\mathrm{\&delta;}\right)---\left(2\right)$

In formula, K is the understeering coefficient of vehicle m is vehicle mass, and a is that vehicle centroid arrives front axle distance, and b is that vehicle centroid arrives rear axle distance, and L=a+b is wheel base, k ₁for front tyre cornering stiffness, k ₂for rear tire cornering stiffness, μ _sfor vehicle side is to adhesion value;

Expect that side slip angle controls, in scope little as far as possible, to get β _d=0.

Before yaw moment does not get involved wagon control, the torque of vehicle generally distributes between antero posterior axis according to acceleration demand, and the propulsive effort of note four-wheel is F _xi, can think that the propulsive effort of longitudinal direction of the left and right sides is equal; For producing yaw moment, the necessary Differential Driving of propulsive effort of left and right sides, the change note Δ F of four wheel drive power _xi, then the propulsive effort after Four wheel alignment needed for reality is F _xi ^*=F _xi+ Δ F _xi; (3)

Wherein i=1,2,3,4 represent the near front wheel, off front wheel, left rear wheel, off hind wheel respectively, are just with vehicle forward direction.

After yaw moment electronic control unit obtains information of vehicles, Δ F required under just needing to calculate different conditions by Q computing _xi, represent the action of four wheels, the impact of different action sequences on result is inconsistent.Concrete, by arranging Δ F under different conditions _xicost function use Q computing to obtain restraining minimum action sequence namely to reach the object improving steering stability.

Adopt the intensified learning Q algorithm based on multiple agent, will in span different conditions is become with combination of two after Δ β decile, by selecting different conditions, calculate Q value and the Q (s of the different actions (i.e. the changing value of the propulsive effort of four motors) under this state, a), again by selecting next action and state updating Q value, its principle upgraded is Q (s, a)=g+rmin Q (s ', a ') wherein, (s, a) is the Q value of current action and state to Q, Q (s ', a ') be the Q value of a upper action and state, g is cost immediately, and r is discount factor.So by repeatedly circulation select to restrain obtain the minimum action sequence of one group of Q value reach dbjective state and Δ β and trend towards 0, so can obtain the action sequence of Least-cost, meet the requirement of control system.

As shown in Figure 3, the diagram of circuit of multiple agent Q computing, comprises the following steps:

1. initialization Q shows;

2. due to turn left and during right steering in foreign steamer different, so foreign steamer in first judging according to corner value, getting δ >0 is left-hand rotation, like this can to the action action of left and right wheels being changed into interior foreign steamer.

3. will determine the content of action and state before calculation cost, this will calculate following value most according to auto model:

$\left\{\begin{array}{c}\\ \max \mathrm{\&Delta;\&beta;}={\mathrm{\&beta;}}_{\max }\\ \max \mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}={\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}_{\max }-{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}_d\\ \max \mathrm{\&Delta;}{F}_{\mathrm{xi}}=\frac{k_{\mathrm{\&beta;}}\max \mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\max \mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4}\end{array}\right.---\left(4\right)$

Max Δ β is side slip angle error maximum permissible value, for yaw-rate error maximum permissible value, max Δ F _xithe maximum permissible value of each wheel drive force change;

4. cost function is defined

Cost function setting principle is that the cost of inner rear wheel and outer front-wheel is less, and by this two-wheeled mean allocation demand yaw torque, be the operation point of Least-cost in 1/2 of demand yaw torque, its concrete form is as follows:

$\left\{\begin{array}{c}\mathrm{if\&delta;}>0\\ g\left(\mathrm{\&Delta;}{F}_{x1}\right)=a_1{({\mathrm{\&Delta;F}}_{x1}+\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_1\\ g\left(\mathrm{\&Delta;}{F}_{x4}\right)=a_1{(\mathrm{\&Delta;}{F}_{x4}-\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_1\\ g\left(\mathrm{\&Delta;}{F}_{x2}\right)=a_2{(\mathrm{\&Delta;}{F}_{x2}-\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_2\\ g\left(\mathrm{\&Delta;}{F}_{x3}\right)=a_2{({\mathrm{\&Delta;F}}_{x3}+\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_2\\ \mathrm{if\&delta;}<0\\ g\left(\mathrm{\&Delta;}{F}_{x1}\right)=a_2{(\mathrm{\&Delta;}{F}_{x1}+\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_2\\ g\left(\mathrm{\&Delta;}{F}_{x4}\right)=a_2{(\mathrm{\&Delta;}{F}_{x4}-\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_2\\ g\left(\mathrm{\&Delta;}{F}_{x2}\right)=a_1{(\mathrm{\&Delta;}{F}_{x2}-\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_1\\ g\left(\mathrm{\&Delta;}{F}_{x3}\right)=a_1{(\mathrm{\&Delta;}{F}_{x3}+\frac{k_{\mathrm{\&beta;}}\mathrm{\&Delta;\&beta;}+k_{\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}}{d/4})}^2+b_1\end{array}\right.---\left(5\right)$

Wherein a ₁, a ₂, b ₁, b ₂for artificially setting constant and ensureing a ₁>a ₂, b ₁>b ₂,

β is side slip angle, F _xifor the propulsive effort of four-wheel, Δ F _xifor the changing value of the propulsive effort of four-wheel, g (Δ F _xi) be each cost immediately of taking turns action, β _dfor expecting side slip angle, Δ β=β-β _dfor side slip angle deviation, k _β, for proportionality coefficient, d is width between two-wheeled, and δ to be wheel steering angle δ be wheel steering angle overlooks conter clockwise for just (i.e. working direction left-hand rotation) with vehicle up direction, for producing the propulsive effort changing value needed for 1/2 demand yaw torque.

Concrete, the cost function coordinate schematic diagram arranged is as Fig. 4 and Fig. 5, as can be seen from the figure the cost of inner rear wheel and outer front-wheel is less, at the operation point Least-cost of half demand yaw torque, the all driving costs of taking turns of reduction along with demand torque all will decline, the rule of these realistic systems.

5. r is discount factor, and reflection laststate is to the influence degree of NextState, and algorithm the convergence speed is relevant with the size of r value, concrete, gets r=0.8 here.

6. plot number is set, often organize different action sequences to reach dbjective state and be the three unities, arrange plot in order to test of many times and obtain the action sequence of restraining, plot number arranges the upper limit, reach higher limit or Q table restrained just can stop calculate, identification reach control objectives.

7. select a state at random, (state, write a Chinese character in simplified form s) is by side slip angle error delta β and yaw-rate error to described state various combination.Concrete, by side slip angle error delta β 10 deciles between ± max Δ β; Yaw-rate error ? between 10 deciles, then by Δ β and combination of two becomes 100 final states, each state comprise different Δ β and

Dbjective state is for working as $\left\{\begin{array}{c}\left|\mathrm{\&Delta;\&beta;}\right|\&le;\min \left|\mathrm{\&Delta;\&beta;}\right|\\ \left|\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}\right|\&le;\min \left|\mathrm{\&Delta;}\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}\right|\end{array}\right.---\left(6\right)$

namely actual side slip angle and yaw velocity trend towards expectation value and reach within error allowed band.

To whole vehicle model be used when estimated vehicle velocity and side slip angle, in auto model, not comprise inclination and pitch freedom, but consider the load transfer that longitudinal direction of car and lateral acceleration cause.The present invention adopts 7 degree of freedom car models as follows:

$m(\stackrel{\&CenterDot;}{u}-\mathrm{\&gamma;v})=\underset{i=1}{\overset{4}{\mathrm{\&Sigma;}}}{F}_{\mathrm{xi}}-\mathrm{mg}\sin \mathrm{\&xi;}-F_w-F_f---\left(7\right)$

$m(\stackrel{\&CenterDot;}{v}+\mathrm{\&gamma;u})=\underset{i=1}{\overset{4}{\mathrm{\&Sigma;}}}{F}_{\mathrm{yi}}---\left(8\right)$

$\left[\begin{array}{c}{F}_{\mathrm{xi}}\\ F_{\mathrm{yi}}\end{array}\right]=\left[\begin{array}{cc}\cos {\mathrm{\&delta;}}_i& -\sin {\mathrm{\&delta;}}_i\\ \sin {\mathrm{\&delta;}}_i& \cos {\mathrm{\&delta;}}_i\end{array}\right]\left[\begin{array}{c}{F}_{\mathrm{xwi}}\\ F_{\mathrm{ywi}}\end{array}\right]---\left(9\right)$

$I_Z\stackrel{\&CenterDot;}{\mathrm{\&gamma;}}=a(F_{y1}+F_{y2})-b(F_{y3}+F_{y4})+M_Z---\left(10\right)$

$M_z=\frac{d}{2}(\mathrm{\&Delta;}{F}_{x2}-\mathrm{\&Delta;}{F}_{x1}+\mathrm{\&Delta;}{F}_{x4}-\mathrm{\&Delta;}{F}_{x3})---\left(11\right)$

$\left\{\begin{array}{c}{F}_{z1}=\frac{\mathrm{mgb}}{2l}-\frac{m\stackrel{\&CenterDot;}{u}{h}_g}{2l}+\frac{m\stackrel{\&CenterDot;}{v}b{h}_g}{\mathrm{lb}}\\ F_{z2}=\frac{\mathrm{mgb}}{2l}-\frac{m\stackrel{\&CenterDot;}{u}{h}_g}{2l}-\frac{m\stackrel{\&CenterDot;}{v}b{h}_g}{\mathrm{lb}}\\ F_{z3}=\frac{\mathrm{mga}}{2l}+\frac{m\stackrel{\&CenterDot;}{u}{h}_g}{2l}+\frac{m\stackrel{\&CenterDot;}{v}a{h}_g}{\mathrm{lb}}\\ F_{z4}=\frac{\mathrm{mga}}{2l}+\frac{m\stackrel{\&CenterDot;}{u}{h}_g}{2l}-\frac{m\stackrel{\&CenterDot;}{v}{\mathrm{ah}}_g}{\mathrm{lb}}\end{array}\right.---\left(12\right)$

Vehicle parameter is as follows:

M---vehicle mass;

U---longitudinal velocity and x direction speed, ---x directional acceleration;

V---side velocity and y direction speed, ---y directional acceleration;

I _z---vehicle is around the rotor inertia of z-axis;

A---front axle is to the horizontal throw of barycenter, and b---rear axle is to the horizontal throw of barycenter, and l=a+b is wheel base;

D---wheelspan (getting wheel base identical), h _g---height of center of mass;

I=1,2,3,4 represent respectively left front, right front, left back, right back;

F _xwi, F _ywi---each wheel at tyre axis system along the longitudinal direction of axletree and side direction tire force;

F _xi, F _yi---respectively take turns the component in x, y direction when tire in vehicle axis system is subject to the antagonistic force on ground

F _zi---the vertical load of each wheel;

M _z---yaw moment;

δ _ifor wheel turning angle (ignoring camber steer effect), wherein δ ₁=δ ₂=δ, δ ₃=δ ₄=0;

Mg sin ξ is grade resistance, and ξ is grade angle;

F _wfor frontal resistance, its expression formula is:

$F_w=\frac{1}{2}{C}_D{A}_f{\mathrm{\&rho;}}_a{u}^2---\left(13\right)$

Wherein coefficient is: C _dfor aerodynamic drag factor, A _ffor the wind area of vehicle, these two coefficients all have very large relation, ρ with the profile of vehicle _afor density of air;

F _ffor rolling resistance, its expression formula is:

F _f=mgf （14）

Wherein, f is coefficient of rolling resistance.

8. define Q (s, a)=g+rminQ (s ', a ') wherein Q (s, a) is the Q value of current state and action, the Q value that Q (s ', a ') is a upper action and state, and g is cost immediately.Setting like this, can under cannot determining to take which type of coordination strategy cost minimum in each state artificially, use automatic cycle to constantly update Q value and find optimum scheme, take into account interacting between the state of front and back simultaneously, ensure convergence and validity.

Random select a motor and perform an action obtain cost g immediately, (action writes a Chinese character in simplified form a) for by Δ F for wherein said action _xi± max Δ F _xiwithin 20 deciles, each action represents the acceleration or deceleration of different wheel, and the action under each state has 4, and namely four wheels all can self contained function.

9. next state is determined according to current state and performed action; It is different that different states performs identical action cost, is determined by cost function.Then initialization Q matrix, and obtain state action cost matrix.

Outer circulation is to constantly can repeatedly access each state, show to make Q to be converged in Inner eycle, with the optional initial condition of outer circulation for starting point, constantly perform an action, obtain cost, find next state, upgrade the Q value that current state performs current action, arrive new state, again perform an action and upgrade Q value, arrive dbjective state always and just exit Inner eycle.From the Q matrix finally obtained, be just easy to find optimum coordination strategy, make accumulative Least-cost.

Above-mentioned example is only for illustration of the present invention, the strongest at the alerting ability of all-wheel powered electronlmobil application.Applicable equally for non-all-wheel powered automobile, only need change cost function, as only controlled the propulsive effort of front turbine or electrical motor for forward drive vehicle, i.e. Δ F _x1, Δ F _x2just can bear, brake-power control only be done to trailing wheel namely Δ F is set _x3<0, Δ F _x4<0 meets the demands equally.

Claims (9)

1., based on an individual drive electronlmobil yaw moment control method for multiple agent, it is characterized in that, comprise following operation:

1) each motor coordination controller using yaw moment electronic control unit as individual drive electronlmobil, yaw moment electronic control unit is turned to by state observation and measurement module Real-time Obtaining chaufeur and is intended to and motoring condition;

2) signal that yaw moment electronic control unit analysis collection is next, obtains vehicle running state, calculates the error of the yaw velocity of reality and demand and the error delta β of the vehicle centroid sideslip angle of reality and demand, then use the intensified learning Q algorithm based on multiple agent to obtain the adjustment of the torque needed for the four-wheel action sequence of individual drive electronlmobil;

3) action sequence is sent to each drive motor controller of individual drive electronlmobil by multiple agent, and in conjunction with the rate request of vehicle, send instruction and perform the action sequence obtained to each drive motor, the yaw moment needed for generation, correct the attitude of vehicle.

2. as claimed in claim 1 based on the individual drive electronlmobil yaw moment control method of multiple agent, it is characterized in that, described state observation and measurement module, between yaw moment electronic control unit and electric machine controller, all transmit the signal needed for control process by CAN.

3. as claimed in claim 1 based on the individual drive electronlmobil yaw moment control method of multiple agent, it is characterized in that, described chaufeur turns to intention input to comprise:

δ is wheel steering angle, and to turn left for just, signal is from corner displacement sensor;

U is vehicular longitudinal velocity, signal from longitudinal speed sensor,

for yaw velocity is from yaw-rate sensor;

β is side slip angle, estimates that vehicular longitudinal velocity u and vehicle side obtain to speed v by auto model, $\mathrm{\&beta;}=\arctan \frac{v}{u};$

μ _sfor vehicle side is to adhesion value;

When side slip angle and yaw velocity computing value exceed maxim, get respective maxim as its estimated valve, the maxim of side slip angle and yaw velocity is got as follows:

A _ymaxfor lateral acceleration maxim; M/s and m/s ²be respectively the unit of speed and acceleration/accel.

4. as claimed in claim 1 based on the individual drive electronlmobil yaw moment control method of multiple agent, it is characterized in that, producing with working direction the yaw velocity turned left just is, to overlook conter clockwise from vehicle up direction for moment positive dirction, the unstable attitude of vehicle judges as follows:

1) δ >0, turn left, revolver is interior wheel, oversteer, the yaw moment that demand is negative;

2) δ >0, turn left, revolver is interior wheel, understeering, the yaw moment that demand is positive;

3) δ <0, turn right, right wheel is taken turns for interior, oversteer, the yaw moment that demand is positive;

4) δ <0, turn right, right wheel is taken turns for interior, understeering, the yaw moment that demand is negative;

Wherein, δ is deflection angle, for yaw velocity deviation, for yaw velocity, for expecting yaw velocity;

When understeering, increase the propulsive effort that outer front-wheel reduces inner rear wheel simultaneously, when oversteer, reduce the propulsive effort that outer front-wheel increases inner rear wheel simultaneously, maximum to the contribution of replying steady-state quantities.

5., as claimed in claim 1 based on the individual drive electronlmobil yaw moment control method of multiple agent, it is characterized in that, described expectation side slip angle is β _d, expect that yaw velocity is wherein, function for wheel steering angle δ and speed of a motor vehicle u:

${\stackrel{.}{\mathrm{\&gamma;}}}_d\&le;\min \left\{\right|\frac{u}{L(1+{\mathrm{Ku}}^2)}\mathrm{\&delta;}|,|\frac{0.8{\mathrm{\&mu;}}_s}{v}\left|\right\}\mathrm{sign}\left(\mathrm{\&delta;}\right)$

In formula, K is the understeering coefficient of vehicle m is vehicle mass, and a is that vehicle centroid arrives front axle distance, and b is that vehicle centroid arrives rear axle distance, and L=a+b is wheel base, k ₁for front tyre cornering stiffness, k ₂for rear tire cornering stiffness, μ _sfor vehicle side is to adhesion value; V is vehicle side velocity;

Expect that side slip angle controls, in scope little as far as possible, to get β _d=0.

6. as claimed in claim 1 based on the individual drive electronlmobil yaw moment control method of multiple agent, it is characterized in that, for producing yaw moment, the propulsive effort of left and right sides needs Differential Driving, the change note Δ F of four wheel drive power _xi, then the propulsive effort after Four wheel alignment needed for reality is F _xi ^*=F _xi+ Δ F _xi;

Wherein i=1,2,3,4 represent the near front wheel, off front wheel, left rear wheel, off hind wheel respectively, are just with vehicle forward direction;

After yaw moment electronic control unit obtains information of vehicles, Δ F required under adopting the intensified learning Q algorithm based on multiple agent to calculate different conditions _xi, by arranging Δ F under different conditions _xicost function use Q computing to obtain restraining minimum action sequence:

Will in span becoming different conditions with combination of two after Δ β decile, by selecting different conditions, calculating the Q value Q (s of the different actions under this state, a), then by selecting next action and state updating Q value, its principle upgraded is Q (s, a)=g+rminQ (s ', a ') wherein, (s, a) is the Q value of current action and state to Q, Q (s ', a ') be the Q value of a upper action and state, g is cost immediately, and r is discount factor; Selecting to restrain to obtain the minimum action sequence of one group of Q value by repeatedly circulating, reaching dbjective state namely trend towards 0 with Δ β, obtain the action sequence of Least-cost, meet control overflow.

7., as claimed in claim 6 based on the individual drive electronlmobil yaw moment control method of multiple agent, it is characterized in that, the described intensified learning Q algorithm based on multiple agent, comprises the following steps:

1) initialization Q shows;

2) judge interior foreign steamer according to corner value, the action of left and right wheels is changed into the action of interior foreign steamer;

3) content of action and state to be determined before calculation cost, calculate following value most according to auto model:

$\left\{\begin{array}{c}\max \mathrm{\&Delta;\&beta;}={\mathrm{\&beta;}}_{\max }\\ \max \mathrm{\&Delta;}\stackrel{.}{\mathrm{\&gamma;}}={\stackrel{.}{\mathrm{\&gamma;}}}_{\max }-{\stackrel{.}{\mathrm{\&gamma;}}}_d\\ \max {\mathrm{\&Delta;F}}_{\mathrm{xi}}=\frac{k_{\mathrm{\&beta;}}\max \mathrm{\&Delta;\&beta;}+k_{\stackrel{.}{\mathrm{\&gamma;}}}\max \mathrm{\&Delta;}\stackrel{.}{\mathrm{\&gamma;}}}{d/4}\end{array}\right.$

Max Δ β is side slip angle error maximum permissible value, for yaw-rate error maximum permissible value, max Δ F _xithe maximum permissible value of each wheel drive force change; for proportionality coefficient, d is width between two-wheeled;

4) define cost function, cost function setting principle is that the cost of inner rear wheel and outer front-wheel is less, by this two-wheeled mean allocation demand yaw torque, is the operation point of Least-cost in 1/2 of demand yaw torque,

5) discount factor r is set;

6) plot number is set;

7) Stochastic choice state, by side slip angle error delta β decile between ± max Δ β; Yaw-rate error ? between decile, then by Δ β and combination of two becomes final state, each state comprise different Δ β and

Dbjective state is $\left\{\begin{array}{c}\left|\mathrm{\&Delta;\&beta;}\right|\&le;\min \left|\mathrm{\&Delta;\&beta;}\right|\\ \left|\mathrm{\&Delta;}\stackrel{.}{\mathrm{\&gamma;}}\right|\&le;\min \left|\mathrm{\&Delta;}\stackrel{.}{\mathrm{\&gamma;}}\right|\end{array}\right.$

Min| Δ β | ≈ 0, represent that actual side slip angle and yaw velocity trend towards expectation value and reach within error allowed band;

8) define Q (s, a)=g+rminQ (s ', a '), wherein (s, a) is the Q value of current state and action to Q, the Q value that Q (s ', a ') is a upper action and state, and g is cost immediately;

Random select a motor and perform an action obtain cost g immediately, determine next state according to current state and performed action; It is different that different states performs identical action cost, is determined by cost function, then initialization Q matrix, and obtains state action cost matrix;

9) constantly repeatedly access each state, show to make Q to be converged in Inner eycle; Constantly perform an action, obtain cost, find next state, upgrade the Q value that current state performs current action, arrive new state, again perform an action and upgrade Q value, arrive dbjective state always and just exit Inner eycle; From the Q matrix finally obtained, find optimum coordination strategy, make accumulative Least-cost.

8., as claimed in claim 7 based on the individual drive electronlmobil yaw moment control method of multiple agent, it is characterized in that, described cost function is:

<math> <mfenced open='{' close=''> <mtable> <mtr> <mtd> <mi>if&amp;delta;</mi> <mo>></mo> <mn>0</mn> </mtd> </mtr> <mtr> <mtd> <mi>g</mi> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mrow> <mi>x</mi> <mn>1</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>a</mi> <mn>1</mn> </msub> <msup> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mrow> <mi>x</mi> <mn>1</mn> </mrow> </msub> <mo>+</mo> <mfrac> <mrow> <msub> <mi>k</mi> <mi>&amp;beta;</mi> </msub> <mi>&amp;Delta;&amp;beta;</mi> <mo>+</mo> <msub> <mi>k</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </msub> <mi>&amp;Delta;</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </mrow> <mrow> <mi>d</mi> <mo>/</mo> <mn>4</mn> </mrow> </mfrac> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msub> <mi>b</mi> <mn>1</mn> </msub> </mtd> </mtr> <mtr> <mtd> <mi>g</mi> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mrow> <mi>x</mi> <mn>4</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>a</mi> <mn>1</mn> </msub> <msup> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mrow> <mi>x</mi> <mn>4</mn> </mrow> </msub> <mo>-</mo> <mfrac> <mrow> <msub> <mi>k</mi> <mi>&amp;beta;</mi> </msub> <mi>&amp;Delta;&amp;beta;</mi> <mo>+</mo> <msub> <mi>k</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </msub> <mi>&amp;Delta;</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </mrow> <mrow> <mi>d</mi> <mo>/</mo> <mn>4</mn> </mrow> </mfrac> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msub> <msub> <mi>b</mi> </msub> <mn>1</mn> </msub> </mtd> </mtr> <mtr> <mtd> <mi>g</mi> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mrow> <mi>x</mi> <mn>2</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>a</mi> <mn>2</mn> </msub> <msup> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mrow> <mi>x</mi> <mn>2</mn> </mrow> </msub> <mo>-</mo> <mfrac> <mrow> <msub> <mi>k</mi> <mi>&amp;beta;</mi> </msub> <mi>&amp;Delta;&amp;beta;</mi> <mo>+</mo> <msub> <mi>k</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </msub> <mi>&amp;Delta;</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </mrow> <mrow> <mi>d</mi> <mo>/</mo> <mn>4</mn> </mrow> </mfrac> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msub> <msub> <mi>b</mi> <mn>2</mn> </msub> </msub> </mtd> </mtr> <mtr> <mtd> <mi>g</mi> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mi>x</mi> </msub> <mn>3</mn> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>a</mi> <mn>2</mn> </msub> <msup> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mrow> <mi>x</mi> <mn>3</mn> </mrow> </msub> <mo>+</mo> <mfrac> <mrow> <msub> <mi>k</mi> <mi>&amp;beta;</mi> </msub> <mi>&amp;Delta;&amp;beta;</mi> <mo>+</mo> <msub> <mi>k</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </msub> <mi>&amp;Delta;</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </mrow> <mrow> <mi>d</mi> <mo>/</mo> <mn>4</mn> </mrow> </mfrac> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msub> <msub> <mi>b</mi> <mn>2</mn> </msub> </msub> </mtd> </mtr> <mtr> <mtd> <mi>if&amp;delta;</mi> <mo>&lt;0</mo> </mtd> </mtr> <mtr> <mtd> <mi>g</mi> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mrow> <mi>x</mi> <mn>1</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>a</mi> <mn>2</mn> </msub> <msup> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mrow> <mi>x</mi> <mn>1</mn> </mrow> </msub> <mo>+</mo> <mfrac> <mrow> <msub> <mi>k</mi> <mi>&amp;beta;</mi> </msub> <mi>&amp;Delta;&amp;beta;</mi> <mo>+</mo> <msub> <mi>k</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </msub> <mi>&amp;Delta;</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </mrow> <mrow> <mi>d</mi> <mo>/</mo> <mn>4</mn> </mrow> </mfrac> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msub> <msub> <mi>b</mi> </msub> <mn>2</mn> </msub> </mtd> </mtr> <mtr> <mtd> <mi>g</mi> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mrow> <mi>x</mi> <mn>4</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>a</mi> <mn>2</mn> </msub> <msup> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mrow> <mi>x</mi> <mn>4</mn> </mrow> </msub> <mo>-</mo> <mfrac> <mrow> <msub> <mi>k</mi> <mi>&amp;beta;</mi> </msub> <mi>&amp;Delta;&amp;beta;</mi> <mo>+</mo> <msub> <mi>k</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </msub> <mi>&amp;Delta;</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </mrow> <mrow> <mi>d</mi> <mo>/</mo> <mn>4</mn> </mrow> </mfrac> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msub> <msub> <mi>b</mi> </msub> <mn>2</mn> </msub> </mtd> </mtr> <mtr> <mtd> <mi>g</mi> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mrow> <mi>x</mi> <mn>2</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>a</mi> <mn>1</mn> </msub> <msup> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mrow> <mi>x</mi> <mn>2</mn> </mrow> </msub> <mo>-</mo> <mfrac> <mrow> <msub> <mi>k</mi> <mi>&amp;beta;</mi> </msub> <mi>&amp;Delta;&amp;beta;</mi> <mo>+</mo> <msub> <mi>k</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </msub> <mi>&amp;Delta;</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </mrow> <mrow> <mi>d</mi> <mo>/</mo> <mn>4</mn> </mrow> </mfrac> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msub> <msub> <mi>b</mi> <mn>1</mn> </msub> </msub> </mtd> </mtr> <mtr> <mtd> <mi>g</mi> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mrow> <mi>x</mi> <mn>3</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>a</mi> <mn>1</mn> </msub> <msup> <mrow> <mo>(</mo> <mi>&amp;Delta;</mi> <msub> <mi>F</mi> <mrow> <mi>x</mi> <mn>3</mn> </mrow> </msub> <mo>+</mo> <mfrac> <mrow> <msub> <mi>k</mi> <mi>&amp;beta;</mi> </msub> <mi>&amp;Delta;&amp;beta;</mi> <mo>+</mo> <msub> <mi>k</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </msub> <mi>&amp;Delta;</mi> <mover> <mi>&amp;gamma;</mi> <mo>.</mo> </mover> </mrow> <mrow> <mi>d</mi> <mo>/</mo> <mn>4</mn> </mrow> </mfrac> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msub> <msub> <mi>b</mi> <mn>1</mn> </msub> </msub> </mtd> </mtr> </mtable> </mfenced></math>

Wherein a ₁, a ₂, b ₁, b ₂for setting constant and a ₁>a ₂, b ₁>b ₂;

β is side slip angle, F _xifor the propulsive effort of four-wheel, Δ F _xifor the changing value of the propulsive effort of four-wheel, g (Δ F _xi) be each cost immediately of taking turns action, β _dfor expecting side slip angle, Δ β=β-β _dfor side slip angle deviation, for proportionality coefficient, d is width between two-wheeled, and δ is wheel steering angle, for producing the propulsive effort changing value needed for 1/2 demand yaw torque.

9. as claimed in claim 7 based on the individual drive electronlmobil yaw moment control method of multiple agent, it is characterized in that, change cost function can meet non-all-wheel powered automobile:

For forward drive vehicle, only the propulsive effort of front turbine or electrical motor is controlled, i.e. Δ F _x1, Δ F _x2just can bear, brake-power control only be done to trailing wheel namely Δ F is set _x3<0, Δ F _x4<0;

For rear drive vehicle, then make contrary setting.