CN105946858A - Method for optimizing parameters of four-driving electric car state observer based on genetic algorithm - Google Patents

Abstract

The invention discloses a method for optimizing parameters of a four-driving electric car state observer based on a genetic algorithm. By means of the method, the problem that parameter adjustment of an electric car state observer is difficult is solved. The method includes the following steps of building a car single-wheel rolling model and a simplified three-degree-of-freedom car model, designing a longitudinal tire force observer through a sliding-mode observer method with the wheel rotation angular velocity and driving moment of measurement information of a car sensor as input, designing a front shaft lateral tire force sliding-mode observer and a rear shaft lateral tire force sliding-mode observer correspondingly with the estimated longitudinal tire force value, the front wheel rotation angle, the lateral acceleration and the yaw velocity as input, finally designing a car speed full-dimension state observer with the estimated longitudinal and lateral tire force values, the longitudinal and lateral accelerations, the yaw acceleration and the front wheel rotation angle of a car as input, and carrying out observer parameter optimization on estimation modules through the genetic algorithm on the basic of the designed modular car state observer.

Description

Four-drive electric car state observer parameter optimization method based on genetic algorithm

Technical field

The present invention relates to a kind of modularity four-wheel driving electric vehicle state observer parameter optimization method based on genetic algorithm , belong to vehicle state estimation technical field.

Background technology

As the representative of new-energy automobile, electric automobile for gasoline combustion is as the orthodox car of power, cleaning, Environmental protection, the aspect such as energy-conservation occupy obvious advantage.Therefore, the recoverable amount of electric automobile is in the trend increased year by year, and it is handled Stability and active safety sex chromosome mosaicism have also been obtained and pay close attention to widely.

The active safety control system of electric automobile can be effectively improved vehicle handling stability, thus reduces vehicle accident Occur.And the premise that its various control logics are effectively implemented is accurately to obtain the running condition information of vehicle.Yet with life The restriction of the factor such as cost and measurement error of producing, in volume production car, Some vehicles running condition information cannot be directly by vehicle-mounted biography Sensor measurement obtains.Therefore, utilize measurable car status information design observer that the car status information that cannot measure is entered Row estimation has been increasingly becoming study hotspot.

In vehicle state estimation problem, vehicle-state observer parameter is to affect its key factor estimating accuracy, observer Parameter regulation problem is also its technological difficulties.Traditional observer regulation generally uses the manually side of regulation based on great many of experiments Method, this control method not only workload is very big, and it cannot be guaranteed that the parameter regulated is the optimal parameter of current working.Cause This, it is necessary to observer parameter is optimized by a kind of intelligent optimization algorithm of design.

Summary of the invention

For solving electronic vehicle attitude observer parameter regulation difficult problem, the present invention provides a kind of 4 wheel driven based on genetic algorithm electricity Electrical automobile state observer parameter optimization method, as a example by modular four-wheel driving electric vehicle state observer, uses heredity Algorithm is observed the optimization of device parameter.Wherein, modular vehicle-state observer is by longitudinal tire force sliding mode observer, side Constitute to tire force sliding mode observer and car speed omnidirectional vision.

The present invention is achieved by the following technical solutions:

A kind of four-drive electric car state observer parameter adjusting method based on genetic algorithm, comprises the following steps:

Step one, set up the Three Degree Of Freedom auto model of vehicle single-wheel roll modeling and simplification；

Step 2, modular vehicle state Observer Design: with vehicle sensors metrical information vehicle wheel rotation angular velocity and driving force Square, as input, uses the sliding mode observer method longitudinal tire force design longitudinal tire force sliding mode observer to four wheels；Again with Longitudinal tire force estimated value, front wheel angle, lateral acceleration and yaw velocity, as input, separately design axle lateral Tire force sliding mode observer；Finally with longitudinal and lateral tire force estimated value, longitudinal direction and lateral acceleration, yaw acceleration and car Front wheel angle as input, the longitudinal speed of design, lateral speed and the car speed omnidirectional vision of yaw velocity；

Step 3, modular vehicle state observer based on step 2 design, use genetic algorithm respectively to each estimation module It is observed device parameter optimization.

Further, in described step 3 use genetic algorithm respectively each estimation module is observed device parameter optimization include with Lower step:

3.1) bivariant standard genetic algorithm is used the most respectively the longitudinal tire force sliding mode observer parameter of four, vehicle wheel to be entered Row optimizes, then using the front axle longitudinal tire force estimated value through parameter optimization as input tire force sliding mode observer lateral to front axle Parameter is optimized, and finally tire force sliding mode observer parameter lateral to rear axle is optimized；

3.2) using through the front axle longitudinal direction of parameter optimization and axle lateral tire force estimated value as input, use univariate Standard genetic algorithm carries out parameter optimization to yaw rate state observer；

3.3) using through the longitudinal direction of parameter optimization and lateral tire force estimated value as input, use multi-objective genetic algorithm to vehicle Longitudinally, laterally speed observer parameter is optimized, and obtains Pareto optimal solution set.

Owing to have employed above-mentioned technical scheme, the invention has the beneficial effects as follows:

(1) problem manually regulating difficulty for vehicle-state observer parameter, it is proposed that one is applicable to modularity four-wheel drive electricity The genetic algorithm parameter optimization method of electrical automobile state observer.

(2) use high-fidelity vehicle dynamics simulation software veDYNA that the observer parameter optimized has been carried out validation verification, knot Fruit shows that observer parameter optimization method proposed by the invention has certain effect, it is possible to ensure the standard of observer estimated result Really property.

Accompanying drawing explanation

Fig. 1 vehicle single-wheel rolls kinetic model；

Stress schematic diagram overlooked by Fig. 2 vehicle；

Fig. 3 modular vehicle state observer structure chart；

Fig. 4 genetic algorithm performs step；

Fig. 5 the near front wheel longitudinal tire force sliding mode observer parameter optimization result

Fig. 6 longitudinally, laterally speed state observer parameter optimization Pareto forward position

Fig. 7 longitudinal tire force simulation result

Fig. 8 lateral tire force simulation result

Fig. 9 longitudinally, laterally speed and yaw velocity simulation result

Table 1 vehicle-state observer parameter optimization result

Table 2 four-wheel driving electric vehicle parameter

Table 3 longitudinally, laterally speed state observer parameter optimization Pareto optimal solution

Detailed description of the invention

Below in conjunction with the accompanying drawings, technical scheme proposed by the invention is further elaborated and illustrates.

The invention provides a kind of modularity four-wheel driving electric vehicle state observer parameter optimization method based on genetic algorithm, The method includes following step:

Step one, set up the vehicle Three Degree Of Freedom model of vehicle single-wheel roll modeling and simplification

1. set up vehicle single-wheel roll modeling

For design longitudinal tire force sliding mode observer, Vehicular system is reduced to vehicle single-wheel roll modeling as shown in Figure 1.

Single-wheel can be obtained by Fig. 1 and roll shown in kinetics equation such as formula (1):

$J{\stackrel{\·}{\mathrm{\ω}}}_i=-R_{eff}{F}_{xi}+T_i,(i=1,2,3,4),---\left(1\right)$

Wherein, J is the rotary inertia of wheel, units/kg m², ω_iFor the rotational angular velocity of each wheel, unit rad/s, R_eff For the effective radius of tire, unit m, T_iFor the driving moment of each wheel, unit Nm.

2. set up the Three Degree Of Freedom auto model simplified

Fig. 2 is that stress schematic diagram overlooked by vehicle, studies for convenience, the present invention in view of vehicle longitudinally, laterally and yaw The stressing conditions in direction, is reduced to Three Degree Of Freedom model by whole vehicle model.Setting up coordinate system on vehicle, initial point is positioned at automobile Barycenter, the direction that vehicle advances is x-axis positive direction, and level is y-axis positive direction to the left, and z-axis positive direction is true by right-hand screw rule Fixed, as shown in Figure 2.The Three Degree Of Freedom auto model kinetics equation such as formula (2) that application Newton's second law can obtain simplifying is shown

$\begin{array}{c}m{\stackrel{\·}{V}}_x=F_x+{\mathrm{mrV}}_y\\ m{\stackrel{\·}{V}}_y=F_y-{\mathrm{mrV}}_x^2\\ I_z\stackrel{\·}{r}=M_z\end{array}---\left(2\right)$

Wherein, m is automobile gross mass, units/kg, and r is the yaw velocity of automobile, unit rad/s, V_xAnd V_yIt it is vehicle Longitudinal direction under bodywork reference frame and side velocity, unit m/s, I_zFor car load around the rotary inertia of vehicle axis system z-axis, single Position kg m², F_xAnd F_yRepresent longitudinal direction of car and lateral tire force, unit N, M respectively_zFor vehicle around z-axis rotating torque, single Position Nm.

According to power and torque equilibrium equation, longitudinal direction of car and lateral tire force F_x、F_yWith vehicle around z-axis rotating torque M_zPermissible It is expressed as:

$\begin{array}{c}{F}_x=(F_{x1}+F_{x2}){\mathrm{cos\δ}}_f-(F_{y1}+F_{y2}){\mathrm{sin\δ}}_f+F_{x3}+F_{y4},\\ F_y=(F_{x1}+F_{x2}){\mathrm{sin\δ}}_f+(F_{y1}+F_{y2}){\mathrm{cos\δ}}_f+F_{x3}+F_{y4},\\ M_z=l_F(F_{x1}+F_{x2}){\mathrm{sin\δ}}_f+l_F(F_{y1}+F_{y2}){\mathrm{cos\δ}}_f-l_R(F_{y3}+F_{y4})\end{array}---\left(3\right)$

Wherein, F_xi/F_yi(i=1 ..., 4) it is respectively four longitudinal directions taken turns and lateral tire force, unit N, δ_fIt is vehicle front wheel angle, Unit rad, l_FAnd l_RIt is the vehicle centroid distance away from front/rear axle respectively, unit m.

Step 2, modular vehicle state Observer Design: with vehicle sensors metrical information vehicle wheel rotation angular velocity and driving force Square, as input, uses the sliding mode observer method longitudinal tire force design longitudinal tire force sliding mode observer to four wheels；Again with Longitudinal tire force estimated value, front wheel angle, lateral acceleration and yaw velocity, as input, separately design axle lateral Tire force sliding mode observer；Finally with longitudinally and lateral tire force estimated value and longitudinal direction and lateral acceleration, yaw acceleration and Vehicle front wheel angle, as input, designs the car speed omnidirectional vision of longitudinally, laterally speed and yaw velocity.Will Each observer module of above-mentioned design carries out integrated to obtain modular vehicle state observer, and its structure chart is as shown in Figure 3. For convenience the estimation problem of the present invention is introduced, first can make as follows by vehicle sensors parameter measured directly Illustrate:

Driving moment T of (1) four wheel_i(i=1,2,3,4) is although not directly measuring, but other can measure letter can to pass through vehicle Breath (engine moment T_e, engine speed ω_e, pressure of wheel cylinder p_t) be calculated, therefore can be considered as directly surveying Amount information；(2) the angular signal δ of steering wheel for vehicle can be obtained by photoelectric encoder measurement, and then can pass through relational expression δ_f=δ/I_swIt is calculated the front wheel angle δ of vehicle_f, I_swFor steering gear ratio；The rotational angular velocity of (3) four wheels ω_i(i=1,2,3,4) can be obtained by wheel speed sensors measurement；(4) longitudinal direction of car and lateral acceleration a_x、a_yCan be passed by acceleration Sensor measurement obtains；(5) yaw rate r can be obtained by gyroscope measurement.

Modular vehicle state Observer Design specifically includes following steps:

1, longitudinal tire force Design of Sliding Mode Observer

Roll kinetics equation according to single-wheel, provide first-order system as follows:

$\left\{\begin{array}{c}{\stackrel{\·}{\mathrm{\ω}}}_1=\frac{1}{J}{T}_i-\frac{R_{eff}}{J}{F}_{xi}\\ y={\mathrm{\ω}}_i\end{array}\right.---\left(4\right)$

Wherein, T_iIt is system input, ω_iMeasurement output as system is also system mode simultaneously, when system mode changes, Unknown worm amount F_xiChange the most therewith.Here, F_xiBe exactly our state to be estimated, then this estimation problem can be described as The process of system Unknown worm is gone out by measurement output estimation.

Theoretical according to sliding mode observer, here definition systematic error isIt is systematic error that the present invention chooses sliding-mode surface, I.e.And choose liapunov function:

V=S²/2 (5)

To formula (4) derivation, can obtain:

$\stackrel{\·}{V}=S\stackrel{\·}{S}=S\stackrel{\·}{\widetilde{\mathrm{\ω}}}=S(\stackrel{\·}{\mathrm{\ω}}-\stackrel{\·}{\widehat{\mathrm{\ω}}})---\left(6\right)$

According to State Observer Theory, formula (3) is configured as form, wherein L_xiIt it is observer gain.

$\stackrel{\·}{\widehat{\mathrm{\ω}}}=\frac{1}{J}{T}_i-\frac{R_{eff}}{J}{\hat{F}}_{xi}+L_{xi}(x-\hat{x})---\left(7\right)$

Formula (4) and formula (7) are substituted in (6), can obtain:

$\stackrel{\·}{V}=S\frac{R_{eff}}{J}{\hat{F}}_{xi}+S\[-\frac{R_{eff}}{J}{F}_{xi}-L_{xi}(x-\hat{x})\]---\left(8\right)$

Where it is assumed thatMeet with lower inequality:

$||-\frac{R_{eff}}{J}{F}_{xi}-L_{xi}(x-\hat{x})||\≤{\mathrm{\ρ}}_{xi}---\left(9\right)$

In above-mentioned hypothesis, F_xiMeet Bounded Conditions, if then ρ_xiTake sufficiently large value, it assumes that can set up.By formula (9) It is brought in formula (8), can obtain:

$\stackrel{\·}{V}\≤S\frac{R_{eff}}{J}{\hat{F}}_{xi}+\left|S\right|{\mathrm{\ρ}}_{xi}---\left(10\right)$

Now, if takingWherein sign (S) is sign function, and then can formula (10) be expressed as:

$\stackrel{\·}{V}\≤-{\mathrm{S\ρ}}_{xi}sign\left(S\right)+\left|S\right|{\mathrm{\ρ}}_{xi}=0---\left(11\right)$

By above-mentioned derivation, the sliding mode observer form of present invention design is as follows:

$\begin{array}{c}\stackrel{\·}{\widehat{\mathrm{\ω}}}=\frac{1}{J}{T}_i-\frac{R_{eff}}{J}{\hat{F}}_{xi}+L_{xi}(x-\hat{x})\\ {\hat{F}}_{xi}=-\frac{J}{R_{eff}}{\mathrm{\ρ}}_{xi}sign\left(S\right)\end{array}---\left(12\right)$

Systematic error derivative table is shown as by convolution (4) and formula (12) further:

$\stackrel{\·}{\widetilde{\mathrm{\ω}}}=\stackrel{\·}{\mathrm{\ω}}-\stackrel{\·}{\widehat{\mathrm{\ω}}}=-\frac{R_{eff}}{J}{\hat{F}}_{xi}-{\mathrm{\ρ}}_{xi}sign\left(S\right)-L_{xi}(\mathrm{\ω}-\widehat{\mathrm{\ω}})---\left(13\right)$

It is t when the time₁, when system reaches stable, can obtainTherefore:

$-\frac{R_{eff}}{J}{\hat{F}}_{xi}-{\mathrm{\ρ}}_{xi}sign\left(S\right)-L_{xi}(\mathrm{\ω}-\widehat{\mathrm{\ω}})=0---\left(14\right)$

Then according to formula (14), Unknown worm amount F_xiEstimated value can be to be expressed as form:

${\hat{F}}_{xi}=-\frac{J}{R_{eff}}{\mathrm{\ρ}}_{xi}sign\left(S\right)-\frac{J}{R_{eff}}{L}_{xi}(\mathrm{\ω}-\widehat{\mathrm{\ω}})---\left(15\right)$

Formula (15) is exactly to the present invention is directed to Unknown worm amountThe sliding mode observer of design, wherein, L_xiIt is feedback oscillator, ρ_xiIt is Sliding formwork gain.

Owing to time lag, Spatial lag and system inertia etc. affect, easily there is chattering phenomenon in sliding mode system, and this will increase Estimation difference thus affect estimated result.In order to weaken the impact of buffeting, the present invention uses saturation function (16) to replace sign function sign(S)。

${\mathrm{sign}}_{eq}(S,\mathrm{\φ})=\frac{S}{\left|S\right|+\mathrm{\φ}}---\left(16\right)$

Wherein, S represents that estimation difference, φ ＞ 0 are used for reconciling function sign_eqThe slope of (S, φ).

Formula (16) is brought in formula (15), can be as follows to obtain longitudinal tire force sliding mode observer form:

${\hat{F}}_{xi}=-\frac{J}{R_{eff}}{\mathrm{\ρ}}_{xi}\frac{{\mathrm{\ω}}_i-{\widehat{\mathrm{\ω}}}_i}{||{\mathrm{\ω}}_i-{\widehat{\mathrm{\ω}}}_i||+\mathrm{\φ}}-\frac{J}{R_{eff}}{L}_{xi}({\mathrm{\ω}}_i-{\widehat{\mathrm{\ω}}}_i)---\left(17\right)$

2, lateral tire force Design of Sliding Mode Observer

According to the vehicle Three Degree Of Freedom kinetics equation simplified, it is contemplated that vehicle is along the lateral motion equations of y-axis and turning around z-axis Square equilibrium equation, can obtain following vehicle two degrees of freedom kinetics equation:

$\left\{\begin{array}{c}{\mathrm{ma}}_y=F_{yf}{\mathrm{cos\δ}}_f+F_{yr}+F_{xf}{\mathrm{sin\δ}}_f\\ I_z\stackrel{\·}{r}=(F_{yf}{\mathrm{cos\δ}}_f+F_{xf}{\mathrm{sin\δ}}_f)l_F-F_{yr}{l}_R\end{array}\right.---\left(18\right)$

Wherein, a_yFor vehicle lateral acceleration, unit m/s, F_yf=F_y1+F_y2For the lateral tire force of front axle, unit N, F_yr=F_y1+F_y2For the lateral tire force of rear axle, unit N, F_xf=F_x1+F_x2For front axle longitudinal tire force, unit N.

By the front axle lateral tire force F in formula (18)_yfTire force F lateral with rear axle_yrUncoupling, obtains:

$\left\{\begin{array}{c}{F}_{yf}=\[I_z\stackrel{\·}{r}+l_R{\mathrm{ma}}_y-F_{xf}{\mathrm{sin\δ}}_f(l_F+l_R)\]\frac{1}{(l_F+l_R){\mathrm{cos\δ}}_f}\\ F_{yr}=(l_F{\mathrm{ma}}_y-I_z\stackrel{\·}{r})\frac{1}{l_F+l_R}\end{array}\right.---\left(19\right)$

Tire force F lateral for front-wheel_yf, formula (19) is turned to the first-order system of shape such as formula (4):

$\left\{\begin{array}{c}\stackrel{\·}{r}=-\frac{l_{R}m}{I_z}{a}_y+\frac{(l_F+l_R){\mathrm{cos\δ}}_f}{I_z}{F}_{yf}+\frac{(l_F+l_R){\mathrm{sin\δ}}_f}{I_z}{F}_{xf}\\ y=r\end{array}\right.---\left(20\right)$

Wherein, r is system mode, is also systematic survey output simultaneously, a_yInput for system, F_yfFor system Unknown worm amount, It is also intended to the state estimated simultaneously.

According to longitudinal tire force Design of Sliding Mode Observer process, the front axle lateral tire force sliding mode observer form of present invention design is such as Under:

${\hat{F}}_{yf}=\frac{I_z}{{\mathrm{cos\δ}}_f(l_F+l_R)}{\mathrm{\ρ}}_{yf}\frac{r-\hat{r}}{\left|r-\hat{r}\right|+\mathrm{\φ}}+\frac{I_z}{{\mathrm{cos\δ}}_f(l_F+l_R)}{L}_{yf}(r-\hat{r})---\left(21\right)$

Wherein, L_yfFor the feedback oscillator of front axle lateral tire force sliding mode observer, ρ_yfFor front axle lateral tire force sliding mode observer Sliding formwork gain.Convolution (20) and formula (21) are it can be seen that when tire force lateral to front axle is estimated, need with front axle The value of longitudinal tire force is as input.

In like manner can be as follows with design rear axle lateral tire force sliding mode observer form:

${\hat{F}}_{yr}=\frac{-I_z}{(l_F+l_R)}{\mathrm{\ρ}}_{yr}\frac{r-\hat{r}}{\left|r-\hat{r}\right|+\mathrm{\φ}}-\frac{I_z}{(l_F+l_R)}{L}_{yr}(r-\hat{r})---\left(22\right)$

Wherein, L_yrFor the feedback oscillator of rear axle lateral tire force sliding mode observer, ρ_yrFor rear axle lateral tire force sliding mode observer Sliding formwork gain.

3, car speed omnidirectional vision design

According to the equilibrium equation of power, longitudinally, laterally the relation between acceleration and vehicle tyre power can be described as:

$\begin{array}{c}{\mathrm{ma}}_x=F_x\\ {\mathrm{ma}}_y=F_y\end{array}---\left(23\right)$

Wherein, a_x、a_yIt is respectively longitudinal direction and lateral acceleration, unit m/s of vehicle².According to formula (2) and formula (23), vertical Can be further represented as to, lateral speed and yaw velocity:

$\begin{array}{c}{\stackrel{\·}{V}}_x=a_x+{\mathrm{rV}}_y\\ {\stackrel{\·}{V}}_y=a_y-{\mathrm{rV}}_x\\ \stackrel{\·}{r}=M_z/I_z\end{array}---\left(24\right)$

Obtain owing to longitudinally, laterally acceleration and yaw velocity directly can be measured by vehicle sensors, therefore select these three Measure and export as systematic survey, and using they differences with its estimated value as the correction term of car speed observer, based on non-linear Full micr oprocessorism structure, can be designed that the car speed full micr oprocessorism of longitudinal direction of car, lateral speed and yaw velocity is expressed Shown in formula such as formula (25):

$\begin{array}{c}{\stackrel{\·}{\hat{V}}}_x=a_x+r{\hat{V}}_y+K_x(a_x-{\hat{a}}_x)\\ {\stackrel{\·}{\hat{V}}}_y=a_y-r{\hat{V}}_x+K_y(a_y-{\hat{a}}_y)\\ \stackrel{\·}{\hat{r}}={\hat{M}}_z/I_z+K_r(r-\hat{r})\end{array}---\left(25\right)$

Wherein, K_i(i=x, y r) represent observer gain.Utilize tire force estimated value, can be by longitudinally, laterally acceleration estimation ValueAnd vehicle is around z-axis rotating torque estimated valueIt is expressed as:

$\begin{array}{c}{\hat{a}}_x=\frac{1}{m}\[({\hat{F}}_{x1}-{\hat{F}}_{x2}){\mathrm{cos\δ}}_f-({\hat{F}}_{y1}+{\hat{F}}_{y2}){\mathrm{sin\δ}}_f+{\hat{F}}_{x3}+{\hat{F}}_{x4}\],\\ {\hat{a}}_y=\frac{1}{m}\[({\hat{F}}_{x1}+{\hat{F}}_{x2}){\mathrm{sin\δ}}_f+({\hat{F}}_{y1}+{\hat{F}}_{y2}){\mathrm{cos\δ}}_f+{\hat{F}}_{y3}+{\hat{F}}_{y4}\],\\ {\hat{M}}_z=l_F({\hat{F}}_{x1}+{\hat{F}}_{x2}){\mathrm{sin\δ}}_f+l_F({\hat{F}}_{y1}+{\hat{F}}_{y2}){\mathrm{cos\δ}}_f-l_R({\hat{F}}_{y3}+{\hat{F}}_{y4})\end{array}---\left(26\right)$

Step 3, modular vehicle state observer based on step 2 design, use genetic algorithm respectively to each estimation module It is observed device parameter optimization.Scope and the optimum results of parameters optimization are as shown in table 1:

Table 1 vehicle-state observer parameter optimization result

Wherein parameters optimization scope is the more conservative scope be given by empirical value when manually regulating, to ensure it comprises Optimized parameter, specifically includes following steps:

1. use the bivariant standard genetic algorithm longitudinal tire force sliding mode observer parameter the most respectively to four, vehicle wheel ρ_xi/L_xi(i=1,2,3,4) is optimized, then using the front axle longitudinal tire force estimated value through parameter optimization as input to front isometric To tire force sliding mode observer parameter ρ_yf/L_yrIt is optimized, finally tire force sliding mode observer parameter ρ lateral to rear axle_yr/L_yr It is optimized.

According to the longitudinal tire force sliding mode observer designed by step 2 and axle lateral tire force sliding mode observer, L, ρ is the observer parameter needing to optimize.For tire force sliding mode observer Parametric optimization problem, due to longitudinally and laterally tire force Sliding mode observer has similar structure, and therefore the present invention is only given as a example by the near front wheel longitudinal tire force sliding mode observer, utilizes The process that observer parameter is optimized by genetic algorithm.

When observer parameter is optimized, use high-fidelity dynamics simulation software veDYNA, select the electronic vapour of four-wheel drive Car is as emulation vehicle, and vehicle parameter is as shown in table 2:

Table 2 four-wheel driving electric vehicle parameter

Allowing vehicle travel under conventional high attachment two-track lineman's condition, concrete operating mode is set to: in the road of surface friction coefficient μ=0.8 Lu Shang, vehicle accelerates by static, when car speed accelerates to 80km/h, carries out two-track line operation, remains a constant speed afterwards Linear motion.Wherein, it is contemplated that onboard sensor measurement error in reality, sensor measurement information vehicle wheel rotation angle speed is given respectively Degree ω_i, driving moment T_i, longitudinal acceleration a_x, lateral acceleration a_y, yaw acceleration r and vehicle front wheel angle δ_fAdd Amplitude is the zero-mean white noise of 0.0001.

Genetic algorithm (Genetic Algorithm, be called for short GA) be a class use for reference biosphere natural selection and natural genetic mechanism with Machine searching algorithm.Breeding, intersection and the gene mutation phenomenon occurred in genetic algorithm simulation natural selection and natural genetic process, The most all retain one group of candidate solution, and choose from Xie Qunzhong preferably individual by fitness function, utilize genetic operator These individualities are combined by (select, intersect and make a variation), produce the candidate solution group of a new generation, repeat this process, until meeting Till certain convergence index, concrete execution step is as shown in Figure 4.First genetic algorithm program is initialized, design parameter Be set to: population scale is 10, evolutionary generation be 20 elite numbers be 2, crossover probability is 0.8, and mutation probability is 0.2.

Fitness function is the unique criterion instructing the direction of search, and how selecting it is the key issue in GA.Carry out left front During wheel longitudinal tire force sliding mode observer parameter optimization, the present invention chooses estimation differenceAverage and mean square deviation make Evaluation index for parameter optimization.In order to make optimum results more accurately rationally, first according to formula (27), the two is normalized Process,

$G\left(x_i\right)=\frac{x_i-x_{min}}{x_{\max }-x_{min}}---\left(27\right)$

Wherein, G (x_i) ∈ [0 1], x_minAnd x_maxBeing respectively the minima in one group of data and maximum, the present invention selects to adapt to Degree function is as follows:

$\min J_{F_{x1}}={\mathrm{\Γ}}_{mx1}M\left(G\right({\tilde{F}}_{x1i}\left)\right)+{\mathrm{\Γ}}_{ex1}\mathrm{\σ}\left(G\right({\tilde{F}}_{x1i}\left)\right)---\left(28\right)$

Wherein, Γ_mx1And Γ_ex1It is average and the mean square deviation weight factor of the near front wheel longitudinal tire force error respectively, M () and D () point Wei not ask for the function of average and mean square deviation,

$M\left(y_i\right)=\frac{\underset{i=1}{\overset{N}{\Σ}}{y}_i}{N},\mathrm{\σ}\left(y_i\right)=\sqrt{\frac{\underset{i=1}{\overset{N}{\Σ}}{(y_i-{\stackrel{\‾}{y}}_i)}^2}{N}},---\left(29\right)$

Wherein, N is total number of single variable, and N=t/s, t are simulation time, and s is simulation step length.

Genetic algorithm specifically performs step as shown in Figure 4.During emulation, choose Γ_mx1=0.5, Γ_ex1=0.5, simulation time is 23s, Simulation step length is 0.01.The near front wheel longitudinal tire force sliding mode observer parameter optimization result based on genetic algorithm as it is shown in figure 5, During optimization, along with the increase of population algebraically, fitness function converges on a minima, obtains now by optimizing us Sliding formwork gain be L_x1=25.3026, feedback oscillator is ρ_x1=682.3490, corresponding fitness function value is 0.13376.

Longitudinal tire force and the axle lateral tire force observer parameter of its excess-three wheel are all optimized by said process, its Parameter optimization result is as shown in table 1.

2. using through the front axle longitudinal direction of parameter optimization and axle lateral tire force estimated value as input, use univariate mark GA-like Arithmetic carries out parameter optimization to vehicle yaw acceleration state observer.

For yaw rate observer Parametric optimization problem, needing the parameter optimized is K_r, when program parameter initializes, Each parameter with above-mentioned arrange identical.Shown in the fitness function such as formula (30) selected, wherein estimation differenceConventional high Under attachment two-track lineman's condition, its parameter optimization result is as shown in table 1.

$\min J_r={\mathrm{\Γ}}_{mr}M\left(G\right({\tilde{r}}_i\left)\right)+{\mathrm{\Γ}}_{er}\mathrm{\σ}\left(G\right({\tilde{r}}_i\left)\right)---\left(30\right)$

3. vehicle is indulged as input, employing multi-objective genetic algorithm through the longitudinal direction of parameter optimization and lateral tire force estimated value It is optimized to, lateral speed observer parameter, and obtains Pareto optimal solution set:

From the longitudinally, laterally speed observer form shown in formula (25), longitudinally and laterally both speed is mutually coupled, right Longitudinal speed will be using the estimated value of lateral speed as input when estimating, meanwhile, and also will be with when lateral speed is estimated The estimated value of longitudinal speed is as input.Therefore, to when longitudinally and laterally speed observer parameter is optimized, single goal Standard genetic algorithm is the most applicable.In view of the relation that influences each other between the two estimated value, the present invention utilizes Matlab workbox In gamultiobj function multi-objective optimization question is solved.Gamultiobj function uses controlled elite genetic algorithm, This algorithm be nondominated sorting genetic algorithm II (nondominated sorting genetic algorithm II, NSGA-II) variant.Its ultimate principle is: find the vector being made up of optimized variable in feasible zone so that one group is mutually rushed Prominent object function minimizes the most simultaneously, and limits Pareto by arranging optimum front end coefficient (Pareto Fraction) (Pareto) number individual on forward position (elite is individual), is converged on Pareto leading surface so that solving.Formula (31) is longitudinal direction and side Object function to speed observer parameter optimization

$\begin{array}{c}{f}_1=\min J_{v_x}={\mathrm{\Γ}}_{m}M\left(G\right({\tilde{v}}_{xi}\left)\right)+{\mathrm{\Γ}}_e\mathrm{\σ}\left(G\right({\tilde{v}}_{xi}\left)\right)\\ f_2=\min J_{v_y}={\mathrm{\Γ}}_{m}M\left(G\right({\tilde{v}}_{yi}\left)\right)+{\mathrm{\Γ}}_e\mathrm{\σ}\left(G\right({\tilde{v}}_{yi}\left)\right)\end{array}---\left(31\right)$

Multi-objective genetic algorithm parameter is set to: optimum front end coefficient is 0.3, and population scale is 50, and evolutionary generation is 50, hands over Fork probability is 0.8, mutation probability 0.2.Under conventional high attachment two-track lineman's condition, through the Pareto forward position that parameter optimization obtains As shown in Figure 6, the target function value of institute's optimized variable value and correspondence thereof is as shown in table 3:

Table 3 longitudinally/laterally speed state observer parameter optimization Pareto optimal solution

Sequence number	Kx	Ky	f1	f2
					1	0.0018	0.0099	0.363410	0.376690
2	0.0041	0.0096	0.368920	0.376650
					3	0.0043	0.0098	0.373136	0.376647
4	0.0044	0.0098	0.376021	0.376645
					5	0.0046	0.0094	0.379915	0.376643
6	0.0050	0.0094	0.388216	0.376638
					7	0.0053	0.0098	0.393134	0.376632
8	0.0057	0.0087	0.398318	0.376631
					9	0.0058	0.0095	0.400976	0.376628
10	0.0061	0.0092	0.404255	0.376626
					11	0.0064	0.0093	0.408570	0.376622
12	0.0070	0.0097	0.416050	0.376618
					13	0.0076	0.0091	0.422540	0.376616
14	0.0096	0.0098	0.435631	0.376608

As seen from Figure 6, two object functions are conflicting, and the reduction of one of them target function value then can cause separately The increase of one target function value, therefore, then needs to weigh two object functions in Pareto forward position, selects one group and is suitable for Solution.In table 3 it can be seen that in the 14 groups of optimal solutions enumerated, object function f₁Value relatively and object function f₂'s Changing greatly of value.Gap between its minima with maximum is the biggest, and therefore the present invention considers that emphatically excursion is bigger Object function f₁Value, select one group of object function f₁The less solution of value.The final solution selected is such as sequence number 1 institute in table 3 Show, i.e. K_x=0.0018, K_y=0.0099.

The off-line that of the present invention four-drive electric car state observer parameter optimization method of based on genetic algorithm is given below is imitated True checking.

In order to verify the effectiveness of observer parameter optimization method, first the parameter of institute's optimization in table 1 is input to modular In vehicle-state observer, and using the four-wheel driving electric vehicle in veDYNA as emulation vehicle.And it is double with conventional high attachment Shifting lineman's condition, as emulation operating mode, is verified and the vehicle-state observer estimation effect under identical operating mode during observer parameter optimization. Concrete experimental result and analysis are given below.

In view of the symmetry of Vehicular system, for longitudinal tire force, only provide the simulation result of left side longitudinal tire force.Fig. 7-9, For the simulation result figure under this operating mode.Fig. 7 be respectively the near front wheel and left rear wheel longitudinal tire force observer estimated result with VeDYNA output actual value correlation curve and its estimation difference.Fig. 8 respectively front-wheel and trailing wheel lateral tire force observer are estimated Meter result and veDYNA output actual value correlation curve and its estimation difference.Fig. 9 the most longitudinally/laterally speed and yaw angle speed Degree observer estimated result and veDYNA output actual value correlation curve and its estimation difference.By simulation result figure it can be seen that For estimated vehicle-state, the observer estimated value through parameter optimization can preferably be followed the tracks of the most defeated by veDYNA The actual value gone out, and have less estimation difference, this illustrates that observer parameter based on genetic algorithm proposed by the invention is excellent Change method has certain effectiveness.

Claims (4)

1. a four-drive electric car state observer parameter adjusting method based on genetic algorithm, it is characterised in that include with Lower step:

Step one, set up the Three Degree Of Freedom auto model of vehicle single-wheel roll modeling and simplification；

Step 2, modular vehicle state Observer Design: with vehicle sensors metrical information vehicle wheel rotation angular velocity and driving force Square, as input, uses the sliding mode observer method longitudinal tire force design longitudinal tire force observer to four wheels；Again with longitudinal direction Tire force estimated value, front wheel angle, lateral acceleration and yaw velocity, as input, separately design the lateral tire of axle Power sliding mode observer；Before finally with longitudinal and lateral tire force estimated value, longitudinal direction and lateral acceleration, yaw acceleration and vehicle Wheel corner is as input, design longitudinal direction speed, lateral speed and the car speed omnidirectional vision of yaw velocity；

Step 3, modular vehicle state observer based on step 2 design, use genetic algorithm respectively to each estimation module It is observed device parameter optimization.

A kind of four-drive electric car state observer parameter adjusting method based on genetic algorithm, its Being characterised by, vehicle single-wheel roll modeling and the Three Degree Of Freedom auto model of simplification that described step one is set up be:

1.1) vehicle single-wheel roll modeling:

$J{\stackrel{\·}{\mathrm{\ω}}}_i=-R_{eff}{F}_{xi}+T_i,(i=1,2,3,4),$

Wherein, J is the rotary inertia of wheel, and ω is the rotational angular velocity of each wheel, R_effFor the effective radius of tire, T is The driving moment of each wheel；

1.2) the Three Degree Of Freedom auto model simplified

Longitudinal direction of car and lateral tire force F_x、F_yWith vehicle around z-axis rotating torque M_zCan be expressed as:

F_x=(F_x1+F_x2)cosδ_f-(F_y1+F_y2)sinδ_f+F_x3+F_x4,

F_y=(F_x1+F_x2)sinδ_f+(F_y1+F_y2)cosδ_f+F_y3+F_y4,

M_z=l_F(F_x1+F_x2)sinδ_f+l_F(F_y1+F_y2)cosδ_f-l_R(F_y3+F_y4)

Wherein, F_xi/F_yi(i=1 ..., 4) it is respectively the longitudinal direction of four wheels and lateral tire force, δ is vehicle front wheel angle, l_FAnd l_RPoint It not that vehicle centroid is away from front axle and the distance of rear axle, unit m.

A kind of four-drive electric car state observer parameter adjusting method based on genetic algorithm, its Being characterised by, the modular vehicle state observer of described step 2 design specifically includes:

2.1) longitudinal tire force sliding mode observer, form is:

${\hat{F}}_{xi}=-\frac{J}{R_{eff}}{\mathrm{\ρ}}_{xi}\frac{{\mathrm{\ω}}_i-{\widehat{\mathrm{\ω}}}_i}{|{\mathrm{\ω}}_i-{\widehat{\mathrm{\ω}}}_i|+\mathrm{\φ}}-\frac{J}{R_{eff}}{L}_{xi}({\mathrm{\ω}}_i-{\widehat{\mathrm{\ω}}}_i)$

Wherein, L_xiIt is feedback oscillator, ρ_xiIt is sliding formwork gain, φ ＞ 0, ω_i(i=1,2,3,4) angle of rotation of four wheels it is respectively Speed, unit rad/s, J is the rotary inertia of wheel, units/kg m²；R_effFor the effective radius of tire, unit m.

2.2) lateral tire force sliding mode observer, including:

Front axle lateral tire force synovial membrane observer is:

${\hat{F}}_{yf}=\frac{I_z}{{\mathrm{cos\δ}}_f(l_F+l_R)}{\mathrm{\ρ}}_{yf}\frac{r-\hat{r}}{|r-\hat{r}|+\mathrm{\φ}}+\frac{I_z}{{\mathrm{cos\δ}}_f(l_F+l_R)}{L}_{yf}(r-\hat{r})$

Wherein, L_yfFor the feedback oscillator of front axle lateral tire force sliding mode observer, ρ_yfFor front axle lateral tire force sliding mode observer Sliding formwork gain；

Rear axle lateral tire force synovial membrane observer:

${\hat{F}}_{yr}=\frac{-I_z}{(l_F+l_R)}{\mathrm{\ρ}}_{yr}\frac{r-\hat{r}}{|r-\hat{r}|+\mathrm{\φ}}-\frac{I_z}{(l_F+l_R)}{L}_{yr}(r-\hat{r})$

Wherein, L_yrFor the feedback oscillator of rear axle lateral tire force sliding mode observer, ρ_yrFor rear axle lateral tire force sliding mode observer Sliding formwork gain；

2.3) car speed omnidirectional vision design:

Select longitudinal acceleration, lateral acceleration and yaw velocity to export as systematic survey, and by they and its estimated value it Difference, as the correction term of car speed observer, based on non-linear full micr oprocessorism structure, designs longitudinal direction of car, lateral speed And the car speed full micr oprocessorism of yaw velocity, expression formula is:

${\stackrel{\·}{\hat{V}}}_x=a_x+r{\hat{V}}_y+K_x(a_x-{\hat{a}}_x)$

${\stackrel{\·}{\hat{V}}}_y=a_y+r{\hat{V}}_x+K_y(a_y-{\hat{a}}_y)$

$\stackrel{\·}{\hat{r}}={\hat{M}}_z/I_z+K_r(r-\hat{r})$

Wherein, K_i(i=x, y, r) represent observer gain, utilizes tire force estimated value, can be by longitudinally, laterally acceleration estimation ValueAnd vehicle is around z-axis rotating torque estimated valueIt is expressed as:

${\hat{a}}_x=\frac{1}{m}\[({\hat{F}}_{x1}+{\hat{F}}_{x2})cos\mathrm{\δ}-({\hat{F}}_{y1}+{\hat{F}}_{y2})\sin \mathrm{\δ}+{\hat{F}}_{x3}+{\hat{F}}_{x4}\],$

${\hat{a}}_y=\frac{1}{m}\[({\hat{F}}_{x1}+{\hat{F}}_{x2})sin\mathrm{\δ}+({\hat{F}}_{y1}+{\hat{F}}_{y2})cos\mathrm{\δ}+{\hat{F}}_{y3}+{\hat{F}}_{y4}\],$

${\hat{M}}_z=a({\hat{F}}_{x1}+{\hat{F}}_{x2})sin\mathrm{\δ}+a({\hat{F}}_{y1}+{\hat{F}}_{y2})cos\mathrm{\δ}-b({\hat{F}}_{y3}+{\hat{F}}_{y4})$

Wherein, a_x、a_yIt is respectively longitudinal direction and lateral acceleration, unit m/s of vehicle²。

A kind of four-drive electric car state observer parameter adjusting method based on genetic algorithm, its It is characterised by, described step 3 uses genetic algorithm respectively each estimation module is observed device parameter optimization and include following step Rapid:

3.1) bivariant standard genetic algorithm is used the most respectively the longitudinal tire force sliding mode observer parameter of four, vehicle wheel to be entered Row optimizes, then using the front axle longitudinal tire force estimated value through parameter optimization as input tire force sliding mode observer lateral to front axle Parameter is optimized, and finally tire force sliding mode observer parameter lateral to rear axle is optimized

3.2) using through the front axle longitudinal direction of parameter optimization and axle lateral tire force estimated value as input, use univariate Standard genetic algorithm carries out parameter optimization to yaw rate state observer, and the observer parameter of optimization is input to Yaw rate observer；

3.3) using through the longitudinal direction of parameter optimization and lateral tire force estimated value as input, use multi-objective genetic algorithm to vehicle Longitudinally, laterally speed observer parameter is optimized, and obtains Pareto optimal solution set.