阅读说明：本技术 一种基于栈式自编码的主动悬架系统及其工作方法 (Active suspension system based on stack type self-coding and working method thereof ) 是由 孙玲玲 侯力文 牛宁 于 2018-07-13 设计创作，主要内容包括：本发明涉及一种基于栈式自编码的主动悬架系统及其工作方法。所述基于栈式自编码的主动悬架系统,包括控制单元和分别设置在四个车轮上的信号采集单元；所述信号采集单元包括,车身加速度传感器、车轮加速度传感器、轮速传感器、空气弹簧和磁流变减振器；控制单元与四个信号采集单元连接。本发明所述基于栈式自编码的主动悬架系统,将栈式自编码器用于汽车行驶模式的识别,实现汽车空气弹簧模式的选择和磁流变减震器模式的选择；根据路况寻优出最佳的调节因子,保证车辆的平顺性和操纵稳定性；节约能源的同时实现控制效果最优。(The invention relates to an active suspension system based on stack type self-coding and a working method thereof. The active suspension system based on the stack type self-coding comprises a control unit and signal acquisition units respectively arranged on four wheels; the signal acquisition unit comprises a vehicle body acceleration sensor, a wheel speed sensor, an air spring and a magneto-rheological shock absorber; the control unit is connected with the four signal acquisition units. According to the active suspension system based on the stacked self-coding, the stacked self-coder is used for identifying the automobile running mode, so that the selection of an automobile air spring mode and the selection of a magneto-rheological damper mode are realized; optimizing an optimal adjusting factor according to the road condition to ensure the smoothness and the operation stability of the vehicle; the control effect is optimal while the energy is saved.)

1. An active suspension system based on stack type self-coding is characterized by comprising a control unit and signal acquisition units respectively arranged on four wheels; the signal acquisition unit comprises a vehicle body acceleration sensor, a wheel speed sensor, an air spring and a magneto-rheological shock absorber; the control unit is connected with the four signal acquisition units.

2. The stack self-encoding based active suspension system of claim 1, wherein the signal acquisition unit is connected to the control unit sequentially via a charge amplifier and a data acquisition instrument.

3. A method of operating an active suspension system according to any one of claims 1-2, comprising the steps of:

1) constructing a stack type self-coding neural network;

1.1) training a first hidden layer; under different road conditions, the signals collected by the four signal collecting units are used as input signals to train the stacked self-encoder; the stacked self-encoder learns to obtain a first-order characteristic of an input signal;

1.2) training a second hidden layer; taking the first-order characteristic of an input signal as an input, and training a stacked self-encoder to obtain a second-order characteristic;

1.3) taking the second-order features as input of a softmax classifier layer, and training to obtain a model capable of mapping the second-order features to digital labels;

2) carrying out mode identification on the active suspension system through a stack type self-coding network; modes of the active suspension system include a comfort mode and a safety mode;

3) performing corresponding control according to the mode of the active suspension system;

3.1.1) in order to realize the control of a comfortable mode, a seven-freedom-degree vehicle model of a vehicle suspension system is established, wherein a vehicle body has three degrees of freedom of vertical movement, lateral movement and pitching movement, and four wheels respectively have one degree of freedom of vertical movement; wherein a and B are front and rear wheelbases, the front wheelbase is the distance from the vehicle center of mass to the front axle, the rear wheelbase is the distance from the vehicle center of mass to the rear axle, B is the wheelbase, z_sIs the vertical displacement of the car body, theta is the pitch angle of the car body,

3.1.2) in order to realize the control of the safe mode, a four-freedom-degree half-car model of a vehicle suspension system is established, wherein a car body has two degrees of freedom of vertical and lateral inclination, and wheels on the left side and the right side respectively have one degree of freedom of vertical motion; m is_sThe vehicle body mass; z is a radical of_sl、z_srRespectively displacement of the left side and the right side of the vehicle body; m is_tl、m_trThe vertical displacement of the left and right wheels is z_tl、z_tr；h_sDistance of center of mass of sprung mass to roll axis, q_l、q_rRespectively inputting road surfaces of wheels at the left side and the right side;k_sl、k_srspring rates of left and right suspension springs, c_sl、c_srDamping coefficients of the shock absorbers on the left side and the right side are respectively; h is the distance from the center of mass of the sprung mass to the ground;

3.2.1) for the air springs, when the identified mode is the comfortable mode, the vertical acceleration at the position of the mass center of the vehicle body is taken as an evaluation index, and the optimized parameter is the rigidity k of the four air springs_sf1、k_sf2、k_sr1、k_sr2When the spring stiffness optimization based on the differential evolution algorithm is carried out, the obtained system fitness function is as follows:

the constraints are as follows:

considering the dynamic load of the tire between the wheel and the road surface and the dynamic deflection of the suspension of the vehicle body; when the vehicle is running, the tires are prevented from leaving the ground, and the dynamic load of the tires and the road surface is smaller than the static load, i.e.

Wherein f is_ktf1、f_ktf2、f_ktr1、f_ktr2Four tire static loads; obtaining the static loads of the four tires according to a static force balance and moment balance equation of the vehicle as follows:

wherein g is the acceleration of gravity, m_tfMass of individual tires of the front row, m_trIndividual tire mass for the rear row;

moving the suspension by z due to the stroke limitation of the suspension mechanism_si-z_ti(i ═ f1, f2, r1, r2) are limited to a certain range; the damage to the comfort caused by the impact on the limiting block is avoided; the corresponding constraint conditions are:

wherein S is_maxLimiting the allowable working space of the vehicle suspension;

finally, the optimized four air spring stiffness is obtained and used as the spring stiffness of the comfort mode;

3.2.2) for the air spring, when the identified driving mode is the safe mode, the acceleration of the roll angle of the vehicle body is used as an evaluation index for optimization, and the optimization parameter is the rigidity k of the air spring_sf1、k_sf2、k_sr1、k_sr2When the spring stiffness optimization based on the differential evolution algorithm is carried out, the obtained system fitness function is as follows:

the constraint conditions comprise the constraint conditions of the step 3.2.1) and a rollover factor LTR of the vehicle;

finally, the optimized spring stiffness is the air spring stiffness in the safety mode;

3.2.3) for the magnetorheological damper, when the identified running mode is the comfortable mode, simultaneously considering the riding comfort and the operation stability of the vehicle;

the ceiling damping control is realized within a certain range by adopting an equivalent method;

in the formula, F_siThe damping force of the ceiling is used as the damping force,

using equivalent method to make F_MRRealizing the damping control of the ground shed in a certain range;

the ground shed control strategy is as follows:

in the formula: f_giDamping force of the ground shed, beta_giIs the damping coefficient of the ground shed;

the hybrid damping control strategy integrates the characteristics of ceiling control and ground ceiling control, and gives consideration to the smoothness and the operation stability; the comprehensive control force output is as follows:

F_h＝αF_s+(1-α)F_g

in the formula: f_hThe output force of the magneto-rheological shock absorber is shown, and alpha is an adjusting factor;

and (3) obtaining an optimal regulating factor by adopting a differential evolution algorithm, and setting a system fitness function as follows when the parameter of the regulating factor is optimized:

in the formula (I), the compound is shown in the specification,in order to optimally control the root mean square value of the vertical acceleration of the sprung mass and the root mean square value of the vertical acceleration of the unsprung mass,

3.2.4) for the magneto-rheological damper, when the identified running mode is a safe mode, the control of the vehicle side inclination is realized by generating an additional moment by utilizing the sliding mode variable structure control;

control of rolling movement of vehicles by magneto-rheologicalThe shock absorber realizes the continuous adjustment of the semi-active suspension damping and generates an additional moment M for preventing the vehicle from rolling_R(ii) a The equation of motion is:

I_xxis the rolling moment of inertia of the vehicle, a_yIs a set frame lateral acceleration; at a side inclination angle

defining an integral sliding mode surface:

in the formula (I), the compound is shown in the specification,

further:

in the formula, k₁And k₂Is a non-zero positive integer, if the system output roll angle is desired

if the sliding mode control is in an ideal state, the sliding mode control is carried out

meanwhile, combining the formula (1) to obtain:

wherein the content of the first and second substances,in order to ensure the dynamic quality of the sliding mode, an exponential approach law is selected as an approaching condition:

wherein epsilon and c are both positive constants;

combining formulas (3), (6) and (7), obtaining:

it follows that the additional moment required to bring the suspension back to the desired roll angle is:

control of the roll f_MRlAnd f_MRrThe damping forces of the left and right magneto-rheological shock absorbers are respectively, and the specific decision process is as follows:

A) when in useAnd is

B) when in use

C) when in use

D) when in use

by the control strategy, the output damping force of the front and rear magneto-rheological shock absorbers in the safe mode can be obtained.

4. The method for operating an active suspension system according to claim 3, wherein in step 2), the comfort mode and the safety mode are defined by using carsim software to set simulation conditions;

① a double-moving-line simulation working condition with the vehicle speed of 10 km/h-100 km/h, the speed interval of 1km/h, a control mode of a safe mode when the speed is more than or equal to 30km/h, and a control mode of a comfortable mode when the speed is less than 30 km/h;

② the speed is 10 km/h-100 km/h highway simulation working condition, the speed interval is 1km/h, at this moment, the control mode is comfort mode;

③ the speed is 10 km/h-100 km/h fishhook simulation working condition, the speed interval is 1km/h, the control mode is safe mode when the speed is more than or equal to 30km/h, and the control mode is comfortable mode when the speed is less than 30 km/h.

Technical Field

The invention relates to an active suspension system based on stack type self-coding and a working method thereof, belonging to the technical field of control of an automobile active suspension system.

Background

Safe, fast and comfortable driving is a generally pursued target of automobiles, and frequent traffic safety accidents and consequent active safety problems are always important social concerns. Riding comfort and operating stability are two important performance indexes of the automobile. At present, people mainly aim at riding comfort in the research on automobile suspensions. Under special conditions, the suspension becomes an important actuator for active safety, particularly for improving roll stability. In designing an active suspension, it is considered to suppress vibration of sprung mass and wheel bounce, while suppressing roll motion of the vehicle, to improve ride comfort and handling stability of the automobile. The existing active suspension can not realize real-time switching control of control modes under different road conditions, wastes energy and can not enable the performance to reach the optimum.

Stacked self-coding is one type of deep learning, an unsupervised learning algorithm that applies back-propagation to set a target value as an input. The encoder attempts to learn an approximation of the identity function so that the output is similar to the input. The encoding steps of each layer are run in forward order, giving the encoding steps of a stacked auto-encoder, and the decoding steps are run in reverse order, giving the decoding stack of each auto-encoder. In this way, the activation value of the deepest hidden unit can be used as the characteristic of the classifier, and the stacked self-coding neural network is applied to the reconstruction data identification classification.

The stack type self-coding neural network is a neural network model formed by multiple layers of sparse self-encoders, namely the output of a previous self-encoder is used as the input of a next self-encoder.

And training the hidden layer of the first layer, namely the output of the hidden layer of the first self-encoder is used as the second layer, namely the input of the second self-encoder, so as to train the weight and the bias of the second self-encoder, and sequentially, so that the parameters in the stacked self-encoding neural network can be trained.

Disclosure of Invention

Aiming at the defects of the prior art, the invention provides an active suspension system based on stacked self-coding.

The invention also provides a working method of the active suspension system.

The technical scheme of the invention is as follows:

an active suspension system based on stack type self-coding comprises a control unit and signal acquisition units respectively arranged on four wheels; the signal acquisition unit comprises a vehicle body acceleration sensor, a wheel speed sensor, an air spring and a magneto-rheological shock absorber; the control unit is connected with the four signal acquisition units.

The wheel acceleration sensor measures the vertical acceleration of the unsprung mass, namely the vertical acceleration of a wheel below the suspension; the vehicle body acceleration sensor measures the vertical acceleration of the sprung mass, namely the vertical acceleration of the vehicle body above the suspension; the wheel speed sensor measures the rotating speed of the wheel;

according to the invention, the signal acquisition unit is preferably connected with the control unit sequentially through the charge amplifier and the data acquisition instrument.

A working method of the active suspension system comprises the following steps:

1) constructing a stack type self-coding neural network;

1.1) training a first hidden layer; under different road conditions, the signals collected by the four signal collecting units are used as input signals to train the stacked self-encoder; the stacked self-encoder learns to obtain a first-order characteristic of an input signal;

1.2) training a second hidden layer; taking the first-order characteristic of an input signal as an input, and training a stacked self-encoder to obtain a second-order characteristic;

1.3) taking the second-order features as input of a softmax classifier layer, and training to obtain a model capable of mapping the second-order features to digital labels;

the stack type self-coding network is used for identifying the mode of the active suspension system;

2) carrying out mode identification on the active suspension system through a stack type self-coding network; modes of the active suspension system include a comfort mode and a safety mode;

3) performing corresponding control according to the mode of the active suspension system;

3.1.1) to achieve control of comfort mode, seven degree of freedom vehicles with vehicle suspension systems are builtThe model is characterized in that the vehicle body has three degrees of freedom of vertical direction, side inclination and pitching, and four wheels respectively have one degree of freedom of vertical motion; wherein a and B are front and rear wheelbases, the front wheelbase is the distance from the vehicle center of mass to the front axle, the rear wheelbase is the distance from the vehicle center of mass to the rear axle, B is the wheelbase, z_sIs the vertical displacement of the car body, theta is the pitch angle of the car body,

is the vehicle body roll angle, z_sf1、z_sf2、z_sr1、z_sr2Is the vertical displacement, z, of the sprung mass_tf1、z_tf2、z_tr1、z_tr2Is the unsprung mass vertical displacement, k_sf1、k_sf2、k_sr1、k_sr2Spring rate of air spring, c_sf1、c_sf2、c_sr1、c_sr2Damping coefficient, k, for magnetorheological vibration dampers_tf、k_trThe rigidity of the front and rear tires, q_f1、q_f2、q_r1、q_r2Inputting for the road surface;

3.1.2) in order to realize the control of the safe mode, a four-freedom-degree half-car model of a vehicle suspension system is established, wherein a car body has two degrees of freedom of vertical and lateral inclination, and wheels on the left side and the right side respectively have one degree of freedom of vertical motion; m is_sThe vehicle body mass; z is a radical of_sl、z_srRespectively displacement of the left side and the right side of the vehicle body; m is_tl、m_trThe vertical displacement of the left and right wheels is z_tl、z_tr；h_sDistance of center of mass of sprung mass to roll axis, q_l、q_rRespectively inputting road surfaces of wheels at the left side and the right side; k is a radical of_sl、k_srSpring rates of left and right suspension springs, c_sl、c_srDamping coefficients of the shock absorbers on the left side and the right side are respectively; h is the distance from the center of mass of the sprung mass to the ground;

3.2.1) for the air spring, when the identified mode is the comfortable mode, the vertical acceleration at the position of the mass center of the vehicle body is taken as an evaluation index, and parameters are optimizedStiffness k for four air springs_sf1、k_sf2、k_sr1、k_sr2When the spring stiffness optimization based on the differential evolution algorithm is carried out, the obtained system fitness function is as follows:

in order to optimize the root mean square value of the vertical velocity of the sprung mass,in order to optimize the root mean square value of the vertical acceleration of the sprung mass,

is the root mean square value of the sprung mass vertical velocity of the passive suspension system,

the mean square root value of the sprung mass vertical acceleration of the passive suspension system is obtained; the sprung mass vertical velocity is obtained by derivation of the sprung mass vertical displacement; the vertical acceleration of the sprung mass is obtained by derivation of the vertical speed of the sprung mass;

the constraints are as follows:

considering the dynamic load of the tire between the wheel and the road surface and the dynamic deflection of the suspension of the vehicle body; when the vehicle is running, the tires are prevented from leaving the ground, and the dynamic load of the tires and the road surface is smaller than the static load, i.e.

Wherein f is_ktf1、f_ktf2、f_ktr1、f_ktr2Four tire static loads; obtaining the static loads of the four tires according to a static force balance and moment balance equation of the vehicle as follows:

wherein g is the acceleration of gravity, m_tfMass of individual tires of the front row, m_trIndividual tire mass for the rear row;

moving the suspension by z due to the stroke limitation of the suspension mechanism_si-z_ti(i ═ f1, f2, r1, r2) are limited to a certain range; the damage to the comfort caused by the impact on the limiting block is avoided; the corresponding constraint conditions are:

wherein S is_maxLimiting the allowable working space of the vehicle suspension; s_max＝0.1m。

Finally, the optimized four air spring stiffness is obtained and used as the spring stiffness of the comfort mode;

3.2.2) for the air spring, when the identified driving mode is the safe mode, the acceleration of the roll angle of the vehicle body is used as an evaluation index for optimization, and the optimization parameter is the rigidity k of the air spring_sf1、k_sf2、k_sr1、k_sr2When the spring stiffness optimization based on the differential evolution algorithm is carried out, the obtained system fitness function is as follows:

in order to optimize the root mean square value of the side dip angle of the rear vehicle body,in order to optimize the root mean square value of the acceleration of the roll angle of the rear vehicle body,

is the vehicle body roll angle root mean square value of the passive suspension system,

the mean square root value of the acceleration of the vehicle body roll angle of the passive suspension system is obtained;

the constraint conditions comprise the constraint conditions of the step 3.2.1) and a rollover factor LTR of the vehicle; the rollover factor LTR is the ratio of the absolute value of the difference between the vertical loads of the left and right tires to the sum of the total vertical loads of the tires;

when the loads of the left tire and the right tire are equal, the value of LTR is 0, and the vehicle is in a safe state; when the rolling occurs, the vertical load of the wheels on two sides is transferred, and when one side of the wheel is away from the ground, LTR is 1; in order to prevent the vehicle from rolling over, the wheels must be prevented from moving off the ground. Once the wheels leave the wheel surfaces, the stable control is difficult to carry out and the vehicle is ensured not to turn over;

finally, the optimized spring stiffness is the air spring stiffness in the safety mode;

3.2.3) for the magnetorheological damper, when the identified running mode is the comfortable mode, simultaneously considering the riding comfort and the operation stability of the vehicle;

the classical ceiling damping control strategy aims at inhibiting the vibration of sprung mass and improving the riding comfort of the automobile, and the hybrid damping control strategy integrates the characteristics of ceiling control and ground ceiling control and gives consideration to the smoothness and the operation stability.

The ideal ceiling damping force is:

but the ceiling damping is an ideal model and cannot be applied to the invention;

the ceiling damping control is realized within a certain range by adopting an equivalent method;

in the formula, F_siThe damping force of the ceiling is used as the damping force,

is the vertical velocity of the sprung mass,

is the vertical velocity of the unsprung mass, z_siIs the vertical displacement of the sprung mass, z_tiIs the vertical displacement of the unsprung mass, F_imaxThe maximum adjustable force is the maximum adjustable force of the magneto-rheological shock absorber; beta is a_siOptimizing and determining a ceiling damping coefficient according to suspension parameters; the sprung mass vertical velocity is obtained by integrating an acceleration signal measured by a sprung mass acceleration sensor; the sprung mass displacement is obtained by performing secondary integration on an acceleration signal measured by a sprung mass acceleration sensor;

in order to restrain wheel jump and improve the operation stability of the automobile, the ground shed control is necessarily introduced on the basis of the ceiling control, and ideal ground shed damping force is as follows:

the ground shed damping is also an ideal model, and can not be directly applied to the invention;

using equivalent method to make F_MRRealizing the damping control of the ground shed in a certain range;

the ground shed control strategy is as follows:

in the formula: f_giDamping force of the ground shed, beta_giIs the damping coefficient of the ground shed;

the hybrid damping control strategy integrates the characteristics of ceiling control and ground ceiling control, and gives consideration to the smoothness and the operation stability; the comprehensive control force output is as follows:

F_h＝αF_s+(1-α)F_g

in the formula: f_hThe output force of the magneto-rheological shock absorber is shown, and alpha is an adjusting factor;

and (3) obtaining an optimal regulating factor by adopting a differential evolution algorithm, and setting a system fitness function as follows when the parameter of the regulating factor is optimized:

in the formula (I), the compound is shown in the specification,

in order to optimally control the root mean square value of the vertical acceleration of the sprung mass and the root mean square value of the vertical acceleration of the unsprung mass,

the mean square root value of the vertical acceleration of the sprung mass and the mean square root value of the vertical acceleration of the unsprung mass of the passive suspension system are obtained; the constraint condition is the constraint condition of the step 3.2.1); after the adjustment factor is obtained, the output damping force of the magneto-rheological shock absorber is finally obtained; the passive suspension system is a suspension system with constant spring stiffness and damping coefficient;

3.2.4) for the magneto-rheological damper, when the identified running mode is a safe mode, the control of the vehicle side inclination is realized by generating an additional moment by utilizing the sliding mode variable structure control;

the roll motion of the vehicle is controlled, the continuous adjustment of the semi-active suspension damping is realized through the magneto-rheological shock absorber, and the additional moment M for preventing the roll of the vehicle is generated_R(ii) a The equation of motion is:

I_xxis the rolling moment of inertia of the vehicle, a_yIs a set frame lateral acceleration; at a side inclination angle

For control purposes, assume a desired system output roll angle of

The tracking error e_rrComprises the following steps:

defining an integral sliding mode surface:

in the formula (I), the compound is shown in the specification,

for tracking error e_rrThe first derivative of (a);

further:

in the formula, k₁And k₂Is a non-zero positive integer, if the system output roll angle is desired

Then, derivation is performed on the sliding mode surface s to obtain:

if the sliding mode control is in an ideal state, the sliding mode control is carried out

Thus, can obtain

By adjusting k₁And k₂Value of (2), make tracking errorDifference (D)

Approaching to zero;

meanwhile, combining the formula (1) to obtain:

wherein the content of the first and second substances,

in order to ensure the dynamic quality of the sliding mode, an exponential approach law is selected as an approaching condition:

wherein epsilon and c are both positive constants;

combining formulas (3), (6) and (7), obtaining:

it follows that the additional moment required to bring the suspension back to the desired roll angle is:

the additional moment obtained in equation (9) is an ideal additional moment for adjusting the roll angle of the sprung mass to a desired value; for the magneto-rheological shock absorber, the additional moment is realized by adjusting the magnitude and the difference of the left damping force and the right damping force, and the additional damping moment is limited by two factors: the speed of relative motion of the left and right sprung and unsprung masses; secondly, adjusting the damped exciting current; when the desired additional moment M is obtained_ROn the basis, firstly, the output damping force of each shock absorber is decided, and then expected exciting current is obtained according to the characteristics of the magneto-rheological shock absorber;

control of the roll f_MRlAnd f_MRrAre respectively provided withThe specific decision process for the damping force of the left and right magneto-rheological shock absorbers is as follows:

A) when in use

And is

Meanwhile, if the right magnetorheological shock absorber outputs the additional damping force, the anti-roll moment is weakened, so that:

B) when in use

And is

In the time, if the left and right magneto-rheological shock absorbers output the additional damping force at the same time, the anti-roll moment is generated, so that:

C) when in useAnd is

In the time, if the left and right magneto-rheological shock absorbers output additional damping force, the roll degree is increased, so that:

D) when in use

And is

In time, if the left magnetorheological shock absorber outputs the additional damping force, the anti-roll moment is weakened, so that:

by the control strategy, the output damping force of the front and rear magneto-rheological shock absorbers in the safe mode can be obtained.

According to the present invention, in step 2), the comfort mode and the safety mode are defined by using carsim software to set simulation conditions;

① a double-moving-line simulation working condition with the vehicle speed of 10 km/h-100 km/h, the speed interval of 1km/h, a control mode of a safe mode when the speed is more than or equal to 30km/h, and a control mode of a comfortable mode when the speed is less than 30 km/h;

② the speed is 10 km/h-100 km/h highway simulation working condition, the speed interval is 1km/h, at this moment, the control mode is comfort mode;

③ the speed is 10 km/h-100 km/h fishhook simulation working condition, the speed interval is 1km/h, the control mode is a safe mode when the speed is more than or equal to 30km/h, and the control mode is a comfortable mode when the speed is less than 30 km/h;

and subsequently, a double-line-shifting simulation working condition of 25 km/h-35 km/h, a fishhook simulation working condition and an expressway simulation working condition are used as test data sets to test the identification and classification effects of the stacked self-coding neural network.

The trained and tested stack type self-coding neural network is applied to vehicle state recognition, and a control mode can be selected in real time according to a driving state.

The invention has the beneficial effects that:

1. according to the active suspension system based on the stacked self-coding, the stacked self-coder is used for identifying the automobile running mode, so that the selection of an automobile air spring mode and the selection of a magneto-rheological damper mode are realized; optimizing an optimal adjusting factor according to the road condition to ensure the smoothness and the operation stability of the vehicle; the control effect is optimal while the energy is saved;

2. the active suspension system based on the stack type self-coding is characterized in that a switching controller is designed through a pattern recognition function of a neural network, so that real-time switching control of an active suspension control mode is realized; and more complex model data is predicted through deep learning, and the defect that a shallow neural network is easy to converge to a local minimum value is effectively overcome.

Drawings

FIG. 1 is a schematic structural view of a vibration model of a semi-vehicle according to the present invention;

FIG. 2 is a schematic view of the construction of the model of the rolling half-car of the present invention;

FIG. 3 is a block diagram of a stacked self-encoding network suspension pattern recognition network training according to the present invention;

FIG. 4 is a seven-degree-of-freedom vehicle dynamics model;

FIG. 5 is a four degree-of-freedom vehicle roll dynamics model;

FIG. 6 is a double shift line work condition simulation setup;

FIG. 7 is a working condition simulation setup of a fishhook;

FIG. 8 is a setting of highway condition road roughness;

FIG. 9 is a control flow diagram of a stacked self-encoding mode recognition system;

wherein, 1, a charge amplifier; 2. a vehicle body acceleration sensor; 3. an air spring; 4. a wheel acceleration sensor; 5. an air spring; 6. a magnetorheological damper; 7. a data acquisition instrument; 8. and a wheel speed sensor.

Detailed Description

The invention is further described below, but not limited thereto, with reference to the following examples and the accompanying drawings.