CN114675537B - Vehicle suspension system model predictive control method based on road condition monitoring - Google Patents
Vehicle suspension system model predictive control method based on road condition monitoring Download PDFInfo
- Publication number
- CN114675537B CN114675537B CN202210312774.2A CN202210312774A CN114675537B CN 114675537 B CN114675537 B CN 114675537B CN 202210312774 A CN202210312774 A CN 202210312774A CN 114675537 B CN114675537 B CN 114675537B
- Authority
- CN
- China
- Prior art keywords
- suspension
- control
- future
- data
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 239000000725 suspension Substances 0.000 title claims abstract description 108
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000012544 monitoring process Methods 0.000 title claims abstract description 22
- 238000003062 neural network model Methods 0.000 claims abstract description 33
- 238000004364 calculation method Methods 0.000 claims abstract description 14
- 238000012549 training Methods 0.000 claims abstract description 7
- 239000013598 vector Substances 0.000 claims description 43
- 230000001133 acceleration Effects 0.000 claims description 22
- 230000008859 change Effects 0.000 claims description 21
- 230000000875 corresponding effect Effects 0.000 claims description 16
- 210000002569 neuron Anatomy 0.000 claims description 16
- 238000013016 damping Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 8
- 230000005484 gravity Effects 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 101100001674 Emericella variicolor andI gene Proteins 0.000 claims 1
- 230000007246 mechanism Effects 0.000 abstract description 18
- 238000013459 approach Methods 0.000 abstract description 2
- 238000006073 displacement reaction Methods 0.000 description 8
- 238000011161 development Methods 0.000 description 2
- 239000012530 fluid Substances 0.000 description 2
- 238000005096 rolling process Methods 0.000 description 2
- 238000004590 computer program Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000007599 discharging Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
- G06F17/13—Differential equations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Computational Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Algebra (AREA)
- Operations Research (AREA)
- Medical Informatics (AREA)
- Automation & Control Theory (AREA)
- Life Sciences & Earth Sciences (AREA)
- Databases & Information Systems (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Vehicle Body Suspensions (AREA)
Abstract
The invention provides a vehicle suspension system model predictive control method based on road condition monitoring, which takes random road condition data and random control signals as the input of an active suspension system model and takes the discrete state of the system as the output, so that a large amount of training data is generated offline to train a neural network model, and the model can approach the mechanism model of the active suspension system with high precision; the trained neural network model is used for replacing a mechanism model of an active suspension system, the road condition monitoring data and the feasible solution of the minimization problem, which are given by the laser radar, are used as the neural network model to be input, the discrete predicted value of the system state is calculated and used for solving the minimization problem, and therefore the optimal control signal is obtained to control the electro-hydraulic system to generate corresponding control force to maintain the stability of the vehicle body. Compared with a mechanism model directly used, the neural network model can exert the advantages of strong nonlinear approximation capability and rapid calculation capability, thereby realizing high-speed operation and high-frequency implementation control of model predictive control.
Description
Technical Field
The invention relates to the technical field of vehicle suspension systems, in particular to a vehicle suspension system model predictive control method based on road condition monitoring.
Background
Today, with rapid development, the frequency of use of vehicles is gradually increased, and passengers desire more and more comfort for driving the vehicles. Vibration reduction is clearly an essential link for improving the riding experience of passengers. The suspension system is an important part of the vehicle underbody construction and has the main functions of bearing the dynamic load of the vehicle body, reducing the vibration impact of the road surface unevenness excitation on the vehicle body, and keeping the tires in good contact with the road surface as much as possible so as to improve the performance of the vehicle. Therefore, the suspension system is optimized and researched, so that the comfort, stability and safety of the vehicle can be effectively improved, and favorable conditions are provided for the development of modern vehicles. With the improvement of life quality of people, the requirements of people on automobiles are more focused on the driving smoothness of the automobiles.
With the proposal of the idea of active suspension, the control of the active suspension to ensure the running stability of the automobile has been paid attention to by a plurality of automobile industries. The rigidity and the damping of the active suspension can be adjusted in real time along with the change of an external input source, so that the impact on a human body under different working conditions can be effectively restrained. The control of the active suspension depends on real-time monitoring of road condition information, which is obtained from a laser radar arranged in front of the automobile. Therefore, the road condition information data obtained by the laser radar provides an input source for actively changing the control force for the control system of the vehicle active suspension. In the existing advanced control technology scheme, model predictive control can calculate control signal output from the viewpoint of optimizing future states, and has been successfully and widely applied to various industries. The optimization idea of model predictive control is widely accepted and appreciated by engineering technicians.
However, model predictive control requires a solution minimization problem of real-time reciprocation iteration, and when calculating an objective function value, a mechanism model of a control object needs to be continuously called to predict a system state for a period of time in the future. Since the control requirements of the active suspension system of the vehicle respond rapidly, if the control of the active suspension is implemented by applying the model predictive control concept, the on-line calculation load problem of the mechanism model of the suspension system must be considered. The number of differential equations is large and the nonlinearity degree is obvious from the view of the mechanism model of the active suspension system, so that the use of the mechanism model of the active suspension to predict the future system state tends to bring a large calculation amount. Therefore, how to reduce the predicted calculation amount of the future system state is a key to effectively exert the advantages of the model predictive control technique.
Disclosure of Invention
The technical problems to be solved by the invention are as follows: the invention provides a vehicle suspension system model predictive control method based on road condition monitoring, which is characterized in that random road condition data and random control signals are used as input of an active suspension system model, and discrete states of the system are used as output, so that a large amount of training data is generated offline for training the neural network model; and secondly, the trained neural network model is used for replacing a mechanism model of an active suspension system, the feasible solution of the minimization problem is used as the neural network model input according to road condition monitoring data given by a laser radar, and a discrete predicted value of the system state is calculated and used for solving the minimization problem, so that an optimal control signal is obtained to control an electro-hydraulic system to generate corresponding control force to maintain the stability of a vehicle body.
The technical scheme adopted for solving the technical problems is as follows: a vehicle suspension system model predictive control method based on road condition monitoring comprises the following steps:
Step 1: the method comprises the steps of determining state variables of an active suspension system, establishing a state space model of the vehicle active suspension system, determining a maximum value u max and a minimum value u min of a control signal change range, and determining a single-step change maximum value delta u max of the control signal.
The control signal is directly applied to the electro-hydraulic system telling the electro-hydraulic system to provide the same force. Since any hardware device has upper and lower limits on the operating range, the force that the electro-hydraulic system can provide is also maximum and minimum, i.e., the control signal maximum u max and minimum u min. In addition, since the electro-hydraulic system generates force by fluid squeezing, there is also a range of force generated by charging and discharging fluid in a single step time. The maximum and minimum values of these ranges are all technical indicators of the electro-hydraulic system hardware itself.
Because the rolling motion and the pitching motion of the four-wheel vehicle have the same motion law, in practical application, a corresponding mechanism model is usually built aiming at a half vehicle suspension system by utilizing Newton's mechanical law, and the basic principle of building the mechanism model is that according to the Newton's mechanical law, namely: the mass multiplied by the acceleration equals the resultant force. Therefore, the mechanism model firstly established by the method is specifically shown as a formula ①:
Wherein, when the index i is equal to 1, the front suspension is represented; when i is equal to 2, the rear suspension is represented; represents the vertical acceleration of the suspension, the displacement variation of the suspension/> X i represents the vertical displacement of the suspension, b i represents the horizontal distance between the center of gravity of the suspension and the center of gravity of the vehicle body,/>Representing body pitch angle (when the body is level,/>) X 0 represents the vertical displacement of the vehicle body,/> Representing the vertical speed of the vehicle body,/>Representing pitch rate of vehicle body,/>Represents the vertical speed of the vehicle body, u i represents the acting force (i.e. control signal) of the electro-hydraulic system, the vertical deformation displacement deltay i＝yi-xi,yi of the tire represents the vertical road surface protrusion height of the suspension (y i is less than 0 if the road surface is concave), and v-Representing pitch acceleration of the vehicle body,/>Representing the vertical acceleration of the vehicle body; m and M represent the unsprung and sprung masses, respectively, of the suspension system, I 0 represents moment of inertia, spring force F s(Δxi) of the suspension and damping force/>The calculation mode of (2) is as follows:
In ② above, when When, sign function/>When/>When, sign function/>K lin and k non represent the linear and nonlinear spring coefficients of the suspension springs, respectively, and c lin,csys and c non represent the linear and nonlinear damping coefficients of the suspension damper, respectively, with the reference numerals i=1, 2. Similarly, the elastic force T s(Δyi) generated by the tire due to deformation is calculated as follows:
Ts(Δyi)＝ζlin·Δyi+ζnon·(Δyi)3 ③
In the above formula, ζ lin and ζ non represent the linear elastic coefficient and the nonlinear elastic coefficient of the tire, respectively; i=1, 2, i is equal to 1, and represents the front suspension, and i is equal to 2, and represents the rear suspension.
Since i is equal to 1 or 2, the mechanism model in equation ① is actually composed of 4 differential equations, 8 state variables θ 1,θ2,…,θ8 can be determined, which in turn represent respectivelyThe following state space model can be established for the active suspension system, and the model specifically comprises 8 state equations, namely:
Wherein Δθ 1,Δθ2,…,Δθ8 represents the variation corresponding to the 8 state variables θ 1,θ2,…,θ8, respectively, Δt represents the single step time, and the index i=1, 2.
As can be seen from the formula ②, F s(θ7+b1·θ4-θ3) in the formula ④ represents the spring force of the suspension, F d(θ5+b1·θ2-θ1) represents the damping force of the suspension, and the spring force F s(θ7+b1·θ4-θ3) and the damping force F d(θ5+b1·θ2-θ1) of the suspension are calculated as follows:
In formula ⑤, sign (θ 5+b1·θ2-θ1) =1 when θ 5+b1·θ2-θ1 is not less than 0; when θ 5+b1·θ2-θ1 < 0, sign (θ 5+b1·θ2-θ1) = -1.
Similarly, as can be seen from the equation ③, the elastic force T s(yi-θ4+i) generated by the tire due to deformation is calculated as follows:
Ts(yi-θ4+i)＝ζlin·(yi-θ4+i)+ζnon·(yi-θ4+i)3 ⑥
in the above formula ⑥, ζ lin and ζ non represent the linear elastic coefficient and the nonlinear elastic coefficient of the tire, respectively; i=1, 2, i is equal to 1, and represents the front suspension, and i is equal to 2, and represents the rear suspension.
Step 2: n sets of input data and N sets of output data are generated according to steps 2.1 to 2.4.
Step 2.1: initializing the number of data groups j=1, setting 8 state variables θ 1,θ2,…,θ8 to be equal to 0, and setting the initial value of the control signal u i
Step 2.2: randomly generating a vertical road surface protrusion height y 1 of the front suspension and a vertical road surface protrusion height y 2 of the rear suspension from the interval [ -0.02,0.02], respectively, andAfter the control signal u 1 of the front suspension and the control signal u 2 of the rear suspension are randomly generated, 12 data in the j-th group of input data are set to be equal to theta 1,θ2,…,θ8,y1,y2,u1,u2 in sequence.
Step 2.3: delta theta 1,Δθ2,…,Δθ8 is calculated using a state space model of the active suspension system, and then according to the formula theta b＝θb+Δθb, b=1, 2, 3..8, after updating the state variable theta 1,θ2,…,θ8, 8 data in the j-th set of output data are sequentially set equal to theta 1,θ2,…,θ8, respectively.
Step 2.4: judging whether the data group number j is smaller than N; if yes, setting j=j+1, and returning to the step 2.2 to continue generating data; if not, N groups of input data and N groups of output data are obtained.
It is noted that, since the single step time Δt in the formula ④ is known, according to the state equation for Δθ 1/Δt in the formula ④, Δθ 1 is calculated as follows:
Similarly, Δθ 2,Δθ3,…,Δθ8 can be calculated sequentially from the 2 nd to 8 th state equations in the equation ④.
Step 3: establishing a neural network model, determining the number of neurons in the middle layer of the neural network model as H, and training the neural network model by utilizing N groups of input data and N groups of output data to obtain parameters of the neural network model, wherein the parameters specifically comprise: h reference vectors for middle layer neuronsThe H smoothing factors δ 1,δ2,…,δH of the middle layer neurons output 8 coefficient vectors w 1,w2,…,w8 of the neurons.
Step 4: when the vehicle runs, the laser radar arranged in front of the vehicle is utilized to monitor road condition data in front of the vehicle in real time, and the monitoring values of the protrusion heights of the vertical road surface at L control moments in the future are obtained according to the speed conversionAndWherein/>Respectively representing the monitoring value of the vertical road surface protrusion height of the front suspension at the 1 st control moment in the future, the 2 nd control moment in the future, … th control moment in the future; /(I)The monitoring values of the vertical road surface protrusion height of the rear suspension at the future 1 st control moment, the future 2 nd control moment, … nd control moment and the future L th control moment are respectively represented.
Step 5: solving the minimization problem with constraint conditions according to a formula ⑧, thereby obtaining predicted values of the control signal u i at L control moments in the future
Wherein the range of the weight coefficient lambda is 0 < lambda < 1, and the change value of the vertical acceleration at the alpha-th control moment in the futureVariation value/>, of pitch angle acceleration at future alpha-th control momentThe value of the change in the control signal u i at the future alpha-th control time/>Forgetting factor mu (α) satisfies the condition mu (1)≥μ(2)≥…≥μ(L),/>And/>Correspond to predicted values representing state variables θ 1 and θ 2, respectively, at future α -th control time,/>And/>Corresponds To the predicted values of the control signals u 1 and u 2 at the future alpha-th control moment, respectively, s.t. represents the abbreviation of the constraint English word Subject To,/>Represents an arbitrary setting of α=1, 2, …, L; when a=1 is used for the purpose of providing,And/>Initial values representing state variables θ 1 and θ 2, respectively,/>And/>The initial values of the control signals u 1 and u 2 are respectively indicated.
In the process of solving the minimization problem in the formula ⑧, a set of predicted values of the control signal are obtained during each optimization iteration, namely: combined with the monitoring value/>, of the height of the vertical pavement protrusion And/>And then the neural network model trained in the step 3 is called, and the neural network model can be calculated to obtain the neural network model used for calculation in the step 5And/>Predicted value/>And/>
Predictive valueAnd/>The specific calculation process comprises the following steps:
Step 5.1: setting α=1, initializing the input vector z, will An input vector z of 12 x 1 dimensions is composed, wherein/>Respectively correspond to initial values representing state variables theta 1,θ2,…,θ8.
Step 5.2: calculating an output vector corresponding to the input vector z by using the neural network model trained in the step 3
Calculating an output vector corresponding to the input vector z by using the trained neural network modelThe specific implementation process of (2) is shown in step 5.21 to step 5.23.
Step 5.21: the output value of the middle layer neuron, β 1,β2,…,βH, is calculated according to equation ⑨:
In equation ⑨, e represents a natural constant, The upper label T denotes the transpose of the vector, and the lower label h=1, 2, …, H.
Step 5.22: after beta 1,β2,…,βH is formed into a real number vector beta of H multiplied by 1 dimension, the real number vector beta is formed according to the formulaThe output value gamma 1,γ2,…,γ8 of the output layer neuron is calculated.
Step 5.23: combining gamma 1,γ2,…,γ8 into an 8 x1 dimensional output vector
Step 5.3: setting predicted values of 8 state variables theta 1,θ2,…,θ8 at future alpha-th control momentIn turn equal to the output vector/>Respectively according to the formula/>AndCalculating the change value/>, at the future alpha-th control momentAnd/>It should be noted that here the predictor/>8, But when calculating the change value, only the first two predicted values, namely the predicted values/>, of the state variables θ 1 and θ 2 are takenAnd/>The change value is calculated.
Step 5.4: judging whether the control moment meets the condition alpha < L; if so, update α according to the formula α=α+1, then update the input vector z, and thenForming an input vector z with 12 multiplied by 1 dimensions, and returning to the step 5.2; if not, set the initial value/>, of the state variableWhich in turn is equal to the predicted value/>, of the future L-th control instantAnd outputs the variation value/>, of the vertical acceleration and the pitch acceleration at the 1 st to the L th control momentsAnd/>Substituting the variation values of the vertical acceleration and the pitch acceleration into a formula ⑧ to obtain the predicted value/>, of the control signal u i at L control moments in the future
Step 6: the solved predicted values will be optimized according to equation ⑧And/>And the corresponding actions are executed through the electro-hydraulic systems.
By implementing the steps described above, the advantages of the method of the invention are presented below:
the method of the invention generates a large amount of input data and output data for training the neural network model by randomly generating control signals and vertical pavement protrusion height data. As long as the data volume is large enough to contain various possibilities, the corresponding neural network model can approach the mechanism model of the active suspension system with high accuracy. Compared with the method for calculating the predicted value and the variation value of the state variable by directly using the mechanism model, the method has the advantages that the neural network model can exert the advantages of strong nonlinear approximation capability and quick calculation capability, so that the high-speed operation and high-frequency implementation control of model prediction control are realized.
Drawings
The invention is further described below with reference to the drawings and examples.
Fig. 1 is a schematic structural view of a preferred embodiment of the present invention.
Detailed Description
The present invention will now be described in detail with reference to the accompanying drawings. The figure is a simplified schematic diagram illustrating the basic structure of the invention only by way of illustration, and therefore it shows only the constitution related to the invention.
As shown in fig. 1, the vehicle suspension system model predictive control method based on road condition monitoring of the invention comprises the following steps:
Step 1: the method comprises the steps of determining state variables of an active suspension system, establishing a state space model of the vehicle active suspension system, determining a maximum value u max and a minimum value u min of a control signal change range, and determining a single-step change maximum value delta u max of the control signal.
Because the rolling motion and the pitching motion of the four-wheel vehicle have the same motion law, in practical application, a corresponding mechanism model is usually built aiming at a half vehicle suspension system by utilizing Newton's mechanical law, and the basic principle of building the mechanism model is that according to the Newton's mechanical law, namely: the mass multiplied by the acceleration equals the resultant force. Therefore, the mechanism model firstly established by the method is specifically shown as a formula ①:
Wherein, when the index i is equal to 1, the front suspension is represented; when i is equal to 2, the rear suspension is represented; represents the vertical acceleration of the suspension, the displacement variation of the suspension/> X i represents the vertical displacement of the suspension, b i represents the horizontal distance between the center of gravity of the suspension and the center of gravity of the vehicle body,/>Representing body pitch angle (when the body is level,/>) X 0 represents the vertical displacement of the vehicle body,/> Representing the vertical speed of the vehicle body,/>Representing pitch rate of vehicle body,/>Represents the vertical speed of the vehicle body, u i represents the acting force (i.e. control signal) of the electro-hydraulic system, the vertical deformation displacement deltay i＝yi-xi,yi of the tire represents the vertical road surface protrusion height of the suspension (y i is less than 0 if the road surface is concave), and v-Representing pitch acceleration of the vehicle body,/>Representing the vertical acceleration of the vehicle body; m and M represent the unsprung and sprung masses, respectively, of the suspension system, I 0 represents moment of inertia, spring force F s(Δxi) of the suspension and damping force/>The calculation mode of (2) is as follows:
In ② above, when When, sign function/>When/>When, sign function/>K lin and k non represent the linear and nonlinear spring coefficients of the suspension springs, respectively, and c lin,csys and c non represent the linear and nonlinear damping coefficients of the suspension damper, respectively, with the reference numerals i=1, 2. Similarly, the elastic force T s(Δyi) generated by the tire due to deformation is calculated as follows:
Ts(Δyi)＝ζlin·Δyi+ζnon·(Δyi)3 ③
in the above formula, ζ lin and ζ non represent the linear elastic coefficient and the nonlinear elastic coefficient of the tire, respectively; i=1, 2, i is equal to 1, and represents the front suspension, and i is equal to 2, and represents the rear suspension.
Since i is equal to 1 or 2, the mechanism model in equation ① is actually composed of 4 differential equations, 8 state variables θ 1,θ2,…,θ8 can be determined, which in turn represent respectivelyThe following state space model can be established for the active suspension system, and the model specifically comprises 8 state equations, namely:
Wherein Δθ 1,Δθ2,…,Δθ8 represents the variation corresponding to the 8 state variables θ 1,θ2,…,θ8, respectively, Δt represents the single step time, and the index i=1, 2.
As can be seen from the formula ②, F s(θ7+b1·θ4-θ3) in the formula ④ represents the spring force of the suspension, F d(θ5+b1·θ2-θ1) represents the damping force of the suspension, and the spring force F s(θ7+b1·θ4-θ3) and the damping force F d(θ5+b1·θ2-θ1) of the suspension are calculated as follows:
In formula ⑤, sign (θ 5+b1·θ2-θ1) =1 when θ 5+b1·θ2-θ1 is not less than 0; when θ 5+b1·θ2-θ1 < 0, sign (θ 5+b1·θ2-θ1) = -1.
Similarly, as can be seen from the equation ③, the elastic force T s(yi-θ4+i) generated by the tire due to deformation is calculated as follows:
Ts(yi-θ4+i)＝ζlin·(yi-θ4+i)+ζnon·(yi-θ4+i)3 ⑥
in the above formula ⑥, ζ lin and ζ non represent the linear elastic coefficient and the nonlinear elastic coefficient of the tire, respectively; i=1, 2, i is equal to 1, and represents the front suspension, and i is equal to 2, and represents the rear suspension.
Step 2: n sets of input data and N sets of output data are generated according to steps 2.1 to 2.4.
Step 2.1: initializing the number of data groups j=1, setting 8 state variables θ 1,θ2,…,θ8 to be equal to 0, and setting the initial value of the control signal u i
Step 2.2: randomly generating a vertical road surface protrusion height y 1 of the front suspension and a vertical road surface protrusion height y 2 of the rear suspension from the interval [ -0.02,0.02], respectively, andAfter the control signal u 1 of the front suspension and the control signal u 2 of the rear suspension are randomly generated, 12 data in the j-th group of input data are set to be equal to theta 1,θ2,…,θ8,y1,y2,u1,u2 in sequence.
Step 2.3: delta theta 1,Δθ2,…,Δθ8 is calculated using a state space model of the active suspension system, and then according to the formula theta b＝θb+Δθb, b=1, 2, 3..8, after updating the state variable theta 1,θ2,…,θ8, 8 data in the j-th set of output data are sequentially set equal to theta 1,θ2,…,θ8, respectively.
Step 2.4: judging whether the data group number j is smaller than N; if yes, setting j=j+1, and returning to the step 2.2 to continue generating data; if not, N groups of input data and N groups of output data are obtained.
It is noted that, since the single step time Δt in the formula ④ is known, according to the state equation for Δθ 1/Δt in the formula ④, Δθ 1 is calculated as follows:
Similarly, Δθ 2,Δθ3,…,Δθ8 can be calculated sequentially from the 2 nd to 8 th state equations in the equation ④.
Step 3: establishing a neural network model, determining the number of neurons in the middle layer of the neural network model as H, and training the neural network model by utilizing N groups of input data and N groups of output data to obtain parameters of the neural network model, wherein the parameters specifically comprise: h reference vectors for middle layer neuronsThe H smoothing factors δ 1,δ2,…,δH of the middle layer neurons output 8 coefficient vectors w 1,w2,…,w8 of the neurons.
Step 4: when the vehicle runs, the laser radar arranged in front of the vehicle is utilized to monitor road condition data in front of the vehicle in real time, and the monitoring values of the protrusion heights of the vertical road surface at L control moments in the future are obtained according to the speed conversionAndWherein/>Respectively representing the monitoring value of the vertical road surface protrusion height of the front suspension at the 1 st control moment in the future, the 2 nd control moment in the future, … th control moment in the future; /(I)The monitoring values of the vertical road surface protrusion height of the rear suspension at the future 1 st control moment, the future 2 nd control moment, … nd control moment and the future L th control moment are respectively represented.
Step 5: solving the minimization problem with constraint conditions according to a formula ⑧, thereby obtaining predicted values of the control signal u i at L control moments in the future
Wherein the range of the weight coefficient lambda is 0 < lambda < 1, and the change value of the vertical acceleration at the alpha-th control moment in the futureVariation value/>, of pitch angle acceleration at future alpha-th control momentThe value of the change in the control signal u i at the future alpha-th control time/>Forgetting factor mu (α) satisfies the condition mu (1)≥μ(2)≥…≥μ(L),/>And/>Correspond to predicted values representing state variables θ 1 and θ 2, respectively, at future α -th control time,/>And/>Corresponds To the predicted values of the control signals u 1 and u 2 at the future alpha-th control moment, respectively, s.t. represents the abbreviation of the constraint English word Subject To,/>Represents an arbitrary setting of α=1, 2, …, L; when α=1,/>And/>Initial values representing state variables θ 1 and θ 2, respectively,/>And/>The initial values of the control signals u 1 and u 2 are respectively indicated.
In the process of solving the minimization problem in the formula ⑧, a set of predicted values of the control signal are obtained during each optimization iteration, namely: combined with the monitoring value/>, of the height of the vertical pavement protrusion And/>And then the neural network model trained in the step 3 is called, and the neural network model can be calculated to obtain the neural network model used for calculation in the step 5And/>Predicted value/>And/>
Predictive valueAnd/>The specific calculation process comprises the following steps:
Step 5.1: setting α=1, initializing the input vector z, will An input vector z of 12 x 1 dimensions is composed, wherein/>Respectively correspond to initial values representing state variables theta 1,θ2,…,θ8.
Step 5.2: calculating an output vector corresponding to the input vector z by using the neural network model trained in the step 3
Calculating an output vector corresponding to the input vector z by using the trained neural network modelThe specific implementation process of (2) is shown in step 5.21 to step 5.23.
Step 5.21: the output value of the middle layer neuron, β 1,β2,…,βH, is calculated according to equation ⑨:
In equation ⑨, e represents a natural constant, The upper label T denotes the transpose of the vector, and the lower label h=1, 2, …, H.
Step 5.22: after beta 1,β2,…,βH is formed into a real number vector beta of H multiplied by 1 dimension, the real number vector beta is formed according to the formulaThe output value gamma 1,γ2,…,γ8 of the output layer neuron is calculated.
Step 5.23: combining gamma 1,γ2,…,γ8 into an 8 x1 dimensional output vector
Step 5.3: setting predicted values of 8 state variables theta 1,θ2,…,θ8 at future alpha-th control momentIn turn equal to the output vector/>Respectively according to the formula/>AndCalculating the change value/>, at the future alpha-th control momentAnd/>It should be noted that here the predictor/>8, But when calculating the change value, only the first two predicted values, namely the predicted values/>, of the state variables θ 1 and θ 2 are takenAnd/>The change value is calculated.
Step 5.4: judging whether the control moment meets the condition alpha < L; if so, update α according to the formula α=α+1, then update the input vector z, and thenForming an input vector z with 12 multiplied by 1 dimensions, and returning to the step 5.2; if not, set the initial value/>, of the state variableWhich in turn is equal to the predicted value/>, of the future L-th control instantAnd outputs the variation value/>, of the vertical acceleration and the pitch acceleration at the 1 st to the L th control momentsAnd/>Substituting the variation values of the vertical acceleration and the pitch acceleration into a formula ⑧ to obtain the predicted value/>, of the control signal u i at L control moments in the future
Step 6: the solved predicted values will be optimized according to equation ⑧And/>And the corresponding actions are executed through the electro-hydraulic systems.
Note that "=" in the formula regarding variable update indicates that the calculation result on the right side is assigned to the variable on the left side, and a representation of a computer program is adopted, including but not limited to the formula θ b＝θb+Δθb, j=j+1, α=α+1, and "=" in other formulas indicates equality.
While the foregoing is directed to the preferred embodiment of the present invention, other and further embodiments of the invention may be devised without departing from the basic scope thereof, and the scope thereof is determined by the claims that follow. The technical scope of the present invention is not limited to the description, but must be determined according to the scope of claims.
Claims (1)
1. A vehicle suspension system model predictive control method based on road condition monitoring is characterized by comprising the following steps of: the method comprises the following steps:
Step 1: determining a state variable theta 1,θ2,…,θ8 of the active suspension system, establishing a state space model for the vehicle active suspension system according to the state variable theta 1,θ2,…,θ8, determining a maximum value u max and a minimum value u min of the control signal change range, and determining a single-step change maximum value deltau max of the control signal; wherein, the state space model of the vehicle active suspension system is expressed as:
Where M and M represent the unsprung and sprung masses, respectively, of the vehicle suspension system, and I 0 represents the moment of inertia; i=1, 2, when the index i is equal to 1, representing the front suspension, and when i is equal to 2, representing the rear suspension; b i denotes the horizontal distance between the center of gravity of the suspension and the center of gravity of the vehicle body; u i denotes a control signal of the suspension; y i represents the vertical road surface protrusion height of the suspension; Δt represents a single step time; Δθ 1,Δθ2,…,Δθ8 represents the amounts of change corresponding to the 8 state variables θ 1,θ2,…,θ8, respectively; f s(θ7+b1·θ4-θ3) represents the spring force of the suspension, F d(θ5+b1·θ2-θ1) represents the damping force of the suspension, and the spring force F s(θ7+b1·θ4-θ3) and the damping force F d(θ5+b1·θ2-θ1) of the suspension are calculated as follows:
In formula ⑤, sign (θ 5+b1·θ2-θ1) =1 when θ 5+b1·θ2-θ1 is not less than 0; when θ 5+b1·θ2-θ1 < 0, sign (θ 5+b1·θ2-θ1)＝-1;klin and k non represent the linear elastic coefficient and the nonlinear elastic coefficient of the suspension spring, c lin,csys and c non represent the linear damping coefficient, the symmetrical damping coefficient and the nonlinear damping coefficient of the suspension damper, respectively; the calculation method of the elastic force T s(yi-θ4+i generated by the tire due to deformation) is as follows:
Ts(yi-θ4+i)＝ζlin·(yi-θ4+i)+ζnon·(yi-θ4+i)3 ⑥
in the above formula ⑥, ζ lin and ζ non represent the linear elastic coefficient and the nonlinear elastic coefficient of the tire, respectively; i=1, 2, i being equal to 1, representing the front suspension, and i being equal to 2, representing the rear suspension;
step 2: generating N sets of input data and N sets of output data comprises the steps of:
Step 2.1: initializing the number of data groups j=1, setting 8 state variables θ 1,θ2,…,θ8 to be equal to 0, and setting the initial value of the control signal u i i＝1,2；
Step 2.2: randomly generating a vertical road surface protrusion height y 1 of the front suspension and a vertical road surface protrusion height y 2 of the rear suspension from the interval [ -0.02,0.02], respectively, andI=1, 2, and then 12 data in the j-th group of input data are sequentially equal to θ 1,θ2,…,θ8,y1,y2,u1,u2 after the control signal u 1 of the front suspension and the control signal u 2 of the rear suspension are randomly generated;
Step 2.3: using a state space model of the active suspension system to calculate delta theta 1,Δθ2,…,Δθ8, and then according to the formula theta b＝θb+Δθb, b=1, 2, 3..8, after updating the state variable theta 1,θ2,…,θ8, setting 8 data in the j-th set of output data to be equal to theta 1,θ2,…,θ8 respectively in sequence;
Step 2.4: judging whether the data group number j is smaller than N; if yes, setting j=j+1, and returning to the step 2.2 to continue generating data; if not, N groups of input data and N groups of output data are obtained;
Step 3: determining the number of middle-layer neurons of the neural network model as H, and training the neural network model by utilizing N groups of input data and N groups of output data to obtain parameters of the neural network model, wherein the parameters specifically comprise: h reference vectors for middle layer neurons H smoothing factors δ 1,δ2,…,δH of the middle layer neurons, 8 coefficient vectors w 1,w2,…,w8 of the neurons are output;
Step 4: when the vehicle runs, the laser radar arranged in front of the vehicle is utilized to monitor road condition data in front of the vehicle in real time, and the monitoring values of the protrusion heights of the vertical road surface at L control moments in the future are obtained according to the speed conversion AndWherein/>Respectively representing the monitoring value of the vertical road surface protrusion height of the front suspension at the 1 st control moment in the future, the 2 nd control moment in the future, … th control moment in the future; /(I)Respectively representing the monitoring value of the vertical road surface protrusion height of the rear suspension at the 1 st control moment in the future, the 2 nd control moment in the future, … th control moment in the future;
Step 5: solving the minimization problem with constraint conditions according to a formula ⑧, thereby obtaining predicted values of the control signal u i at L control moments in the future
Wherein the range of the weight coefficient lambda is 0 < lambda < 1, and the change value of the vertical acceleration at the alpha-th control moment in the futureVariation value/>, of pitch angle acceleration at future alpha-th control momentThe value of the change in the control signal u i at the future alpha-th control time/>I=1, 2, forgetting factor μ (α) satisfies the condition μ (1)≥μ(2)≥…≥μ(L),/>And/>Correspond to predicted values representing state variables θ 1 and θ 2, respectively, at future α -th control time,/>And/>Corresponds To the predicted values of the control signals u 1 and u 2 at the future alpha-th control moment, respectively, s.t. represents the abbreviation of the constraint English word Subject To,/>Represents an arbitrary setting of α=1, 2, …, L; when a=1 is used for the purpose of providing,And/>Initial values representing state variables θ 1 and θ 2, respectively,/>And/>Initial values representing control signals u 1 and u 2, respectively;
Wherein the predicted value And/>The calculation process of (1) comprises the following steps:
step 5.1: after setting α=1, it will An input vector z of 12 x1 dimensions is composed;
step 5.2: calculating an output vector theta corresponding to the input vector z by using the trained neural network model in the step 3;
the specific implementation process for calculating the output vector theta corresponding to the input vector z by using the trained neural network model comprises the following steps:
Step 5.21: the output value of the middle layer neuron, β 1,β2,…,βH, is calculated according to equation ⑨:
in equation ⑨, e represents a natural constant, i.e. e is approximately equal to 2.718281828459, The upper label T represents the transpose of the vector, the lower label h=1, 2, …, H;
Step 5.22: after beta 1,β2,…,βH is formed into a real number vector beta of H multiplied by 1 dimension, the real number vector beta is formed according to the formula Calculating an output value gamma 1,γ2,…,γ8 of the output layer neuron;
step 5.23: gamma 1,γ2,…,γ8 is combined into an output vector theta [ gamma 1,γ2,…,γ8]T of 8 multiplied by 1 dimension;
step 5.3: setting predicted values of 8 state variables at future alpha-th control moment In turn equal to 8 data in the output vector θ, each according to the formula/>And/>Calculating the change value/>, at the future alpha-th control momentAnd/>
Step 5.4: judging whether the control moment meets the condition alpha < L; if so, after alpha is updated according to the formula alpha=alpha+1, the method comprises the following steps ofForming an input vector z with 12 multiplied by 1 dimensions, and returning to the step 5.2; if not, set the initial value/>, of the state variableIn turn equal to/>And output/>And
Step 6: solving the optimizationAnd/>And the corresponding actions are executed through the electro-hydraulic systems.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210312774.2A CN114675537B (en) | 2022-03-28 | 2022-03-28 | Vehicle suspension system model predictive control method based on road condition monitoring |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210312774.2A CN114675537B (en) | 2022-03-28 | 2022-03-28 | Vehicle suspension system model predictive control method based on road condition monitoring |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114675537A CN114675537A (en) | 2022-06-28 |
| CN114675537B true CN114675537B (en) | 2024-05-17 |
Family
ID=82076938
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210312774.2A Active CN114675537B (en) | 2022-03-28 | 2022-03-28 | Vehicle suspension system model predictive control method based on road condition monitoring |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114675537B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107499271A (en) * | 2017-09-01 | 2017-12-22 | 浙江万安科技股份有限公司 | A kind of car anti-rollover control system and method based on electronic control air suspension and electric controlled brake system |
| CN110096840A (en) * | 2019-05-17 | 2019-08-06 | 黑龙江科技大学 | The sliding-mode control of vehicle suspension |
| CN112269314A (en) * | 2020-09-30 | 2021-01-26 | 南京航空航天大学 | Automobile chassis decoupling control method based on LSTM neural network |
| CN113978196A (en) * | 2021-12-27 | 2022-01-28 | 北京理工大学深圳汽车研究院(电动车辆国家工程实验室深圳研究院) | Suspension roll restraining method and system for unmanned vehicle |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7715965B2 (en) * | 2004-10-15 | 2010-05-11 | Ford Global Technologies | System and method for qualitatively determining vehicle loading conditions |
| FR2890902B1 (en) * | 2005-09-22 | 2007-12-14 | Peugeot Citroen Automobiles Sa | SUSPENSION CONTROL DEVICE, VEHICLE PROVIDED WITH SAME, METHOD OF OBTAINING AND PROGRAM |
-
2022
- 2022-03-28 CN CN202210312774.2A patent/CN114675537B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107499271A (en) * | 2017-09-01 | 2017-12-22 | 浙江万安科技股份有限公司 | A kind of car anti-rollover control system and method based on electronic control air suspension and electric controlled brake system |
| CN110096840A (en) * | 2019-05-17 | 2019-08-06 | 黑龙江科技大学 | The sliding-mode control of vehicle suspension |
| CN112269314A (en) * | 2020-09-30 | 2021-01-26 | 南京航空航天大学 | Automobile chassis decoupling control method based on LSTM neural network |
| CN113978196A (en) * | 2021-12-27 | 2022-01-28 | 北京理工大学深圳汽车研究院(电动车辆国家工程实验室深圳研究院) | Suspension roll restraining method and system for unmanned vehicle |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114675537A (en) | 2022-06-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110154666B (en) | Vehicle suspension system self-adaptive reverse-thrust control method capable of realizing road condition prediction | |
| CN103754081B (en) | The optimum fuzzy compound controller method of Vehicle Suspension with Non-linear Spring | |
| CN101916113B (en) | Automotive body gesture decoupling control method based on active suspension evaluation indicator | |
| CN108345218A (en) | Vehicle active suspension PID controller design method based on teaching optimization algorithm | |
| CN111444623B (en) | Collaborative optimization method and system for damping nonlinear commercial vehicle suspension dynamics | |
| CN105676694A (en) | Intelligent sampling data output feedback control method for automobile suspension system | |
| Cui et al. | Design optimization of vehicle EHPS system based on multi-objective genetic algorithm | |
| CN109398020B (en) | Nonlinear model-based prediction control method for vehicle hydro-electric coupling type ISD suspension | |
| He et al. | Vertical vibration control of an in‐wheel motor‐driven electric vehicle using an in‐wheel active vibration system | |
| CN112158045A (en) | Active suspension control method based on depth certainty strategy gradient | |
| CN112434407B (en) | Optimization design method for multi-suspension dynamic parameter distribution of tracked vehicle | |
| CN102975587A (en) | Vehicle semiactive suspension based on double controllable dampers and control method thereof | |
| Ahmed | Quarter car model optimization of active suspension system using fuzzy PID and linear quadratic regulator controllers | |
| WO2023108699A1 (en) | Novel lqg control method | |
| Yue et al. | Simulation and experimental study on the active stability of high-speed trains | |
| CN114675537B (en) | Vehicle suspension system model predictive control method based on road condition monitoring | |
| CN107839425B (en) | Vertical vibration and pitching vibration cooperative control method for tracked vehicle | |
| Wu et al. | Pitch motion suppression of electric vehicle active suspensions based on multibody dynamics | |
| CN115099035B (en) | Suspension vibration reduction design method with negative stiffness and inertial capacity cooperation under random displacement excitation | |
| jie Yue et al. | Research on vehicle suspension systems based on fuzzy logic control | |
| CN112590483B (en) | Observer-based multi-target integrated control method for automobile lateral stability and active suspension | |
| CN114660936B (en) | Intelligent control method of active suspension system based on RBF neural network | |
| CN115270493A (en) | Axle coupling vibration analysis method | |
| Wong et al. | Design of a fuzzy preview active suspension system for automobiles | |
| CN113110031A (en) | Fuzzy PID active suspension control system and method based on genetic algorithm optimization |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |