Adaptive cruise control method based on neural network approximation

Abstract

The invention relates to the technical field of vehicle control, in particular to an adaptive cruise control method based on neural network approximation, which comprises the following steps: establishing a nonlinear dynamic model of a vehicle, and describing nonlinear resistance of the vehicle; measuring a vehicle distance, and defining a distance tracking error; constructing a virtual controller to obtain a vehicle speed required for realizing distance control; calculating a neural network parameter adaptive law, and controlling a system to tend to design a hyperplane where a controller is located; and calculating a real design controller for controlling the vehicle to control the running of the vehicle, updating the vehicle information in the running process, and repeating the steps. According to the adaptive cruise control method based on neural network approximation, the output of the controller is adjusted on line in real time, and high robust performance is achieved. And based on the state information of the vehicle, the uncertain driving resistance of the vehicle is learned on line by using a neural network, and the initial error is compensated and corrected, so that the self-adaptive cruise control of the vehicle in any initial state (initial position) is realized.

Description

Technical Field

[0001] The present application relates to the field of vehicle control technology, and in particular to an adaptive cruise control method based on neural network approximation. Background Art

[0002] The adaptive cruise control system senses the environment around the vehicle and automatically adjusts the speed, allowing the controlled vehicle to follow the vehicle in front at the desired distance. It can effectively reduce the driver's workload while ensuring the driver's safety, and improve the level of highway intelligence and operating efficiency. With the development of unmanned driving and intelligent networked vehicles, higher requirements are placed on the control needs and control accuracy of vehicle adaptive cruise control systems.

[0003] Model-based automatic driving control methods include nonlinear model predictive control, linear quadratic Gaussian model control, backstepping control, sliding mode control, high gain scheduling control, nonlinear neural network adaptive control, etc. Although the control methods based on these models can solve the adaptive cruise of the vehicle to a certain extent and achieve the corresponding control objectives, the vehicle operation is subject to various uncertain driving resistances, and these forces are related to each independent component of the car. Using simplified models to design the controller will affect the control accuracy and endanger driving stability and safety. Summary of the invention

[0004] The technical problem to be solved by the present invention is that the control accuracy of the existing vehicle adaptive cruise control model is poor.

[0005] To this end, the present invention provides an adaptive cruise control method based on neural network approximation.

[0006] The technical solution adopted by the present invention to solve its technical problem is: An adaptive cruise control method based on neural network approximation comprises the following steps: Step 1: Establish a nonlinear dynamic model of the vehicle and describe the nonlinear resistance of the vehicle by combining the radial basis function neural network; Step 2: Measure the vehicle spacing, combine it with the expected vehicle spacing, and define the spacing plane error. The spacing plane error can be expressed as a function related to the expected vehicle spacing. After taking the derivative of the spacing plane error, it can be expressed as a function related to the virtual control amount, so as to obtain the virtual control amount. Step 3: Combine the first-order inertia link in the DSC dynamic surface control design and introduce the sampling time constant The nonlinear filter is used to convert the virtual control amount of the spacing surface error into the virtual control amount of the speed surface error, and the speed surface error of the control system is constructed based on the virtual control amount of the speed surface error and the nonlinear filter; Step 4: Combine the new variables in the previous steps, redefine the Lyapunov function, and use the optimization parameters of the radial basis neural network to design the adaptive law and actual control quantity. During the following process, repeat steps 2 to 4, and the vehicle uses the iterative actual control quantity to control the distance between vehicles; Step 5: To prove the stability of the patented algorithm, the stability of the vehicle closed-loop system is analyzed and verified.

[0007] Furthermore, in step one, an initial nonlinear dynamic model of the vehicle is constructed based on the output force of the vehicle, the rolling resistance of the vehicle, the air resistance, and the ramp resistance.

[0008] Furthermore, in step one, radial basis function neural network parallel learning is used to approximate the vehicle driving resistance to describe the nonlinear resistance of the vehicle.

[0009] Furthermore, in step 2, the adaptive cruise control system on the vehicle adaptively adjusts its own speed according to the running speed of the front vehicle so that the two vehicles maintain the desired following distance, and the distance surface error is defined as: ,in, is the real position of the front vehicle, so , is the desired vehicle spacing.

[0010] Furthermore, in step 2, the spacing surface error is defined as , for spacing error Taking the derivative, we get: , where the function It is called virtual control quantity.

[0011] Furthermore, in step three, nonlinear filtering is used to filter the virtual control amount of the spacing surface, and the speed error is calculated using the filtered signal.

[0012] Furthermore, in step 3, the virtual control amount of the spacing surface is , the nonlinear filter is ,in, represents the bounded boundary layer error between the spacing surface and the velocity surface, and s is the virtual controller of the velocity surface error; is the filtering time constant; for The estimated value of mainly represents the bounded interference caused by the pilot vehicle speed, the vehicle speed, the vehicle spacing strategy, etc. is a positive function and satisfies ,and ,in is a positive constant, and is introduced into the nonlinear filter Compensate for the effects caused by boundary layer errors.

[0013] Furthermore, in step three, the velocity surface error is redefined using a nonlinear filter as: .

[0014] Furthermore, in step 4, the Lyapunov function is redefined by combining the new variables in the aforementioned system: ,in, , yes The estimated value of yes The estimated value of , is a constant greater than 0.

[0015] Furthermore, in step 4, the adaptive law is designed based on the virtual control quantity of the velocity surface constructed after filtering and the optimization parameters of the radial basis neural network: , , and the control quantity is designed to be in, is the variance of the point estimate.

[0016] The beneficial effect of the present invention is that the adaptive cruise control based on neural network approximation can adjust the controller output online in real time to achieve stronger robust performance than traditional methods. By using the vehicle's state information, the neural network is used to learn the vehicle's uncertain driving resistance online, and the initial error compensation and correction are performed to achieve the vehicle's adaptive cruise control under any initial state (initial position). BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0018] Figure 1 It is a structural schematic diagram of the desired following vehicle distance measurement in the present invention.

[0019] Figure 2 It is an algorithm structure block diagram of the adaptive cruise control method based on neural network approximation in the present invention. DETAILED DESCRIPTION

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0021] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0022] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0023] An adaptive cruise control method based on neural network approximation comprises the following steps: Step 1: Model the vehicle S1.1 Establishing the nonlinear dynamic model of the vehicle (1) Where: is the real position and speed of the vehicle, yes The derivative of the following vehicle position is the following vehicle speed, yes The derivative of the following vehicle speed is the following vehicle acceleration; M is the real mass of the following vehicle; F is the real output force of the vehicle; Indicates the rolling resistance of the vehicle; Indicates the air resistance of the vehicle; represents the slope resistance; The distribution represents the acceleration due to gravity, the rolling resistance coefficient, the air resistance coefficient, the vehicle's frontal area, the air density, and the road bank angle.

[0024] Since the vehicle's driving resistance is affected by various factors such as vehicle mass, weather changes, and road conditions, Parameters such as have time-varying characteristics; in addition, various resistances are related to the individual components on the vehicle, and there is a part of the vehicle running resistance that cannot be accurately modeled, resulting in a large control error when directly using equation (1) to design the controller.

[0025] S1.2 uses the parallel learning characteristics of radial basis neural network to approximate the vehicle's driving resistance, thereby achieving an effective description of the vehicle's nonlinear resistance.

[0026] It should be noted that Lemma 1: The radial basis neural network can approximate unknown nonlinear smooth functions online, and the input of the neural network is: , where q represents the dimension of the neural network input; is a vector parameter, R represents a real number set, m represents the number of neurons, is a Gaussian function: , where is the center and width of the Gaussian function, and the radial basis neural network can set the neuron approximation , so that: , approximation error By adjusting the ideal vector parameters Make it as small as possible, that is, , is a small positive number, represents the mean of the approximation error, The set of real numbers representing z as a q-dimensional vector.

[0027] Based on the above Lemma 1, the nonlinear driving resistance of the vehicle is approximated, and equation (1) can be set as (2) in, , M is the real mass of the following vehicle; F(t) is the real output force of the vehicle; The nonlinear function in the neural network can approximate the nonlinear driving resistance of the vehicle. , is the target error of the multidimensional Gaussian kernel in the radial basis function, usually set to 0.001, is the weight of each coefficient from the hidden layer to the output layer in the radial basis neural network, To make The smallest "optimal" weight vector, is a vector parameter The transpose of , y is the tracking vehicle distance output by the system.

[0028] Step 2: The sensor measures the distance between vehicles and defines the distance tracking error , and design the virtual control amount of the spacing surface error .

[0029] ACC (Adaptive Cruise Control System) Figure 1 The pilot car in the vehicle adjusts its own speed adaptively to keep the desired following distance between the two vehicles. Its structure is as follows: Figure 1 As shown. The spacing tracking error surface is defined as: (3) in, is the real position of the front vehicle. The goal of adaptive cruise control is to design a controller so that the system error state reaches a sliding plane described by a polynomial tracking speed error and moves on the sliding plane. The control system operates stably, that is, , so that , is the desired vehicle spacing.

[0030] In order to ensure that the vehicle can be controlled to automatically follow the preceding vehicle and run safely at the expected vehicle distance d under uncertain driving resistance, the control objectives are set as follows: (1) The actual vehicle spacing approaches the expected vehicle spacing, and the spacing error is as small as possible, that is, ; (2) The following vehicle can respond quickly to changes in the leading vehicle, and the speed error between the two vehicles is as small as possible, that is, .

[0031] In order to achieve the above control objectives, the spacing surface error is defined as: (4) Pitch error Taking the derivative, we get: (5) In the above formula, As the virtual control input, we can get: (6) In the above formula, k d is a constant. In this way, equation (5) can be transformed into (7) Consider the Lyapunov function (8) The derivative is (9) From formula (9), we can see that the spacing surface error is stable. It is not a control input, so It is called the virtual control quantity, combined with formula (6) (10) Combining (5), we can get It can be expressed as (11) at this time Recorded as (12) Step 3: Use a nonlinear filter to convert the virtual control amount of the spacing surface error Converted into a virtual control quantity s of the speed surface, and constructed the speed surface error in this control system .

[0032] In order to avoid the need to use virtual control variables when designing the controller Differentiation leads to complex interference terms, and the sampling time constant is introduced in combination with the first-order inertia link in the DSC dynamic surface control design The nonlinear filter replaces the differential calculation, which satisfies: (13) Where s is the virtual control amount of the speed surface error, and the speed surface error is defined on this basis And the boundary layer error E is: (14) (15) Speed ​​surface error Indicates vehicle speed The error between the actual value and the expected value (virtual control quantity s) of the boundary layer error E is expressed as The expected value s and the nonlinear filter input The error between .

[0033] The specific form is: (16) in, for The estimated value of It mainly represents the bounded interference caused by the speed of the pilot vehicle, the speed of the own vehicle, the vehicle spacing strategy, etc. is a positive function and satisfies ,and ,in is a positive constant. The nonlinear filter designed by equation 16 is introduced by Compensate for the effect caused by boundary layer errors, thereby making Therefore, the first-order inertia link is used to control the variable Converted into virtual control quantity s.

[0034] Combining equation (2), equation (14), and equation (16) The derivative is (17) Step 4: Based on the previous step, redefine the Lyapunov function and combine the radial basis neural network function to optimize the parameters to design the adaptive law and the actual controller.

[0035] First of all, it should be noted that Lemma 2: In a neural network, w is the weight coefficient of the neural network node. , there are constants satisfy:

[0036] Lemma 3: For given parameters ,and , there is an inequality:

[0037] in is a fixed constant.

[0038] Because of the new quantity in the previous step (17) , , , so the Lyapunov function is redefined as: (18) in, , yes The estimated value of yes The estimated value of , is a constant greater than 0.

[0039] Differentiate equation (14) and substitute equation (13) into it, and we can get: (19) From Lemma 1, we can see that the basis function of the radial basis neural network is a Gaussian function, and the error function is defined as the mean square error, so , is the Euclidean distance deviation, is the variance of the point estimate, and is the approximation , naturally, the smaller the deviation, the smaller the variance. Combining Lemma 3, we can get the following two inequalities: (20)

[0040] Substituting equation (20) into equation (19) yields: (twenty one) The adaptive law of the design parameters is: (twenty two)

[0041] Based on formula (21) and combined with formula (22), the actual control quantity is designed for: (twenty three) Substituting the controller and adaptive law designed by equations (22) and (23) into equation (21), we get (twenty four) Step 5: To prove the stability of the patented algorithm, the stability of the vehicle closed-loop system is analyzed. The main theories are as follows: In the vehicle dynamics system described by equation (1), the virtual control variable designed by equation (10), the nonlinear filter designed by equation (16), the adaptive law designed by equation (22) and the actual controller designed by equation (23) are applied to equation (1) to ensure: 1. The following vehicle can gradually track the target vehicle in front and ensure stable operation at the desired safe distance, that is, ; 2. The following vehicle can quickly respond to changes in the target vehicle in front, and the speed error between the two vehicles is as small as possible, that is, ; 3. All states of the closed-loop system are bounded, that is, the vehicle speed , control input , virtual control quantity , the nonlinear filter s is bounded.

[0042] prove: Due to the new boundary layer error E in the above system equation (24) and the bounded interference caused by the pilot vehicle speed, the own vehicle speed, the vehicle spacing strategy, etc. , so based on equation (24), the Lyapunov function of the system is redefined as: (25) Taking the derivative of the above formula, we can get: (26) The boundary layer error is The derivative of is: , and it is known that Bounded and defined

[0043] Then we can deduce: (27) Here is Lemma 4, Young's inequality: Assume that a, b, c and d are positive constants, if they satisfy , then: , where the conditions for equality are: .

[0044] Using Young's inequality in Lemma 4, we know that: (28) Substitute equation (28) into equation (27) and consider , it can be deduced that: (29) According to Lemma 2, we know that: (30) Substituting equation (30) into equation (29), we can obtain (31) design The adaptive law is: (32) Also define: (33) Then, formula (31) can be rewritten as: (34) Integrating equation (34) on [0, t] yields: (35) When equation (35) is satisfied, it means is bounded. We can then know is also bounded. In addition, from formula (35), we can see that: (36) In this case, using Barbalat's theorem, we know that Therefore, the following vehicle can follow the speed of the target vehicle in front and maintain the desired vehicle spacing to drive safely. The nonlinear filter of the first-order inertia link designed by this patent suppresses the filtering error caused by the traditional linear filter and improves the control accuracy. The proof is completed.

[0045] Example: Step 1: Initialize the parameters, set the number of neurons in Lemma 1 to 5, and the center of the Gaussian function [-1 -0.5 0 0.5 1], width [0.5 0.5 0.5 0.5 0.5], parameter adaptation law , Setting value , virtual control amount of spacing surface The setting coefficient in , design the setting coefficient in the actual control quantity u , the sampling time of the nonlinear filter in equation (16) is Set to 0.1s, where the initial value of the virtual control quantity s is s(0)=0, and the adaptive law Setting coefficient Set to 0.5, the system operation cycle dt is set to 0.001s, and the adaptive law , Initial value of ; Adaptive Law Initial value of =0; positive function in nonlinear filter (16) ; is the The time-varying positive function is applied to the adaptive law , and the actual control quantity u, .

[0046] Step 2: The distance of the vehicle ahead collected by the sensor Subtract the vehicle position The obtained tracking distance , and the expected following distance d corresponding to the set vehicle speed are substituted into formula (4) to obtain the spacing surface error , use formula (6) to calculate the virtual control amount v(t), and use formula (10) to get the virtual control amount . Using formula (16) combined with the initial parameters, the virtual control quantity Converted into the derivative of the virtual control quantity s ,renew , use formula (32) to calculate the initial adaptive law ,renew ; Step 3: Use formula (14) to calculate , make As a neural network The input of , the output of the neural network is combined with equation (22) to obtain the adaptive law and , after updating, it is inserted into formula (23) to obtain the actual control quantity u, and the u value is used as the vehicle acceleration to control the vehicle distance.

[0047] Step 4: Update the system status with Update vehicle speed; use Update the vehicle position to participate in the next calculation.

[0048] The entire algorithm structure diagram is as follows Figure 2 As shown, Figure 2 Where NF is the nonlinear filter of formula (16), and RBFNN is the radial basis neural network.

[0049] Step 1: Difference the actual vehicle spacing y output by the system and the expected vehicle spacing d to obtain the spacing surface error , substituting into formula (10) we get , and then combine it with the s and Substitute the value into the nonlinear filter NF of formula (16) to calculate the derivative of the virtual control quantity s , combined with the s value of the previous moment, update the s value, substitute the s value into formula (15) to calculate E, and then combine it with formula (32) to calculate the adaptive law The derivative of , combined with the previous moment Update the current value value.

[0050] Step 2: Subtract the updated s from the first step from v(t) at the previous moment to get .

[0051] Step 3: Substitute y,v(t), , Input RBFNN radial basis neural network to get , combined with formula (22), we can get the adaptive law and The derivative of and , combined with the previous moment and Worth the update and , and , the first step Substituting the value into formula (23) we can get the actual control quantity u.

[0052] Step 4: Update u based on equation (2) and y values, and add them to the s and The value participates in the iterative calculation at the next moment.

[0053] In summary, during the vehicle driving process, the adaptive law obtained in step five is responsible for making the system stably and quickly approach the hyperplane controlled by the final designed controller under the condition of accepting the relative speed difference between the front vehicle and the vehicle's speed. The designed controller controls the vehicle to always keep a distance from the front vehicle and update the vehicle information in real time.

[0054] This patent uses the vehicle's state information and a neural network to learn the vehicle's uncertain driving resistance online. By compensating and correcting the initial error, it achieves adaptive cruise control of the vehicle in any initial state (initial position).

[0055] According to the Hecht-Nielsen theorem, it can be proved that under a given accuracy, the RBF neural network can approximate any nonlinear function with arbitrary accuracy. The adaptive cruise control based on neural network approximation in this patent can adjust the controller output online in real time, achieving stronger robust performance than traditional methods.

[0056] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. An adaptive cruise control method based on neural network approximation, It is characterized in that The following steps are included: Step 1: Establish a nonlinear dynamic model of the vehicle and describe the nonlinear resistance of the vehicle by combining the radial basis function neural network; Step 2: Measure the vehicle spacing, combine it with the expected vehicle spacing, and define the spacing plane error. The spacing plane error can be expressed as a function related to the expected vehicle spacing. After taking the derivative of the spacing plane error, it can be expressed as a function related to the virtual control amount, so as to obtain the virtual control amount. Step 3: Combine the first-order inertia link in the DSC dynamic surface control design and introduce the sampling time constant The nonlinear filter is used to convert the virtual control amount of the spacing surface error into the virtual control amount of the speed surface error, and the speed surface error of the control system is constructed based on the virtual control amount of the speed surface error and the nonlinear filter; Step 4: Combined with the velocity surface error in the previous step, redefine the Lyapunov function, and use the optimization parameters of the radial basis neural network to design the adaptive law and the actual control quantity; During the following process, steps 2 to 4 are repeated, and the vehicle uses the iterative actual control amount to control the distance between vehicles; Step 5: To prove the stability of the patented algorithm, the stability of the vehicle closed-loop system is analyzed and verified.

2. The adaptive cruise control method based on neural network approximation according to claim 1, It is characterized in that In step one, an initial nonlinear dynamic model of the vehicle is constructed based on the vehicle's output force, vehicle rolling resistance, air resistance, and ramp resistance.

3. The adaptive cruise control method based on neural network approximation according to claim 1, It is characterized in that In step one, radial basis function neural network parallel learning is used to approximate the vehicle driving resistance and describe the nonlinear resistance of the vehicle.

4. The adaptive cruise control method based on neural network approximation according to claim 1, It is characterized in that In step 2, the vehicle's built-in adaptive cruise control system adaptively adjusts its own speed according to the running speed of the front vehicle so that the two vehicles maintain the desired following distance. The distance tracking error is defined as: ,in, is the real position of the front vehicle, so , is the desired vehicle spacing.

5. The adaptive cruise control method based on neural network approximation according to claim 4, It is characterized in that In step 2, the spacing surface error is defined as , for spacing error Taking the derivative, we get: , where the function It is called virtual control quantity.

6. The adaptive cruise control method based on neural network approximation according to claim 1, It is characterized in that In step three, nonlinear filtering is used to filter the virtual control quantity of the spacing surface, and the filtered signal is used to calculate the speed surface error.

7. The adaptive cruise control method based on neural network approximation according to claim 1, It is characterized in that In step 3, the virtual control amount of the spacing surface is , the nonlinear filter is ,in, represents the boundary layer error bounded by the spacing surface and the velocity surface, and s is the virtual controller of the velocity surface error; is the filtering time constant; for The estimated value of mainly represents the bounded interference caused by the pilot vehicle speed, the vehicle speed, the vehicle spacing strategy, etc. is a positive function and satisfies ,and ,in is a positive constant, and is introduced into the nonlinear filter Compensate for the effects caused by boundary layer errors.

8. The adaptive cruise control method based on neural network approximation according to claim 7, It is characterized in that In step 3, the speed surface error is redefined as follows using the nonlinear filter and the virtual controller of the pitch surface error: .

9. The adaptive cruise control method based on neural network approximation according to claim 1, It is characterized in that In step 4, redefine the Lyapunov function in combination with the new variables in step 3: ,in, , yes The estimated value of yes The estimated value of , is a constant greater than 0, and the adaptive law of the design parameters is , .

10. The adaptive cruise control method based on neural network approximation according to claim 9, It is characterized in that In step 4, the virtual control quantity based on the filtered velocity surface is combined with the optimization parameters of the neural network to design the adaptive law: , , and design the controller as in, is the variance of the point estimate.

Images