A hierarchical centralized vehicle platoon cooperative driving method based on cloud computing

Abstract

The application discloses a layered centralized vehicle team cooperative driving method based on cloud computing, which comprises the following steps: establishing a vehicle team state space equation; uploading driving information to a cloud server; compensating the driving information; establishing a vehicle team driving optimal control problem according to the compensated driving information; solving the optimal control problem based on a Gauss pseudospectrum method and a sequential quadratic programming algorithm to obtain optimal control amount time series of each vehicle in the vehicle team; issuing the generated optimal control amount time series to a vehicle terminal; compensating the optimal control amount time series; controlling the vehicle to track its driving according to the compensated optimal control amount time series; and continuously repeating the above steps to realize vehicle team cooperative driving. The application adopts a centralized control strategy to increase the global optimization capability of the vehicle team and improve the driving efficiency of the vehicle team; and solves the communication time delay problem of the layered vehicle team controller through a vehicle-cloud data transmission time delay compensator, thereby improving the stability of the vehicle team control.

Description

Technical Field

[0001] This invention belongs to the field of vehicle driving control and relates to vehicle-road cooperative technology, specifically to a hierarchical centralized fleet cooperative driving method based on cloud computing. Background Technology

[0002] In recent years, with the rapid development of technologies such as intelligent connected vehicles, vehicle-road cooperation, and cloud computing, vehicle platooning control is also evolving towards greater complexity and comprehensiveness. Model-predictive platooning control strategies achieve optimal control of the platoon within a finite interval through model prediction, rolling optimization, and feedback correction. This approach enables multi-objective optimization and high control accuracy, and it can effectively handle multi-input multi-output problems. Furthermore, thanks to model-predictive control's gradual approach to the reference curve during continuous rolling cycles, it overcomes the limitations of modern control theory in relying on precise models of complex controlled objects and its insufficient ability to handle constraints.

[0003] Current research applying model predictive control to fleet control has achieved relatively accurate control objectives by establishing complex models, but this requires high-performance computers. Traditional vehicle-side computing is demanding in terms of operating environment (good heat dissipation, no bumps, etc.), which also increases the manufacturing cost per vehicle. Summary of the Invention

[0004] Purpose of the invention: To overcome the shortcomings of existing technologies, this invention provides a hierarchical centralized fleet cooperative driving method based on cloud computing. This method divides the fleet controller into upper and lower layers, with the upper layer controller located in the cloud, thereby reducing vehicle manufacturing costs and avoiding the problem of insufficient vehicle computing power. The centralized control strategy increases the fleet's global optimization capabilities and improves fleet driving efficiency. The vehicle-cloud data transmission latency compensator solves the communication latency problem of the hierarchical fleet controller, improving the stability of fleet control.

[0005] Technical Solution: To achieve the above objectives, this invention provides a hierarchical centralized fleet cooperative driving method based on cloud computing, comprising the following steps:

[0006] S1: Establish the longitudinal state-space equations for the vehicle fleet;

[0007] S2: According to the longitudinal state space equation of the platoon, each following vehicle in the platoon uploads its own driving information to the cloud server.

[0008] S3: After receiving the driving information, the cloud server compensates for the driving information;

[0009] S4: Based on the compensated driving information, establish the optimal control problem for fleet driving;

[0010] S5: Solve the optimal control problem established in step S4 based on the Gaussian pseudospectral method and sequential quadratic programming algorithm to obtain the time series of optimal control quantities for each vehicle in the fleet;

[0011] S6: Include the generated optimal control time series in a data packet and send it to each vehicle in the fleet;

[0012] S7: After receiving the data packet at the vehicle end, compensate for the time series of the optimal control quantity;

[0013] S8: Control the vehicle to follow its movement based on the time series of the compensated optimal control quantity;

[0014] S9: Repeat steps S2 to S8 continuously to achieve coordinated driving of the fleet.

[0015] Furthermore, the method for establishing the longitudinal state-space equation of the vehicle fleet in step S1 is as follows:

[0016] When the convoy is moving, the error state between the i-th vehicle in the convoy (excluding the lead vehicle) and the (i-1)-th vehicle in front of it is: following distance Δd i and relative velocity difference Δv i The expression is as follows:

[0017] Δd i =p i-1 -p i

[0018] Δv i =v i-1 -v i

[0019] In the formula, p i-1 and v i-1 These are the position and speed of the (i-1)th car, respectively, p i and v i These are the position and speed of the i-th car, respectively;

[0020] The longitudinal movement of a platoon with N vehicles can be described by the following state-space equations:

[0021]

[0022] y(t)=x(t)

[0023] In the formula, x(t) is the state error vector of the vehicle at time t, which can be expressed as x(t) = [x1(t), x2(t), ... x N-1 (t)] T The vehicle sequence in the platoon starts from 0, which is the lead vehicle. The state error of the i-th vehicle in the platoon is x. i (t)=[Δdi (t),Δv i (t)] T , Let A be the differential of the state error vector; A be the state control matrix; B be the input matrix; u(t) be the vehicle control input (i.e., acceleration) vector, which can be expressed as u(t) = [u0(t), u1(t), ... u N-1 (t)] T u i (t) is the control input of the i-th vehicle in the convoy, where u0(t) is the known control input planned by the lead vehicle based on future information;

[0024] in,

[0025]

[0026]

[0027] Furthermore, the driving information uploaded in step S2 includes the upload time, the distance to the vehicle in front, and the vehicle's own speed; in step S3, the uploaded driving information is compensated using a discrete numerical integration method through a time delay compensator.

[0028] Furthermore, the specific process of compensating the uploaded driving information using the discrete numerical integration method through the time delay compensator in step S3 is as follows:

[0029] The differential state-space equation for the i-th vehicle in the convoy is:

[0030]

[0031] y(t)=x(t)

[0032] k+t up,k The discrete integral formula for estimating the driving information uploaded by the i-th vehicle in the convoy at time k is:

[0033]

[0034] Among them, t up,k For the data upload latency, [k,k+t] up,k This time period is based on the sampling interval t sp Divide it into (t) up,k / t sp For each small segment, calculate the state increment for each sampling interval, and finally sum them to obtain k+t. up,k The estimated state x of the i-th vehicle at time i i (k+t up,k |k);

[0035] Among them, state differential The calculation uses u i (k+jt sp The values ​​are derived from the previous round of control quantity time series solution results stored in the cloud computer.

[0036] Furthermore, the optimal control problem for convoy driving established in step S4 is expressed as follows:

[0037]

[0038] in,

[0039]

[0040] x(t0)=x initial

[0041] u i ∈[u min ,u max ]

[0042] v i ∈[v min ,v max ]

[0043] In the formula, This is a safety cost item within the process cost, which promotes the convergence of speeds between vehicles, where d0 is the safe stopping distance. c2(Δd) is a variable coefficient; i -h) 2 This is the vehicle distance maintenance cost item in the process cost, which causes the distance between vehicles to tend towards the desired distance h; c3u i 2 To control costs, when the lead car is traveling at a constant speed, this factor causes the acceleration of following vehicles in the convoy to tend towards zero, promoting a balanced state of constant speed among all vehicles in the convoy. Simultaneously, this factor prevents unconstrained changes in acceleration, ensuring passenger comfort. c1, c2, and c3 are all positive values, representing the weighting coefficients of each cost factor. As constraints on the differential state-space equations of the vehicle fleet, x initial Let u be the convoy state at initial time t0. min ,u max These are the upper and lower limits of vehicle acceleration, v min ,v max These represent the upper and lower limits of vehicle speed, respectively.

[0044] As the following distance Δd i The decrease of the variable coefficient This increases the size of the facility, thereby increasing the penalties for safety-related costs.

[0045] Furthermore, the vehicle distance tending towards the desired vehicle distance h is a variable time distance, expressed as follows:

[0046]

[0047] In the formula, v free This refers to the free-flow vehicle speed.

[0048] Furthermore, in step S5, the Gaussian pseudospectral method is used to discretize the optimal control problem to obtain the time series of the optimal control quantities for each vehicle in the fleet. The specific process is as follows:

[0049] A1: Interpolation point selection: The zeros of the Legendre polynomial defined on the interval [-1, 1] are used as interpolation points to discretize the optimal control problem. The M-order Legendre polynomial is expressed as follows:

[0050]

[0051] Let P M If (τ) = 0, the zeros τ1, τ2…τ located in the range (-1, 1) can be obtained. M That is, the Legendre-Gauss (LG) point;

[0052] A2: Interval Transformation: Transform the time interval t∈[t0,t...] of the optimal control problem. f The transformation formula is used to convert the value to τ∈[-1,1].

[0053]

[0054] The original optimal control problem, after being transformed to the interval τ∈[-1,1], takes the following form:

[0055]

[0056] st

[0057]

[0058] B(x(τ0),t0,x(τ f ),t f ) = 0

[0059] C(x(τ),u(τ),τ;t0,t f )≤0

[0060] in, The constraints for the transformed differential state-space equations are B(x(τ0),t0,x(τ)). f ),t f ) = 0 is x(t0) = x initialThe transformed boundary constraints are C(x(τ),u(τ),τ;t0,t). f )≤0 is u i ∈[u min ,u max ] and v i ∈[v min ,v max Process constraints after conversion;

[0061] A3: Discretization Approximation of the Objective Function: For a continuous integral objective function on the interval [-1, 1], the approximate expression discretized using the Gauss-Legendary quadrature formula is as follows:

[0062]

[0063] In the formula, w m Let X(τ) be the Gaussian integral weight of the m-th term. m ) and U(τ m ) for in τ m Discretized state and control input variables; τ0=-1 is a selected initial point, τ1,τ2…τ M M LG points were selected;

[0064] A4: Global Interpolation Approximation State and Control Variables: Constructing M+1 M-order interpolation basis functions l m (τ)(m=0,1…M):

[0065]

[0066] st

[0067]

[0068] The state variables are approximately represented by the Lagrange interpolation polynomial constructed from the basis functions described above:

[0069]

[0070] The control variables are approximately represented by the Lagrange interpolation polynomial constructed from the basis functions described above:

[0071]

[0072] A5: State-space equation constraint transformation: To transform the differential constraints of the vehicle system dynamics into algebraic constraints, the approximate expressions of the state variables are differentiated:

[0073]

[0074] In the formula, D jm ∈R M×(M+1)The differential matrix is ​​represented as follows:

[0075]

[0076] In the formula, j takes values ​​of 1, 2, ..., M, thus enabling the differential state-space equation to be applied at the interpolation nodes τ. j The discrete approximation at a given point, when substituted into the differential state-space equations, can be transformed into an algebraic constraint:

[0077]

[0078] Similarly, for boundary constraints and path constraints at the interpolation point τ j Discretize to transform these two constraints:

[0079] B(X(τ0),t0,X(τ f ),t f ) = 0

[0080] C(X(τ j ),U(τ j ),τ j ;t0,t f )≤0

[0081] A6: Solving the nonlinear programming problem: After the above steps, the optimal control problem of the convoy cooperative driving is transformed into a discrete NLP problem. The NLP problem is solved by using the SNOPT software package based on the sequential quadratic programming algorithm, and the optimal control quantity time series that satisfies the constraints of the differential state space equation, process and boundary constraints can be obtained.

[0082] Furthermore, in step S6, the data packet includes the optimal control quantity time sequence and the transmission time. In step S7, the optimal control quantity time sequence is compensated by a delay compensator. The specific compensation method is as follows: after receiving the data packet, the vehicle end compares the current time with the transmission time in the data packet and only extracts the optimal control quantity time sequence packet after the receiving time as the control signal.

[0083] Furthermore, in step S8, the vehicle is controlled to track its movement by a lower-level vehicle-end controller, which is a PI controller. Its control principle is: the system controls the target u... d (k) and the actual output value of the system u r Subtracting (k) yields the system deviation err(k), then proportional and integral calculations are performed using the system deviation, and the sum is used to obtain the control quantity:

[0084] err(k)=u d (k)-u r (k)

[0085]

[0086] In the formula, V signal is the output control quantity of the PI controller, and is the voltage signal controlling the drive / brake device; err(k) is the PI controller error input at time k, err(k-1) is the error input at time k-1, and the proportional and integral coefficients of the PI control are K and K, respectively. p and K i .

[0087] Furthermore, in step S8, the voltage signal V output by the vehicle through the lower-level vehicle-side controller... signal The motor or brakes of the vehicle control the vehicle body to drive it. The output direction of the signal is determined by judging the positive or negative sign of the voltage signal. A positive voltage signal is output to the motor, and a negative voltage signal is output to the brake.

[0088] The motor is modeled using the following piecewise function:

[0089]

[0090] In the formula, T c P is the torque value on the external characteristic curve of the motor. r n is the rated power of the motor. r is the rated speed of the motor, and n is the actual speed of the motor;

[0091] The vehicle will automatically adjust the voltage signal to obtain the desired motor output torque. The relationship between the voltage signal of the accelerator pedal and the actual output torque of the motor is as follows:

[0092]

[0093] In the formula, V acc The throttle voltage signal, V max This is the maximum voltage signal;

[0094] Ground braking force is limited as follows:

[0095]

[0096] In the formula, For adhesion, The adhesion coefficient, on dry asphalt pavement, ranges from 0.7 to 0.8; therefore, the maximum braking torque T of the vehicle in the braking device model... bmax The constraint is represented as follows:

[0097]

[0098] During braking, the vehicle also automatically adjusts the voltage signal from the brake pedal to control the clamping pressure of the brake pads and calipers in order to obtain the desired braking torque. The relationship between the brake voltage signal and the vehicle's braking torque is as follows:

[0099]

[0100] In the formula, V brake It is the braking voltage signal, T bmax This is the vehicle's maximum braking torque.

[0101] This invention also provides a hierarchical centralized fleet cooperative driving system to realize a cloud-based hierarchical centralized fleet cooperative driving method. The system includes an upper-layer cloud controller, a lower-layer vehicle-end controller, and a vehicle-to-cloud data transmission delay compensator. The upper-layer cloud controller is a model predictive controller, deployed in the cloud. It establishes a centralized fleet state-space equation and an objective function J to obtain an optimal control problem. The problem is solved using the Gaussian pseudospectral method and a sequential quadratic programming algorithm to obtain a series of optimal control quantity time sequences. The input to the lower-layer vehicle-end controller is the optimal control quantity time sequence output by the upper-layer cloud controller. The lower-layer controller uses a PI controller to control the vehicles to follow the input control quantity time sequence. The vehicle-to-cloud data transmission delay compensator is used to compensate for the delay caused by vehicle-to-everything (V2X) communication between the upper and lower-layer controllers, compensating for both uplink and downlink delays in vehicle-to-cloud data transmission. The fleet driving method provided by this invention divides the fleet controller into upper and lower layers in a hierarchical manner, with the upper layer controller mounted in the cloud, thereby reducing vehicle manufacturing costs and avoiding the problem of insufficient vehicle computing power; the centralized control strategy increases the fleet's global optimization capabilities and improves fleet driving efficiency; and the vehicle-cloud data transmission delay compensator solves the communication delay problem of the hierarchical fleet controller, improving the stability of fleet control.

[0102] This invention benefits from the development of vehicle-to-everything (V2X) and cloud computing technologies. A control architecture with the upper-level controller located in the cloud makes it possible to obtain the optimal control time series for convoy cooperative driving through cloud computing. While significantly reducing the cost per vehicle, the high-performance cloud computer also removes the complexity limitations on the upper-level control algorithm. Furthermore, cloud-controlled convoys are easier to centrally control; compared to distributed control, centralized control can achieve globally optimal control.

[0103] Beneficial effects: Compared with the prior art, this invention assumes that the upper-level model predictive controller is located in the cloud and solves the problem of uplink and downlink transmission latency through a latency compensator, avoiding the drawbacks caused by single-vehicle intelligence. It can utilize the higher computing power of the cloud server for fleet control and removes the limitation of computational complexity on the upper-level model predictive controller. Secondly, the cloud-based hierarchical fleet control method can uniformly process and calculate the information of all vehicles in the fleet, thereby realizing a centralized fleet control method and achieving global optimal control of the fleet. Attached Figure Description

[0104] Figure 1 This is a schematic flowchart of the method of the present invention;

[0105] Figure 2 For the variable coefficient of the safety cost term Vehicle spacing Δd i The change curve;

[0106] Figure 3 This is a graph showing the evolution of data transmission latency.

[0107] Figure 4 A comparison chart of the mean absolute error curves in the distance tracking of the three following vehicles;

[0108] Figure 5 This is a graph showing the average absolute error of the speed tracking of the three following vehicles. Detailed Implementation

[0109] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading this invention, any modifications of the invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0110] This invention provides a hierarchical centralized fleet cooperative driving system, comprising an upper-layer cloud controller, a lower-layer vehicle-end controller, and a vehicle-to-cloud data transmission delay compensator. The upper-layer cloud controller is a model predictive controller, deployed in the cloud. It establishes a centralized fleet state-space equation and objective function J to obtain an optimal control problem, which is solved using the Gaussian pseudospectral method and sequential quadratic programming algorithm to obtain a series of optimal control quantity time sequences. The lower-layer vehicle-end controller receives the optimal control quantity time sequence output by the upper-layer cloud controller and uses a PI controller to control vehicles to follow the input control quantity time sequence. The vehicle-to-cloud data transmission delay compensator compensates for the delay caused by vehicle-to-everything (V2X) communication between the upper and lower-layer controllers, compensating for both uplink and downlink delays in V2X data transmission.

[0111] like Figure 1As shown, based on the above-mentioned hierarchical centralized fleet cooperative driving system, this invention provides a cloud-based hierarchical centralized fleet cooperative driving method, comprising the following steps:

[0112] Step (1): Establish the longitudinal state space equation of the fleet through the upper-level cloud controller;

[0113] When the convoy is moving, the error state between the i-th vehicle in the convoy (excluding the lead vehicle) and the (i-1)-th vehicle in front of it is: following distance Δd i and relative velocity difference Δv i ;

[0114] Δd i =p i-1 -p i

[0115] Δv i =v i-1 -v i

[0116] In the formula, p i-1 and v i-1 These are the position and speed of the (i-1)th car, respectively, p i and v i These are the position and speed of the i-th vehicle, respectively. The longitudinal movement of a platoon of N vehicles can be described by the following state-space equation:

[0117]

[0118] y(t)=x(t)

[0119] In the formula, x(t) is the state error vector of the vehicle convoy at time t, which can be expressed as x(t) = [x1(t), x2(t), x3(t)]. T The vehicle sequence in the platoon starts from 0, which is the lead vehicle. The state error of the i-th vehicle in the platoon is x. i (t)=[Δd i (t),Δv i (t)] T , Let A be the differential of the state error vector; let A be the state control matrix; let B be the input matrix; and let u(t) be the vehicle control input (i.e., acceleration) vector, which can be expressed as u(t) = [u0(t), u1(t), u2(t), u3(t)]. T u i (t) represents the control input of the i-th vehicle in the convoy, where u0(t) is the known control input planned by the lead vehicle based on future information.

[0120]

[0121] Step (2): Each vehicle in the convoy uploads its driving information to the cloud server:

[0122] Driving information data packets uploaded by each following vehicle Including upload time k, distance Δd from the vehicle in front. i and its own speed v i .

[0123] Step (3): After receiving the driving information in step (2), the cloud server compensates for it through the time delay compensator:

[0124] The differential state-space equation for the i-th vehicle in the convoy is:

[0125]

[0126] y(t)=x(t)

[0127]

[0128] k+t up,k The discrete integral formula for estimating the driving information uploaded by the i-th vehicle in the convoy at time k is:

[0129]

[0130] Among them, t up,k For the data upload latency, [k,k+t] up,k This time period is based on the sampling interval t sp Divide it into (t) up,k / t sp For each small segment, calculate the state increment for each sampling interval, and finally sum them to obtain k+t. up,k The estimated state x of the i-th vehicle at time i i (k+t up,k |k).

[0131] State Differentiation The calculation uses u i (k+jt sp The values ​​are derived from the previous round of control quantity time series solution results stored in the cloud computer.

[0132] Step (4): Based on the compensated results in step (3), establish the optimal control problem for fleet driving through the fleet cooperative driving optimal control model of the upper-layer cloud controller:

[0133] The optimal control problem for convoy driving is as follows:

[0134]

[0135] st

[0136]

[0137] x(t0)=x initial

[0138] u i ∈[u min ,u max ]

[0139] v i ∈[v min ,v max ]

[0140] In the formula, This is a safety cost item within the process cost, which promotes the convergence of speeds between vehicles, where d0 is the safe stopping distance. c2(Δd) is a variable coefficient. i -h) 2 This is the vehicle spacing maintenance cost item in the process cost, which promotes the distance between vehicles to approach the desired distance h. i 2 To control costs, when the lead car is traveling at a constant speed, this factor causes the acceleration of following vehicles in the convoy to tend towards zero, promoting a balanced state of constant speed among all vehicles in the convoy. Simultaneously, this factor prevents unconstrained changes in acceleration, ensuring passenger comfort. c1, c2, and c3 are all positive values, representing the weighting coefficients of each cost factor. The constraints are given to the differential state-space equations of the vehicle fleet. initial Let u be the convoy state at initial time t0. min ,u max These are the upper and lower limits of vehicle acceleration, v min ,v max These represent the upper and lower limits of vehicle speed, respectively.

[0141] In this embodiment, when d0 is 7m, the change of the variable coefficient is as follows: Figure 2 As shown, with the following distance Δd i The decrease of the variable coefficient This increases the size of the facility, thereby increasing the penalties for safety-related costs.

[0142] The desired distance h between vehicles in a convoy is a variable time interval, and its expression is:

[0143]

[0144] In the formula, v free The free-flow velocity is taken as 33.3 m / s in this embodiment.

[0145] Step (5): The upper-level controller solves this optimal control problem based on the Gaussian pseudospectral method and the sequential quadratic programming algorithm to obtain the time series of the optimal control quantities for each vehicle in the fleet. The specific process is as follows:

[0146] (1) Interpolation point selection: The zeros of the Legendre polynomial defined on the interval [-1, 1] are used as interpolation points to discretize the optimal control problem. The M-order Legendre polynomial is expressed as follows:

[0147]

[0148] Let P M If (τ) = 0, the zeros τ1, τ2…τ located in the range (-1, 1) can be obtained. M That is, the Legendre-Gauss (LG) point.

[0149] (2) Interval transformation: Transform the time interval of the optimal control problem t∈[t0,t... f The transformation formula is used to convert the value to τ∈[-1,1].

[0150]

[0151] The original optimal control problem, after being transformed to the interval τ∈[-1,1], takes the following form:

[0152]

[0153] st

[0154]

[0155] B(x(τ0),t0,x(τ f ),t f ) = 0

[0156] C(x(τ),u(τ),τ;t0,t f )≤0

[0157] in, The constraints for the transformed differential state-space equations are B(x(τ0),t0,x(τ)). f ),t f ) = 0 is x(t0) = x initial The transformed boundary constraints are C(x(τ),u(τ),τ;t0,t). f )≤0 is u i ∈[u min ,u max ] and v i ∈[v min ,v max The process constraints after the transformation.

[0158] (3) Discretization approximation of the objective function: For a continuous integral objective function on the interval [-1, 1], the approximate expression of discretization using the Gauss-Legend quadrature formula is as follows:

[0159]

[0160] In the formula, w m Let X(τ) be the Gaussian integral weight of the m-th term. m ) and U(τ m ) for in τ m Discretized state and control input variables; τ0=-1 is a selected initial point, τ1,τ2…τ M M LG points were selected.

[0161] (4) Global interpolation approximation of state and control variables: Construct M+1 M-order interpolation basis functions l m (τ)(m=0,1…M):

[0162]

[0163] st

[0164]

[0165] The state variables are approximately represented by the Lagrange interpolation polynomial constructed from the basis functions described above:

[0166]

[0167] The control variables are approximately represented by the Lagrange interpolation polynomial constructed from the basis functions described above:

[0168]

[0169] (5) State-space equation constraint transformation: In order to transform the differential constraints of the vehicle system dynamics into algebraic constraints, the approximate expressions of the state variables are differentiated:

[0170]

[0171] In the formula, D jm ∈R M×(M+1) The differential matrix is ​​represented as follows:

[0172]

[0173] In the formula, j takes values ​​of 1, 2, ..., M. This allows for the interpolation of the differential state-space equation at the interpolation node τ. j The discrete approximation at a given point. Substituting this into the differential state-space equations, we can transform it into an algebraic constraint:

[0174]

[0175] Similarly, for boundary constraints and path constraints at the interpolation point τ j Discretize to transform these two constraints:

[0176] B(X(τ0),t0,X(τ f ),t f ) = 0

[0177] C(X(τ j ),U(τ j ),τ j ;t0,t f )≤0

[0178] (6) Solving the nonlinear programming problem: After the above steps, the optimal control problem of the convoy cooperative driving is transformed into a discrete NLP problem. The NLP problem is solved by using the SNOPT software package based on the sequential quadratic programming algorithm, and the optimal control quantity time series that satisfies the constraints of the differential state space equation, process and boundary constraints can be obtained.

[0179] Step (6): The optimal control time series generated in step (5) is included in a data packet and sent to each vehicle in the fleet; the data packet includes the optimal control time series U. i (k) and the time of issuance k:

[0180] U i (k)=[u i (k),u i (k+t sp ...,u i (k+N x t sp )],N x <N p

[0181]

[0182] In the formula, N P To predict the interval length, N x For data packets The shortest interval of the time series of the control quantity is the longest.

[0183] Step (7): After receiving the data packet at the vehicle end, the time series of the optimal control quantity is compensated by the time delay compensator;

[0184] Due to transmission delay, the i-th car will arrive at k+t. dw,k Receive data packets at all times At this point, the delay compensation strategy will be applied to the control quantity time series U. i Choose k+t in (k) dw,k The control quantity after a certain time is used as the control signal for the next round.

[0185] Step (8): Based on the optimal control quantity time series u after compensation in step (7) d The lower-level vehicle-side controller controls the vehicle to track its movement:

[0186] The lower-level vehicle-end controller is a PI controller, whose basic principle is that the system control target u d (k) and the actual output value of the system u r Subtracting (k) yields the system deviation err(k), then proportional and integral calculations are performed using the system deviation, and the sum is used to obtain the control quantity:

[0187] err(k)=u d (k)-u r (k)

[0188]

[0189] In the formula, V signal is the output control quantity of the PI controller, and is the voltage signal controlling the drive / brake device; err(k) is the PI controller error input at time k, err(k-1) is the error input at time k-1, and the proportional and integral coefficients of the PI control are K and K, respectively. p and K i The vehicle outputs a voltage signal V through the lower-level vehicle-side controller. signal It controls the vehicle's motor or brakes to drive the vehicle body. The output direction of the signal is determined by the sign of the voltage signal: a positive voltage signal is output to the motor, and a negative voltage signal is output to the brake.

[0190] The motor is modeled using the following piecewise function:

[0191]

[0192] In the formula, T c P is the torque value on the external characteristic curve of the motor. r n is the rated power of the motor. r is the rated speed of the motor, and n is the actual speed of the motor.

[0193] The vehicle will automatically adjust the voltage signal to obtain the desired motor output torque. The relationship between the voltage signal of the accelerator pedal and the actual output torque of the motor is as follows:

[0194]

[0195] In the formula, Vacc The throttle voltage signal, V max This is the maximum voltage signal.

[0196] Ground braking force is limited as follows:

[0197]

[0198] In the formula, For adhesion, The adhesion coefficient, on dry asphalt surfaces, ranges from 0.7 to 0.8 in this embodiment. Therefore, the maximum braking torque T of the vehicle in the braking device model... bmax The constraint is represented as follows:

[0199]

[0200] During braking, the vehicle automatically adjusts the voltage signal from the brake pedal to control the clamping pressure of the brake pads and calipers, thereby obtaining the desired braking torque. The relationship between the brake voltage signal and the vehicle's braking torque is as follows:

[0201]

[0202] In the formula, V brake It is the braking voltage signal, T bmax This is the vehicle's maximum braking torque.

[0203] Step (9): Repeat steps (2) to (8) continuously to achieve coordinated driving of the fleet.

[0204] Based on the above, in order to verify the effectiveness and practical effect of the present invention, this embodiment will conduct an example simulation of the above solution, as follows:

[0205] In this embodiment, the predetermined values ​​for the upper-layer cloud controller and the lower-layer vehicle controller are shown in Table 1 below:

[0206] Table 1. Preset values ​​for the upper-layer cloud controller and the lower-layer vehicle controller.

[0207]

[0208]

[0209] The simulation platform was built using Matlab and Simulink. The upper-level cloud controller and the vehicle-to-cloud data transmission delay compensator were simulated using .m programs, while the lower-level vehicle controller, the vehicle's motor, brakes, and the single-vehicle model were simulated using Simulink.

[0210] In this embodiment, three methods are employed: a distributed platoon control method, an IDM-based platoon control method, and a cloud-based model prediction control method. The simulation environment is configured as follows: data transmission latency is [details to be inserted here]. Figure 3 The initial positions and speeds of the four vehicles in the convoy are set as follows:

[0211]

[0212] In the formula, h represents the desired vehicle spacing. The convoy is in a normal driving state at the initial moment, that is, the vehicles in the convoy are at the same speed and the distance between the front and rear vehicles is the desired vehicle spacing.

[0213] To verify the performance of the fleet controller in the face of traffic oscillations, the lead car exhibited acceleration and deceleration behaviors in the simulation. The lead car decelerated at times 5s and 40s, with the deceleration rate set to -1m / s². 2 (lasting 5 seconds) and -2 m / s 2 (Continued for 4 seconds); acceleration is performed at 20 seconds and 54 seconds, with acceleration set to 1 m / s² respectively. 2 (lasting 5 seconds) and 2 m / s 2 (Duration 4s). The prediction interval N for each round in this simulation. P Duration is 10 seconds, control interval N C Duration 0.5s, sampling time t sp The time is 0.01s, a total of 160 rounds of rolling updates are performed, the simulation time is 80s, and the other parameter values ​​are set according to the default values ​​in the above text.

[0214] Figure 4 and Figure 5 The mean absolute tracking error curves of the three following vehicles in the platoon were compared under centralized and distributed control strategies, as well as under an IDM-based control strategy, for both distance and speed. Combined with... Figure 4 and Figure 5 As shown in Table 2, for vehicle distance tracking error, the centralized control strategy has the best control effect, followed by the distributed control strategy, and the IIDM-based control strategy is the worst. The average vehicle distance tracking error of the centralized control strategy is 0.123m, which is 54.28% and 77.39% lower than the latter two, respectively. Furthermore, the centralized control strategy provided by this invention also converges to the desired vehicle distance the fastest, which is 10s and 4s faster than the latter two, respectively.

[0215] Table 2 Comparison of data results under three control strategies

[0216]

[0217]

[0218] For speed tracking error, the error curves under centralized and distributed control strategies are basically similar, and the peak error values ​​are also basically equal. However, the peak error value of the IDM-based control strategy is much larger than the former two. The average speed tracking errors of centralized, distributed, and IDM-based strategies are 0.182 m / s, 0.181 m / s, and 0.329 m / s, respectively. It can be seen that although the average error of centralized is higher than that of distributed, the difference is not significant. Centralized is only 0.55% higher than distributed, while the speed tracking error of IDM-based is much larger than that of centralized, exceeding it by 80.77%. In addition, the centralized control strategy of this invention also converges to the desired vehicle speed the fastest, 10 seconds and 2 seconds faster than distributed and IDM-based strategies, respectively.

[0219] Furthermore, the error mean curve of the distributed control strategy showed obvious oscillations, while the error curves of the centralized and IDM-based control strategies changed more smoothly.

[0220] In summary, the centralized control strategy of this invention has a better tracking control effect. Although its speed tracking performance is slightly higher than that of the distributed system, it is superior to the other two in terms of vehicle distance tracking performance, desired vehicle distance, and desired speed convergence speed.

Claims

1. A hierarchical centralized fleet cooperative driving method based on cloud computing, characterized in that, Includes the following steps: S1: Establish the longitudinal state-space equations for the vehicle fleet; S2: According to the longitudinal state space equation of the platoon, each following vehicle in the platoon uploads its own driving information to the cloud server. S3: After receiving the driving information, the cloud server compensates for the driving information; S4: Based on the compensated driving information, establish the optimal control problem for fleet driving; S5: Solve the optimal control problem established in step S4 based on the Gaussian pseudospectral method and sequential quadratic programming algorithm to obtain the time series of optimal control quantities for each vehicle in the fleet; S6: Include the generated optimal control time series in a data packet and send it to each vehicle in the fleet; S7: After receiving the data packet at the vehicle end, compensate for the time series of the optimal control quantity; S8: Control the vehicle to follow its movement based on the time series of the compensated optimal control quantity; S9: Repeat steps S2 to S8 continuously to achieve coordinated driving of the convoy; The specific process of compensating the uploaded driving information using the discrete numerical integration method through the time delay compensator in step S3 is as follows: The differential state-space equation for the i-th vehicle in the convoy is: ； Always The discrete integral formula for estimating the driving information uploaded by the i-th vehicle in the convoy at time step is: ； in, For data upload latency, This time period is based on the sampling interval. Divide it into For each small segment, the state increment is calculated sequentially for each sampling interval, and finally summed to obtain the result. Estimated state of the i-th vehicle at time i ; Among them, state differential The calculation used The values ​​are derived from the time series solution results of the previous round of control variables stored in the cloud computer; The optimal control problem for convoy driving established in step S4 is expressed as follows: ； in, ； ； ； ； In the formula, This is a safety cost item within the process cost, which promotes the convergence of speeds between vehicles. For safe parking distance, These are variable coefficients; This is the vehicle distance maintenance cost item in the process cost, which causes the distance between vehicles to tend towards the desired distance. ; To control costs, this item causes the acceleration of following vehicles in the convoy to approach 0 when the lead car is traveling at a constant speed, thus promoting the equilibrium of the vehicles in the convoy to travel at a constant speed. At the same time, this item also avoids changes in acceleration that are not subject to penalty constraints. All are positive values, representing the weighting coefficients of each cost item. The constraints are the differential state-space equations of the vehicle fleet. Initial time The condition of the team These are the upper and lower limits of vehicle acceleration, respectively. These represent the upper and lower limits of vehicle speed, respectively.

2. The hierarchical centralized fleet cooperative driving method based on cloud computing according to claim 1, characterized in that, The method for establishing the longitudinal state-space equation of the vehicle fleet in step S1 is as follows: When the convoy is moving, the error state between the i-th vehicle in the convoy (excluding the lead vehicle) and the (i-1)-th vehicle in front of it is: following distance. and relative speed difference The expression is as follows: ； In the formula, and These are the position and speed of the (i-1)th car, and These are the position and speed of the i-th car, respectively; The longitudinal movement of a platoon with N vehicles is described by the following state-space equations: ； In the formula, for The state error vector of the vehicle at time t is expressed as: The vehicle sequence in the platoon starts from 0, which is the lead vehicle. The state error of the i-th vehicle in the platoon is... , It is the differential of the state error vector; For the state control matrix, The input matrix; The vehicle control input vector is expressed as: , The control input for the i-th vehicle in the convoy, where The known control inputs are planned by the lead vehicle based on future information; in, ； ； ； 。 3. The hierarchical centralized fleet cooperative driving method based on cloud computing according to claim 2, characterized in that, The driving information uploaded in step S2 includes the upload time, distance to the vehicle in front, and vehicle speed; in step S3, the uploaded driving information is compensated using a discrete numerical integration method through a time delay compensator.

4. The hierarchical centralized fleet cooperative driving method based on cloud computing according to claim 3, characterized in that, The vehicle distance tends towards the desired vehicle distance. For variable time intervals, the expression is as follows: ； In the formula, This refers to the free-flow vehicle speed.

5. A hierarchical centralized fleet cooperative driving method based on cloud computing according to claim 4, characterized in that, In step S5, the Gaussian pseudospectral method is used to discretize the optimal control problem, resulting in the time series of the optimal control quantities for each vehicle in the fleet. The specific process is as follows: A1: Interpolation point selection: The zeros of the Legendre polynomial defined on the interval [-1, 1] are used as interpolation points to discretize the optimal control problem. The M-order Legendre polynomial is expressed as follows: ； in, Represents the interpolation point; make The zero point located in the range (-1,1) can be obtained. That is, the Legendre-Gauss (LG) point; A2: Interval Transformation: Changing the time interval of the optimal control problem Converted to using the conversion formula The conversion formula is as follows: ； The original optimal control problem is transformed into an interval The following format is as follows: ； st ； ； ； in, The constraints are for the transformed differential state-space equations. for Boundary constraints after transformation for and Process constraints after conversion; A3: Discretization Approximation of the Objective Function: For a continuous integral objective function on the interval [-1, 1], the approximate expression discretized using the Gauss-Legendary quadrature formula is as follows: ； In the formula, Let m be the Gaussian integral weight. and In order to be in Discretized state and control input variables; For an initial point to be selected, M LG points were selected; A4: Global Interpolation Approximation State and Control Variables: Constructing M+1 M-order interpolation basis functions : ； st ； The state variables are approximately represented by the Lagrange interpolation polynomial constructed from the basis functions described above: ； The control variables are approximately represented by the Lagrange interpolation polynomial constructed from the basis functions described above: ； A5: State-space equation constraint transformation: In order to transform the differential constraints of the vehicle system dynamics into algebraic constraints, the approximate expressions of the state variables are transformed. Differentiate: ； In the formula, The differential matrix is ​​represented as follows: ； In the formula, Choosing values ​​1, 2, ..., M, allows for the interpolation of the differential state-space equation at the interpolation nodes. The discrete approximation at a given point is then substituted into the differential state-space equations to transform them into algebraic constraints: ； Similarly, boundary constraints and path constraints are applied at the interpolation points. Discretize to transform these two constraints: ； ； A6: Solving the nonlinear programming problem: After the above steps, the optimal control problem of the convoy cooperative driving is transformed into a discrete NLP problem. The SNOPT software package based on the sequential quadratic programming algorithm is used to solve the NLP problem, and the optimal control quantity time series that satisfies the constraints of the differential state space equation, process and boundary constraints is obtained.

6. The hierarchical centralized fleet cooperative driving method based on cloud computing according to claim 1, characterized in that, In step S6, the data packet includes the optimal control quantity time sequence and the transmission time. In step S7, the optimal control quantity time sequence is compensated by a time delay compensator. The specific compensation method is as follows: after receiving the data packet, the vehicle end compares the current time with the transmission time in the data packet and only extracts the optimal control quantity time sequence packet after the receiving time as the control signal.

7. The hierarchical centralized fleet cooperative driving method based on cloud computing according to claim 1, characterized in that, In step S8, the vehicle is controlled to track its movement by a lower-level vehicle-end controller. The lower-level vehicle-end controller is a PI controller, and its control principle is: system control target Compared with the actual output value of the system Subtraction yields the systematic deviation Then, proportional and integral calculations are performed using the system deviation, and the results are summed to obtain the control quantity: ； ； In the formula, The output control quantity of the PI controller is the voltage signal used to control the drive / brake equipment; It is the current moment. The error input of the PI controller, For a moment The error input, and the proportional and integral coefficients of the PI control are respectively and .

8. The hierarchical centralized fleet cooperative driving method based on cloud computing according to claim 1, characterized in that, In step S8, the vehicle outputs a voltage signal through the lower-level vehicle-side controller. The motor or brakes of the vehicle control the vehicle body to drive it. The output direction of the signal is determined by judging the positive or negative sign of the voltage signal. A positive voltage signal is output to the motor, and a negative voltage signal is output to the brake. The motor is modeled using the following piecewise function: ； In the formula, This refers to the torque value on the motor's external characteristic curve. This refers to the rated power of the motor. This refers to the rated speed of the motor. This refers to the actual speed of the motor; The vehicle will automatically adjust the voltage signal to obtain the desired motor output torque. The relationship between the voltage signal of the accelerator pedal and the actual output torque of the motor is as follows: ； In the formula, This is the throttle voltage signal. This is the maximum voltage signal; Ground braking force is limited as follows: ； In the formula, For adhesion, The adhesion coefficient; the maximum braking torque of the vehicle in the braking system model. The constraint is represented as follows: ； In the formula, Represents the rated lever arm; During braking, the vehicle also automatically adjusts the voltage signal from the brake pedal to control the clamping pressure of the brake pads and calipers in order to obtain the desired braking torque. The relationship between the brake voltage signal and the vehicle's braking torque is as follows: ； In the formula, It is the braking voltage signal. This is the vehicle's maximum braking torque.

Images