Vehicle formation data driving prediction cloud control method under time-varying topology

Abstract

The invention discloses a vehicle formation data-driven prediction cloud control method under time-varying topology, and the method comprises the steps: carrying out the modeling of a vehicle formation under a linear parametric system, enabling a model to describe the change of a topological structure of the vehicle formation, and carrying out the control through a data-driven prediction control algorithm, therefore, the topology change caused by vehicle departure and the like can be effectively dealt with. Besides, by deploying a control algorithm in a cloud server, sufficient computing resources of the cloud can be utilized to solve the optimization problem, the real-time performance of the control process is ensured, and therefore an effective solution can be provided when the vehicle-mounted hardware is difficult to meet the control requirement due to the increase of the vehicle formation scale.

Description

Technical Field

[0001] The present invention belongs to the field of automatic control technology, and specifically relates to a vehicle formation data-driven predictive cloud control method under time-varying topology. Background Art

[0002] With the continuous deepening of research on data-driven predictive control theory, researchers have proposed a data predictive control method for linear time-invariant systems based on Willem's fundamental lemma, which is named the DeePC (Data-EnablEd Predictive Control) method. Compared with model predictive control and traditional subspace identification-based data-driven predictive control methods, the advantage of the DeePC method is that it does not rely on prior knowledge such as the specific model of the system, does not require identification of model parameters, and can control the system using only input and output data. Subsequent research has extended the DeePC method to linear parameter-varying systems, further improving its control capabilities for dynamic time-varying systems.

[0003] While some research has applied the DeepPC method to vehicle platooning control, it is based on a linear time-invariant system and is not suitable for the topological changes that may occur in real-world scenarios. Furthermore, as the size of a platoon increases, the load on data storage, communication, and computing tasks increases, and on-board hardware may struggle to meet control requirements. Integrating cloud control is an effective approach to addressing this challenge. Cloud control technology aims to introduce cloud computing to ensure real-time control and deploy control methods in the cloud, thereby achieving autonomous intelligent control of the system.

[0004] In summary, current research on data-driven predictive control for vehicle platoons cannot cope with topology changes caused by reasons such as vehicle leaving the platoon or vehicle cutting in line. Furthermore, it cannot provide a good solution when the size of the platoon increases, making it difficult for the on-board hardware to meet control requirements. Summary of the Invention

[0005] In view of this, the present invention provides a data-driven predictive cloud control method for vehicle formations under time-varying topology, which can handle and adapt to scenarios where the topology structure of vehicle formations changes, and ensures the real-time control by combining cloud computing.

[0006] The present invention provides a data-driven predictive cloud control method for vehicle formations under time-varying topology, comprising the following steps:

[0007] Step 1: For a vehicle formation consisting of m intelligent connected vehicles, the control goal is to control the distance and speed of each vehicle to the target distance s * With target speed v * , the linear parameter-varying system state space equation model of vehicle formation under time-varying topology is established as follows:

[0008]

[0009] Among them, x(k+1) is the state of the system at time (k+1), x(k) is the state of the system at time k, y(k) is the output of the system at time k, C is the output matrix, A(p k ) is the system matrix, p k is the scheduling signal, B′(p k ) is the input matrix, u′(k) is the expanded input vector of the system;

[0010] Step 2: Based on the linear parameter-variable system state-space equation model constructed in Step 1, the DeePC method is used and slack variables σ1 and σ2 are introduced to establish a robust solver for time-varying topology vehicle formations, as shown below:

[0011]

[0012] And ω j (k)=0,j∈[0,m-1]

[0013] in, is the cost function of the optimization problem, u [k-n,k-1] is the initial input trajectory of the system, is the initial output trajectory of the system containing noise, is the prediction input obtained by solving the optimization problem, For The corresponding predicted output is, is the upper bound of the output noise, H L+n is the L+n order Hankel matrix, is a quadratic function of input and output, is the input scheduling output sequence collected offline from the vehicle formation, g(k) is the solution of the equation in the first constraint condition, is the output affected by noise, is the upper bound of noise, λ g 、 and are the weights of the corresponding variable's squared two-norm;

[0014] Step 3. Deploy the robust solver on the cloud server and deploy the vehicle formation on the local computer. Before executing the control task, the local computer sends the input scheduling and output sequence obtained from the vehicle formation offline to the cloud server. The cloud server constructs the Hankel matrix in the optimization problem based on the received sequence. During the execution of the control task, at each sampling time k, the local computer sends the initial input scheduling and output sequence of the vehicle formation to the cloud server. The cloud server uses the robust solver to solve the optimization problem, and then sends the optimal control quantity at time k+1 to the local computer. The local computer uses the received control quantity to complete the control of the vehicle formation. Then, the local computer obtains the input scheduling and output sequence at the current time and uses it to update the initial input scheduling and output sequence of the vehicle formation. A new round of control process is executed at the next sampling time until the set number of cycles is reached or a stop command is received.

[0015] Furthermore, the local computer sends the input scheduling output sequence obtained from the vehicle formation offline to the cloud server by using the gRPC protocol to send data.

[0016] Furthermore, the p k for:

[0017]

[0018] At time k, if vehicle i is in the platoon, then otherwise

[0019] Furthermore, the B′(p k ) is B′(p k )=[B(p k ) H d ],B(p k )=diag(p k )B d ,

[0020] Beneficial effects:

[0021] The present invention models vehicle platoons under a linear parameter-varying system. The model can describe changes in the topological structure of the vehicle platoon, thereby effectively responding to topological changes caused by reasons such as vehicle leaving the platoon. In addition, by deploying the control algorithm on a cloud server, the sufficient computing resources of the cloud can be used to solve the optimization problem, ensuring the real-time nature of the control process. This can provide an effective solution when the increase in the size of the vehicle platoon makes it difficult for the on-board hardware to meet the control requirements. DETAILED DESCRIPTION

[0022] The present invention is described in detail with reference to the following examples.

[0023] The symbols involved in the present invention are: The vector expression of a sequence is defined as:

[0024]

[0025] The L-order Hankel matrix generated by a sequence is defined as:

[0026]

[0027] The definition of the diagonalization operation diag(x) represents the diagonalization of the column vector x into a diagonal matrix.

[0028] The block diagonal Coriolis product operation is defined as:

[0029]

[0030] in The representative dimension is n ψ the identity matrix, Represents Korneck product.

[0031] Defining Operations for:

[0032]

[0033] Where p represents a scheduling signal of a linear parameter-variable system.

[0034] Define unit column quantity represents a p-dimensional column vector with only the rth row being 1 and the rest being 0.

[0035] The present invention provides a data-driven predictive cloud control method for vehicle formations under time-varying topology, which mainly includes the following steps:

[0036] Step 1: For a vehicle formation consisting of m intelligent connected vehicles, the control goal is to control the distance and speed of each vehicle to the target distance s * With target speed v * , the linear parameter-varying system state space equation model of vehicle formation under time-varying topology is established as follows:

[0037]

[0038] Among them, x(k+1) is the state of the system at time (k+1), x(k) is the state of the system at time k, y(k) is the output of the system at time k, C is the output matrix and has is a p-dimensional unit column vector with only the rth row being 1 and the rest of the rows being 0; A(p k ) is the system matrix and A(pk )=diag(p k )A d , p k is a scheduling signal and At time k, if vehicle i is in the platoon, then otherwise B′(p k ) is the input matrix and B′(p k )=[B(p k ) H d ],B(p k )=diag(p k )B d , u′(k) is the extended input vector of the system and u′(k)=[u(k) T ∈ T ] T , u(k) is the control input vector of the system at time k and u i (k) represents the control input of vehicle i.

[0039] Specifically, the construction process of the linear parameter-varying system state space equation model of vehicle platooning under time-varying topology is as follows:

[0040] For a vehicle formation consisting of m intelligent connected vehicles, the control goal is to control the distance and speed of each vehicle to the target distance s * With target speed v * The vehicle formation state, input and output are defined as follows:

[0041]

[0042] Among them, x(t), u(t), y(t) represent the state, input and output of the system at time t respectively. Indicates the distance between the i-th vehicle and the preceding vehicle relative to the target distance s * The error, Indicates the speed of the i-th vehicle relative to the target speed v * The error, u i (t) represents the control input of the i-th vehicle, i.e., the acceleration. Based on this, the continuous state space equation of the vehicle platooning system can be obtained without considering the topological change:

[0043]

[0044] The definitions of the matrices in the state space equation are as follows:

[0045]

[0046] in,

[0047]

[0048] and represents a p-dimensional unit column vector with only the rth row set to 1 and the rest set to 0, and ∈(t) represents the error in the speed of the leader vehicle followed by the platoon relative to the target speed. Discretizing the above state-space equation yields the following discrete state-space equation:

[0049]

[0050] in, Δt>0 is the sampling time interval, and this part is the existing work.

[0051] The scheduling signal p k Defined as:

[0052]

[0053] At time k:

[0054]

[0055] Define the following matrix:

[0056] A(p k )=diag(p k )A d ,B(p k )=diag(p k )B d

[0057] B′(p k )=[B(p k )H d ],u′(k)=[u(k) T ∈ T ] T ,

[0058] Finally, the state space equation of the vehicle formation under the time-varying topology expressed by a linear parametric system is obtained:

[0059]

[0060] In addition, considering that vehicle i leaves the team, the vehicle following it (i.e., vehicle i+1) will change from vehicle i to vehicle i-1, which will result in a distance error of vehicle i+1. A mutation occurs, and we express this spacing mutation in the following form:

[0061] If ω i(k-1)=1 and ω i (k) = 0

[0062] in, represents the transpose of the unit column vector, y(k) + represents the output of the vehicle formation after the i-th vehicle leaves the formation, y(k) - represents the output of the vehicle platoon before the i-th vehicle leaves the platoon.

[0063] Step 2: Based on the linear parameter-varying system state-space equation model constructed in Step 1, the DeePC method is used and the influence of bounded additive noise on the system output is considered. Slack variables σ1 and σ2 are introduced to establish a robust LPV-DPC solver for time-varying topology vehicle formations, as shown below:

[0064]

[0065] And ω j (k)=0,j∈[0,m-1]

[0066] in, is the cost function of the optimization problem, u [k-n,k-1] is the initial input trajectory of the system, is the initial output trajectory of the system containing noise, is the prediction input obtained by solving the optimization problem, For The corresponding predicted output is, is the upper bound of the output noise, H L+n is the L+n order Hankel matrix, is a quadratic function of input and output, is the input-dispatching-output sequence collected offline from the vehicle formation, g(k) is the solution of the equation in the first constraint condition, is the output affected by noise, is the upper bound of noise, λ g 、 and are the weights of the corresponding variable's squared norm.

[0067] The establishment process of the above data-driven predictive control optimization problem is as follows:

[0068] Step 2.1: Use the DeepPC method (Linear Parameter Varying System–Data-enabled Predictive Control, LPV-DPC) to formulate a data-driven predictive control optimization problem for a time-varying topology vehicle platoon, as follows:

[0069]

[0070] And ω j (k)=0,j∈[0,m-1]

[0071] Step 2.2: Considering the possible influence of noise in practical applications, the influence of bounded additive noise on the system output is introduced. By introducing slack variables σ1 and σ2, the following robust LPV-DPC method is established:

[0072]

[0073] Step 3: Deploy the robust LPV-DPC solver constructed in Step 2 on the cloud server and deploy the vehicle formation on the local computer. Before executing the control task, the local computer will collect the input-scheduling-output sequence from the vehicle formation offline. The data is sent to the cloud server via the gRPC protocol. The cloud server constructs the Hankel matrix in the optimization problem based on the received input-scheduling-output sequence. During the execution of the control task, at each sampling time k, the local computer converts the initial input-scheduling-output sequence {u [-n,-1] (k),p [-n,-1] (k),y [-n,-1] (k)} is sent to the cloud server, which uses the robust LPV-DPC solver to solve the optimization problem and then sends the optimal control quantity at time k+1 to the local computer. The local computer uses the received control quantity to complete the control of the vehicle formation. Then, the local computer obtains the input-scheduling-output sequence at the current moment and uses it to update the initial input-scheduling-output sequence of the vehicle formation. A new round of control process is executed at the next sampling moment until the set number of cycles is reached or a stop command is received.

[0074] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A data-driven predictive cloud control method for vehicle platooning under time-varying topology, characterized in that: The following steps are involved: Step 1: For a vehicle formation consisting of m intelligent connected vehicles, the control goal is to control the distance and speed of each vehicle to the target distance s * With target speed v * , the linear parameter-varying system state space equation model of vehicle formation under time-varying topology is established as follows: Among them, x(k+1) is the state of the system at time (k+1), x(k) is the state of the system at time k, y(k) is the output of the system at time k, C is the output matrix, A(p k ) is the system matrix, p k is the scheduling signal, B′(p k ) is the input matrix, u′(k) is the expanded input vector of the system; Step 2: Based on the linear parameter-variable system state-space equation model constructed in Step 1, the DeePC method is used and slack variables σ1 and σ2 are introduced to establish a robust solver for time-varying topology vehicle formations, as shown below: in, is the cost function of the optimization problem, u [k-n,k-1] is the initial input trajectory of the system, is the initial output trajectory of the system containing noise, is the prediction input obtained by solving the optimization problem, For The corresponding prediction output is, is the upper bound of the output noise, H L+n is the L+n order Hankel matrix, is a quadratic function of input and output, is the input scheduling output sequence collected offline from the vehicle formation, g(k) is the solution of the equation in the first constraint condition, is the output affected by noise, is the upper bound of the noise, and are the weights of the corresponding variable's second norm squared; Step 3. Deploy the robust solver on the cloud server and deploy the vehicle formation on the local computer. Before executing the control task, the local computer sends the input scheduling and output sequence obtained from the vehicle formation offline to the cloud server. The cloud server constructs the Hankel matrix in the optimization problem based on the received sequence. During the execution of the control task, at each sampling time k, the local computer sends the initial input scheduling and output sequence of the vehicle formation to the cloud server. The cloud server uses the robust solver to solve the optimization problem, and then sends the optimal control quantity at time k+1 to the local computer. The local computer uses the received control quantity to complete the control of the vehicle formation. Then, the local computer obtains the input scheduling and output sequence at the current time and uses it to update the initial input scheduling and output sequence of the vehicle formation. A new round of control process is executed at the next sampling time until the set number of cycles is reached or a stop command is received.

2. The data-driven predictive cloud control method for vehicle formations under time-varying topology according to claim 1 is characterized in that: The local computer sends the input scheduling output sequence obtained from the vehicle formation offline to the cloud server by using the gRPC protocol to send data.

3. The data-driven predictive cloud control method for vehicle formations under time-varying topology according to claim 1 is characterized in that: The p k for: At time k, if vehicle i is in the platoon, then otherwise 4. The data-driven predictive cloud control method for vehicle platooning under time-varying topology according to claim 1 is characterized in that: The B′(p k ) is B′(p k )=[B(p k ) H d ],B(p k )=diag(p k )B d ,

Images