A logistics AGV platoon cooperative control method based on a DMG-MPC algorithm

Abstract

The application relates to a logistics AGV formation cooperative control method based on a DMG-MPC algorithm, which comprises the following steps of defining initialization parameters, constructing a model, solving optimal temporary variables by using estimated gradients, updating local variables of a current node and projecting in combination with neighbor nodes, updating auxiliary variables, obtaining the gradient of the current node by updating a gradient estimator, designing an iterative control algorithm and performing inner and outer double-layer iterations. By adopting a double-layer structure, the non-convex cooperative optimization problem constructed in a multi-AGV system is solved while the feasibility and calculation efficiency of distributed solving are ensured, the momentum method is combined with the gradient tracking method, a faster convergence rate is obtained, the projection operation is adopted to strictly guarantee the feasibility of AGV dynamics constraints, a plurality of AGVs can realize full autonomous cooperative navigation and dynamic obstacle avoidance in a warehouse environment, in an emergency working condition such as a device fault or a passage blockage scene, transportation tasks are re-distributed and a cooperative bypass path is generated, and the overall loading and unloading efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of distributed multi-agent control system, and particularly relates to a logistics AGV formation cooperative control method based on a DMG-MPC algorithm. BACKGROUND

[0002] The cooperative control demand of the automatic warehouse system for large-scale AGV cluster is increasingly urgent. In the modern logistics hub, dozens or even hundreds of automatic guided vehicles need to perform cooperative handling tasks among high-density shelf channels, which puts forward strict requirements for the real-time trajectory planning technology of the multi-vehicle system: on the one hand, the calculation complexity brought by the non-convex obstacle avoidance constraint between vehicles needs to be solved, and on the other hand, the dynamic disturbance such as forklift insertion and goods sliding needs to be coped with, while the timely control response cycle needs to be met to ensure the safety of operation. The traditional cooperative control method is limited when coping with the non-convex coupled constraints of the multi-vehicle system: the distributed solving framework is inefficient in dealing with high-dimensional constraints, which restricts the real-time control performance. The technical evolution trend shows that efficient multi-agent cooperative control needs to integrate constraint feasibility and calculation real-time performance. However, the current method does not achieve a balance between convergence speed and numerical stability when solving the distributed non-convex optimization problem with dynamics constraints. This limitation restricts the large-scale application of logistics AGV systems in complex scenarios such as narrow channel operation and high-density cluster. SUMMARY

[0003] In view of the deficiencies in the prior art, the present application provides a logistics AGV formation cooperative control method based on a DMG-MPC algorithm, which solves the problem of insufficient processing efficiency of the traditional logistics AGV system when solving complex non-convex obstacle avoidance constraints between vehicles. The present application combines the momentum method with the gradient tracking method to obtain a faster convergence rate, and uses projection operation to strictly guarantee the feasibility of AGV dynamics constraints. The method can avoid obstacles that may exist in the environment during AGV travel, while ensuring safety and path efficiency.

[0004] To solve the above technical problems, the present application provides the following technical scheme: a logistics AGV formation cooperative control method based on a DMG-MPC algorithm, comprising the following steps:

[0005] S1, initializing the parameters of the DMG-MPC algorithm, importing the position information and target position information of each node of the AGV, performing data processing, and constructing AGV dynamics constraints;

[0006] S2, constructing a model predictive control problem, obtaining a problem formula, and using an SCA algorithm to convex the constructed problem formula;

[0007] S3, based on the distributed structure, using the estimated gradient to solve the optimal temporary variable of the model predictive control problem at each node;

[0008] S4, updating the local variable of the current node and projecting the local variable into the constraint set in combination with the optimal temporary variable of the neighbor node on each node;

[0009] S5, updating the auxiliary variable of the current node by using the Nesterov momentum method on each node;

[0010] S6, updating the gradient estimator based on the gradient tracking method in combination with the auxiliary variable and the gradient of the neighbor node on each node, and obtaining the gradient of the current node;

[0011] S7, repeating steps S2-S6 for external iteration, and repeating steps S3-S6 for internal iteration, until the iteration ends when the algorithm ending condition is met.

[0012] Further, in step S1, the specific process includes the following steps:

[0013] S11, the parameters of the DMG-MPC algorithm are initialized, including the number of AGVs, that is, the number of nodes m, error tolerance γ1, γ2, maximum iteration number, gradient information Where is the gradient calculation function;

[0014] S12, the position information of each node of the AGV includes horizontal and vertical coordinates, horizontal and vertical velocities, and acceleration, and the target position information includes horizontal and vertical coordinates;

[0015] S13, the AGV dynamics constraints include agent position constraints, control input constraints, and non-convex constraints of the avoidance condition between each pair of AGVs.

[0016] Further, the AGV adopts a bidirectional ring topology structure for the communication network.

[0017] Further, in step S2, the specific process includes the following steps:

[0018] S21, constructing a model predictive control problem Wherein at time step t, given Optimize the control input sequence by constructing the following coupled cost function Where u i (s; t) represents the predicted input sequence at time t+s; denotes the set of natural numbers, and in the above control problem:

[0019]

[0020] In the formula, x i (t) represents the initial state of the i-th node at time t, and the control input is ui x i (s;t;u i ) denotes the state solution of the node at time t+s; denotes the coupling cost function; V fi (·) denotes the terminal cost; m denotes the number of nodes; N denotes the prediction horizon length.

[0021] S22, convexifying the problem formula J i (x(t);u) in step S21 using the SCA algorithm, denoted as

[0022] Further, in step S3, the specific process includes the following steps:

[0023] S31, updating the estimated gradient of the kth iteration The update formula is expressed as:

[0024] The update formula is expressed as:

[0025]

[0026] where is the gradient calculation function, denotes the auxiliary variable of the kth iteration of node i, is the gradient of node j at the kth iteration, and m denotes the number of nodes;

[0027] S32, updating the optimal temporary variable of the model predictive control problem using the estimated gradient The update formula is:

[0028]

[0029] For the distributed structure, a local copy is introduced for each node where denotes the input sequence of node j assumed by node i, and then is defined as the combination of the original variable u i of node i and the variable , where denotes the problem formula after the convexification process in step S22, denotes the auxiliary variable of the kth iteration of node i, denoted as the optimal temporary variable of the kth iteration of node i, denotes the local constraint.

[0030] Further, in step S4, the specific process includes the following steps:

[0031] S41. Update the local variable using the optimal temporary variable of the neighboring node. The update formula is as follows:

[0032]

[0033] Where α is the positive step size, w ij These are elements in the weight matrix W of the time-varying network, reflecting the weights of node i on the information of its neighbor node j. These are the auxiliary variables and optimal temporary variables of the neighboring nodes in the k-th iteration. This represents the local variable obtained from the (k+1)th iteration of the current node after the update.

[0034] S42, the result obtained in step S41 Projecting onto the constraint set is denoted as

[0035] Furthermore, step S5 specifically includes:

[0036] The momentum obtained using Nesterov's method and the projection in step S42 The auxiliary variable is updated using the following formula:

[0037]

[0038] Where β is the momentum parameter. This represents the auxiliary variable for the k-th iteration of the current node. This represents the auxiliary variable obtained from the (k+1)th iteration of the current node after the update.

[0039] Furthermore, step S6 specifically includes:

[0040] The gradient estimator is updated using the gradient tracking method, and the update formula is as follows:

[0041]

[0042] Where w ij These are elements in the weight matrix W, reflecting the weights of the current node i on the information of its neighbor node j. It is the gradient of neighbor node j at iteration k. It is the gradient of the current node i at iteration k+1, and m represents the number of nodes.

[0043] Furthermore, in step S7, the specific process includes the following steps:

[0044] S71. Each repetition of steps S2-S6 is denoted as one external iteration k. After each external iteration, the error between the estimated solution produced by the current iteration and the estimated solution of the previous iteration needs to be evaluated, i.e.:

[0045]

[0046] wherein γ1 represents an error tolerance; if the set error range is not reached, then the iteration is repeated until the set error range is reached or a set maximum number of iterations is reached, wherein and respectively represent the estimated solution of node i at the (k+1)th and kth external iteration;

[0047] S72, each time step S3-S6 is completed once, it is recorded as an internal iteration k in , each time an internal iteration is completed, the error of the estimated solution of the output generated by the current iteration and the estimated solution of the last iteration is evaluated, that is:

[0048]

[0049] wherein γ2 represents an error tolerance; if the set error range is not reached, then the iteration is repeated until the set error range is reached or a set maximum number of iterations is reached, wherein and respectively represent the estimated solution of node i at the kth external iteration k in +1 and k in internal iteration.

[0050] By means of the above technical scheme, the application provides a logistics AGV formation cooperative control method based on a DMG-MPC algorithm, which has at least the following beneficial effects:

[0051] (1) The DMG-MPC algorithm provided by the application combines the advantages of Nesterov momentum and gradient tracking mechanism method, compared with existing distributed algorithms, the application greatly improves the image convergence rate while ensuring the convergence of cooperative control, and has better performance and efficiency in processing cooperative control;

[0052] (2) The DMG-MPC algorithm provided by the application has scalability, and can not only be applied to logistics AGV formation cooperative control, but also be directly applied to all optimization scenes with non-convex objective functions and non-convex constraints;

[0053] (3) The convergence rate of the DMG-MPC algorithm provided by the application has theoretical guarantee;

[0054] (4) The logistics AGV formation cooperative control method based on the DMG-MPC algorithm provided by the application breaks through the limitation of the processing efficiency of high-dimensional constraints of the traditional distributed solving framework, which restricts the real-time control performance, and promotes the large-scale application of the logistics AGV system in complex scenes such as narrow channel operation and high-density cluster;

[0055] (5)The application solves the non-convex collaborative optimization problem constructed in the multi-AGV system by adopting a double-layer architecture, ensures the feasibility and computational efficiency of distributed solving, combines the momentum method with the gradient tracking method, obtains a faster convergence rate, adopts the projection operation to strictly guarantee the feasibility of the AGV dynamics constraint, so that multiple AGVs can realize full autonomous collaborative navigation and dynamic obstacle avoidance in the warehouse environment, avoid accidents, and in the emergency working conditions such as equipment failure or channel blockage scene, re-distribute the transportation task and generate a cooperative detour path, and improve the overall loading and unloading efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0056] The drawings described herein are used to provide further understanding of the present application, and form a part of the present application, the illustrative embodiments of the present application and the description thereof are used to explain the present application, and do not constitute improper limitation on the present application. In the drawings:

[0057] Figure 1 A flowchart of the logistics AGV formation collaborative control method based on the DMG-MPC algorithm of the present application;

[0058] Figure 2 The AGV communication network in the present application adopts a bidirectional ring topology;

[0059] Figure 3 A schematic diagram of the simulation process of the embodiment of the present application;

[0060] Figure 4 The convergence curve of the embodiment of the present application when the time step t is 5;

[0061] Figure 5 The convergence curve of the embodiment of the present application when the time step t is 10;

[0062] Figure 6 The convergence trajectory of the coupling cost function of the embodiment of the present application. DETAILED DESCRIPTION

[0063] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below in combination with the drawings and specific embodiments. The realization process of how to apply technical means to solve technical problems and achieve technical effects of the present application can be fully understood and implemented.

[0064] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0065] Please refer to Figures 1-6 This illustration shows a specific implementation of the present embodiment. This embodiment adopts a two-layer architecture and combines the momentum method with the gradient tracking method. At the same time, it uses projection operations to strictly ensure the feasibility of AGV dynamic constraints. While ensuring the feasibility and computational efficiency of distributed solution, it solves the non-convex collaborative optimization problem in the multi-AGV system, realizing fully autonomous collaborative navigation and dynamic obstacle avoidance of multiple AGVs in the warehouse environment. In emergency situations such as equipment failure or channel blockage, it can reallocate transportation tasks and generate collaborative detour paths to improve the overall loading and unloading efficiency.

[0066] Please refer to Figure 1 This embodiment proposes a collaborative control method for logistics AGV formation based on the DMG-MPC algorithm, which includes the following steps:

[0067] S1. Initialize the parameters of the DMG-MPC algorithm, import the position information and target position information of each AGV node, perform data processing, and construct AGV dynamic constraints.

[0068] As a preferred embodiment of step S1, the specific process includes the following steps:

[0069] S11. The parameters for initializing the DMG-MPC algorithm include the number of AGVs (i.e., the number of nodes m), error tolerances γ1 and γ2, maximum number of iterations, and gradient information. in It is the gradient calculation function;

[0070] S12, The position information of each AGV includes the coordinates of the i-th AGV. ( The horizontal coordinate is... (Vertical coordinates), horizontal and vertical velocities ( Indicates horizontal velocity. (representing vertical velocity) and acceleration u i The discrete-time state equation for each AGVi is defined as follows: In this embodiment, the initial position is set to p1 = [-1, -1]T , p2 = [-3, -1] T , p3 = [-3, -3] T , p4 = [-1, -3] T , the target position is set to p d1 = [2, 2] T , p d2 = [3, 1] T , p d3 = [2, 0] T , p d4 = [1, 1] T .

[0071] S13, the intelligent agent position constraint is set to The control input constraint is set to [-2, 2], a non-convex constraint of the collision avoidance condition between each pair of AGVs, and the mathematical expression of the non-convex constraint of the collision avoidance condition is:

[0072] wherein

[0073] wherein, ||p i (t) - p j (t) || is the Euclidean distance between AGVs i and j, is a collision avoidance safety threshold, and the physical meaning of the expression of the non-convex constraint of the collision avoidance condition is that the square of the distance between two AGVs must be not less than

[0074] More specifically, the communication network of the AGV adopts a bidirectional ring topology structure, as shown in Figure 2 .

[0075] S2, a model predictive control problem is constructed, a problem formula is obtained, and the SCA algorithm is used to convex the constructed problem formula;

[0076] As a preferred embodiment of step S2, the specific process includes the following steps:

[0077] S21, a model predictive control problem is constructed wherein at time step t, given The control input sequence is optimized by constructing the following coupling cost function wherein u i (s; t) represents the predicted input sequence at time t+s; represents a natural number set, and in the above control problem,

[0078]

[0079] In the formula, x i(t) represents the initial state of the i-th AGV at time t, with control input u i , then x i i (s; t, u fi ) represents the state solution of the AGV at time t + s; represents the coupling cost function; V i (·) represents the terminal cost; m represents the number of AGVs; and N represents the prediction horizon length.

[0080] represents the coupling cost function, which is defined as:

[0081] where, for Q i = I2, and Q ij = diag{1, 0, 1, 0};

[0082] In the formula, x di represents the state of the AGVi when reaching the target, represents the expected relative state difference (to ensure that the AGVi and the AGVj in the formation maintain a fixed distance); Q i = diag{50, 200, 50, 200} is a weight matrix, and the physical meaning of the diagonal elements corresponds to the weight of the state vector in order, and the position error weight is set to 50 and the speed error weight is set to 200; R i = I2 is an input weight matrix, and I2 represents a 2x2 unit matrix, Q ij = diag{1, 0, 1, 0} is a position coordination error weight matrix, which is designed to ensure that the speed of each AGV is allowed to be adjusted independently (such as the speed difference between the inside and outside when turning) on the premise of maintaining a fixed relative position, to avoid instability of the formation due to the speed synchronization constraint.

[0083] V fi (·) represents the terminal cost, which is defined as:

[0084] In the formula, P i = 50Q i is a weight matrix of the terminal cost, which is used to amplify the importance of state tracking at the terminal time; in this embodiment, the terminal control law is determined by constructing a terminal constraint set X f to participate in the model optimization implicitly, to ensure that the optimization closed loop is stable, where K i is a parameter that determines how to adjust the terminal control input according to the relative state difference, and the terminal control law κ i (x i ) forces u​i (N)=κ i (x i (N)), ensuring that the state converges to x. di , where u i (N) represents the optimal control input calculated by the i-th AGV at the last time step (i.e., time t+N) in the prediction time domain, x i (N) is the state of the AVG system at the end of the prediction time domain; the prediction time domain length is set to 10 time steps, i.e., N = 10;

[0085] S22. Use the SCA algorithm to solve the problem formula J in step S21. i Convexation is performed on (x(t), u) denoted as

[0086] S3. At each node, i.e. each AGV, based on the distributed structure, the optimal temporary variables for the model predictive control problem are solved by estimating the gradient.

[0087] As a preferred embodiment of step S3, the specific process includes the following steps:

[0088] S31, Update k-th in Gradient estimation in the next iteration Its formula is expressed as:

[0089]

[0090] in It is the gradient calculation function. Indicates the k-th AGVi in Auxiliary variables for the next iteration is the gradient of AGVj at k iterations, and m is the number of AGVs;

[0091] S32. Using the estimated gradient The optimal temporary variable for updating the model predictive control problem is given by the following formula:

[0092]

[0093] For a distributed structure, for each AGV Introducing local copies in Let AGVj be the input sequence of the AGVi hypothesis. Defined as the primitive variable u of AGVi i and variables The combination Formula for representing the problem The model obtained after the convexity process in step S22 Indicates the k-th AGViin Auxiliary variables for the next iteration Let AGVi be the kth digit. in The optimal temporary variable for the next iteration. This indicates a local constraint.

[0094] S4. On each AGV, update the local variables of the current AGV by combining the optimal temporary variables of the neighboring AGVs and project the local variables into the constraint set;

[0095] As a preferred embodiment of step S4, the specific process includes the following steps:

[0096] S41. Update the current AGV (AGVi) at k using the optimal temporary variable of the neighboring AGV (AGVj). in The update formula for the local variables in the next iteration is:

[0097]

[0098] Where α is the positive step size, w ij These are elements in the weight matrix W of the time-varying network, reflecting the weights of the current AGVi on its neighbor AGVj. It is the kth in Auxiliary variables and optimal temporary variables for the next iteration of AGVj. This represents the local variable of AGVi obtained in the (k+1)th iteration after the update.

[0099] S42, the result obtained in step S41 Projecting onto the constraint set is denoted as

[0100] S5. On each AGV, update the auxiliary variables of the current AGV using the Nesterov momentum method;

[0101] As a preferred embodiment of step S5, the specific process includes the following steps:

[0102] The momentum obtained using Nesterov's method and the projection in step S42 The auxiliary variable is updated using the following formula:

[0103]

[0104] Where β is the momentum parameter. Describes the auxiliary variable for the k-th iteration of AGVi. This represents the auxiliary variable obtained from the (k+1)th iteration of AGVi after the update.

[0105] S6, updating the gradient estimator based on the gradient tracking method on each AGV combined with the auxiliary variables and the gradients of the neighbor AGVs to obtain the gradient of the current AGV;

[0106] As a preferred embodiment of step S6, the specific process comprises the following steps:

[0107] The updating formula of the gradient estimator based on the gradient tracking method is:

[0108]

[0109] where w ij is an element in the weight matrix W, reflecting the weight of the information of the neighbor AGVj of the current AGVi, is the gradient of AGVj at the k in th iteration, is the gradient of AGVi at the k in +1th iteration, and m represents the number of AGVs.

[0110] S7, repeating steps S2-S6 for external iteration and repeating steps S3-S6 for internal iteration until the iteration ends when the algorithm termination condition is met;

[0111] S71, each time the steps S2-S6 are repeated, it is recorded as one external iteration k, and each time the external iteration is completed, the error between the estimated solution of the output generated by the current iteration and the estimated solution of the last iteration needs to be evaluated, that is:

[0112]

[0113] where γ1 represents the error tolerance; if the set error range is not reached, the iteration is repeated until the set error range or the set maximum number of iterations is reached, where and represent the estimated solution of AGVi at the k+1th and kth external iteration, respectively.

[0114] S72, each time the steps S3-S6 are repeated, it is recorded as one internal iteration k in , and each time the internal iteration is completed, the error between the estimated solution of the output generated by the current iteration and the estimated solution of the last iteration needs to be evaluated, that is:

[0115]

[0116] where γ2 represents the error tolerance; if the set error range is not reached, the iteration is repeated until the set error range or the set maximum number of iterations is reached, where and represent the estimated solution of AGVi at the k in +1th and kth external iteration in the kth external iteration, respectively.in estimated solution at the sub-internal iteration.

[0117] In this embodiment, the operation steps of DMG-MPC (Distributed Momentum-Gradient Model Predictive Control) algorithm in logistics AGV formation cooperative control are provided, and the projection operation is adopted to strictly guarantee the feasibility of AGV dynamics constraints, and the operation simulation process is as shown in Figure 3 As shown in the figure, four AGVs are set to start from their respective starting positions and reach their respective destinations to handle related matters, while satisfying the anti-collision constraint and keeping within the safety zone (dotted line boundary). This shows that the proposed algorithm successfully achieves the control target while ensuring safety.

[0118] This embodiment also provides a convergence rate analysis of the DMG-MPC algorithm for the optimization problem, Figure 4 and Figure 5 The convergence curve for performing the task shows the convergence of the DMG-MPC algorithm at time steps t = 5, t = 10, and outer loop iteration k = 30. By quantifying the gap between the intermediate solution and the optimal solution, the convergence of the internal iteration is verified. As shown in Figure 4 and Figure 5 The optimality gap of the proposed DMG-MPC algorithm shows a downward trend and achieves linear convergence speed. Figure 6 The convergence trajectory of the coupled cost function is shown, indicating that all AGVs have approached their designated target positions at the 10th time step. The above analysis shows that the algorithm has theoretical feasibility in handling the optimization problem.

[0119] In summary, the method proposed in the present application focuses on real-time decision-making and safety optimization, can efficiently solve large-scale online optimization problems, is suitable for various scenarios, and solves the problems of slow speed, poor accuracy, and safety of existing logistics cooperative control algorithms in application.

[0120] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the skilled in the art can combine and combine the different embodiments or features of the embodiments or examples described in the present specification without contradiction.

[0121] The logic and / or steps represented in flow diagrams or otherwise described herein, for example, can be considered as a sequence of executable instructions, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions, or a combination of the above.

[0122] The above embodiments introduce the present application in detail, the principles and embodiments of the present application are described by applying specific examples, the above example is only used to help understand the method of the present application and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, the specific embodiment and application range will be changed, and the above description should not be understood as the limitation of the present application.

Claims

1. A logistics AGV platoon cooperative control method based on a DMG-MPC algorithm, characterized in that, The method comprises the following steps: S1, initializing the parameters of the DMG-MPC algorithm, importing the position information of each node of the AGV and the target position information, performing data processing, and constructing the AGV dynamics constraint; The specific process of step S1 comprises the following steps: S11, the parameters of the initialization DMG-MPC algorithm, including the number of AGVs, that is, the number of nodes , error tolerance , , maximum iteration number, gradient information , wherein is a gradient calculation function; S12, the position information of each node of the AGV comprises horizontal and vertical coordinates, horizontal and vertical velocities, and acceleration, and the target position information comprises horizontal and vertical coordinates; S13, the AGV dynamics constraint comprises an agent position constraint, a control input constraint, and a non-convex constraint of an anti-collision condition between each pair of AGVs; S2, constructing a model predictive control problem, obtaining a problem formula, and convexifying the constructed problem formula by using an SCA algorithm; S3, on each node, based on a distributed structure, an estimated gradient is used to solve the optimal temporary variable of the model predictive control problem; S4, on each node, the local variable of the current node is updated in combination with the optimal temporary variable of the neighbor node and the local variable is projected into the constraint set; S5, on each node, the auxiliary variable of the current node is updated by using a Nesterov momentum method; S6, on each node, the gradient estimator is updated based on a gradient tracking method in combination with the auxiliary variable and the gradient of the neighbor node, and the gradient of the current node is obtained; S7, repeating steps S2-S6 for external iteration, and repeating steps S3-S6 for internal iteration, until the iteration ends when the algorithm ending condition is met. 2.The DMG-MPC algorithm-based logistics AGV platoon cooperative control method according to claim 1, characterized in that: The AGV adopts a bidirectional ring topology structure for the communication network.

3. The DMG-MPC algorithm-based logistics AGV platoon cooperative control method according to claim 1, characterized in that: The specific process of step S2 comprises the following steps: S21, constructing a model predictive control problem where at time step , given , the control input sequence is optimized by constructing the following coupled cost function where denotes the predicted input sequence at time denotes the set of natural numbers, in the above control problem: ； where , denotes the initial state of the th node at time with control input , then denotes the state solution of the th node at time denotes the coupling cost function, defined as: wherein for , , , , and ; wherein denotes the state of the nth AGV when it reaches the target position, denotes the desired relative state difference; is a weight matrix; is an input weight matrix, denotes a 2x2 identity matrix, is a position coordination error weight matrix; denotes the terminal cost, defined as: ； In the above formula, is a weight matrix of terminal cost; denotes the number of nodes; denotes the prediction time domain length; S22, using the SCA algorithm to solve the problem formula in step S21 convexification is denoted as .

4. The logistics AGV platoon cooperative control method based on the DMG-MPC algorithm according to claim 3, characterized in that: The specific process of step S3 comprises the following steps: S31、updating the first estimated gradient of the next iteration whose update formula is expressed as: ； wherein is a gradient computation function, denotes a node the first auxiliary variable of the is a node the gradient at iteration, denotes the number of nodes; S32, using the estimated gradient The optimal temporary variable of the model predictive control problem is updated, and the update formula is: ； For a distributed structure, for each node Introducing local copies where denotes a node Assuming an input sequence for a node is defined as the node controlling the input sequence and the input sequence is defined as the combination of denotes the problem formulation the model obtained after the convexification process in step S22, denotes the auxiliary variable of the th iteration of a node denotes the optimal temporary variable of the th iteration of a node denotes the local constraints.

5. The DMG-MPC algorithm-based logistics AGV platoon cooperative control method according to claim 4, characterized in that: The specific process of step S4 comprises the following steps: S41, the local variable is updated by using the optimal temporary variable of the neighbor node, and the update formula is: ； wherein is a positive step size, is an element in the weight matrix W of the time-varying network, reflecting the weight of the node to its neighbor node information, , is the th iteration of the neighbor node, is an auxiliary variable and an optimal temporary variable of the th iteration of the current node; S42, the obtained projected into the constraint set is denoted by .

6. The DMG-MPC algorithm-based logistics AGV platoon cooperative control method according to claim 5, characterized in that: The specific process of step S5 comprises the following steps: Using Nesterov momentum and the projection obtained in step S42 Updating the auxiliary variables, the specific update formula is: ； wherein is a momentum parameter, denotes an auxiliary variable of the current node at the denotes an updated auxiliary variable of the current node at the denotes an updated auxiliary variable of the current node at the denotes an updated auxiliary variable of the current node at the 7. The DMG-MPC algorithm-based logistics AGV platoon cooperative control method according to claim 6, characterized in that: The specific process of step S6 comprises the following steps: The gradient estimator is updated based on the gradient tracking method, and the update formula is: ； in These are elements in the weight matrix W of the time-varying network, reflecting the current node's... Its neighboring nodes Information weight, Neighboring nodes exist The gradient at the next iteration The current node exist The gradient at the next iteration Indicates the number of nodes. Indicates the current node's... Auxiliary variables for the next iteration This indicates that the updated current node is the [number]th node. Auxiliary variables for each iteration.

8. The DMG-MPC algorithm-based logistics AGV platoon cooperative control method according to claim 7, characterized in that: The specific process of step S7 comprises the following steps: S71, each time steps S2-S6 is completed, it is called an outer iteration Each time an outer iteration is completed, the error between the estimated solution produced by the current iteration and the estimated solution of the last iteration is evaluated, i.e. ； wherein denotes the error tolerance; if the set error range is not reached, the iteration is repeated until the set error range is reached or a set maximum number of iterations is reached, wherein and denote the nodes the estimated solution at the and external iterations; S72, each time steps S3-S6 are completed once, it is called an internal iteration each time an internal iteration is completed, the error between the estimated solution of the output generated by the current iteration and the estimated solution of the last iteration is evaluated, i.e. ； wherein denotes the error tolerance; if the set error range is not reached, the iteration is repeated until the set error range is reached or a set maximum number of iterations is reached, wherein and denote the nodes In the first external iteration and the estimated solution at the i-th internal iteration.

Images