A multi-agv distributed path planning method based on a prediction model

Abstract

The present application relates to the technical field of computer, disclose a kind of multi-AGV distributed path planning method based on prediction model, comprising: S1, obtain target task;S2, the speed of automatic guided vehicle is discretized, determine global decision variable and constraint, obtain total transport time target function and obtain global hybrid logic dynamic model;S3, based on global hybrid logic dynamic model obtains prediction transport time target function and obtains local hybrid logic dynamic model;S4, based on local hybrid logic dynamic model obtains prediction model, the information of each automatic guided vehicle and its corresponding sub-target task are input prediction model to obtain the path planning of each automatic guided vehicle.The method of the present application has the technical advantages of reducing the amount of calculation, simplifying the calculation process, can obtain the shortest target completion time under the premise of guaranteeing no collision between AGVs, improve efficiency.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a distributed path planning method for multiple AGVs (Automatic Guided Vehicles) based on a predictive model. Background Technology

[0002] Automated Guided Vehicles (AGVs) are intelligent mobile robots typically used in industrial environments to transport materials or perform specific tasks in manufacturing systems, warehouses, container terminals, and other similar settings. Due to the requirements of Industry 4.0, AGVs must be more intelligent, autonomous, and efficient, enabling manufacturing systems to be more adaptable and flexible in complex and dynamic operating environments. The control system employed must effectively plan the AGV's path, handling vehicle coordination and avoiding conflicts, collisions, and deadlocks—essentially a conflict-free path planning problem for AGVs.

[0003] However, current AGV fleet path planning faces two challenges. First, each AGV should be intelligent enough to correctly respond to changes in the operating environment (such as machine failures or changes in delivery points) to reduce economic losses caused by these uncertainties. Second, multiple AGVs should cooperate to more effectively avoid collisions. The current trend in AGV path planning is decentralization, with each AGV independently deciding its actions. However, there is a lack of literature on online control and distributed computing. Therefore, there is an urgent need for a multi-AGV path planning method that improves adaptability and avoids collisions while ensuring short task completion times. Summary of the Invention

[0004] To address the technical problems of insufficient flexibility and adaptability in existing AGV path planning technologies, this invention provides a multi-AGV distributed path planning method based on a predictive model, which improves the flexibility, robustness, and scalability of AGV path planning in manufacturing and logistics environments.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a multi-AGV distributed path planning method based on a prediction model, comprising:

[0006] S1. Obtain the target task, and determine the sub-target task for each automated guided vehicle based on the target task;

[0007] S2. Discretize the speed of the automated guided vehicle running on the path, determine the global decision variables and impose constraints based on the target task, obtain the total transportation time objective function and construct the model to obtain the global hybrid logic dynamic model, and obtain the global planning problem based on the total transportation time objective function.

[0008] S3. Based on the global hybrid logic dynamic model, local decision variables are introduced and constrained to obtain the objective function for predicting transportation time and to construct the model, thereby obtaining a local hybrid logic dynamic model. Based on the objective function for predicting transportation time, a local programming problem is obtained.

[0009] S4. Based on the local hybrid logic dynamic model, a prediction model is obtained. The information of each automated guided vehicle and its corresponding sub-target task are input into the prediction model to obtain the path planning for each automated guided vehicle.

[0010] Preferably, step S2 includes:

[0011] The target task includes a start node, a delivery node, and a planned time. The planned time is discretized into several time intervals. Based on the target task, a mathematical description is performed to obtain global decision variables.

[0012] The total transportation time objective function is obtained by applying spatial and spatiotemporal constraints to the global decision variables, including collision avoidance constraints and spatiotemporal continuity constraints.

[0013] Preferably, in step S2, the objective function for total transportation time is optimized, and the objective function for total transportation time is solved and defined as a global programming problem.

[0014] Preferably, step S3 includes:

[0015] By introducing a local programming time window parameter, the global programming problem is decomposed into several local programming problems, which are then solved repeatedly using the local programming time window.

[0016] Preferably, in step S3, the objective function for predicting transportation time is optimized, and the objective function for predicting transportation time is solved and defined as a local programming problem.

[0017] Preferably, in step S4, the prediction model is a distributed prediction model, and obtaining the prediction model based on the local hybrid logic dynamic model includes:

[0018] Based on the aforementioned local hybrid logic dynamic model, an augmented Lagrange function is constructed by introducing two non-negative Lagrange multipliers and a penalty function.

[0019] The quadratic penalty term is expanded using the first-order Taylor formula, decomposing the global programming problem into several subproblems and obtaining the sub-prediction transport time objective function for each subproblem. Based on the sub-prediction transport time objective function of the subproblems, the dual problem of each subproblem is obtained through mathematical description. A distributed prediction model is constructed using a distributed algorithm based on the alternating direction multiplier method of serial iteration.

[0020] Preferably, the distributed prediction model constructed in step S4 using the distributed algorithm based on the alternating direction multiplier method of serial iteration includes: using the distributed algorithm to solve the dual problem of the subproblem to obtain the local decision variables of the automated guided vehicle at any time, wherein the Lagrange multipliers and penalty parameters are updated according to the corresponding formulas.

[0021] This invention discretizes the speed of an AGV running along a path. Under this condition, it first determines global decision variables to establish a global hybrid logic dynamic model (referred to as the global model) describing the AGV's operation, avoiding collisions between AGVs through constraints between variables. Then, the global model is transformed into a predictive model (referred to as the local model), determining control actions by minimizing the desired objective along a finite backward horizontal line, used for dynamically planning a collision-free path for the AGV. Based on this, an augmented Lagrange function and decomposition method, a Lagrange multiplier update formula, and a penalty function are designed to construct a distributed path planning method based on ADMM for model solving. The Gurobi solver is used to obtain the path results. Compared with existing technologies, this invention has technical advantages such as reducing computational load, simplifying the calculation process, improving adaptability, achieving the shortest target completion time while ensuring collision-free AGV operation, and improving efficiency. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating a multi-AGV distributed path planning method based on a prediction model in Example 2.

[0023] Figure 2 This is a schematic diagram of the distributed basic structure of the AGV in Example 2.

[0024] Figure 3 This is a pseudocode diagram of the ADMM framework of the AGV in Example 2.

[0025] Figure 4 This is a schematic diagram of the test scenario for the AGV in Example 2. Detailed Implementation

[0026] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Example 1

[0028] A distributed path planning method for multiple AGVs based on a predictive model includes:

[0029] S1. Obtain the target task, and determine the sub-target task for each automated guided vehicle based on the target task;

[0030] S2. Discretize the speed of the automated guided vehicle running on the path, determine the global decision variables and impose constraints based on the target task, obtain the total transportation time objective function and construct the model to obtain the global hybrid logic dynamic model, and obtain the global planning problem based on the total transportation time objective function.

[0031] S3. Based on the global hybrid logic dynamic model, local decision variables are introduced and constrained to obtain the objective function for predicting transportation time and to construct the model, thereby obtaining a local hybrid logic dynamic model. Based on the objective function for predicting transportation time, a local programming problem is obtained.

[0032] S4. Based on the local hybrid logic dynamic model, a prediction model is obtained. The information of each automated guided vehicle and its corresponding sub-target task are input into the prediction control model to obtain the path planning for each automated guided vehicle.

[0033] The steps S1 to S4 above consider the problem of conflict-free pickup and delivery routes for multiple AGVs. The transmission process is regarded as a dynamic system. First, global decision variables are determined to establish a global hybrid logic dynamic model describing the operation of AGVs. Constraints between variables are used to avoid collisions between AGVs. Based on this, a model predictive control method is proposed to address the uncertainties generated during transportation.

[0034] Following the above embodiment, step S2 includes: the target task includes a start node, a delivery node, and a planning time; the planning time is discretized into several time intervals; a mathematical description is performed based on the target task to obtain global decision variables;

[0035] The total transportation time objective function is obtained by applying spatial and spatiotemporal constraints to the global decision variables, including collision avoidance constraints and spatiotemporal continuity constraints.

[0036] A set of pickup and delivery requests are transported between specified node pairs using a fleet of AGVs. After obtaining the target task, the start time of the transport requests and tasks can be determined. The problem to be solved is to determine the path that each vehicle should take between two consecutive start and delivery nodes in order to achieve conflict-free paths between multiple AGVs during transport and to minimize the target task completion time.

[0037] In industrial environments, multiple AGVs are needed to transport various materials within a connected network of guided paths. During material transport, a collision-free vehicle path needs to be planned for each AGV from its starting point to its delivery point. Starting and delivery points are pre-assigned to each AGV based on the target task.

[0038] In manufacturing or warehousing systems, AGVs are primarily used to transport materials between the starting point and destination of a task in a mesh routing environment. In a mesh routing environment, the map is represented as a directed graph G = (N, E), where N is the set of nodes representing loading and unloading locations and the locations where the AGV can change direction; E = {(i, j) | i ∈ N, j ∈ N} is the set of arcs representing the connecting paths between two adjacent nodes. Each vehicle's task is to transport materials from its starting point to its destination. The following assumptions are made regarding this problem:

[0039] The geometry of the AGV is small enough that the AGV is considered as a point occupying a specific node;

[0040] Each AGV is assigned tasks from its starting node to its delivery node, and these nodes are different;

[0041] The speed of the AGV is constant, and the turning time can be included in the corresponding travel time;

[0042] Each AGV can wait or change direction at a node in the path network, and each node can be occupied by a maximum of one AGV at any given time.

[0043] The lanes are bidirectional, and at any given time, a lane can be occupied by a maximum of one AGV.

[0044] When an AGV completes a specific task, it stops at the end node;

[0045] Each loading / unloading node is connected, and the number of AGVs is less than the number of nodes.

[0046] Based on the above assumptions, a global hybrid logic dynamic model is established, which discretizes the planning time T×Δt into a series of time gaps {Δt,2Δt,...,T×Δt}. The entire path process of the AGV is divided into T time gaps for processing. Collision-free operation between AGVs is achieved through reasonable planning of time windows.

[0047] First, the symbols and corresponding descriptions of the model are introduced, as shown in Table 1.

[0048] Table 1 Model Symbols and Descriptions

[0049]

[0050]

[0051] The global decision variables for this model are as follows:

[0052] State variables

[0053] Global control variables

[0054] in,

[0055] 0-1 variables, This indicates that AGV k arrives at node i at time t; otherwise,

[0056] This indicates that AGV k leaves node i at time t and... Move to node j at time t. If AGV k is always at the same node from t to t+1, then nodes i and j can be the same.

[0057] The global hybrid logic dynamic model achieves collision-free guidance paths between AGVs through constraints between variables. These constraints include spatial and spatiotemporal constraints, as detailed below:

[0058] I. Spatial Constraints:

[0059]

[0060]

[0061]

[0062] Among them, constraint (1) corresponds to the spatial constraint of the AGV's path starting point, and constraints (2) and (3) correspond to the spatial constraint of the AGV's path ending point.

[0063] II. Collision Avoidance Constraints:

[0064]

[0065]

[0066]

[0067] Among them, constraint (4) ensures that each AGV stays on only one node at any time; constraint (5) ensures that each node can be occupied by at most one AGV; constraint (6) requires that the connection between node i and node j is unidirectional, and each connection can be occupied by at most one AGV at any time.

[0068] III. Spatiotemporal continuity constraint:

[0069]

[0070]

[0071] Constraints (7) to (8) ensure the spatiotemporal continuity of AGV k moving node i to adjacent node j.

[0072] In constraint (8), It is a non-linear term; redefine an auxiliary variable.

[0073] Its update method is Constraint (8) can be rewritten as constraint (9):

[0074]

[0075] In constraint (8), the nonlinear term This can be expressed using the following three linear inequalities:

[0076]

[0077]

[0078]

[0079] Based on the above model, the objective function J1 for total transportation time is determined as follows:

[0080]

[0081] Following the above embodiment, in step S2, the objective function J1 for the total transportation time is optimized. The objective function for the total transportation time is solved and defined as a global programming problem. The global programming problem P1 can be solved by a commercial solver, such as Gurobi.

[0082] Following the above embodiments, step S3 includes:

[0083] By introducing a local programming time window parameter, the global programming problem is decomposed into several local programming problems, which are then solved repeatedly using the local programming time window.

[0084] This embodiment, based on the established global model, further proposes a local model and provides the corresponding model predictive control formula. Model predictive control is a control strategy that uses a local model to determine the control action by minimizing the desired objective along a finite backward horizontal line, used for dynamically planning the collision-free path of an AGV. For the predictive model, local decision variables and new parameters need to be introduced, and several constraints of the global hybrid logic dynamic model also need to be adjusted because the initial constraints of the global planning are designed to fit the framework of the model predictive controller. The introduced new parameters are shown in Table 2.

[0085] Table 2 Model Symbols and Descriptions

[0086]

[0087] Introducing local decision variables:

[0088] Local control variables C is set to

[0089] in, 0-1 variables, This indicates that AGV k is within the local time window T. p Reach node j; otherwise,

[0090] The state variables here are consistent with the local model.

[0091] This embodiment decomposes the global planning problem P1 into a smaller local planning problem P2, using a local planning time window T. p (T p Repeat the solution process if the value is much smaller than H.

[0092] For the local hybrid logic dynamic model, constraints (2) to (3) are modified to ensure that each AGV reaches the local planning range T. p The local termination node within. The modified constraints are as follows:

[0093]

[0094]

[0095] Among them, constraint (14) ensures that each AGV selects a local termination node in the local problem under consideration; constraint (15) guarantees that each AGV at t=T p It can only stay at a local termination node (j∈N) L )superior.

[0096] For the constraints (1) to (9) of the global hybrid logic dynamic model, when considering local programming, the domain of t needs to be modified from t∈{0,1,...,H} to t∈{0,1,...,T}. p}

[0097] For the total transportation time objective function J1, a local version, namely the predicted transportation time objective function J2, is also needed, defined as follows:

[0098]

[0099] Following the above embodiment, in step S3, the objective function for predicting transportation time is optimized, and the objective function for predicting transportation time is solved and defined as a local programming problem.

[0100] The optimization problem of predicting transportation time objective function J2 is denoted as P2. This is a mixed integer linear programming problem that can be solved by commercial solvers such as Gurobi.

[0101] Using the proposed model predictive control strategy, the local hybrid logic dynamic model only considers the AGV's path within a local range T when performing AGV path planning. p The system performs path planning within the network, identifies local endpoints, and approximates the distance between the global endpoint and the local endpoints offline. This embodiment uses pre-calculated distance data to estimate the model's target value from the global route map configuration.

[0102] Although this method cannot guarantee the optimality of the obtained solution, if a suitable T is chosen... p Its accuracy can be quite high, and its calculation speed is faster than global planning.

[0103] Example 2

[0104] Example 1 introduced the design of a model predictive controller based on a developed hybrid logic dynamic model, where the AGV path is dynamically determined in a centralized manner. However, due to industrial planning requirements (including flexibility, robustness, and scalability), and limitations in memory, communication, and computing power, distributed computing is needed to improve the performance of the entire AGV system. Distributed computing decomposes the centralized problem into smaller problems, with each AGV making its own decisions based on its interactions with other AGVs, thereby reducing the overall computational load and significantly improving computational efficiency.

[0105] Based on Example 1, this example designs a distributed algorithm based on the alternating direction multiplier method to obtain a better solution within a reasonable computation time.

[0106] See Figure 1-4 The difference between this embodiment and Embodiment 1 is that in step S4, the prediction model is a distributed prediction model, and the process of obtaining the prediction model based on the local hybrid logic dynamic model includes:

[0107] Based on the aforementioned local hybrid logic dynamic model, an augmented Lagrange function is constructed by introducing two non-negative Lagrange multipliers and a penalty function.

[0108] The quadratic penalty term is expanded using the first-order Taylor formula, decomposing the global programming problem into several subproblems and obtaining the sub-prediction transport time objective function for each subproblem. Based on the sub-prediction transport time objective function of the subproblems, the dual problem of each subproblem is obtained through mathematical description. A distributed prediction model is constructed using a distributed algorithm based on the alternating direction multiplier method of serial iteration.

[0109] Following the above embodiment, the construction of the distributed prediction model using the distributed algorithm based on the alternating direction multiplier method of serial iteration in step S4 includes: using the distributed algorithm to solve the dual problem of the subproblem to obtain the local decision variables of the automated guided vehicle at any time, wherein the Lagrange multipliers and penalty parameters are updated according to the corresponding formulas.

[0110] The following are the decomposable optimization formulas and distributed planning algorithms for distributed model predictive control.

[0111] First, for multiple AGVs, the collision avoidance constraint (5) and constraint (6) are coupled. Two non-negative Lagrange multipliers λ1 and λ2 are introduced to relax these two coupled constraints. The augmented Lagrange function formula for the local programming problem P2 is constructed as follows:

[0112]

[0113] Where ε is the penalty function, introduced

[0114]

[0115]

[0116] The augmented Lagrangian function of the mixed-integer linear programming problem is transformed into:

[0117]

[0118] Because of the introduction of a quadratic penalty term, minimizing L c The problem cannot be separable into AGV-level subproblems. To maintain separability, a linearization technique is used for the cross-penalty portion surrounding the estimated optimal solution. A first-order Taylor expansion is applied to the quadratic penalty term. It can be represented as:

[0119]

[0120] In formula (19) Items can revolve around points After performing a first-order Taylor linearization expansion, it can be expressed as:

[0121]

[0122] in It can be considered as a constant. It can also be converted in the same way.

[0123] Therefore, the original centralized local programming problem P2 can be decomposed into a series of subproblems. The objective function of AGV k is:

[0124]

[0125] The objective function of the dual problem is defined as the objective function. right The minimum value obtained. For each pair (λ1, λ2), we need to find the minimum value. smallest Different values ​​of (λ1, λ2) correspond to different objective functions of the dual problem. The dual problem of each subproblem of AGV k is defined as follows:

[0126]

[0127] After presenting the decomposed dual problem, the following is the implementation process of the proposed distributed algorithm based on the serial iterative alternating direction multiplier method. For the decomposed subproblems, the decision variables of AGV k are calculated during iteration p. The formula is as follows:

[0128]

[0129] The Lagrange multiplier update formula is as follows:

[0130]

[0131] The penalty parameter is transformed using an adaptive method, as shown below:

[0132]

[0133] Where r(p,t) and s(p,t) are the original residual and the dual residual, respectively. r(p,t) and s(p,t) are defined as follows:

[0134]

[0135] in It is the average preprocessing result of each subproblem in the state vector at time t in the p-th iteration.

[0136]

[0137] The algorithm continues to execute until r(p,t)≤r lim Or s(p,t)≤s lim or p = p max r lim and s lim 10 -3 .

[0138] This embodiment constructs an augmented Lagrange objective function based on a local model and introduces a corresponding decomposition method to break down the original problem into a series of smaller sub-problems. The AGV path planning method proposed in this embodiment can achieve path planning for multiple AGVs under the condition of a short target task completion time. In this method, each time an AGV moves one step, a local problem is solved. In the solution of the current step, the distributed solution of sub-problems and the iterative updating of parameters (Lagrange multipliers, penalty functions, residuals, etc.) constitute the algorithm's flow. Iteration stops when the residuals meet the stopping condition or the maximum number of iterations is reached. The AGV moves one step based on the result and performs the next round of solution until all AGVs reach the destination. The algorithm then ends, and the results are recorded based on the actual paths taken by the AGVs.

[0139] The method in this embodiment utilizes Gurobi to obtain high-quality path results. Furthermore, this invention considers the movement time of the AGV and performs dynamic programming, which simplifies the calculation process and improves the solution speed and efficiency.

[0140] Based on Example 1, this embodiment provides an efficient distributed optimization technique using the Alternating Direction Method of Multipliers (ADMM), which can reduce computational load and is of great value for real-time decision-making.

[0141] Test case

[0142] See Figure 4 Test scenario: The experiment relies on a map G(N,E), a strongly connected map consisting of multiple nodes and directed arcs, which can simulate various types of warehouse or system layouts. The distance between two adjacent nodes is set to a unit distance, and the unit time interval Δt is set to 1 second. Based on this, different degrees of randomness are set for the scenario. To effectively evaluate the algorithm, the scenario is further subjected to 30 random simulations under different task conditions, and the average of the results is analyzed.

[0143] Five methods were used for comparative evaluation experiments: Centralized MPC (C-MPC) in Example 1, Distributed MPC based on ADMM (DMPC) in Example 2, Priority MPC (P-MPC), Static Programming (SP), and Dynamic Priority Programming (DPP). P-MPC, based on the MPC framework, plans the path of each AGV according to a fixed priority and uses the planned path as a collision constraint; SP is a mainstream planning method implemented through a spatiotemporal model; DPP is a state-of-the-art dynamic method that uses a dynamically updated priority list to resolve AGV conflicts based on offline paths. All three MPC methods are based on the proposed MLD model, which predicts the feasible locations of the AGVs.

[0144] First, the calculation results of C-MPC and DMPC methods were evaluated and compared with those of SPP and DPP methods. The experimental results are shown in Table 3.

[0145] Table 3

[0146]

[0147]

[0148] Table 3 provides the performance of five test methods for the test scenarios. As can be seen from the table, the DMPC method proposed in this invention achieves a good balance between the objective function value and computational complexity. There is a small gap (approximately 2%) in productivity targets (costs 1 and 2) between the proposed DMPC and static planning (SP). These gaps are caused by the approximation of global costs when using MPC planning. However, SP has high computational complexity due to considering all roadmap information, and the computation time for large-scale scenarios can exceed 1000 seconds. Furthermore, SP assumes no interruptions during transportation. The MPC planner of this invention has a significant advantage in computational efficiency, and the proposed DMPC method computes a solution in approximately 10 seconds in large-scale scenarios (compared to approximately 1000 seconds using SP), which has the potential for real-time applications in industrial environments. Compared to the large gap in computation time, the gap in productivity metrics relative to costs 1 and 2 is relatively small.

[0149] The embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A multi-AGV distributed path planning method based on a prediction model, characterized in that, include: S1. Obtain the target task, and determine the sub-target task for each automated guided vehicle based on the target task; S2. Discretize the speed of the automated guided vehicle running on the path, determine the global decision variables and impose constraints based on the target task, obtain the total transportation time objective function and construct the model to obtain the global hybrid logic dynamic model, and obtain the global planning problem based on the total transportation time objective function. Step S2 includes: The target task includes a start node, a delivery node, and a planned time. The planned time is discretized into several time intervals. Based on the target task, a mathematical description is performed to obtain global decision variables. The total transportation time objective function is obtained by applying spatial and spatiotemporal constraints to the global decision variables, wherein the spatiotemporal constraints include collision avoidance constraints and spatiotemporal continuity constraints. The global decision variables include the state variable x(t) and the global control variable u(t). ， ， in, For 0-1 variables, This indicates that the k-th AGV arrives at node i at time t; otherwise, , This indicates that the k-th AGV leaves node i at time t and... Move to node j at any time; S3. Based on the global hybrid logic dynamic model, local decision variables are introduced and constrained to obtain the objective function for predicting transportation time and to construct the model, thereby obtaining a local hybrid logic dynamic model. Based on the objective function for predicting transportation time, a local programming problem is obtained. S4. Based on the local hybrid logic dynamic model, a prediction model is obtained. The information of each automated guided vehicle and its corresponding sub-target task are input into the prediction model to obtain the path planning for each automated guided vehicle. In step S4, the prediction model is a distributed prediction model, and obtaining the prediction model based on the local hybrid logic dynamic model includes: Based on the aforementioned local hybrid logic dynamic model, an augmented Lagrange function is constructed by introducing two non-negative Lagrange multipliers and a penalty function. The quadratic penalty term is expanded using the first-order Taylor formula, decomposing the global programming problem into several subproblems and obtaining the sub-prediction transportation time objective function for each subproblem. The dual problem of each subproblem is obtained by mathematically describing the sub-prediction transportation time objective function. A distributed prediction model is constructed using a distributed algorithm based on the alternating direction multiplier method of serial iteration. The distributed prediction model constructed using the distributed algorithm based on the alternating direction multiplier method of serial iteration in step S4 includes: using the distributed algorithm to solve the dual problem of the subproblem to obtain the local decision variables of the automated guided vehicle at any time, wherein the Lagrange multipliers and penalty parameters are updated according to the corresponding formulas.

2. The multi-AGV distributed path planning method based on a prediction model according to claim 1, characterized in that, In step S2, the objective function for total transportation time is optimized, and the objective function for total transportation time is solved and defined as a global programming problem.

3. The multi-AGV distributed path planning method based on a prediction model according to claim 1, characterized in that, Step S3 includes: By introducing a local programming time window parameter, the global programming problem is decomposed into several local programming problems, which are then solved repeatedly using the local programming time window.

4. The multi-AGV distributed path planning method based on a prediction model according to claim 1, characterized in that, In step S3, the objective function for predicting transportation time is optimized, and the objective function for predicting transportation time is solved and defined as a local programming problem.

Images