Two-stage Optimization Implementation Method for AGV Task Allocation and Path Planning

Abstract

A two-stage optimization implementation method for AGV task allocation and path planning. In the task allocation stage, all possible travel types of AGVs in the workshop are listed through itinerary planning to determine the possible travel information under different task combinations; the task allocation plan is quickly obtained through the itinerary-based AGV task allocation model. In the path planning stage, the time window algorithm is used to initialize, update, and arrange the time windows for the map resources occupied by AGVs, and adjustment strategies such as material package exchange, priority advancement, and reserved duration relaxation are carried out for the situation where the material delivery time cannot be met due to obstacle avoidance and waiting, so as to achieve conflict-free path planning for AGVs. The present invention finds the optimal scheduling scheme for AGVs within a limited time by decomposing the two sub-problems of task allocation and path planning.

Description

Technical Field

[0001] The present invention relates to a technology in the field of automated warehousing, specifically a two-stage optimization implementation method for AGV task allocation and path planning. Background Art

[0002] An Automatic Guided Vehicle (AGV) refers to a transport vehicle equipped with automatic navigation devices such as electromagnetic or optical devices, capable of traveling along a specified navigation path, and having safety protection and various transfer functions. The scheduling optimization of multiple AGVs refers to, under the constraint of a production plan, allocating distribution tasks to multiple AGVs, planning the distribution paths of each AGV, and at the same time avoiding conflicts such as collisions and deadlocks between AGVs. The goal is to minimize the use cost of AGVs on the premise of meeting the production plan.

[0003] In existing research that simultaneously considers AGV task allocation and path planning, it is basically achieved by simplifying the solution of some of the problems. In addition, existing research assumes that AGVs only perform single deliveries, and rarely considers the problem of AGVs performing multiple round-trip deliveries during the distribution process. This is because when considering multiple round-trip deliveries in the AGV scheduling optimization process, the modeling and solution processes become very complex. Summary of the Invention

[0004] In view of the deficiency of the existing technology in having no method for AGV task allocation and path planning that considers multiple round-trip deliveries, the present invention proposes a two-stage optimization implementation method for AGV task allocation and path planning, decomposes the two sub-problems of task allocation and path planning, and finds the optimal scheduling plan for AGVs within a limited time.

[0005] The present invention is realized through the following technical solutions:

[0006] The present invention relates to a two-stage optimization implementation method for AGV task allocation and path planning applied to the scheduling optimization of multiple AGVs in a grid-based workshop map. First, in the task allocation stage, based on the characteristics of the workshop map, all possible travel types of AGVs in the workshop are listed through itinerary planning, and the possible travel information under different task combinations is determined; and for multiple distribution tasks in a distribution batch in the workshop, a task allocation plan is quickly obtained through an AGV task allocation model based on the itinerary. In the path planning stage, the task allocation plan obtained by the solution is transformed into time window planning work, and a time window algorithm is used to initialize, update, and arrange the time windows of the map resources occupied by AGVs, and adjustment strategies such as material package exchange, priority advancement, and reserved duration relaxation are performed on the situation where the material delivery time cannot be met due to obstacle avoidance and waiting, so as to achieve conflict-free path planning for AGVs.

[0007] The described travel plan includes: classifying the AGV travel according to the number of distribution kits, the characteristics of the driving road, and the combination of target temporary storage areas, and estimating the total driving duration of each travel.

[0008] The classification of the AGV travel is based on the rasterized map of one-way driving roads. Set the number of station temporary storage areas as S, the loading capacity of the AGV as Q, and use J to represent the set of all possible travel types of the AGV. Then, according to the enumeration, the number of all possible travel types is Describe the travel type information as a vector α of length S. The i-th value in the vector is the number of AO kits sent to the temporary storage area i. The vector also determines all the target temporary storage areas of the AGV; in the task allocation model, through the difference in the combination of travel distribution kits, calculate the number value N j , obtain the travel type under this combination, and realize the one-to-one correspondence between the kit combination and the travel type. For example, α = [0, Q, …, 0, 0] means that this travel needs to send Q kits to the temporary storage area 2.

[0009] To establish the correspondence between the kit combination and the travel in AGV distribution, first assign number values {D 1 , D 2 , …, D S} to the S temporary storage areas. Then, the number value of the target temporary storage area of the i-th kit is D i ∈ {D 1 , D 2 , …, D s}. The number value of the j-th travel type is To distinguish travel types, by designing the values of {D 1 , D 2 , …, D s}, make the calculated number values of each travel type different.

[0010] When S = 3 and Q = 2, there are a total of |J| = 9 travels, including: α 1 = [1, 0, 0], α 2 = [2, 0, 0], α 3 =

[0011] [0, 1, 0], α 4 = [0, 2, 0], α 5 = [0, 0, 1], α 6 = [0, 0, 2], α 7 = [1, 1, 0], α 8 = [1, 0, 1], α 9 = [0, 1, 1]. Take the set D = {1, 3, 7}, then V = {1, 2, 3, 6, 7, 14, 4, 8, 10}, and the number value N corresponding to each travel type in the setj are all different.

[0012] The estimation of the total driving duration of each trip is specifically as follows: According to all the nodes passed by the loop corresponding to the trip, calculate the total duration required for the AGV to complete the loop without external interference, and use it as the estimated value of the total driving duration of each trip.

[0013] Taking the second trip α of the above 9 trips 2 = [2, 0, 0] as an example, the loop of this trip type includes: starting from the in-line library, driving to the temporary storage area 3 for unloading, moving to another one-way driving road after unloading is completed, and finally returning to the in-line library. The nodes involved in the second trip in the grid map include:

[0014] where n i , n j is a turning node, is a unloading node, the straight-line time = 0.2 min, the turning (steering) time = 0.2 min, the unloading time = 0.2 min, assuming k 3 = 12, at this time there are a total of 2k 2 + 2 = 26 nodes in the node set N 3 The estimated running duration C 2 of the second trip = 0.2×26 + 0.2×3 + 0.2 = 6 min.

[0015] The so-called rapid solution through the trip-based AGV task allocation model means: taking multiple packages with different target station temporary storage areas and different latest delivery times under a delivery batch as input, solving through the AGV task allocation model, and allocating each package to each delivery batch of each AGV.

[0016] The decision variables of the trip-based AGV task allocation model are: x ikr , w jkr , c kr , z kr , y k and t i , where: x ikr is a 0-1 variable, indicating whether the i-th package is completed by the r-th trip of the k-th AGV, 1 if yes and 0 if no; w jkr is a 0-1 variable, indicating whether the r-th trip of the k-th AGV is the j-th trip type, 1 if yes and 0 if no; c kr is a continuous variable, representing the running duration of the r-th trip of the k-th AGV; z kr is a continuous variable, representing the total running duration of the first r trips of the k-th AGV; yk is a 0-1 variable, indicating whether the k-th AGV is enabled, where 1 means enabled and 0 means not enabled; t i is a continuous variable, representing the actual delivery time of the i-th packet;

[0017] The objective function of the trip-based AGV task allocation model is: min VC·∑ k∈K y k +TC·∑ k∈K ∑ r∈R c kr , that is, to minimize the total sum of AGV input and operation costs, where: VC and TC are the unit input cost and the operation cost per unit time of the AGV respectively;

[0018] The constraints of the trip-based AGV task allocation model include:

[0019] 1) AGV vehicle enabling constraint: where: I is the set of packets, K is the set of AGV vehicles, and R is the set of actual delivery trips of each AGV;

[0020] 2) AGV trip type constraint. Each trip of the AGV is one of all the trip types obtained from the trip planning:

[0021] where: C j is the estimated running duration of the AGV for the j-th trip type;

[0022] 3) Latest delivery time constraint for packets. The actual delivery time of the packet needs to be before the required time: where T i is the delivery duration of the i-th packet from the in-line library to the temporary storage area of the target workstation, and L i is the latest delivery time of the i-th packet;

[0023] 4) Constraint on the value range of decision variables: x ikr , w jkr , y k are 0-1 variables, c kr ≥0, z kr ≥0, t i ≥0.

[0024] The time window algorithm described above specifically includes:

[0025] ① Renumber all the trips of all AGVs in the task allocation result to obtain the trip set S, and through the vector M s =(Btype,A,t rqt, k, rank) for all trip information, where: s is the trip number, s ∈ S; Btype is the trip type number corresponding to trip s, Btype ∈ J; A is the set of all delivery package numbers for trip s; t rqt is the time when trip s starts to be executed; k is the AGV number corresponding to trip s; rank is the execution priority of trip s, and the initial value is the remaining delivery time of all delivery packages corresponding to trip s, that is, the sum of the differences between the latest delivery time and the actual delivery time of all packages ∑ i∈A (L i -t i ). The smaller the value, the more urgent the work is, and the more necessary it is to arrange the time window preferentially. M s The set of all nodes passed by the corresponding trip in order is M s At node n d ∈ N s The time window function on it is T w,sd =(k, M s , q, t im,d , t out,d ), where: q is the sequential number of node n d in the node set N s ; t in,d is the time when vehicle k enters node n d ; t out,d is the time when vehicle k leaves node n d . For the time window of node n d , it satisfies: t out,d =t in,d +h d . When node n d is the starting node, the time to enter the starting node is equal to the start time of work; when node n d is not the starting node, the time to enter the node is equal to the time when AGV leaves the (q - 1)-th node in the path, that is For multiple time windows on a node, they can be in the form of a vector: T w,d =[T w,1d , T w,2d ,…, T w,Pd , where: P is the total number of trips passing through node n d in all the paths planned by the current trips, P ≤ i - 1. When there are multiple time windows for a job on an edge, the entry time in the newly added job time window must satisfy:

[0026] 1) The time to enter this edge must be greater than the leaving time of AGV from the previous edge; 2) The length of the idle time window of this edge is sufficient for vehicle k to drive away from this edge within this time. Assume in job M sPreviously, there were (s - 1) work arrangements, and among these (s - 1) works, P works had planned itineraries passing through node n d To find a long enough free time window to arrange new work, the free time window coefficient is determined by the following formula:

[0027] After C is determined, the entry time of the time window on node n d is determined by the following formula: When the first P time windows on node n d are arranged closely and it is impossible to arrange a new time window in the previous free time window, the new time window needs to be arranged after the Pth time window, and the corresponding function of the entry time is: t in,d =(t out,d ) P . After the time window planning is completed, the actual delivery time of AO material package q is given by the following formula: Where: s q is the number of the itinerary corresponding to AO material package q d q is the node number corresponding to the target temporary storage area of AO material package q

[0028] ② Time window planning, specifically including:

[0029] Step1. Define all AGV itinerary information according to the solution results obtained from the AGV task allocation model

[0030] Step2. Initialization of the time window. Find the set of nodes passed by the itinerary according to the itinerary type, and arrange an ideal time window distribution for each node

[0031] Step3. Update of the time window. After arranging the node time windows in the ideal situation, check whether there are the same nodes between the itineraries of different works. If there are no same nodes, end the time window planning; when there are the same nodes, calculate the time window vector of this node

[0032] Step4. Insertion of the time window. According to the obtained time window vector, calculate the free time window coefficient, determine the entry time of the time window insertion, and update the time windows of all nodes after the same node of this work

[0033] Step5. Calculation of the actual delivery time. Calculate the actual delivery time of each material package through the formula

[0034] The so-called packet exchange means: for packets that violate the latest delivery time constraint, search for packets with the same target temporary storage area, and by exchanging the delivery vehicle numbers and delivery schedules of the two packets, realize the interchange of the actual delivery times of the packets. When the latest delivery time constraints of both packets can be satisfied after the exchange, the exchange is carried out. Specifically, it includes:

[0035] i) Check the path planning result of the time window algorithm to obtain the set I′ of packets that violate the constraint, the number K of packets in the set, and let p = 1.

[0036] ii) Obtain the target temporary storage area of the p-th packet in the set I′, and search for the set I of packets with the same target temporary storage area in the set I of all packets p .

[0037] iii) Traverse the set I of packets p , and judge whether there exists a packet q that satisfies the condition that the actual delivery time of packet q is less than the latest delivery time of packet p, and the actual delivery time of packet p is less than the latest delivery time of packet q. If it exists, go to step iv); otherwise, end the packet exchange strategy and go to the packet exchange strategy.

[0038] iv) Exchange the delivery vehicle numbers and itinerary rounds of packets p and q, and remove the p-th packet from the set I′. Judge whether the set I′ is an empty set. When I′ is an empty set, end the packet exchange strategy and the adjustment is completed; when I′ is not an empty set, let p = p + 1 and go to step ii).

[0039] The so-called priority advance means: if the packet exchange strategy fails, for the packets that violate the constraint, find the itinerary where the packet is located, advance the priority of the time window planning corresponding to the itinerary, re-perform the time window planning, and at the same time use the packet exchange strategy for auxiliary adjustment. Specifically, it includes:

[0040] i) Check the path planning result of the time window algorithm to obtain the set I′ of packets that violate the constraint, the number K of packets in the set. When I′ is an empty set, end the priority advance strategy and the adjustment is completed; when I′ is not an empty set, go to step ii).

[0041] ii) Search to obtain the itinerary number s of the first packet in the set I′ p .

[0042] iii) When s p = 1, end the priority advance strategy and go to the reserved duration relaxation strategy; when s p ≠ 1, then exchange the priorities of itinerary s p and itinerary (s p - 1), and go to step iv).

[0043] (IV) Rearrange all the trips in the trip set P through the time window algorithm again, and adjust them using the material package exchange strategy. If the adjustment fails, go to step (I).

[0044] The relaxation of the reserved duration mentioned above means: multiplying and relaxing the estimated driving duration of the trips in the task allocation stage to reserve more time for the conflicts of AGVs, re-solving the task allocation model, and running the time window algorithm for path planning. At the same time, the material package exchange strategy and the priority advance strategy are used for auxiliary adjustment. The initial value of the relaxation multiple k is 1.1, which specifically includes:

[0045] a. Set the relaxation multiple k = 1.1.

[0046] b. Relax the estimated driving duration of each trip type in the task allocation stage, and let C j = kC j , and re-call the solution method for task allocation and path planning.

[0047] c. Re-check the path planning result of the time window algorithm to obtain the set I' of material packages that violate the constraints, and the number K of material packages in the set. When I' is an empty set, end the reserved duration relaxation strategy and the adjustment is completed; when I' is non-empty, adjust it through the material package exchange strategy and the priority advance strategy to obtain the final set I'' of material packages that violate the constraints. When I'' is an empty set, end the reserved duration relaxation strategy and the adjustment is completed. Otherwise, let k = 1.1k, and go to step b.

[0048] The present invention relates to a system for implementing the above method, including: a task allocation unit, a path planning unit, and an adjustment strategy unit. Among them: the task allocation unit, according to the material package information of multiple different target station temporary storage areas and different latest delivery times under a delivery batch, obtains the preliminary task allocation result of the AGV through trip planning and model solution. The path planning unit defines the delivery trip information of each AGV under the preliminary task allocation result of the AGV, and arranges the order of the AGV occupying the map resources through the time window algorithm, so as to plan the actual running path of the AGV and avoid the conflicts and collisions of the AGV. The adjustment strategy unit, aiming at the situation where the path planning unit may violate the latest delivery time constraint, applies the material package exchange, priority advance, and reserved duration relaxation strategies to adjust the material packages that violate the latest delivery time constraint, and finally obtains the AGV task allocation and path planning result that meets the constraints.

[0049] Technical Effects

[0050] Compared with the existing conventional technical means, the present invention solves the multi-AGV scheduling problem in stages by dividing modules, and simultaneously optimizes the task allocation and path planning through the adjustment strategy, reducing the total distribution cost by 13%.

[0051] Compared with the prior art, in view of the phenomenon that the AGV makes multiple round trips in the workshop, in the task allocation stage, the present invention proposes the concept of "journey", classifies the AGV journeys through journey planning, greatly reduces the complexity of modeling, and on this basis, establishes a task allocation model based on "journey"; in the path planning stage, a time window algorithm with the journey schedule as the input and three conflict resolution strategies are designed to achieve efficient planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a flowchart of the method of the present invention;

[0053] Figure 2 is a rasterized map of the application workshop in the embodiment;

[0054] Figure 3 is a schematic diagram of the time window planning result in the embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] As Figure 1 shown, the present embodiment relates to a method for optimizing and implementing AGV task allocation and path planning in two stages, including:

[0056] In the task allocation stage, all possible journey types of the AGV in the workshop are listed through journey planning, and the possible journey information under different task combinations is determined; the task allocation plan is quickly obtained through the AGV task allocation model based on the journey;

[0057] In the path planning stage, the time window algorithm is used to initialize, update and arrange the time window of the map resources occupied by the AGV, and adjustment strategies such as material package exchange, priority advance, and reserved duration relaxation are carried out for the situation where the material delivery time cannot be met due to obstacle avoidance and waiting, so as to realize conflict-free path planning of the AGV. The specific implementation process is as follows:

[0058] ① Journey planning:

[0059] As Figure 2 shown is a rasterized map of the general assembly workshop of a civil airliner. The driving roads of the AGV have been pre-planned and rasterized in advance. The grid point size is the same as the size of the AGV, and there is a QR code at the center of each grid point for positioning. There are two AGV driving roads in the workshop, and both are one-way driving.

[0060] In this embodiment, each node is sequentially numbered according to the actual driving direction of the AGV, and the node set N =

[0061] {n 1 , n 2 , …, n m}, where They are the material temporary storage areas on the sides of three workstations respectively. The AGV travels through the inner road to the temporary storage area of the target workstation to deliver the AO material package. After the delivery task is completed, it returns to the docking point near the AGV charging pile through the outer road.

[0062] There are 3 temporary storage areas for workstations in the final assembly workshop. The upper limit of the number of material packages delivered by each AGV is 2. A total of 9 types of travel routes are calculated, as shown in Table 1.

[0063] Table 1

[0064]

[0065]

[0066] ② Solving the AGV task allocation model based on travel routes:

[0067] In this embodiment, the maximum number of AGVs in the AGV set K is 6; the maximum number of travel routes in the AGV travel route set R is 10; the unit input cost VC of the AGV is 100; the unit operation cost TC is 2; there are a total of 94 material packages in the material package set I, and the specific information of each material package is shown in Table 2.

[0068] Table 2

[0069]

[0070]

[0071] By using the GUROBI solver to solve the model, the task allocation results of each AGV and the specific information of each trip of the AGV are obtained, as shown in Table 3.

[0072] Table 3

[0073]

[0074] ③ Time window algorithm planning:

[0075] Step1: Renumber all the trips of all AGVs in the task allocation results to obtain the trip set S, and use the vector M s =(Btype,A,t rqt ,k,rank) to represent all the trip information, where: s is the trip number, s∈S; Btype is the trip type number corresponding to trip s, Btype∈J; A is the set of all delivery material package numbers of trip s; t rqt is the start time of trip s; k is the AGV number corresponding to trip s; rank is the execution priority of trip s. Finally, the converted all trip information is shown in Table 4.

[0076] Table 4

[0077]

[0078] Step 2: According to the travel information vector M of each travel s in sequence s , calculate the entry time and departure time of each node occupied by travel s through the time window algorithm, and arrange the order of occupied nodes for all travels by inserting and updating the time window, finally obtaining the conflict-free time window planning result for multiple AGVs. As Figure 2 shown, the time windows of each node under the optimal task allocation result obtained by the time window algorithm are as follows. The abscissa is the delivery time axis, and the ordinate is the number value of each node in the grid map. Different colors represent the travels of different AGVs. The color of each color block is the corresponding AGV, the ordinate of the color block is the node number, and the horizontal length of the color block is the time window length of the corresponding AGV occupying this node during this trip. Figure 3 shown, the time windows of each node under the optimal task allocation result obtained by the time window algorithm are as follows. The abscissa is the delivery time axis, and the ordinate is the number value of each node in the grid map. Different colors represent the travels of different AGVs. The color of each color block is the corresponding AGV, the ordinate of the color block is the node number, and the horizontal length of the color block is the time window length of the corresponding AGV occupying this node during this trip.

[0079] Step 3: Calculate the actual delivery time of each package after planning through the time window algorithm, as shown in Table 5. Among them: t i is the planned delivery time of package i in the task allocation stage, obtained by solving the task allocation model; is the actual delivery time of package i after completing the path planning through the time window algorithm; L i is the latest delivery time of package i.

[0080] Table 5:

[0081]

[0082]

[0083] ④ Adjustment strategy:

[0084] During the solution process of the two-stage algorithm, the actual delivery times of packages numbered 3, 23, and 24 violated the latest delivery time constraint. For these packages, the adjustment was completed through the package exchange strategy. The specific process is as follows:

[0085] Step 1: According to the path planning result, obtain the set of packages I' = {3, 23, 24} that violate the constraint, and the number of packages K = 3 in the set, and let p = 1.

[0086] Step 2: The target temporary storage area of the p-th package (i.e., package 3) in the set I' is temporary storage area 3. Search for the set of packages I 1= {1, 2, 3, 4, 5, 6, 20, 21, 22, 23, 24, 25, 26, 67, 68, 69, 74, 75, 76, 77, 80, 81, 82, 85, 86, 87, 88, 90, 91}。

[0087] Step3: Traverse the set I of material packages 1 , and determine whether there exists a material package q such that the actual delivery time of material package q is less than the latest delivery time of material package p, and the actual delivery time of material package p is less than the latest delivery time of material package q. It is found that material package 1 satisfies the above conditions, and go to Step4.

[0088] Step4: Exchange the delivery vehicle number and trip round of the material package numbered 3 and the material package numbered 1, and remove the material package numbered 3 from the set I'. Since I' is not empty, let p = p + 1, and repeat the above exchange process. The final exchanged numbers are as follows: exchange material package 3 and material package 1, exchange material package 23 and material package 2, exchange material package 24 and material package 21. After the exchange is completed, I' is an empty set, and the adjustment is completed.

[0089] The final AGV task allocation is shown in Table 6.

[0090] Table 6:

[0091]

[0092] The final optimized cost includes: A total of 5 AGVs are used, and the input cost is 500; the operation cost of the AGV is 760; the total distribution cost is 1260.

[0093] Using the original first-in-first-out scheduling method in the workshop, some material packages violate the latest delivery time constraint when 5 AGVs are used for distribution. Therefore, 6 AGVs are selected for distribution, and the AGV distribution plan is shown in Table 7.

[0094] Table 7:

[0095]

[0096] The cost obtained by the original workshop scheduling method includes: A total of 6 AGVs are used, and the input cost is 600; the operation cost of the AGV is 768; the total distribution cost is 1368. Compared with the original workshop scheduling method, the AGV scheduling algorithm of this patent reduces the AGV distribution cost by 8%, and the method of this patent reduces the input cost of the AGV, and has more cost advantages in the case of multiple distribution batches.

[0097] The above specific embodiments can be locally adjusted in different ways by those skilled in the art without departing from the principles and purposes of the present invention. The protection scope of the present invention is subject to the claims and is not limited by the above specific embodiments. All implementation solutions within its scope are subject to the present invention.

Claims

1. A two-stage optimization implementation method for AGV task allocation and path planning, characterized in that, firstly, in the task allocation stage, based on the characteristics of the workshop map, all possible travel types of AGVs in the workshop are listed through itinerary planning, and the possible travel information under different task combinations is determined; for multiple distribution tasks in a distribution batch in the workshop, a task allocation plan is quickly obtained through an AGV task allocation model based on itinerary; in the path planning stage, the obtained task allocation plan is converted into time window planning work, and a time window algorithm is used to initialize, update and arrange time windows for the map resources occupied by AGVs, and adjustment strategies such as material package exchange, priority advancement, and reserved duration relaxation are carried out for the situation where the material delivery time cannot be met due to obstacle avoidance and waiting, so as to achieve conflict-free path planning of AGVs; The rapid solution through the AGV task allocation model based on itinerary means: taking multiple material packages with different target workstation temporary storage areas and different latest delivery times in a distribution batch as input, solving through the AGV task allocation model, and allocating each material package to each distribution batch of each AGV; The decision variables of the above-mentioned travel-based AGV task allocation model are: x ikr , w jkr , c kr , z kr , y k and t i , where: x ikr is a 0-1 variable, indicating whether the i-th material package is completed by the r-th trip of the k-th AGV. If so, it is 1; if not, it is 0; w jkr is a 0-1 variable, indicating whether the r-th trip of the k-th AGV is of the j-th trip type. If so, it is 1; if not, it is 0; c kr is a continuous variable, representing the running duration of the r-th trip of the k-th AGV; z kr is a continuous variable, representing the total running duration of the first r trips of the k-th AGV; y k is a 0-1 variable, indicating whether the k-th AGV is enabled. If so, it is 1; if not, it is 0; t i is a continuous variable, representing the actual delivery time of the i-th material package; The objective function of the travel-based AGV task allocation model is: min VC·∑ k∈K y k +TC·∑ k∈K ∑ r∈ R c kr , that is, to minimize the total AGV investment and operating costs, where: VC and TC are the unit investment cost and the operating cost per unit time of the AGV, respectively; The constraints of the AGV task allocation model based on itinerary include: 1) AGV vehicle activation constraints: Where: I is the set of material packages, K is the set of AGV vehicles, and R is the set of actual delivery trips for each AGV; 2) Constraints on AGV travel types. Each trip of the AGV is one of all travel types obtained from the travel plan: Where: C j is the estimated running time of the AGV for the j-th travel type; 3) Constraint on the latest delivery time of the ingredient package. The actual delivery time of the ingredient package needs to be before the required time: where T i is the delivery duration of the i-th ingredient package from the line-side warehouse to the temporary storage area at the target workstation, and L i is the latest delivery time of the i-th ingredient package; 4) Constraints on the value range of decision variables: x ikr , w jkr , y k are 0-1 variables, c kr ≥ 0, z kr ≥ 0, t i ≥ 0.

2. The two-stage optimization implementation method for AGV task allocation and path planning according to claim 1, characterized in that, the itinerary planning includes: classifying AGV itineraries according to the number of distribution material packages, the characteristics of the driving roads, and the combination of target temporary storage areas, and estimating the total driving duration of each itinerary, that is, calculating the total duration required for the AGV to drive through the loop corresponding to the itinerary without external interference as the estimated value of the total driving duration of each itinerary.

3. The two-stage optimization implementation method for AGV task allocation and path planning according to claim 1, characterized in that, the time window algorithm specifically includes: ①Renumber all the trips of all AGVs in the task assignment result to obtain a trip set S, and use the vector M s =(Btype, A, y rqt , k, rank) to represent all trip information, where: s is the trip number, s ∈ S, Btype is the trip type number corresponding to trip s, Btype ∈ J, A is the set of all delivery package numbers of trip s, t rqt is the start execution time of trip s, k is the AGV number corresponding to trip s, rank is the execution priority of trip s, and the initial value is the remaining delivery time of all delivery packages corresponding to trip s, that is, the sum of the differences between the latest delivery time and the actual delivery time of all packages ∑ i∈A (L i -t i ), M s The set of all nodes passed by the corresponding trips in order is M s At node n d ∈ N s The time window function on it is T w,sd =(k, M s , q, t in,d , t out,d ), where: q is the sequential number of node n d in the node set N s , t in,d is the time when vehicle k enters node n d , t out,d is the time when vehicle k leaves node n d , for the time window of node n d , it satisfies: t out,d =t in,d +h d , when node n d is the starting node, the time to enter the starting node is equal to the start time of work, when node n d is not the starting node, the time to enter the node is equal to the time when the AGV leaves the (q - 1)-th node in the path, that is For multiple time windows on a node, they can be in the form of a vector: T w,d =[T w,1d , T w,2d ,…, T w,Pd , where: P is the total number of trips passing through node n d in the currently planned paths of all trips, P ≤ i - 1; ② Time window planning, specifically including: Step1, Define all AGV itinerary information according to the solution result obtained by the AGV task allocation model; Step2, Initialization of time windows, find the set of nodes passed by the itinerary according to the itinerary type, and arrange an ideal time window distribution for each node; Step3, Update of time windows, after arranging the node time windows in the ideal situation, check whether there are the same nodes between the itineraries in different workshops; if there are no same nodes, end the time window planning; when there are the same nodes, calculate the time window vector of this node; Step4, Insertion of time windows, calculate the free time window coefficient according to the obtained time window vector, determine the entry time of time window insertion, and update the time windows of all nodes after the same node of this work; Step5, Calculate the actual delivery time, and calculate the actual delivery time of each material package through the formula.

4. The two-stage optimization implementation method for AGV task allocation and path planning according to claim 3, characterized in that, When there are multiple working time windows on an edge, the entry time in the newly added working time window must satisfy: 1) The time to enter this edge must be greater than the departure time of the AGV from the previous edge; 2) The length of the idle time window of this edge is sufficient for vehicle k to drive away from this edge within this time; Assume that before job M s there have been (s - 1) job arrangements, and among these (s - 1) jobs, P jobs have planned routes passing through node n d . To find a long enough idle time window to arrange a new job, the idle time window coefficient is determined by the following formula: After C is determined, the entry time of the time window on node n d is determined by the following formula: When node n d The first P time windows on are arranged closely, and it is impossible to arrange a new time window in the previous free time window. Then the new time window needs to be arranged after the P-th time window. The corresponding function for entering time is: t in,d =(t out,d ) P ; After the time window planning is completed, the actual delivery time of the AO material package q is given by the following formula: where: s q is the number of the itinerary corresponding to the AO material package q, and d q is the node number corresponding to the target temporary storage area of the AO material package q.

5. The two-stage optimization implementation method for AGV task allocation and path planning according to claim 1, characterized in that, The so-called packet exchange means: for packets that violate the constraints, search for packets with the same target temporary storage area, and by exchanging the delivery vehicle numbers and delivery schedules of the two packets, realize the interchange of the actual delivery times of the packets. When the latest delivery time constraints of both packets can be satisfied after the exchange, the exchange is carried out; The so-called priority advancement means: when the packet exchange strategy fails, for packets that violate the constraints, find out the itinerary where the packet is located, advance the priority of the time window planning work corresponding to the itinerary, re-plan the time window, and at the same time use the packet exchange strategy for auxiliary adjustment. The so-called reserved duration relaxation means: multiply and relax the estimated driving duration of the itinerary in the task allocation stage, reserve more time for the conflicts of AGVs, re-solve the task allocation model, and run the time window algorithm for path planning. At the same time, use the packet exchange strategy and the priority advancement strategy for auxiliary adjustment.

6. The AGV task allocation and path planning two-stage optimization implementation method according to claim 5, characterized in that, the so-called packet exchange specifically includes: i) Check the path planning result of the time window algorithm, obtain the set I' of packets that violate the constraints, and the number K of packets in the set, and let p = 1; ii) Obtain the target temporary storage area of the p-th packet in the set i′, and search for the packet set I in the entire packet set I that has the same target temporary storage area as it p ; iii) Traverse the set of ingredient packs I p , and determine whether there exists an ingredient pack q such that the actual delivery time of ingredient pack q is less than the latest delivery time of ingredient pack p, and the actual delivery time of ingredient pack p is less than the latest delivery time of ingredient pack q; if there is, go to step iv), otherwise, end the ingredient pack exchange strategy and go to the ingredient pack exchange strategy; iv) Exchange the delivery vehicle numbers and itinerary rounds of packets p and q, remove the p-th packet from the set I', and judge whether the set I' is an empty set. When I' is an empty set, end the packet exchange strategy and the adjustment is completed. When I' is non-empty, let p = p + 1, and go to step ii).

7. The AGV task allocation and path planning two-stage optimization implementation method according to claim 5, characterized in that, the so-called priority advancement specifically includes: a) Check the path planning result of the time window algorithm, obtain the set I' of packets that violate the constraints, and the number K of packets in the set. When I' is an empty set, end the priority advancement strategy and the adjustment is completed. When I' is non-empty, go to step b); (ii) Search for the travel number s where the first ingredient packet in set I' is located p ; (III) When s p = 1, end the priority advance strategy and switch to the reserved duration relaxation strategy; when s p ≠ 1, then swap the priority of trip s p with the priority of trip (s p - 1), and go to step (IV); iv) Re-arrange all the itineraries in the itinerary set P through the time window algorithm, and use the packet exchange strategy for adjustment. When the adjustment fails, go to step a).

8. The AGV task allocation and path planning two-stage optimization implementation method according to claim 5, characterized in that, the so-called reserved duration relaxation specifically includes: a. Set the relaxation multiple k = 1.1; b. Relax the estimated driving duration of each trip type in the task assignment phase, and let C j = kC j , and then re - call the solution method to perform task assignment and path planning; c. Re-check the path planning result of the time window algorithm, obtain the set I' of packets that violate the constraints, and the number K of packets in the set. When I' is an empty set, end the reserved duration relaxation strategy and the adjustment is completed. When I' is non-empty, adjust through the packet exchange strategy and the priority advancement strategy to obtain the final set I'' of packets that violate the constraints. When I'' is an empty set, end the reserved duration relaxation strategy and the adjustment is completed. Otherwise, let k = 1.1k, and go to step b).

9. A system for implementing the AGV task allocation and path planning two-stage optimization implementation method described in any one of claims 1 to 8, characterized in that, comprising: A task allocation unit, a path planning unit, and an adjustment strategy unit, where: The task allocation unit obtains the preliminary task allocation result of the AGV through itinerary planning and model solving based on the information of multiple packs with different target workstation staging areas and different latest delivery times under a delivery batch. The path planning unit defines the delivery itinerary information of each AGV under the preliminary task allocation result of the AGV, arranges the order of AGVs occupying map resources through the time window algorithm, so as to plan the actual running path of the AGV and avoid conflicts and collisions of the AGV. The adjustment strategy unit, aiming at the situation where the path planning unit may violate the latest delivery time constraint, applies strategies such as pack exchange, priority advancement, and reserved duration relaxation to the packs that violate the latest delivery time constraint, and finally obtains the AGV task allocation and path planning result that meets the constraints.

Images