A multi-AGV path planning method, system, storage medium and product

Abstract

The present invention belongs to the field of intelligent logistics, and specifically relates to a multi-AGV path planning method, system, storage medium and product, S1: constructing a warehouse topology map; S2: extracting feature information of each sub-path segment; S3: designing a time window prediction model, and updating the parameters of the time window prediction model; S4 solving the first K shortest paths from the starting point to the end point of each AGV, and generating a candidate path set C of n AGVs; S5: introducing the chromosome structure of the genetic algorithm into the ant lion optimization algorithm, constructing an indefinite length spatiotemporal composite multi-chromosome encoding suitable for multi-AGV path planning, which is used to represent the positions of ants and ant lions, wherein each position represents a multi-AGV path planning scheme; S6-S11: using the ant lion optimization algorithm to solve the multi-AGV path planning scheme of the elite ant lion, and controlling each AGV to perform tasks; in a multi-AGV storage environment, the possibility of mutual interference between the AGVs can be effectively reduced, path congestion can be reduced, and work efficiency can be improved.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent logistics, and specifically relates to a multi-AGV path planning method, system, storage medium and product. Background Art

[0002] At present, many scholars have proposed path planning algorithms based on the time window principle to effectively coordinate the path planning of multiple AGVs and resolve path conflicts. These methods effectively reduce the occurrence of path conflicts through time allocation, priority setting, real-time path replanning and intelligent collaboration. However, most of the current time window solution methods use simple physical models for calculation. It is generally assumed that the travel speed of the AGV is fixed, and the travel time is calculated at a fixed speed. Although some studies have taken into account the change in speed, they still ideally assume that the acceleration or deceleration is fixed. For example, the travel time of the AGV is calculated using static factors such as fixed straight travel, turning speed, and acceleration and deceleration. This type of processing method ignores or simplifies the complex motion state change process of the AGV in actual work, resulting in inaccurate time window solution.

[0003] In addition, the current multi-AGV path optimization goal is generally to minimize the running time or energy consumption, and the time and space distance between the paths is rarely considered, resulting in a short time and space distance between the conflict-free paths. AGVs are prone to interfere with each other, resulting in poor system robustness and weak anti-interference ability. Although the introduction of conflict resolution methods such as dynamic adjustment or path replanning can effectively alleviate local path conflicts, it is also easy to have a negative impact on the overall system. For example, when multiple AGVs need to replan their paths at the same time, new conflicts may arise or the real-time performance of the system may decrease. In addition, frequent path adjustments may also increase the energy consumption and wear of AGVs, thereby affecting the long-term stability and reliability of the system. Summary of the invention

[0004] The object of the present invention is to provide a multi-AGV path planning method, system, storage medium and product that considers the path time and space distance and can improve work efficiency.

[0005] In order to achieve the above object, the present invention adopts such technical solution:

[0006] A multi-AGV path planning method comprises the following steps:

[0007] S1: Build a warehouse topology map based on the storage environment;

[0008] S2: Obtain the travel data of each AGV within a period T, split the travel data into sub-path segments and extract the feature information of each sub-path segment;

[0009] S3: designing a time window prediction model based on a neural network, using the feature information to train the time window prediction model, and updating the parameters of the time window prediction model;

[0010] S4: Solve the first K shortest paths from the starting point to the end point of each AGV and generate a candidate path set C for n AGVs;

[0011] S5: Introduce the chromosome structure of genetic algorithm into the antlion optimization algorithm, and construct an indefinite length spatiotemporal composite multi-chromosome encoding suitable for multi-AGV path planning, which is used to represent the positions of ants and antlions, where each position represents a multi-AGV path planning solution;

[0012] S6: extracting input features from the multi-AGV path planning scheme, and generating a time window sequence for each path using the time window prediction model;

[0013] S7: discretize the time window sequence and the path node sequence to generate a path spatiotemporal sequence;

[0014] The generation method of the path spatiotemporal sequence mentioned in step S7 is as follows:

[0015] (1);

[0016] (2);

[0017] in, is the path node sequence of the nth AGV, is the number of the mth path node, is the time window sequence of the nth AGV, The nth AGV reaches the path node time, The nth AGV leaves the path node time;

[0018] node With Node The discretization operation between is as follows:

[0019] Compute Node With Node Time interval and nodes With Node The number of space-time path points that need to be inserted :

[0020] (3);

[0021] (4);

[0022] in, For Node With Node The time interval, in seconds, For Node The x-axis coordinate of For Node The y-axis coordinate of is the discretized time step in seconds;

[0023] Compute Node and The k1th point inserted between Time coordinates and the x-axis coordinate , y-axis coordinate :

[0024] (5);

[0025] (6);

[0026] (7);

[0027] in, ;

[0028] node The discretization operation at is as follows:

[0029] Compute Node Duration of stay , and nodes The number of space-time path points after discretization ;

[0030] (8);

[0031] (9);

[0032] Compute Node The coordinates of the k2th space-time path point after discretization , represents the time coordinate of the k2th point, represents the x-axis coordinate of the k2th point, Represents the y-axis coordinate of the k2th point:

[0033] (10);

[0034] (11);

[0035] (12);

[0036] in, ;

[0037] S8: Calculate the path space-time distance based on the discretized path space-time sequence;

[0038] The path space-time distance in step S8 is calculated as follows:

[0039] (13);

[0040] in, Represents any two space-time path points and The space-time distance between

[0041] Calculate the space-time distance between any two paths and the total space-time distance of all paths :

[0042] (14);

[0043] (15);

[0044] in, Represents the spatiotemporal path sequence of the i-th AGV, which is composed of the spatiotemporal path points after the corresponding path node sequence is discretized. N represents the spatiotemporal path sequence The number of space-time path points in , represent The mth space-time path point in represent The mth space-time path point in The shortest distance between any space-time path points in ;

[0045] S9: Calculate the fitness values ​​of the ant lions and sort them, and select the ant lion with the smallest fitness value as the elite ant lion;

[0046] S10: The ants walk randomly under the influence of the corresponding ant lion and the elite ant lion, calculate the fitness value of the ants, and update the ant population;

[0047] In this embodiment, the calculation function of the fitness value mentioned in steps S9 and S10 is:

[0048] (16);

[0049] (17);

[0050] (18);

[0051] (19);

[0052] in, represents the fitness value of the antlion, represents the location of the i-th antlion, represents the fitness value of the ant, represents the position of the i-th ant, Represents the conflict penalty weight. Since avoiding conflicts is the main goal, a relatively high value needs to be set to ensure that avoiding conflicts is given priority in path planning. Represents the number of path conflicts; Represents the weight of the total completion time, which is mainly used to minimize the total path length, so we can choose a Smaller but still significant values, Represents the end point of the total path, Represents the starting point of the total path, Represents the total completion time; Represents the weight of the total path time-space distance. In order to encourage the increase of the total path time-space distance, this weight should be set to a negative value. represents the total space-time distance of all paths, Indicates whether there is a conflict between paths, 1 means there is a conflict, 0 means there is no conflict. Indicates that the i-th vehicle arrives at the node time, Indicates that the i-th vehicle leaves the node time, Indicates that the i-th vehicle occupies the node time window, Indicates that the i-th vehicle occupies a connected node With Node The time window of the path, Indicates that the i-th car and the j-th car are at the node There is a node conflict at Indicates that the i-th car and the j-th car are at the connecting node With Node There is a conflict between the paths. Indicates that the i-th car and the j-th car are at the connecting node With Node There is a pursuit conflict in the path;

[0053] Among them, the specific steps of the ant randomly walking under the influence of the corresponding ant lion and the elite ant lion in step S10 are as follows:

[0054] S10-1: Generate a random number a in the range of 0 to 1. If a>0.4, the ant moves to the position of the elite ant lion, otherwise, the ant moves to the position of the corresponding ant lion. The movement is performed through the crossover operation. The positions of the ant and the ant lion are composed of multiple chromosomes. The crossover operation can only be performed between corresponding chromosomes. In order to ensure the continuity of the path, the intersection point must be selected at the common path node of the two chromosomes. When there are multiple common path nodes between the corresponding chromosomes, one of them is randomly selected as the crossover node. The corresponding chromosome refers to the chromosome in the ant and the ant lion that represents the same AGV path. Two new positions can be obtained through the crossover operation;

[0055] S10-2: Randomly mutate the waiting time of each path node;

[0056] S10-3: Calculate the fitness values ​​of the two new positions. If the fitness value of the ant lion is greater than the fitness value of the corresponding ant, capture the ant, update the ant lion position to the position of the ant, and update the ant lion population;

[0057] S11: If the maximum number of iterations is reached, the elite ant lion is output, and each AGV is controlled to perform the task according to the multi-AGV path planning scheme represented by the location of the elite ant lion. Otherwise, jump to step S5.

[0058] Preferably, in step S1, the potential driving path of the AGV is simplified to a straight line in the warehouse topology map, and the path intersections, parking spaces and charging points are defined as nodes, and the nodes are numbered in sequence.

[0059] Preferably, in step S2, all AGV historical travel data within period T is obtained and split into sub-path segments according to the constructed warehouse topology map. Each sub-path segment records in detail the starting point, end point, departure time, node waiting time, path length, action type, and the time of arrival at and leaving the end point for training the time window prediction model.

[0060] Preferably, in step S3, the feature information is first processed by One-Hot encoding and normalization, and then the processed feature information is input into the time window prediction model. The time window prediction model uses a three-layer hidden layer structure to extract and compress feature information layer by layer. The first hidden layer contains 200 neurons, and ReLU is used as the activation function to perform weighted summation and nonlinear transformation on the input feature information. The extracted first feature is the basic and direct relationship. The second hidden layer contains 40 neurons, and the ReLU activation function is also used to further compress and filter the first feature, remove redundant information, retain key features, and improve the generalization ability of the model. The third hidden layer includes 10 neurons, and the third hidden layer is divided into two parts, each of which contains 5 neurons, and customized feature extraction is performed for two tasks respectively, one part is used for arrival time prediction, and the other part is used for departure time prediction to ensure that each task can obtain core information. The third hidden layer still uses the ReLU activation function to avoid gradient disappearance and maintain the nonlinear learning ability of the model. The output layer uses a linear activation function to generate continuous time values ​​and output the arrival time of the next node respectively. and departure time , used for AGV path planning and time scheduling optimization.

[0061] Preferably, step S5 also includes setting the maximum number of iterations of the ant lion optimization algorithm, and the population size of ants and ant lions, randomly generating the initial positions of ants and ant lions, the initial positions of ants and ant lions are composed of the path node sequence of each AGV and the waiting time sequence of the path nodes of each AGV, and the initial positions of each ant and each ant lion correspond one to one, and the path plan selects the shortest paths of different AGVs from the candidate path set C for random combination, and the waiting time of the node is randomly set in the range greater than or equal to 0.

[0062] Preferably, the method for extracting input features in step S7 is to split the path of each AGV into sub-path segments, and obtain the starting point number of the sub-path segment. , sub-path segment end point number , time to leave the starting point of the sub-path segment , whose initial value is the waiting time of the node where the total path starting point is located. For the subsequent sub-path segments, the time of leaving the sub-path segment starting point is the time of leaving the end point of the previous sub-path segment. Whether the path starting point in the sub-path segment is the total path starting point, if yes, then =1 ,on the contrary =0 , Is the end point of the path in the sub-path segment the end point of the total path? If yes, E=1 ,on the contrary =0 , the action of the AGV at the end of the path in the sub-path segment, straight =0 , turn =1 , turn around =2 ,wait =3 , subpath segment length , load operation F=1 , no-load operation F=0 The waiting time of the node at the end of the subpath is .

[0063] Preferably, in step S6, the indefinite length spatiotemporal composite multi-chromosome encoding is that the position of each ant and ant lion in the population is composed of multiple chromosomes, the number of chromosomes is determined by the number of AGVs, the nth chromosome represents the path of the nth AGV, each chromosome is composed of a series of path nodes and node waiting times, the path nodes and the waiting time at this node together constitute a gene of the chromosome, and the length of the chromosome depends on the number of path nodes from the starting point to the end point.

[0064] A multi-AGV path planning system includes a memory and a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement any of the methods described above.

[0065] A computer-readable storage medium stores a computer program, which implements any of the above methods when executed.

[0066] A computer product comprises a computer program, which implements any of the above methods when executed by a processor.

[0067] By adopting the aforementioned design scheme, the beneficial effects of the present invention are: first, a travel time estimation model is designed based on limited input features, and a more accurate time window can be obtained compared with the traditional time window solution method; second, on the basis of generating an accurate time window, a multi-AGV path collaborative planning method is designed with the goal of optimizing the path time-space distance, which increases the time-space distance of conflict-free multi-AGV path planning, and can effectively reduce the possibility of mutual interference between AGVs in a multi-AGV warehousing environment, reduce path congestion, ensure smoother communication and coordination between AGVs, and at the same time optimize system energy consumption and improve work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 is a flow chart of the present invention;

[0069] Figure 2 A topological map of a storage environment in an embodiment of the present invention;

[0070] Figure 3 This is a network structure diagram of the time window prediction model of the present invention;

[0071] Figure 4 The coordinate diagram of the starting point and the end point of each AGV in the embodiment of the present invention;

[0072] Figure 5 A schematic diagram of the indefinite length spatiotemporal composite multi-chromosome encoding of the present invention;

[0073] Figure 6 A schematic diagram of the crossover operation in the random walk of ants in the present invention;

[0074] Figure 7 It is a spatiotemporal three-dimensional path diagram of the multi-AGV path planning solution of the elite ant lion in an embodiment of the present invention. DETAILED DESCRIPTION

[0075] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only partial embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] A multi-AGV path planning method, such as Figure 1 As shown, the following steps are included:

[0077] S1: Construct a warehouse topology map according to the storage environment; in this embodiment, the topology map is used to describe Figure 2 In the warehouse environment shown, the potential driving paths of the AGV are simplified into straight lines in the topological map. Key locations such as path intersections, parking spaces, and charging points are defined as nodes. The nodes are numbered in sequence from 1 to 65, for a total of 65 nodes.

[0078] S2: Obtain the travel data of each AGV within a period T, split the travel data into sub-path segments and extract the feature information of each sub-path segment;

[0079] In this embodiment, the travel data in step S2 mainly includes the following information: the type of task, i.e., loaded or unloaded, the sequence of path nodes that the AGV needs to pass through to complete the task, the time of arrival and departure at each path node, the action performed at each path node, the coordinates of each path node, and the arrangement of the waiting time at each node. Each collected travel data is split into sub-path segments containing only two adjacent path nodes, and the characteristic information of each sub-path segment in each travel data is extracted, including the starting point number of the sub-path segment. , sub-path segment end point number ,S Indicates whether the path starting point of the sub-path segment is the total path starting point, E stands for Whether the path end point of the sub-path segment is the total path end point, and the path end point action in the sub-path segment A , load state F, the above features need to be One-Hot encoded, sub-path segment length , time to leave the starting point of the sub-path segment , the waiting time of the node where the subpath end point is located , normalization is required.

[0080] S3: Design a time window prediction model based on neural network, such as Figure 3 As shown, the feature information extracted in step S2 is used to train the time window prediction model and update the time window prediction model parameters;

[0081] In this embodiment, in step S3, the feature information is first processed by One-Hot encoding and normalization, and then the processed feature information is input into the time window prediction model. The time window prediction model uses a three-layer hidden layer structure to extract and compress feature information layer by layer. The first hidden layer contains 200 neurons and uses ReLU (Rectified Linear Unit) is used as the activation function to perform weighted summation and nonlinear transformation on the input feature information. The first feature extracted is the basic and direct relationship, such as the simple association between path length and time. The second hidden layer contains 40 neurons and also uses the ReLU activation function to further compress and filter the first feature, remove redundant information, retain key features, and improve the generalization ability of the model. The third hidden layer includes 10 neurons, and the third hidden layer is divided into two parts, each containing 5 neurons, which are customized for two tasks. Feature extraction is performed, one part is used to predict the arrival time, and the other part is used to predict the departure time, ensuring that each task can obtain core information. The third hidden layer still uses the ReLU activation function to avoid gradient disappearance and maintain the nonlinear learning ability of the model. The output layer uses a linear activation function to generate continuous time values ​​and output the arrival time of the next node respectively. and departure time , used for path planning and time scheduling optimization of AGV, the model can efficiently capture the complex patterns of path and time and provide accurate time prediction by sharing the features of the first two layers and performing task separation in the third layer.

[0082] Among them, the training and updating of model parameters mentioned in step S3 means that the model calculates the output results through forward propagation, uses the mean square error as the loss function to evaluate the prediction error, calculates the gradient through back propagation, and uses the optimization algorithm Adam to continuously update the weight matrix and bias terms, so that the prediction error of the model is gradually reduced. After multiple rounds of training, the model can accurately predict the arrival and departure time of AGV, providing support for path planning and time scheduling optimization.

[0083] S4: Solve the first K shortest paths from the starting point to the end point of each AGV and generate the candidate path C of n AGVs;

[0084] In this embodiment, the starting and ending positions of the four AGVs are known as follows: Figure 4 As shown in the figure, they all perform load tasks, transporting goods from the starting point to the end point, and use the Eppstein algorithm to solve the first k shortest paths from the starting point to the end point of each AGV, generating a candidate path set C of 4 AGVs. Its time complexity is 0(a+blog b+k), where a=80 is the number of edges in the topological map, b=63 is the number of nodes, and k=50 is the number of shortest paths to be found. Compared with other methods, it has better performance and practicality;

[0085] S5: The chromosome structure of the genetic algorithm is introduced into the antlion optimization algorithm to construct an indefinite-length spatiotemporal composite multi-chromosome encoding suitable for multi-AGV path planning, which is used to represent the positions of ants and antlions, where each position represents a multi-AGV path planning scheme; the multi-AGV path planning scheme here refers to a combination of candidate paths randomly selected by each AGV from its own candidate path C.

[0086] In this embodiment, step S5 also includes setting the maximum number of iterations of the ant lion optimization algorithm and the population size of ants and ant lions, randomly generating the initial positions of the ants and ant lions, the initial positions of the ants and ant lions are composed of the path node sequence of each AGV and the waiting time sequence of the path nodes of each AGV, and the initial positions of each ant and each ant lion correspond one to one;

[0087] In this embodiment, the maximum number of iterations of the antlion optimization algorithm is set to 100, and the populations of ants and antlions are each set to 1000.

[0088] In this embodiment, the one-to-one correspondence between ants and ant lions mentioned in step S5 means that when the population is initialized, each ant and the corresponding ant lion start from the same position, that is, the positions overlap, and the initial positions of the ants and ant lions are composed of the sequence of each AGV path node and the waiting time sequence of each AGV path node, wherein the path plan selects the shortest paths of different AGVs from the candidate path set C for random combination, and the waiting time of the node is randomly set in the range greater than or equal to 0;

[0089] Introducing the chromosome structure of genetic algorithm, a composite multi-chromosome encoding with indefinite length and space-time suitable for multi-AGV path planning is designed. Figure 5 As shown, it is used to represent the positions of ants and ant lions, where each position represents a multi-AGV path planning scheme;

[0090] In this embodiment, the indefinite length spatiotemporal composite multi-chromosome encoding mentioned in step S5 means that the position of each ant and ant lion in the population is composed of multiple chromosomes, the number of chromosomes is determined by the number of AGVs, the nth chromosome represents the path of the nth AGV, each chromosome is composed of a series of path nodes and node waiting times, the path nodes and the waiting time at the nodes together constitute a gene of the chromosome, and the length of the chromosome depends on the number of path nodes from the starting point to the end point;

[0091] S6: Extract input features from the multi-AGV path planning scheme and generate a time window sequence for each path using the time window prediction model;

[0092] In this embodiment, the method for extracting input features in step S6 is to split the path of each AGV into sub-path segments, and obtain the starting point number of the sub-path segment. , sub-path segment end point number , time to leave the starting point of the sub-path segment , whose initial value is the waiting time of the node where the total path starting point is located. For the subsequent sub-path segments, the time of leaving the sub-path segment starting point is the time of leaving the end point of the previous sub-path segment. Whether the path starting point in the sub-path segment is the total path starting point, if yes, then =1 ,on the contrary =0 , Is the end point of the path in the sub-path segment the end point of the total path? If yes, E=1 ,on the contrary =0 , the action of the AGV at the end of the path in the sub-path segment, straight =0 , turn =1 , turn around =2 ,wait =3 , subpath segment length , load operation F=1 , no-load operation F=0 The waiting time of the node at the end of the subpath is ;

[0093] S7: discretize the time window sequence and the path node sequence to generate a path spatiotemporal sequence;

[0094] In this embodiment, the path spatiotemporal sequence mentioned in step S7 is generated as follows:

[0095] (1);

[0096] (2);

[0097] in, is the path node sequence of the nth AGV, is the number of the mth path node, is the time window sequence of the nth AGV, The nth AGV reaches the path node time, The nth AGV leaves the path node time;

[0098] node With Node The discretization operation between is as follows:

[0099] Compute Node With Node Time interval and nodes With Node The number of space-time path points that need to be inserted :

[0100] (3);

[0101] (4);

[0102] in, For Node With Node The time interval, in seconds, For Node The x-axis coordinate of For Node The y-axis coordinate of is the discretized time step in seconds;

[0103] Compute Node and The k1th point inserted between Time coordinates and the x-axis coordinate , y-axis coordinate :

[0104] (5);

[0105] (6);

[0106] (7);

[0107] in, ;

[0108] node The discretization operation at is as follows:

[0109] Compute Node Duration of stay , and nodes The number of space-time path points after discretization ;

[0110] (8);

[0111] (9);

[0112] Compute Node The coordinates of the k2th space-time path point after discretization , represents the time coordinate of the k2th point, represents the x-axis coordinate of the k2th point, Represents the y-axis coordinate of the k2th point:

[0113] (10);

[0114] (11);

[0115] (12);

[0116] in, ;

[0117] S8: Calculate the path space-time distance based on the discretized path space-time sequence;

[0118] In this embodiment, the path temporal and spatial distance mentioned in step S8 is calculated as follows:

[0119] (13);

[0120] in, Represents any two space-time path points and The space-time distance between

[0121] Calculate the space-time distance between any two paths and the total space-time distance of all paths :

[0122] (14);

[0123] (15);

[0124] in, Represents the spatiotemporal path sequence of the i-th AGV, which is composed of the spatiotemporal path points after the corresponding path node sequence is discretized. N represents the spatiotemporal path sequence The number of space-time path points in , represent The mth space-time path point in represent The mth space-time path point in The shortest distance between any space-time path points in ;

[0125] S9: Calculate the fitness values ​​of the ant lions and sort them, and select the ant lion with the smallest fitness value as the elite ant lion;

[0126] S10: The ants walk randomly under the influence of the corresponding ant lion and the elite ant lion, calculate the fitness value of the ants, and update the ant population;

[0127] In this embodiment, the calculation function of the fitness value mentioned in steps S9 and S10 is:

[0128] (16);

[0129] (17);

[0130] (18);

[0131] (19);

[0132] in, represents the location of the i-th antlion, represents the position of the i-th ant, Represents the conflict penalty weight. Since avoiding conflicts is the main goal, a relatively high value needs to be set to ensure that avoiding conflicts is given priority in path planning. Represents the number of path conflicts; Represents the weight of the total completion time, which is mainly used to minimize the total path length, so we can choose a Smaller but still significant values, Represents the end point of the total path, Represents the starting point of the total path, Represents the total completion time; Represents the weight of the total path time-space distance. In order to encourage the increase of the total path time-space distance, this weight should be set to a negative value. represents the total space-time distance of all paths, Indicates whether there is a conflict between paths, 1 means there is a conflict, 0 means there is no conflict. Indicates that the i-th vehicle arrives at the node time, Indicates that the i-th vehicle leaves the node time, Indicates that the i-th vehicle occupies the node time window, Indicates that the i-th vehicle occupies a connected node With Node The time window of the path, Indicates that the i-th car and the j-th car are at the node There is a node conflict at Indicates that the i-th car and the j-th car are at the connecting node With Node There is a conflict between the paths. Indicates that the i-th car and the j-th car are at the connecting node With Node There is a pursuit conflict in the path;

[0133] In this embodiment, the specific steps of the ant randomly walking under the influence of the corresponding ant lion and the elite ant lion in step S10 are as follows:

[0134] S10-1: Generate a random number a in the range of 0 to 1. If a>0.4, the ant moves to the location of the elite ant lion. Otherwise, the ant moves to the location of the corresponding ant lion. The movement is performed through a crossover operation, such as Figure 6 As shown in the figure, the positions of ants and antlions are composed of multiple chromosomes. The crossover operation can only be performed between corresponding chromosomes. In order to ensure the continuity of the path, the crossover point must be selected at the common path node of the two chromosomes. When there are multiple common path nodes between the corresponding chromosomes, one of them is randomly selected as the crossover node. The corresponding chromosome refers to the chromosome in the ant and antlion that represents the same AGV path. Two new positions can be obtained through the crossover operation.

[0135] S10-2: Randomly mutate the waiting time of each path node, such as randomly increasing, decreasing or clearing the waiting time, so as to expand the search range in the time dimension;

[0136] S10-3: If the fitness value of the antlion is greater than the fitness value of the corresponding ant, the ant is captured, the position of the antlion is updated to the position of the ant, and the antlion population is updated;

[0137] Calculate the fitness values ​​of the two new positions, and select the position with the smaller fitness as the new position of the ant for update.

[0138] S11: If the maximum number of iterations is reached, the elite ant lion is output, and each AGV is controlled to perform tasks according to the multi-AGV path planning scheme represented by the location of the elite ant lion. The spatiotemporal three-dimensional path diagram of the multi-AGV path planning scheme of the elite ant lion is shown in Figure 7 As shown, otherwise, jump to step S5.

[0139] In this embodiment, a multi-AGV path planning system for implementing the above method is also provided.

[0140] A multi-AGV path planning system includes a memory and a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement any of the methods described above.

[0141] In this embodiment, a storage medium storing a computer program capable of implementing the above method is also provided.

[0142] A computer-readable storage medium stores a computer program, which implements any of the above methods when executed.

[0143] In this embodiment, a product capable of implementing the above method is also provided.

[0144] A computer product comprises a computer program, which implements any of the above methods when executed by a processor.

[0145] In summary, the present invention designs a travel time estimation model based on limited input features, which can obtain a more accurate time window compared to the traditional time window solving method; secondly, on the basis of generating an accurate time window, a multi-AGV path collaborative planning method is designed with the goal of optimizing the path time-space distance, which increases the time-space distance of conflict-free multi-AGV path planning, and can effectively reduce the possibility of mutual interference between AGVs in a multi-AGV warehousing environment, reduce path congestion, ensure smoother communication and coordination between AGVs, and at the same time optimize system energy consumption and improve work efficiency.

[0146] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-AGV path planning method, characterized by: The steps include: S1: Build a warehouse topology map based on the storage environment; S2: Obtain the travel data of each AGV within a period T, split the travel data into sub-path segments and extract the feature information of each sub-path segment; S3: designing a time window prediction model based on a neural network, using the feature information to train the time window prediction model, and updating the time window prediction model parameters; S4: Solve the first K shortest paths from the starting point to the end point of each AGV and generate a candidate path set C for n AGVs; S5: Introduce the chromosome structure of genetic algorithm into the antlion optimization algorithm, and construct an indefinite length spatiotemporal composite multi-chromosome encoding suitable for multi-AGV path planning, which is used to represent the positions of ants and antlions, where each position represents a multi-AGV path planning solution; S6: extracting input features from the multi-AGV path planning scheme, and generating a time window sequence for each path using the time window prediction model; S7: discretize the time window sequence and the path node sequence to generate a path spatiotemporal sequence; The path spatiotemporal sequence in step S7 is generated as follows: （1）； （2）； in, is the path node sequence of the nth AGV, is the number of the mth path node, is the time window sequence of the nth AGV, The nth AGV reaches the path node time, The nth AGV leaves the path node time; node With Node The discretization operation between is as follows: Compute Node With Node Time interval and nodes With Node The number of space-time path points that need to be inserted : （3）； （4）； in, For Node With Node The time interval, in seconds, For Node The x-axis coordinate of For Node The y-axis coordinate of is the discretized time step in seconds; Compute Node and The k1th point inserted between Time coordinates and the x-axis coordinate , y-axis coordinate : （5）； （6）； （7）； in, ; node The discretization operation at is as follows: Compute Node The duration of stay , and nodes The number of space-time path points after discretization ; （8）； （9）； Compute Node The coordinates of the k2th space-time path point after discretization , represents the time coordinate of the k2th point, represents the x-axis coordinate of the k2th point, Represents the y-axis coordinate of the k2th point: （10）； （11）； （12）； in, ; S8: Calculate the path space-time distance based on the discretized path space-time sequence; The path space-time distance in step S8 is calculated as follows: （13）； in, Represents any two space-time path points and The space-time distance between Calculate the space-time distance between any two paths and the total space-time distance of all paths : （14）； （15）； in, Represents the spatiotemporal path sequence of the i-th AGV, which is composed of all the spatiotemporal path points after the discretization of the corresponding path node sequence. represents the spatiotemporal path sequence of the jth AGV, which is composed of all the spatiotemporal path points of the jth AGV path node sequence after discretization. N represents the spatiotemporal path sequence The number of space-time path points in , represent The mth space-time path point in represent The mth space-time path point in The shortest distance between any space-time path points in ; S9: Calculate the fitness values ​​of the ant lions and sort them, and select the ant lion with the smallest fitness value as the elite ant lion; S10: The ants walk randomly under the influence of the corresponding ant lion and the elite ant lion, calculate the fitness value of the ants, and update the ant population; The calculation function of the fitness value in steps S9 and S10 is: （16）； （17）； （18）； （19）； in, represents the fitness value of the antlion, represents the location of the i-th antlion, represents the fitness value of the ant, represents the position of the i-th ant, represents the conflict penalty weight, represents the weight of the total completion time, Represents the end point of the total path, Represents the starting point of the total path, Represents the total completion time; Represents the weight of the total path time-space distance. In order to encourage the increase of the total path time-space distance, this weight should be set to a negative value. represents the total space-time distance of all paths, Indicates whether there is a conflict between paths, 1 means there is a conflict, 0 means there is no conflict. Indicates that the i-th AGV arrives at the node time, Indicates that the i-th AGV leaves the node time, Indicates that the i-th AGV occupies the node time window, Indicates that the i-th AGV occupies a connection node With Node The time window of the path, when When is 1, Indicates that the i-th AGV and the j-th AGV are at the node There is a node conflict at Indicates that the i-th AGV and the j-th AGV are at the connection node With Node There is a conflict between the paths. Indicates that the i-th AGV and the j-th AGV are at the connection node With Node There is a pursuit conflict on the path. When it is 0, Indicates that the i-th AGV and the j-th AGV are at the node There is no node conflict. Indicates that the i-th AGV and the j-th AGV are at the connection node With Node There is no conflict in the paths. Indicates that the i-th AGV and the j-th AGV are at the connection node With Node There is no pursuit conflict on the path; Among them, the specific steps of the ant randomly walking under the influence of the corresponding ant lion and the elite ant lion in step S10 are as follows: S10-1: Generate a random number a in the range of 0 to 1. If a>0.4, the ant moves to the position of the elite ant lion, otherwise, the ant moves to the position of the corresponding ant lion. The movement is performed through the crossover operation. The positions of the ant and the ant lion are composed of multiple chromosomes. The crossover operation can only be performed between corresponding chromosomes. In order to ensure the continuity of the path, the intersection point must be selected at the common path node of the two chromosomes. When there are multiple common path nodes between the corresponding chromosomes, one of them is randomly selected as the crossover node. The corresponding chromosome refers to the chromosome in the ant and the ant lion that represents the same AGV path. Two new positions can be obtained through the crossover operation; S10-2: Randomly mutate the waiting time of each path node; S10-3: Calculate the fitness values ​​of the two new positions. If the fitness value of the ant lion is greater than the fitness value of the corresponding ant, capture the ant, update the ant lion position to the position of the ant, and update the ant lion population; S11: If the maximum number of iterations is reached, the elite ant lion is output, and each AGV is controlled to perform the task according to the multi-AGV path planning scheme represented by the location of the elite ant lion. Otherwise, jump to step S5.

2. The multi-AGV path planning method according to claim 1, characterized in that: In step S1, the potential driving path of the AGV is simplified to a straight line in the warehouse topology map, and the path intersections, parking spaces, and charging points are defined as nodes, and the nodes are numbered in sequence.

3. The multi-AGV path planning method according to claim 1, characterized in that: In step S2, all AGV historical travel data within period T are obtained and split into sub-path segments according to the constructed warehouse topology map. Each sub-path segment records in detail the starting point, end point, departure time, node waiting time, path length, action type, and arrival and departure time of the end point for training the time window prediction model.

4. The multi-AGV path planning method according to claim 1, characterized in that: In step S3, the feature information is first processed by One-Hot encoding and normalization, and then the processed feature information is input into the time window prediction model. The time window prediction model uses a three-layer hidden layer structure to extract and compress feature information layer by layer. The first hidden layer contains 200 neurons, and ReLU is used as the activation function to perform weighted summation and nonlinear transformation on the input feature information. The extracted first feature is the basic and direct relationship. The second hidden layer contains 40 neurons, and the ReLU activation function is also used to compress and filter the first feature, remove redundant information, retain key features, and improve the generalization ability of the model. The third hidden layer includes 10 neurons, and the third hidden layer is divided into two parts, each containing 5 neurons, and customized feature extraction is performed for two tasks respectively. One part is used for predicting the arrival time, and the other part is used for predicting the departure time to ensure that each task can obtain core information. The third hidden layer still uses the ReLU activation function to avoid gradient disappearance and maintain the nonlinear learning ability of the model. The output layer uses a linear activation function to generate continuous time values ​​and output the arrival time of the next node respectively. and departure time , used for AGV path planning and time scheduling optimization.

5. The multi-AGV path planning method according to claim 1, characterized in that: Step S5 also includes setting the maximum number of iterations of the ant lion optimization algorithm, as well as the population size of ants and ant lions, randomly generating the initial positions of ants and ant lions, the initial positions of ants and ant lions are composed of the path node sequence of each AGV and the waiting time sequence of the path nodes of each AGV, and the initial positions of each ant and each ant lion correspond one to one. The path plan selects the shortest paths of different AGVs from the candidate path set C for random combination, and the waiting time of the node is randomly set in the range greater than or equal to 0.

6. The multi-AGV path planning method according to claim 1, characterized in that: The method for extracting input features in step S6 is to split the path of each AGV into sub-path segments and obtain the starting point number of the sub-path segment. , sub-path segment end point number , time to leave the starting point of the sub-path segment , whose initial value is the waiting time of the node where the total path starting point is located. For the subsequent sub-path segments, the time of leaving the starting point of the sub-path segment is the time of leaving the end point of the previous sub-path segment; Is the starting point of the path in the sub-path segment the starting point of the total path? =1 ,on the contrary =0 ; Whether the end point of the path in the sub-path segment is the end point of the total path, if yes, E=1 ,on the contrary =0 ; The action of the AGV at the end of the path in the sub-path segment, straight ahead =0 , turn =1 , turn around =2 ,wait =3 ; Subpath segment length ; Load operation F=1 , no-load operation F=0 ; The waiting time value of the node where the subpath end point is .

7. The multi-AGV path planning method according to claim 1, characterized in that: In step S5, the indefinite length spatiotemporal composite multi-chromosome encoding is that the position of each ant and ant lion in the population is composed of multiple chromosomes, the number of chromosomes is determined by the number of AGVs, the nth chromosome represents the path of the nth AGV, and each chromosome is composed of a series of path nodes and node waiting times. The path nodes and the waiting time at this node together constitute a gene of the chromosome, and the length of the chromosome depends on the number of path nodes from the starting point to the end point.

8. A multi-AGV path planning system, comprising a memory and a processor, and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the method described in any one of claims 1 to 7 is implemented.

10. A computer product, comprising a computer program, characterized in that: When the computer program is executed by a processor, the method described in any one of claims 1 to 7 is implemented.

Images