CN117151596B

CN117151596B - Logistics management method, system and storage medium for storage AGVs (automatic guided vehicle) through Internet of things

Info

Publication number: CN117151596B
Application number: CN202311432728.7A
Authority: CN
Inventors: 王西; 罗春晖; 王成玉
Original assignee: Lingxian Weilai Technology Group Co ltd
Current assignee: Lingxian Weilai Technology Group Co ltd
Priority date: 2023-11-01
Filing date: 2023-11-01
Publication date: 2023-12-29
Anticipated expiration: 2043-11-01
Also published as: CN117151596A

Abstract

The invention discloses an internet of things storage AGV logistics management method, a system and a storage medium; the start point and the end point of the current task are determined. To select the best path, the quality of the different paths is calculated using the objective function F1 and one best path is selected. The quality of the path is composed of a path length L and energy consumption E; the objective function F1 is:the method comprises the steps of carrying out a first treatment on the surface of the w1 and w2 are weights of path length and energy consumption, respectively; for adjusting their relative importance in the objective function; the support vector machine of the invention can predict potential problems in advance through characteristic engineering and correction mechanisms. The system can adjust the path to avoid these problems, thereby reducing the occurrence of faults and errors. And extracting the characteristics of the neighbor solution by using a support vector machine by using characteristic engineering, and correcting. And calculating a correction factor alpha according to the error degree of the previous step, and adjusting the precipitation function F2. This ensures the adaptability of the search process.

Description

Logistics management method, system and storage medium for storage AGVs (automatic guided vehicle) through Internet of things

Technical Field

The invention relates to the technical field of internet of things warehouse logistics management, in particular to an internet of things warehouse AGV logistics management method, an internet of things warehouse AGV logistics management system and a storage medium.

Background

The intelligent warehouse logistics system driven by the Internet of things is an emerging field of the logistics industry, represents advanced technical application in the warehouse and logistics fields, and aims to realize intelligent, automatic and efficient operation. The technology is based on the internet of things technology, and can acquire, transmit, analyze and apply a large amount of data in real time so as to comprehensively optimize warehouse management and logistics operation, thereby improving the overall efficiency and accuracy.

The internet of things technology is composed of key technologies such as sensors, radio Frequency Identification (RFID), wireless communication and the like, equipment, goods and transportation tools inside and outside a warehouse are connected together, and real-time acquisition, transmission and processing of information are achieved. The RFID technology is used for identifying and tracking goods, so that the efficiency and accuracy of goods management are improved, and the goods state in the warehouse becomes visible in real time.

In an intelligent warehouse logistics system, an Automatic Guided Vehicle (AGV) is a key component. An AGV is an unmanned vehicle that is capable of autonomous travel and transport of materials. In intelligent warehouse logistics, AGVs can automatically plan paths, load cargoes and transport the cargoes to specified positions according to logistics requirements, and automatic unloading of the cargoes is achieved. In addition, intelligent logistics robot can carry, classify and vanning goods automatically. In the technical system, one of the core challenges is to realize path planning of the AGV, ensure that the AGV can efficiently execute various tasks, and further realize smooth operation of the warehouse logistics system. Some prior art techniques have thus enabled intelligent path planning for AGVs, including:

(1) Storage supervisory systems (CN 201710417235.4) based on the internet of things: the system adopts three feedback of preset position relations to display and guide the logistics transportation vehicle to reselect the optimal path in real time. However, this conventional technique is limited in that it depends on a preset positional relationship, and thus, when faced with a change in environment or demand, the relationship may no longer be accurate or applicable. In addition, the preset position relationship cannot fully consider the influence of real-time traffic, congestion and other dynamic factors on the path selection, so that the path is not optimized.

(2) Optimization system of automated wisdom logistics scheduling path (cn202310485289. X): the system utilizes an ant colony algorithm to automatically plan a navigation path. The ant colony algorithm is a heuristic algorithm whose performance is highly dependent on the choice of heuristic rules and parameters and therefore not applicable to all environments or situations. In addition, the ant colony algorithm may be in a local optimal solution in the complex and high-dimensional problems, especially under the condition of a large storage environment, the global optimal solution may not be found, and even the local optimal solution may be in the ant colony algorithm, so that a plurality of AGV trolleys cannot perform effective path planning.

To sum up, the drawbacks of the prior art can be summarized as:

(1) The real-time adaptability is not enough: the prior art has insufficient adaptability in the face of real-time changes in environment or demand. This results in a path planning system that has difficulty coping with new situations in time and cannot effectively adjust path selection to meet changing demands.

(2) The flexibility of path planning is not enough: the path planning method in the prior art lacks enough flexibility, and is difficult to flexibly cope with dynamically-changing path selection requirements. This limits the ability of the path planning system to make it inflexible in the face of changing environments and requirements.

Therefore, the method, the system and the storage medium for logistic management of the storage AGVs through the Internet of things are provided.

Disclosure of Invention

In view of the above, the embodiment of the invention hopes to provide a logistic management method, system and storage medium for the storage AGVs of the Internet of things, so as to solve or alleviate the technical problems existing in the prior art, namely, the real-time adaptability and the path planning flexibility are insufficient, and at least provide a beneficial choice for the problems;

the technical scheme of the embodiment of the invention is realized as follows:

first aspect

Logistics management method for storage AGVs through Internet of things

Summary of the technology

The logistic management method for the storage AGVs of the Internet of things aims at improving the path planning efficiency, adaptability and accuracy of the Automatic Guided Vehicles (AGVs) by comprehensively applying a flood filling algorithm and a Support Vector Machine (SVM) algorithm.

(II) overview of the method

The logistic management method of the storage AGVs of the Internet of things mainly comprises the following steps:

s1, executing a flood filling algorithm

In this step, the warehouse area is considered (abstracted) as a flood boundary. And at each time step, performing path planning for the current task of the AGV according to the position information and the transportation task provided by the RFID tag of the goods. This includes the sub-steps of:

s101, defining a target

The start point and the end point of the current task are determined. To select the best path, the quality of the different paths is calculated using the objective function F1 and one best path is selected. The quality of the path is composed of a path length L and energy consumption E;

the objective function F1 is:；

w1 and w2 are weights of path length and energy consumption, respectively; for adjusting their relative importance in the objective function;

the path length L is the distance between adjacent points on the path;；

representing the distance in the x-direction and the y-direction between adjacent points on the path, which is the sum of Euclidean distances between adjacent points on the path; namely: / >；

E represents energy consumption, which is energy consumption between adjacent points on the path; the energy model of the AGV can be calculated; wherein:；

wherein the method comprises the steps ofRepresenting energy consumption between adjacent points on the path;

the path P of the AGV is:；

wherein the method comprises the steps ofFor the start of the task, < >>The task end point is the task end point;

in S101, the alternative implementation is implemented as a result of selecting the path P with the smallest F1 value, i.e. F1 (P) is the smallest.

S102, initializing

The initial path strategy I1 is generated as an initial solution by Dijkstra algorithm. Meanwhile, a flood filling algorithm is performed to determine an initial water level L1 and a precipitation rate R. These parameters affect the convergence and speed of the algorithm during the search process. In the step S102, the Dijkstra algorithm includes:

s1021, executing Dijkstra algorithm, and calculating the shortest path length from the starting point to all other nodes to obtain a path length array D;

s1022, for each two nodes v, determining a path P formed by connecting the two nodes v through an objective function F1;

s1023, selecting a path strategy I1 as a path corresponding to a node v with the minimum F1 (P), namely:；

argmin is a parameter value of the function, and refers to the minimum value calculated by F1 (P);

In the S102, the initial water level L1 is a starting water level of a flood filling algorithm, which determines a size of a search space; the initial water level L1 is set as a path length:；

i.e. the path length of the initial solution. This helps to ensure that the search space starts from a reasonable starting point.

In S102, the precipitation rate R is a regulating parameter, controlling the rate of water level drop. A larger R value will cause the water level to drop rapidly, while a smaller R value will cause the water level to drop slower; more specifically, smaller values of R may make the algorithm more gradual, but may require more iterations to converge. A larger R value may speed up the search process but may result in missing a potentially better path during the water level drop.

S103, executing a precipitation function F2 and performing internal circulation

The precipitation function F2 iterates the initial water level L1 toThe water level L2 is performed and a plurality of neighbor solutions R1 are calculated in each cycle. The best solution R2 is selected from these neighbor solutions and the final path strategy I2 is extracted after the search is completed. The precipitation function F2 includes:；

l2 represents the current water level, and t represents time;

the basic idea of the precipitation function F2 is that the water level L2 gradually decreases over time, approaching or reaching an optimal path solution. The precipitation rate R may be set according to the nature of the problem and the computational resources. A larger R value will result in a faster water level drop, faster search speed, but may miss a more optimal path; smaller values of R will result in slower water level drops, slower search speeds, but more likely to find a better path.

In the step S103, the step of inner-circulating includes:

s1031, evaluating an objective function value of the current path strategy I1 through an objective function F1:；

s1032, generating a neighbor solution R1 of the current path policy I1:；

n (I1) is a generating function for generating a neighbor solution of the path policy I1;

s1033, selecting a best neighbor solution R2 from the neighbor solutions:；

r2 is the best neighbor solution, which will become the next path policy; r1 is a neighbor solution generated in S1032, and is obtained according to various operations such as a mobile node, an added node, or a deleted node.

F1 (R1) is the value of the objective function F1 on the neighbor solution R1; that is, F1 (R1) is used to evaluate the quality of the neighbor solution R1.

This formula shows that of all the neighbor solutions R1, the solution R2 having the smallest objective function value F1 (R1) is selected as the best neighbor solution. This means that a path strategy is selected which has a smaller value under the objective function F1, thus finding a better path.

S1034, comparing the objective function value of the best neighbor solution with the current solution, if the objective function value is better, accepting the neighbor solution R1, otherwise, rejecting:

s10341, calculating an objective function value of the best neighbor solution R2:；

s10342, calculating an objective function value of the current solution I1: ；

S10343, execution:

if it isThe objective function value of the neighbor solution R2 is smaller, i.e. the path policy R2 is better than I1, the following operations are performed: accepting neighbor solutions: i2 =i1=r2; otherwise, if->The objective function value representing the neighbor solution R2 is not as good as I1, i.e. the path policy R2 is not as good as I1, no operation is performed, keeping the current solution unchanged. After this logic and constraint is formulated, it is expressed as:

；

s1035, repeating S1031-S1034. Until the termination condition is met (e.g., a certain run time is reached or the water level drops to a certain extent).

S2, executing a support vector machine algorithm

In this step, a Support Vector Machine (SVM) algorithm is introduced to improve path planning. The method specifically comprises the following steps:

s201, extracting features and outputting a corrected optimal solution R3

For each neighbor solution R1 generated in S103, the relevant features are extracted in the current time step, and then output is performed using the SVM model to correct the optimal solution R3. The modified solution R3 is used as an evaluation threshold T to measure the quality of the initial path strategy I1 or the final path strategy I2.

S2011, feature extraction: for the generated neighbor solution R1, feature extraction is performed, and a feature extraction function E (R1) maps the neighbor solution R1 to a feature vector FR1: ；

S2012, performing output correction on the feature vector FR1 by using the SVM model, to obtain a corrected best solution:；

s2013, the measure of the evaluation threshold T:

and determining whether to accept the correction according to the relation between the corrected optimal solution R3 and the original solution R1. If R3 and R1 differ by more than a threshold T, the correction is accepted, otherwise it is not accepted. After this logic and constraint is formulated, it is expressed as:

；

DD represents the Euclidean function, and the distance difference degree between R3 and R1 is calculated:；

r3 and R1 are two solutions or vectors, ||R3-R1|| represents the Euclidean distance between them. This distance measure is used to measure the similarity or difference between two solutions, with smaller specific values indicating that the two solutions are more similar and larger values indicating that they are less similar.

S202, calculating a correction factor alpha and correcting a precipitation function F2

In the precipitation function F2 of S103, the error degree between the best neighbor solution R2 generated in the previous time step and the corrected best solution R3 is calculated. Then, a correction factor α is calculated and the precipitation function F2 is corrected using α in the next time step.

S2021, calculating an error degree vector between the best neighbor solution R2 and the corrected best solution R3: ；

ME (R2, R3) is a function of calculating the degree of error between R2 and R3;

s2022, map the error degree to a correction factor α between 0 and 1 using Sigmoid function:

；

e is the base of natural logarithm, k is the adjustment parameter controlling the slope of Sigmoid function;

s2023, correcting the precipitation function F2: by increasing or decreasing the precipitation rate R, the convergence speed and course of the search are controlled:；

f2' represents a modified precipitation function; AF2 is a function for adjusting the precipitation function according to a.

S3, executing path strategy

According to the metrics in S201, the controller of the AGV executes a final path strategy I2 to drive the AGV to move.

(III) technical characteristics

The method comprehensively utilizes the flood filling algorithm and the support vector machine algorithm, and improves the adaptability and accuracy of path planning. The application of the Internet of things technology, including RFID tags, provides real-time and accurate data for path planning. Through the use of the SVM model, the path planning can be adjusted according to the real-time situation to provide more optimal path selection. The introduction of the correction factor alpha enables the algorithm to carry out self-adaptive adjustment according to the past performance, and improves the efficiency of path planning.

Second aspect

Thing networking storage AGV commodity circulation management system

The system is an integrated software system and aims to realize automatic execution of the logistic management method of the storage AGVs of the Internet of things. The system comprises the following main components:

(1) A processor: the core part of the system is responsible for executing various tasks. The processor has high computational power and can effectively execute program instructions to realize path planning, data processing and decision making.

(2) Register: and a register coupled to the processor for storing program instructions. The instructions comprise an algorithm and logic of the logistic management method of the storage AGVs of the Internet of things. The program instructions define how the system performs path planning and administration logistics tasks.

(3) Program instructions: the program instructions are code segments stored in the register and executed by the processor to perform the internet of things warehouse AGV logistics management method. These instructions include path planning algorithms, data processing logic, and decision making processes.

Third aspect of the invention

Storage medium

The storage medium contains program instructions for executing the internet of things storage AGV logistics management method. These instructions define the operating logic and algorithms of the overall logistics management system. They cover key aspects of path planning, data processing, decision making, and system control.

Compared with the prior art, the invention has the beneficial effects that:

(1) Intelligent path planning: according to the invention, through a flood filling algorithm, the system regards the storage area as a flood boundary, so that the self-adaption of path planning is realized. The system can calculate and adjust the path of the AGV in real time according to the position information and the transportation tasks provided by the RFID tags of different cargoes, and adapt to environmental changes and changes of task demands.

(2) Adaptive accommodation correction and self-correction mechanisms: the support vector machine of the invention can predict potential problems in advance through characteristic engineering and correction mechanisms. The system can adjust the path to avoid these problems, thereby reducing the occurrence of faults and errors. And extracting the characteristics of the neighbor solution by using a support vector machine by using characteristic engineering, and correcting. The support vector machine model predicts the quality of the path through learning and tuning. The precipitation function F2 is used to control the water level and the search process in the flood filling algorithm. The change in water level triggers the generation of a new path strategy. And calculating a correction factor alpha according to the error degree of the previous step, and adjusting the precipitation function F2. This ensures the adaptability of the search process.

(3) High efficiency: the flood fill algorithm of the present invention allows the system to consider multiple possible paths simultaneously in path planning to select the best path. This increases the efficiency of path planning, ensuring that the AGV can reach the target site in the shortest time.

(4) Self-learning ability: the support vector machine model of the present invention can adapt to new situations and data through continuous training and adjustment. This gives the system the potential to self-learn, which can continuously improve the quality and accuracy of path planning.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are required in the embodiments or the technical descriptions will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic flow chart of the method of the present invention;

FIG. 2 is a schematic diagram of the internal circulation flow of step S103 in the method of the present invention;

FIG. 3 is a flow chart of the SVM model training method of the present invention;

FIG. 4 is a program diagram (first part) of the Dijkstra algorithm of step S102 in the method of the present invention to generate an initial path strategy I1 as an initial solution;

FIG. 5 is a program diagram (second part) of the Dijkstra algorithm of step S102 of the present invention for generating an initial path strategy I1 as an initial solution;

FIG. 6 is a schematic program diagram (first part) of an SVM model according to a sixth embodiment of the present invention;

FIG. 7 is a schematic program diagram (second part) of an SVM model according to the sixth embodiment of the present invention;

fig. 8 is a control program diagram (first portion) of a tenth embodiment of the present invention;

fig. 9 is a control routine diagram (second section) of a tenth embodiment of the present invention;

Detailed Description

In order that the above objects, features and advantages of the invention will be readily understood, a more particular description of the invention will be rendered by reference to the appended drawings. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. This invention may be embodied in many other forms than described herein and similarly modified by those skilled in the art without departing from the spirit of the invention, whereby the invention is not limited to the specific embodiments disclosed below;

it should be noted that, in the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different manner from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.

It will be further appreciated by those of skill in the art that the various example elements and algorithm steps described in connection with the embodiments disclosed herein may be embodied in electronic hardware, in computer software, or in a combination of the two, and that the various example elements and steps have been described generally in terms of function in the foregoing description to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.

It is noted that the steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. The software modules may be disposed in Random Access Memory (RAM), memory, read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

Embodiment one: referring to fig. 1, the embodiment discloses a logistic management method for storage AGVs in internet of things, which includes RFID tags and information of each cargo in a storage internet of things system, and for a controller of each AGV, the following steps are executed:

s1, executing a flood filling algorithm (Great Deluge Algorithm, GDA): this is a path planning algorithm that treats the storage area as a flood boundary, similar to flood diffusion. Based on the location information and the transport tasks provided by the RFID tags of the load, the system will begin planning the path of the AGV. Considering the warehouse area as a flood boundary;

under the current time step, performing path planning on the current task of the AGV based on the position information and the transportation task provided by the RFID tag of the goods; comprising the following steps:

s101, defining a target: determining a current task starting point and a task ending point; calculating the quality of different paths through an objective function F1 and alternatively implementing;

specifically, at this step, the system determines the start and end points of the current task. These target points are determined by the cargo location and the transportation task provided by the RFID tag.

S102, initializing: executing Dijkstra algorithm to generate an initial path strategy I1 as an initial solution; executing a flood filling algorithm to determine an initial water level L1 and a precipitation rate R; these parameters will affect the convergence and speed of the search process;

Specifically, the system uses Dijkstra algorithm to generate an initial path strategy I1 as an initial solution. At the same time, the system executes a flood fill algorithm, determining the initial water level L1 and the precipitation rate R. L1 represents the initial height of the water level and R represents the speed of flood diffusion. These parameters will affect the efficiency and speed of the search process.

S103, executing a precipitation function F2 and performing internal circulation: the precipitation function F2 drives the initial water level L1 to be changed into the execution water level L2, a plurality of neighbor solutions R1 are calculated in each cycle, an optimal neighbor solution R2 is generated, and a final path strategy I2 is extracted after searching is stopped;

specifically, in this step, a precipitation function F2 is used to drive the water level decrease, from L1 to L2. In each cycle, the system calculates a plurality of neighbor solutions R1 and selects one of the best neighbor solutions R2. After stopping the search, the final path policy I2 will be extracted.

S2, executing a support vector machine algorithm:

s201, in the alternation of S103 (the initial water level L1 is driven by the precipitation function F2 to be the execution water level L2), extracting corresponding features of each generated neighbor solution R1, executing output correction optimal solution R3 by using a support vector machine model, and taking the output correction optimal solution as an evaluation threshold T of candidate solutions; the evaluation threshold T metric executes an initial path strategy I1 or a final path strategy I2;

Specifically, in the inner loop of S103, for each generated neighbor solution R1, the system extracts a corresponding feature. These features are extracted from the attributes of solution R1, such as path length, energy consumption, etc. These features will then be used to train and evaluate the support vector machine model.

S202, in the precipitation function F2 of S103, comparing the value of the best neighbor solution R2 generated by the precipitation function F2 in the previous time step with the error degree between the corrected best solution R3, and executing mapping operation to output a correction factor alpha; in the next time step, correcting the precipitation function F2 by using a correction factor alpha;

specifically, in the precipitation function F2, the system compares the best neighbor solution R2 generated in the previous time step with the corrected best solution R3, and calculates the error degree between them. The system then calculates a correction factor alpha. This correction factor alpha will be used for the next time step for adjusting the precipitation function F2, ensuring the suitability of the search process.

S3, according to the operation of the measurement in S201, the controller of the AGV executes the final path strategy I2. This will guide the AGV to move and perform tasks according to the revised path plan.

In the embodiment, the flood filling algorithm in the S1 plays a key role in the logistic management method of the storage AGVs of the Internet of things. The primary purpose of this algorithm is to implement intelligent path planning to enable the AGV to efficiently and accurately perform transport tasks. Flood filling algorithms can be seen as analogizing warehouse areas to geographical boundaries of flood spread, which helps this embodiment divide the space within the warehouse into different areas, similar to the area division on a geographical map.

Specifically, the flood fill algorithm uses the cargo location information provided by the RFID tags on the AGVs. These RFID tags act as a locating tool enabling the system to know the location of the cargo in real time, thereby planning the path. Once the current location of the load and the transport mission are known, the flood fill algorithm begins planning the path of the AGV. It determines where the AGV should start, which areas to pass, and how to reach the destination.

Further, in the path planning process, the flood filling algorithm calculates the quality of different paths by using the objective function F1. The objective function F1 is an evaluation criterion that can evaluate the path based on various factors, such as path length, time, energy consumption, etc.

Further, one of the key advantages of the flood fill algorithm is that it achieves adaptivity and instantaneity when combined with the support vector machine algorithm of S2. The support vector machine model predicts the quality of the path through learning and tuning. The precipitation function F2 is used to control the water level and the search process in the flood filling algorithm. The change in water level triggers the generation of a new path strategy. And calculating a correction factor alpha according to the error degree of the previous step, and adjusting the precipitation function F2. This ensures the adaptability of the search process.

In this embodiment, a Support Vector Machine (SVM) algorithm in S2 is used to correct the generated path policy I2. During actual transportation, various dynamic factors may affect the path quality, and the SVM model can analyze these factors and make targeted correction suggestions to improve the path quality. In S2, for each generated neighbor solution R1, the SVM algorithm extracts relevant features from it by feature extraction. These characteristics may include path length, energy consumption, target location, etc., which are key information used to evaluate path quality.

Specifically, the SVM model performs the output correction optimal solution R3 according to the extracted features. The modified best solution is an improved path strategy that takes into account more factors to make the path more practical. The evaluation threshold T in S2 is a metric for comparing the initial path strategy I1 with the modified best solution R3. By evaluating the measure of the threshold T, the system can decide whether to employ a modified path strategy, which is a path selection mechanism.

Further, support vector machine algorithms increase the adaptivity and intelligence of path planning. The method can correct the path according to the real-time information so as to meet the changing requirements and environmental conditions. This allows for more intelligence and flexibility in the logistics management system.

Embodiment two: the embodiment further discloses a specific implementation mode of S101 of the logistic management method of the storage AGVs of the Internet of things:

in S101, the quality of the path is measured and optimized by the objective function F1. The objective function F1 is intended to evaluate the quality of the path in order to select the best path, taking into account two key factors: path length (L) and energy consumption (E). The objective function F1 is:；

w1 and w2 in the objective function F1 are weights of path length and energy consumption, which are interval values of [0,1], for adjusting the relative importance of these two factors in the objective function. This allows the system to adjust the objectives of the path planning according to specific requirements and constraints. For example, if energy costs are more important, the weight of w2 may be increased to reduce energy consumption. If both are considered to be equally important, w1=w2=0.5. The specific selection is determined by the personnel in practice.

The path length L is the distance between adjacent points on the path:；

in the present embodiment, the objective function F1 is composed of the path length L and the energy consumption E. Path length refers to the accumulation of distances between adjacent points on a path, measured using euclidean distances. Each segment on the path can be calculated by euclidean distance and then accumulated; in particular, the method comprises the steps of,representing the distance in the x-direction and the y-direction between adjacent points on the path, which is the sum of Euclidean distances between adjacent points on the path; namely:the method comprises the steps of carrying out a first treatment on the surface of the Where i is the index.

Energy consumption (E) is the accumulation of energy consumption between adjacent points on the path. Different AGVs have different energy consumption calculation standards and are calculated according to parameters such as battery capacity, battery charging efficiency, load and the like; AGV energy consumption under the different stroke amounts can also be provided according to the manufacturer of AGVs:；

in this embodiment, the path P of the AGV is:；

wherein the method comprises the steps ofFor the start of a task->Is a task end point;

in S101 of the present embodiment, the selection of the path P with the smallest F1 value is performed as a result of selection, that is, F1 (P) is the smallest.

Specifically, the objective of S101 is to select a path P, so that the value of the objective function F1 (P) is minimized. I.e. selecting from the alternative paths the path whose combination of path length and energy consumption is closest to the optimal state. This ensures that the quality of the path is optimized taking into account the path length and the energy consumption.

Further, the objective function F1 in S101 calculates the quality of different paths by adjusting their weights based on the path length and the energy consumption. Finally, a path P with the minimum objective function F1 value is selected to implement path planning of the storage AGV of the Internet of things, so that the best balance between path quality and energy consumption is ensured. The method is a core principle of path planning, and is beneficial to improving the efficiency and resource utilization of warehouse logistics management.

Embodiment III: the embodiment further discloses a specific implementation mode of the S102 of the logistic management method of the storage AGVs of the Internet of things.

In S102, the Dijkstra algorithm includes:

s1021, a Dijkstra algorithm is performed, which gradually calculates the shortest path length from the start point to all other nodes by continuously selecting the node closest to the start point. This generates a path length array D, where Di represents the shortest path length from the start point to the ith node.

S1022, for each pair of nodes v, a path P connecting the two nodes v is determined by the objective function F1. This step ensures that the path length and energy consumption are taken into account to select the optimal path.

In S1023, in S102, this embodiment needs to select a path policy I1 to minimize the objective function F1 (P). Specifically, in this embodiment, in all node pairs, a path corresponding to a node v with the smallest F1 (P) is selected as an initial path policy I1: ；

argmin is a parameter value of a function, and refers to the minimum value calculated by F1 (P); this ensures that the path P selected by the present embodiment is optimal under the objective function F1, as this "minimum" refers to the goal of achieving a minimum in path quality. Representing the lowest energy consumption and the lowest time consumption (because of the shortest path).

In the present embodiment, it is noted that, in S102, the initial water level L1 is the initial water level of the flood filling algorithm, which determines the size of the search space; the initial water level L1 is set as the path length:；

Specifically, in S102, the present embodiment mentions that the initial water level L1 is the initial water level of the flood filling algorithm for limiting the size of the search space.

In this embodiment, it is noted that the Dijkstra algorithm mentioned above in this embodiment is an algorithm for calculating the shortest path from one starting point to all other nodes. In warehouse AGV logistics, the Dijkstra algorithm is used to calculate the shortest path length from the start of the AGV to all other locations in the warehouse.

In this embodiment, it should be noted that the value of L1 is set to the path length of the initial solution, i.e., l1=l (I1), because the path length is generally used to measure the quality of the path, taking the path length of the initial solution as the initial water level helps to ensure that the search space starts from a reasonable starting point.

In this embodiment, a specific implementation of the Dijkstra algorithm includes the following steps:

p1, initializing:

p100, creating an array D, wherein the size of the array D is the number of nodes, and the array D is used for storing the current shortest path length from the starting point to each node.

P101, initializationThe path length from the origin to itself is 0, namely:；

p102, initializing the path length of all other nodes to infinity, indicating unknowns, i.eFor all->。

P2, iteration:

p200, selects an unlabeled node u whose value of D [ u ] should be currently minimal, i.e., D [ u ] = min { D [ i ] }, where i is the unlabeled node and D [ i ] is minimal.

P201 marks node u as accessed.

P202, for each neighbor node v of u, calculate the path length D [ v ] from the origin to v through u]. If it isWhere w (u, v) represents the weight of edge (u, v), update dv]Is that。

P3, repeat P2 until all nodes are marked as accessed or the target node is marked as accessed.

P4, results: array D contains the shortest path length from the start point to each node, where D [ i ] represents the shortest path length from the start point to node i.

Further, referring to fig. 4 to 5, the present embodiment further provides a c++ execution procedure of the existing Dijkstra algorithm, and the principle includes:

(1) Initializing a graph and parameters:

the number of nodes (N) and the start point (start) of the graph are determined.

The graph is represented using a adjacency matrix (graph), where graph i j represents the distance from node i to node j. If the two nodes are not reachable, their distance is set to Infinity (INF).

A number of array distances is created for storing the shortest path length from the start point to each node. Initially, the distances of all nodes are initialized to infinity, except the distance of the starting node is initialized to 0.

(2) The Dijkstra algorithm is implemented:

p1, first create a protected array to mark whether the node has been accessed. Initially, all nodes are not accessed.

P2, the distance from the start point is set to 0, and the distance from the start point to itself is set to 0.

And P3, starting to iterate through all nodes, and selecting an unviewed node u closest to the starting point each time.

P4, marking the node u as accessed.

P5, updating the distance between all the nodes v which are not accessed, and comparing the sum of the current distance from the starting point to the node v and the distance from the starting point to the node u to the node v. If the former is smaller than the latter, the distance of node v is updated.

And P6, continuing iteration, selecting the next non-accessed node closest to the starting point, and repeating the steps (P1-P5).

(3) Outputting the shortest path length:

when the Dijkstra algorithm is completed, the array distance contains the shortest path length from the start point to all other nodes. All that is next required is to output these shortest path lengths, showing the shortest distance from the origin to each node.

The core idea of the Dijkstra algorithm provided in this embodiment is to gradually calculate the shortest path length from the start point to all other nodes by continuously selecting the node closest to the start point. It ensures that among the nodes selected in each step, the node that is the shortest from the start point is selected, and the shortest path is found.

In this embodiment, in order to avoid misunderstanding, it should be noted that, although the Dijkstra algorithm may be directly used as an instructive algorithm for path planning in conventional path planning, in this embodiment, one of the purposes of generating the initial path policy I1 by the Dijkstra algorithm is to satisfy the operation condition of the flood filling algorithm. Flood fill algorithms require a starting water level (initial water level L1) and a precipitation rate (R) to limit the search space and control the speed of the search. In the context of this embodiment, the value of the initial water level L1 is determined by the path length of the initial solution I1. By generating the initial path strategy I1 using the Dijkstra algorithm, the present embodiment can ensure that the flood fill algorithm starts from a reasonable starting point, which helps to optimize the search process and eventually find the best path strategy I2. Thus, the output of the Dijkstra algorithm is part of the flood fill algorithm required to define some parameters of the search space.

Further, after the Dijkstra algorithm is selected, an advantageous starting point can be provided for the flood filling algorithm, which is helpful for optimizing the searching process. The Dijkstra algorithm is used to calculate the shortest path length from one starting point to all other nodes in the graph. It provides an accurate shortest path distance from the start point to each node, so it can find a path strategy I1, where the path length L (I1) is the actual shortest path length. The flood filling algorithm is a heuristic search method whose objective is to find an approximate optimal solution in a given time step, uses the objective function F1 to calculate the quality of the different paths, and selects the paths according to its search strategy. During the searching process, different path strategies I2 can be generated under the driving of the precipitation function F2. I2 is the result of adjusting the initial path strategy I1 by moving the water line in the search space.

Therefore, it can be understood that the Dijkstra algorithm can provide a better initial solution for the flood filling algorithm, and the lower limit of the solution result of S1 is ensured. Even if the flood filling algorithm under the current regulation limit cannot meet the threshold evaluation step as in the seventh embodiment, the embodiment can ensure that the initial path strategy I1 with a better initial solution is used for execution, and the AGV trolley can be guaranteed to normally run as a bottom-covering technical means.

In this embodiment, it is noted that in S102, the precipitation rate R is a value of [0,1] in the interval, and the water level falling speed is controlled. A larger R value will cause the water level to drop rapidly, while a smaller R value will cause the water level to drop slower; more specifically, smaller values of R may make the algorithm more gradual, but may require more iterations to converge. A larger R value may speed up the search process but may result in missing a potentially better path during the water level drop. The present embodiment thus illustratively provides several options:

(1) R=0.2: the method is suitable for path planning in complex environments. A small R value will result in a slower water level drop, facilitating finer searches in complex, high-dimensional problems. This is useful for problems that require a large number of dynamic factors, variations and uncertainties to be considered. A small R value increases the number of search iterations, but may help find a better path.

(2) R=0.5: the method is suitable for general path planning problems. The medium R value balances between speed and accuracy. The method is suitable for general path planning problems, can find a better path strategy in reasonable time, and does not need excessive search iteration.

(3) R=0.8: is suitable for large-scale warehouse logistics environment. A large R value causes the water level to drop rapidly, thereby accelerating the search process. This is very useful for large scale warehouse environments and multi-AGV operation situations because it can reduce search time and improve efficiency. However, a larger R value may result in missing a potentially more optimal path, and thus requires careful selection.

(4) R value increasing from 0.1 to 0.9: the search speed needs to be dynamically adjusted. Incrementing the R value is an adaptive strategy that dynamically adjusts the R value based on the water level conditions during the search. Initially, the R value is small in order to perform a preliminary search. As the search proceeds, the R value gradually increases, accelerating the search process. This applies to the problem of the need to adapt to different search phases in the search process.

Embodiment four: referring to fig. 2, the embodiment further discloses a specific embodiment of S103 of the logistic management method of the storage AGV in the internet of things, and a specific embodiment of the internal circulation operation of S103:

in this embodiment, the precipitation function F2 includes:；

l2 represents the current water level, and t represents time;

specifically, the basic idea of the precipitation function F2 is that the water level L2 gradually decreases over time, thereby approaching or reaching an optimal path solution. The precipitation rate R may be set according to the nature of the problem and the computational resources. A larger R value will result in a faster water level drop, faster search speed, but may miss a more optimal path; smaller values of R will result in slower water level drops, slower search speeds, but more likely to find a better path.

Further, by continuously adjusting the water level L2, the objective of the precipitation function F2 is to find the optimal path solution. This solution is typically determined by comparing the objective function values of the different path strategies. The final path strategy I2 is the output of the precipitation function F2, i.e. the best path solution found.

Illustratively, let the initial condition be l2=l0, where L0 represents the length of the initial path. Meanwhile, the present embodiment sets r=1, which means that the unit time drops by one unit. At time t=0, the value of precipitation function F2 is:；

this means that in the initial state, the value of the water level L2 is equal to the initial path length L0.

Over time, the present embodiment observes a change in precipitation function F2. At t=1, the value of the precipitation function is:；

this means that the water level L2 drops by one unit in a unit time. This reflects the progress of the search process, and the path length L2 is reduced.

At t=2, the value of the precipitation function is:；

at t=3, the value of the precipitation function is:；

continuing in this way, it can be seen that with increasing time, the water level L2 will drop at a rate of r=1. This simulates the path length variation during the search to find a better path solution in a limited time.

In this embodiment, referring to fig. 2, in S103, the steps of inner loop include:

specifically, the present embodiment evaluates the objective function value of the current path policy I1. The objective function F1 takes into account the path length L (I1) and the energy consumption E (I1) and applies weights w1 and w2 to adjust their relative importance in the objective function. This objective function value represents the quality of the current path strategy I1.

S1032, generating a neighbor solution R1 of the current path policy I1:；

specifically, in this step, the present embodiment creates a neighbor solution R1 of the current path policy I1 by generating a function N (I1). The neighbor solution may be obtained by performing different operations in the path policy, such as mobile nodes, add nodes, or delete nodes. These neighbor solutions represent viable variations of the path policy.

S1033, selecting a best neighbor solution R2 from the neighbor solutions:；

specifically, at this step, selecting a best neighbor solution R2 from among the neighbor solutions R1 is a critical decision. The best neighbor solution is selected by comparing the objective function values F1 (R1) of the respective neighbor solutions. The present embodiment selects the neighbor solution R2 with the smallest objective function value as the path policy for the next step. This means that the present embodiment finds a path strategy with smaller values under the objective function F1, which is expected to represent a better path.

Further, R2 is the best neighbor solution, which will become the next path policy; r1 is a neighbor solution generated in S1032, and is obtained according to different operations, such as a mobile node, an added node, or a deleted node (refer to embodiment five for details).

Further, F1 (R1) is the value of the objective function F1 on the neighbor solution R1; that is, F1 (R1) is used to evaluate the quality of the neighbor solution R1.

It will be appreciated that this formula shows that of all neighbor solutions R1, the solution R2 with the smallest objective function value F1 (R1) is selected as the best neighbor solution. This means that the present embodiment selects a path strategy with a smaller value under the objective function F1, thereby finding a better path.

S1034, comparing the objective function value of the best neighbor solution with the current solution, if the objective function value is better, accepting the neighbor solution R1, otherwise rejecting; that is to say this step involves the objective function value F2 for the best neighbor solution R2 and the objective function value of the current solution I1A comparison is made. If->The present embodiment accepts the neighbor solution and updates the current solution to i2=i1=r2 if the objective function value of the neighbor solution R2 is smaller, i.e., the path policy R2 is better than I1. If F2 is not less than +. >The objective function value representing the neighbor solution R2 is not as good as I1, no operation is performed, keeping the current solution unchanged.This operation can be subdivided into three small steps:

s10341, calculating an objective function value of the best neighbor solution R2:；/>

s10342, calculating an objective function value of the current solution I1:；

s10343, execution: if F2< The objective function value of the neighbor solution R2 is smaller, i.e. the path policy R2 is better than I1, the following operations are performed:

accepting neighbor solutions: i2 =i1=r2;

；

otherwise, ifThe objective function value representing the neighbor solution R2 is not as good as I1, i.e. the path policy R2 is not as good as I1, no operation is performed, keeping the current solution unchanged.

S1035, repeating S1031-S1034. Until the termination condition is met (e.g., a certain run time is reached or the water level drops to a certain extent). This process allows the search algorithm to continually try to find a better path strategy in the search space to approximate the best solution.

Specifically, the core principle of the approach to the optimal solution is that the logic of the whole step S103 is iterated continuously, starting from the current path strategy, generating a neighbor solution, selecting the optimal neighbor solution, accepting or rejecting the solution, and repeating the process continuously until the termination condition is met. This helps the search algorithm find a better path solution in a given time.

Further, regarding the specific embodiment of S1035: the present embodiment gives the following two specific schemes:

(1) Scheme one: the termination condition is based on run time: in this case, the present embodiment may set a fixed time limit to ensure that the algorithm completes the inner loop in a given time. This is useful for controlling the execution time of the algorithm to avoid infinite execution.

The termination conditions are expressed as follows: if the run time exceeds the threshold Tmax, the inner loop is terminated, i.e.:；

t is the current run time and Tmax is the preset maximum run time.

(2) Scheme II: the termination condition is based on the water level falling: another method is to set the termination condition based on the water level drop. The present embodiment may define a threshold for the water level to drop, and the inner loop ends once the water level drops below this threshold. The drop in water level is part L2 of the precipitation function F2 (L2, t).

The termination condition may be expressed as follows: if the water level drop L2 reaches or falls below the threshold Lmin, the inner loop is terminated:；

l2 is the current water level and Lmin is a preset minimum water level threshold.

Illustratively, assume that the present embodiment expects the search algorithm to find a path in no more than 10 seconds, and expects the water level to drop to a level of no less than 5% to end the inner loop. The embodiment can be provided with: tmax=10 seconds and lmin=0.05 (indicating a 5% drop in water level).

If after any iteration of the inner loop, 10 seconds is reached or the water level drops below 5%, the algorithm will end the inner loop, return to the current optimal path strategy or perform other operations as needed.

It will be appreciated that these termination conditions may be adjusted to balance search time and result quality according to specific application requirements.

Fifth embodiment: the present embodiment further discloses a specific implementation manner of step S1032 in the fourth embodiment, which includes:

the operation of generating the function N (I1) in S1032 includes a mobile node operation and an add node operation. The purpose of these operations is to generate neighbor solutions for the path policy to find a better path during the search.

In this embodiment, regarding the mobile node operation:

s10321, deleting node from path strategy I1：/>；

S10322 in the new positionAdd node->：/>；

The logic is as follows: selecting a nodeIt is a node in the path policy I1. Selecting a new locationIs->Is positioned adjacent to one of the adjacent positions. A new path strategy R1 is created by adding +.>Move to +.>And the obtained product. The goal of this operation is to change the location of the node in hopes of finding a shorter oneIs provided. In warehouse AGV logistics, the mobile node operation may correspond to a real scenario, such as when the AGV encounters an obstacle in path planning or needs to readjust the path. By moving the node, the AGV may attempt to bypass the obstacle or select a more optimal path.

Demonstrative: when the present embodiment performs a mobile node operation, the present embodiment first selects one nodeIt is a node in the path policy I1. Then, the present embodiment selects a new position +.>Is node->Is positioned adjacent to one of the adjacent positions. By adding nodes->Move to new position +.>The present embodiment creates a new path policy R1. Let the present embodiment derive and demonstrate this process with specific examples. Assume that this embodiment has a path policy I1 as follows:；

wherein (x 1, y 1) is the start point and (x 4, y 4) is the end point. The present embodiment contemplates attempting the mobile node (x 2, y 2) to a new location to generate a new path policy R1.

First, select a nodeThen select a new position +.>Such as the (x 3, y 3) position. Now the present embodiment can apply mobile node operation, node +.>Move to new position +.>：；

In this example, the present embodiment generates a new path policy R1 by moving the positions of the nodes (x 2, y 2) to (x 3, y 3). The purpose of this operation is to explore whether a shorter path can connect the start and end points. The process is as follows:；

wherein: r1 represents a new path policy. I1 Is the initial path policy. { ' delete node>。/>The } represents adding nodes +.>. By constantly applying these operations, the present embodiment can change the structure of the path policy during the search process in hopes of finding a shorter path. The intelligent path planning method is beneficial to realizing intelligent path planning in the logistics management of the storage AGVs of the Internet of things so as to adapt to the continuously changing requirements and environmental conditions.

In the present embodiment, regarding the add node operation:

s10321, adding nodes to path strategy I1：/>。

The logic is as follows: selecting a nodeThe node is not in path policy I1. A position (insert position) is selected between certain nodes on the path. A new path strategy R1 is created by inserting +_at (insert position)>And the obtained product. The goal of this operation is to consider whether a new node can join the path to optimize the path length. />

Therefore, further, in the logistics management of the storage AGV of the Internet of things, the operation of the added node corresponds to the occurrence of a new task or the situation of changing the path demand. By adding nodes, the embodiment can find a better path on the existing path to meet new requirements.

Demonstrative: assuming that an AGV is performing a task, the path from start point A to end point B is: a→c→d→e→b. In path planning, the AGV detects that an obstacle is present at node D and cannot traverse. To bypass this obstacle, the AGV performs an add node operation. The method comprises the following specific steps:

(1) Path policy I1:；

(2) Selecting a node to be addedAnd determining the insertion position: suppose a new node is to be added +>And inserts it between C and D, forming a new path policy R1: />；

By connecting nodesInterposed between C and D, a new path strategy is generated to avoid obstacles. In the add node operation, the present embodiment may represent the new path policy R1 using the following formula: />；

Where R1 is the new path policy. I1 Is the original path strategy, e.g., a→c→d→e→b. {The node to be added is +>。

By this operation, the AGV successfully bypasses the obstacle and selects a more optimal path, namely A→C→n_j→D→E→B. This added node operation allows the AGV to adapt to the dynamic environment and select the best path to improve efficiency and avoid obstacles.

Further, AGVs are typically equipped with various sensors, such as lidar, cameras, ultrasonic sensors, etc., for monitoring their surroundings in real time. These sensors may continually detect obstructions, other AGVs, or environmental changes. The AGV may perform mobile node operations to re-plan the path if an obstacle appears or the environment changes. This includes selecting a new target point, typically around an obstacle or taking a shorter path. Based on the new target point, the AGV rerun the path planning algorithm to generate an updated path strategy. This process involves re-selecting nodes, re-calculating path lengths, and re-evaluating energy consumption. Eventually, the AGV begins to move according to the new path strategy to reach the target position.

Summarizing for this embodiment: the purpose of these operations is to explore different path strategies to find the optimal path. During the search process, by applying these operations, the structure of the path policy can be changed, thereby affecting the length and quality of the path. The final goal is to find the shortest path from the start point to the end point, thereby reducing the transport cost of the AGV and improving efficiency. While these operations are part of the search algorithm for continually adjusting the path policy in the search space in hopes of finding the optimal solution. They can be used to real-time path planning in thing networking warehouse AGV commodity circulation management to adapt to the demand and the environmental condition of dynamic change, ensure that the AGV carries out the task in the most effective mode.

Example six: in executing S2, a trained Support Vector Machine (SVM) model is included; the embodiment further discloses a specific implementation manner of specific construction and training of the Support Vector Machine (SVM) model:

s2001, collecting training data: including a known set of path planning examples G:；

wherein each example element Gi includes a known path policy (path plan) and path quality information associated therewith;

Specifically, each example element Gi is composed of a path policy and path quality information associated therewith. The example set G includes a plurality of such examples so that the SVM model can learn how to predict according to the quality of different path strategies.

S2002, feature engineering: converting the path planning problem P into definition features D of an SVM model, wherein the definition features comprise path length L, energy consumption E and target position O; these features will become inputs to the SVM model:；

feature vector representing ith example element Gi including path length +.>Energy consumption->And target position->The method comprises the steps of carrying out a first treatment on the surface of the These features will be used as inputs to the SVM model.

Specifically, each example element Gi is represented as a feature vectorIncluding path length, energy consumption, and target location.

S2003, tag t preparation: providing each example element Gi with a corresponding tagRepresenting the quality of the path plan; these labels may be a measure about path planning, such as the value of the objective function F1 (Gi); label->Is the output of the SVM model, which needs to be related to the input feature +.>And (5) associating.

S2004, SVM training: taking the definition feature D as input and labelingAs output: the SVM will learn how to input features +. >Associated with the quality of the path planning: />；

The T SVM is a mapping function, meaning a Train SVM (hereinafter also written in the form of "Train SVM"), performs a training process of the SVM, and learns how to map input features to output labels. In particular, it will use definition feature D as input and use a tagAs an output. The SVM will learn how to inputThe ingress characteristics are associated with the quality of the path plan in order to predict the quality of different path strategies in the path plan later. The result of the SVM training will be a trained model that can be used for quality assessment of the unknown path strategy.

In the present embodiment, the build information about the example element Gi includes:

(1) Start and end positions: coordinates of a task start point and a task end point in the example.

(2) Path length: the path length that the AGV actually walks while performing the task.

(3) Energy consumption: the energy consumed when the task is completed can be calculated according to the battery service condition of the AGV.

(4) Execution time: the time required to complete the task.

(5) Encountering an obstacle: whether an obstacle is encountered on the path, and at which locations.

In this embodiment, regarding the meaning of the tag t and its measurement: a specific indicator or score associated with path planning performance. In this context, the objective function F1 (Gi) is used to evaluate the quality of the path planning, i.e. the variant of the disclosure of embodiment two: ；

In this formula: li represents the path length, typically the distance from the start point to the end point. Ei represents energy consumption, calculated based on distance traveled on the path (or provided by the AGV manufacturer). Ti represents execution time, and represents time required for completing a task. w1, w2 and w3 are weights for weighting the relative importance of these factors in the objective function (w 3 is selected in the same way as w1 and w2, i.e. the scheme disclosed in embodiment two).

In this example, the objective function F1 (Gi) combines the length of the path plan, energy consumption, and time together to comprehensively evaluate the quality of the path plan. The specific weights may be adjusted according to the needs of the application to reflect which factors are more important.

In the present embodiment, the label is as followsAnd input features->Associated embodiments: label->And input features->The association between these is established by a training dataset in supervised learning. The training data set includes known path planning example elements Gi, each of which includes a path policy (path plan) and path quality information (a value of an objective function F1 (Gi) or other quality index) associated therewith.

In the present embodiment, regarding the mapping function Train SVM: it is essentially a convex optimization problem with respect to support vector machine model parameters, which can be in the form of:

；

w is a weight vector;is an input feature +.>Vectors transformed by the possible feature map; b is the bias term, xi is the relaxation variable, allowing some samples to violate boundary conditions, C is the regularization parameter, used to trade-off minimizing training error and minimizing model complexity, and i is the index.

The objective of this function is to find a parameter vector w and a bias term b such that the inequality condition described above is met and the value of the objective function is minimized. This will result in a mapping of the input featuresTo the corresponding label->Is provided.

It should be noted that other conventional training algorithms (e.g., SMO, gradient descent, etc.) may further optimize the objective function to find the parameters w and b, and thus construct the SVM model required for the present embodiment.

In this embodiment, a specific framework of an SVM model will be further given, including:

s20001, decision boundary: the SVM attempts to find a hyperplane that separates data points of different categories. In the binary classification problem, this hyperplane classifies data into two categories. Expressed as: ；

Where w is the normal vector, x is the input feature vector, and b is the bias term. This hyperplane separates the different classes of data points, satisfying the following condition:

s200011, positive class sample:；

s200012, negative class sample:；

s20002, interval (Margin): the goal of the SVM is to maximize the classification interval, i.e. to find the hyperplane, and maximize the distance (interval) of the support vectors of the different classes to the hyperplane. The calculation of this interval is expressed by the following formula:；

where w represents the weight. Norm of the weight vector w. Maximizing the spacing helps to improve the robustness of the classifier and can better cope with the classification of unknown data points. By maximizing the spacing, the SVM provides some robustness, which can better cope with classification of unknown data points. C is a penalty parameter;

s20003, using gaussian radial basis functions (RBF kernels): the SVM may use a kernel function to map the nonlinear problem into a high-dimensional feature space, allowing a linearly separable hyperplane to be found in the high-dimensional space. Among these, the gaussian radial basis function (RBF kernel) is a kernel function that can map input features to feature space of infinite dimensions. The RBF core is as follows:；

where K (x, x ') represents the result of the kernel, x and x' are input feature vectors, γ is a parameter of the kernel, Is the square of the euclidean distance between the input features. exp is an exponential function based on a natural constant e.

S20004, support vector: the support vector is the data point in the training data closest to the decision boundary. These data points have a critical impact on the location of the decision boundary as they determine the size and shape of the interval. The support vectors correspond to those samples selected during the training process, which satisfyOr->These two interval boundaries. The importance of the support vectors is that they help define the decision boundaries of the classifier, and that at test time, only the classification decisions of the support vectors are active, while other data points are inactive, thereby improving the computational efficiency of the model.

Thus, in this embodiment, the model training process (see fig. 3) includes:

p1, data preparation: i.e., example set G; providing each example element Gi with a corresponding tagRepresenting the class or metric value to which the example belongs.

P2, model construction: a Gaussian radial basis function (RBF kernel) is used to construct a model, defining super-parameters of the model, including penalty parameter C and parameter gamma of the kernel function.

P3, training a model: training is performed using a training dataset. In training, the SVM attempts to find an optimal hyperplane (S20001 and S20002) to separate examples of different classes in feature space.

P4, optimizing an objective function: the goal of the SVM is to maximize the classification interval while ensuring that all examples are correctly classified. Where the objective function is targeted at the maximum separation.

P5, convergence condition: a convergence condition is set, such as a maximum number of iterations or a small change in classification interval.

P6, prediction: once the model training is completed, the implementation of S201 and S202 may be normally participated. The SVM model of this embodiment should be trained before the AGV cart is formally put into service.

Further, penalty parameter C: the penalty parameter C is an interval value of 0,1 for controlling the complexity and fault tolerance of the model. When the value of C is 0.1-0.4, wider intervals are generated, so that some training examples are allowed to classify errors, but the training examples can be better generalized to test data. When the C value is 0.5 to 0.9, a narrow interval is generated, so as to reduce classification errors.

Preferably, cross-validation may be used to select the optimal value of C in addition to the above-described parameter schemes. By trying different C values and evaluating their performance, the C value that yields the best performance on the test data is selected.

Further, the kernel parameter γ is an interval value of [0,1 ]: the "sharpness" of the kernel function is controlled. Smaller gamma values (0.1-0.4) result in a kernel function with a wider peak, thus smoothing the decision boundary. Larger gamma values (0.5-0.9) result in a kernel function with narrower peaks and more complex.

Similar to C, cross-validation can be used to select the best gamma value.

Further, the convergence condition: the convergence condition typically involves some termination criteria to determine when to stop training. The convergence conditions include:

(1) Up to the maximum number of iterations: a maximum number of training iterations, e.g. 1000, may be set, and when this number is reached, training is stopped.

(2) The classification interval varies little: the change in classification interval may be monitored and if the change is less than a certain threshold, e.g., 0.05, the model may be considered to have converged and training stopped.

In this embodiment, referring to fig. 6 to 7, the embodiment provides an SVM model program as shown in the drawing, which is programmed in c++ language; in this procedure, the SVM type (here C-SVC, multi-category classification) can be set by setType, setKernel sets kernel functions (here RBF kernels), setC and setGamma are hyper-parameters of the SVM. When using the procedure shown in the figure, two Mat objects are created, one for storing training data and the other for storing labels. Each training sample and corresponding label are represented using a Mat object. Training data and labels are added to the trainingData and labels. A TrainData object is then created using the TrainData:: create function, passing training data and labels to it. The SVM model is trained using the train function. The model will learn how to classify the input data according to the training data. The trained SVM model can be used to predict new data points, such as test data in the example. The outcome of the prediction will help decide whether path modification is needed.

Embodiment seven: the embodiment further discloses specific implementation manners of S201 and S202 of the logistic management method of the storage AGV of the internet of things:

in S201, the process of using the SVM model to output the corrected best solution R3 is based on feature extraction and operation of the SVM model for improving the quality of the path strategy to achieve better path planning. This process can be applied in thing networking storage AGV logistics management to make the route planning of AGV more intelligent and adaptive:

s2011, feature extraction: for the generated neighbor solution R1, feature extraction is performed, and the feature extraction function E (R1) maps the neighbor solution R1 as featuresVector FR1:；

for the generated neighbor solution R1, a feature extraction operation is performed to map the neighbor solution R1 into a feature vector FR1. These feature vectors capture some key information of R1 such as path length, energy consumption, and target location. These features are inputs to the SVM model.

S2012, performing output correction on the feature vector FR1 by using the SVM model, to obtain a corrected optimal solution:；

the feature vector FR1 is used to input a trained SVM model. The task of the SVM model is to learn how to relate the input features to the path quality. Specifically, it attempts to find a function by analysis of the training data, mapping the input features to the output correction value R3. This process may train the SVM model according to a known set of path planning examples G (including path quality information), which is already described in embodiment six and will not be described here again.

S2013, evaluating a measure of the threshold T:

and determining whether to accept the correction according to the relation between the corrected optimal solution R3 and the original solution R1. If R3 and R1 differ by more than a threshold T, the correction is accepted, otherwise it is not accepted.

；

DD represents the Euclidean function, and the distance difference between R3 and R1 is calculated.

The logic is as follows: the degree of difference between the modified optimal solution R3 and the original solution R1 is measured. This is achieved by calculating the euclidean distance between R3 and R1. This distance represents the degree of difference between the two solutions. If the difference exceeds a predefined evaluation threshold T, the correction is accepted, otherwise not accepted. The choice of threshold T depends on the nature and requirements of the problem, which can be used to control whether corrections are applied. If the difference exceeds the threshold T, it is stated that the output of the SVM model is a large improvement in path quality, and therefore a correction can be applied. Otherwise, the original solution R1 is still valid.

This process allows for the introduction of intelligence and adaptivity in path planning. Through feature extraction and SVM models, the system can learn how to improve the path planning strategy to adapt to different situations and requirements. The evaluation threshold T allows control of when corrections are applied, ensuring that unnecessary changes are ignored, thereby improving the robustness of the system. This helps thing networking storage AGV commodity circulation management system to deal with the environment and the task that change constantly better.

In the present embodiment, the euclidean operation function DD is as follows:；

r3 and R1 are two solutions or vectors, ||R3-R1|| represents the Euclidean distance between them. This distance measure is used to measure the similarity or difference between two solutions, with smaller specific values indicating that the two solutions are more similar and larger values indicating that they are less similar. It should be noted that in practical problems, different distance measures or difference functions may also be chosen. If the objective of the problem is to minimize the objective function, then typically DD (R3, R1) should be smaller and better, indicating that the new solution R3 is closer to the original solution R1 and thus accepted as a candidate solution.

In the present embodiment, regarding the feature extraction function E (R1):；

l (R1): the path length representing the path strategy R1 is typically the sum of the euclidean distances between adjacent points on the path. The specific calculation manner can be combined with the content of the second embodiment. This can be expressed by the following formula:；

e (R1): the energy consumption representing the path policy R1 is typically the sum of the energy consumption between adjacent points on the path. The specific calculation manner can be combined with the content of the second embodiment.

O (R1): the target position of the path policy R1, i.e. the end point of the task, is indicated. This may be a coordinate point or other relevant information to assist the system in understanding the objectives of the path. The specific calculation manner can be combined with the content of the second embodiment.

This example feature extraction function maps the key information of the path plan into a feature vector for use by the SVM model. The goal of feature extraction is to provide information for the SVM to understand the relationship between path quality and these features, and thus output a modified optimal solution R3.

In the present embodiment, in S2012The processing of the feature vector FR1 using a Support Vector Machine (SVM) model is shown to output a modified best solution R3. The specific operation form of the SVM comprises the following steps:

first step, feature mapping: the feature vector FR1 is mapped to a high-dimensional feature space, in particular using a kernel function (see for example the gaussian radial basis function of embodiment six).

Secondly, training a model: during the training phase, the SVM learns how to relate feature vectors to path quality labels. This involves finding a decision boundary to distinguish between positive and negative categories (labels with good path quality).

Third step, predicting: in the prediction phase, the SVM predicts the label of the unknown data point using the trained model. And R3 was sought.

Illustratively, the present embodiment is provided with a data set G comprising example elements Gi, each of which comprises a feature vector FRi and a path quality label ti. An example set G is as follows: ；

Illustratively, the quality of a new path plan is now predicted, the eigenvectors of the path plan being. The present embodiment may use an already trained SVM model to make predictions:

p1, feature mapping:will be mapped as feature vector by the same feature extraction function E (R1)>. That is, whatever the solution (referred to herein as the path policy) that is input, a corresponding feature vector will be generated after the feature extraction function E (). This applies to a given feature extraction function that maps the input solution (path strategy) to feature vectors of the same type, regardless of the differences in the input path strategy. For example, if there are two different path policies R1 and R2, which represent different path plans, respectively, the feature extraction function E () will process the two path policies and map them into different feature vectors FR1 and FR2, respectively. In other words, the feature extraction function E () is a mapping rule of a path policy to a feature vector, and is converted into a corresponding feature vector by E () regardless of the inputted path policy. This consistent feature extraction process helps ensure that different path strategies all undergo the same feature transformation for subsequent Support Vector Machine (SVM) analysis. By using the same feature extraction function E (), different path strategies can be mapped to the same type of data for analysis and prediction by the SVM model. This helps to maintain consistency and comparability of the data.

P2, prediction: SVM model willAs inputs, the following operations are performed: />；

In this scenario, the SVM model uses the learned relationship to compute the followingMapped to a path quality label R3. This tag may be used to evaluate the quality of the path plan to determine whether to accept the correction (depending on the evaluation threshold T).

In the present embodiment, the evaluation threshold T of S2013 is used to determine whether to accept the path quality label R3 output by the SVM as the correction optimal solution. Specifically, this threshold is used to determine whether the difference between R3 and the original solution R1 is large enough that R3 needs to be accepted as a correction to the path plan. If the Euclidean distance difference between R3 and R1 (representing a change in path quality) is greater than a threshold T, then the system will accept R3 as the more optimal path strategy. This means that the system considers that R3 provides a better path, more appropriate for the current task or situation. Otherwise, if the difference in euclidean distance between R3 and R1 is less than or equal to the threshold T, then the system will not accept R3, leaving the original solution R1 unmodified. This means that the system considers that the path quality is not significantly improved, or that the improvement is not of sufficient magnitude to properly introduce corrections.

It will thus be appreciated that the effect of the evaluation threshold T is to balance the optimisation and stability of the path planning. A larger T may result in excessive correction, while a smaller T may result in too conservative, missing some potential improvement paths. Therefore, the selection of the threshold T requires a comprehensive consideration of system performance, task requirements and practical situations to ensure that suitable improvements are achieved in path planning. By way of example, the present embodiment provides several schemes as follows:

(1) Fixed static threshold: t=0.05; the method is suitable for scenes in which the path planning problem is relatively stable and the path quality improvement amplitude is consistent.

(2) Dynamic adaptive threshold: and automatically adjusting T according to the improvement degree of the last iteration. The nature of the applicable path planning problem may change over time or task demand, requiring a flexible scenario. The dynamic threshold may ensure that the appropriate path modification is obtained under different conditions.

(3) Threshold based on task goal: t is determined according to the specific goal of each task. The method is suitable for target diversification of path planning problems, and different tasks require different path quality requirements. This ensures that the path modification is more tailored to the specific task.

(4) Probability threshold: t=0.1 (probability threshold for path improvement); the path quality variation applicable to the path planning problem has a certain randomness, such as the case of interference by environmental factors. Such a threshold may take into account uncertainty to some extent.

It should be noted in this embodiment that R2 is the best neighbor solution generated in S1033, which has a smaller objective function value (typically better path policy quality). The role of R2 is to act as a candidate replacement policy for the current path policy I1.

In the path planning problem, the goal is to find a better path strategy, i.e. a path with smaller objective function values. R2 represents a potentially better path strategy found during the search. However, whether R2 is accepted depends on the result of the comparison with the current solution I1. If the objective function value of R2 is smaller (path quality is better), then R2 has the opportunity to replace the current solution I1, thereby becoming a new current solution for the next search step.

It will be appreciated that the role of R2 is to provide an alternative path strategy (also described in detail in embodiment three) which may be better than the current solution, but ultimately whether or not to accept depends on the comparison and decision step in S1034. If R2 is accepted in the evaluation, it will replace the current solution I1, thereby finding a better path strategy in the search.

Example eight: the embodiment further discloses a specific implementation manner in S202 of the internet of things warehouse AGV logistics management method:

in S202, the step of calculating the correction factor α and correcting the precipitation function F2 includes:

s2021, calculating an error degree vector between the best neighbor solution R2 and the corrected best solution R3:；

ME (R2, R3) is a function of calculating the degree of error between R2 and R3;

s2022, sigmoid function is used to map the error degree to a correction factor α between 0 and 1. The purpose of this step is to determine the degree of correction based on the magnitude of the error metric:；

e is the base of natural logarithm, k is the adjustment parameter controlling the slope of Sigmoid function; a larger k value will result in a more sensitive correction factor a, while a smaller k value will make a more gradual.

S2023, correcting the precipitation function F2: by increasing or decreasing the precipitation rate R, the convergence speed and course of the search are controlled: the corrected precipitation function F2' may control the convergence rate and course of the search. If α is close to 0, F2' will be closer to the original precipitation function F2 and the search process will be slower. If α is close to 1, F2' will lower the water level faster during the search: ；

Further, a specific adjustment is controlled by a function AF2, which adjusts the precipitation function F2 according to the value of α. For example, if α is small, AF2 may decrease precipitation rate R, making the search more gradual. If α is large, AF2 may increase R to speed up the search. The purpose of this step is to balance the search speed and search quality during the search process to ensure that the system can find a sufficiently good path in a limited time.

In this embodiment, ME (R2, R3) is used to calculate the error degree between R2 and R3, and its specific mathematical function form may use euclidean distance to measure the difference between the two solutions. Euclidean distance is a standard method of calculating the distance between vectors, typically expressed in terms of the L2 norm. The Euclidean distance is formulated as follows:；

ME (R2, R3) represents the degree of error between R2 and R3. R2i and R3i represent the i-th element in the two solutions, respectively. Σ represents summing all elements. This distance metric method considers the difference between the two solutions across each element and then squares the sum of squares of these differences to obtain the final distance value. This can be used to measure the difference between R2 and R3, providing a basis for subsequent correction factor calculations.

In this embodiment, AF2 is a function that adjusts the precipitation function according to α, given that this adjustment is achieved by linear interpolation:；

AF2 (F2, α) represents the corrected precipitation function. F2 Is the original precipitation function. F2' is a correction value of the precipitation function calculated based on α. Alpha is the correction factor calculated previously.

This formula represents a linear interpolation of the original precipitation function F2 according to the value of α, such that the corrected precipitation function F2' is intermediate between the original F2 and the fully corrected one. This allows controlling the convergence speed and process of the search, making the search process more intelligent and adaptive.

In this embodiment, it should be noted that the precipitation function F2 can be corrected after the error degree vector is mapped to α by the Sigmoid function, because α can be regarded as a correction factor reflecting the degree of error degree between the best neighbor solution R2 and the corrected best solution R3. The correction is based on the magnitude and direction of the error vector to adjust the precipitation function, so that the searching process is more adaptive.

The error degree vector reflects the deviation of the best neighbor solution R2 from the corrected best solution R3. This bias can be used to quantify the differences in path policy R2 relative to R3, which may be why R2 is inferior to R3. The SVM model can identify and understand the relationship between the characteristics and the quality of different path strategies by learning training data. When the SVM model is used for path quality correction, the SVM model corrects the path strategy according to the characteristic vector and the learned relation.

Thus, further, the causal relationship between the error vector and the precipitation function F2 is as follows: the error degree vector represents the difference of the best neighbor solution R2 relative to the corrected best solution R3, which to some extent reflects the difference of the quality of R2 relative to R3. The SVM model uses the difference and combines the feature vector to correct the path strategy, thereby improving the path quality. The corrected path strategy R3 is closer to the optimal solution, so that the path planning effect is improved.

Example nine: the embodiment further discloses a specific implementation mode of S3 of the logistic management method of the storage AGVs of the Internet of things:

in this embodiment, for an AGV, the final path policy I2 or the initial path policy I1 is a virtual concept, which is an abstract path calculated by a path planning algorithm. These path strategies do not direct the mechanical transport of the AGV to drive, but rather need to be translated into specific control commands to actually guide the AGV to perform tasks; the conversion path strategy is typically implemented in the control system of the AGV for control commands, and this embodiment provides a standardized workflow to achieve this conversion:

first, the AGV is typically equipped with a controller or control system that receives the path strategy as input and executes corresponding control commands.

Second, the path strategy consists of a series of path points, each of which consists of coordinate and direction information. For example, the path policy may be expressed as i2= [ (x 1, y1, θ1), (x 2, y2, θ2), (xn, yn, θn) ], where (xi, yi) is the coordinates and θi is the direction of the robot.

When in use: the controller of the AGV includes a path tracking algorithm for translating the path strategy into control commands for the mechanical transport. This algorithm can calculate what path the robot needs to move along based on the current position and direction, and the path strategy. The controller is responsible for calculating how the robot should adjust its trajectory at the current position and orientation to approach the next waypoint in the path strategy. In path tracking algorithms, the error between the current robot position and the next path point in the path strategy is typically calculated. Based on the error calculations, the controller generates control commands including speed, direction, or rotational speed of the wheel, etc. The generation of these commands involves methods of control theory, such as proportional-integral-derivative (PID) controllers or other control strategies. The robot's execution unit (e.g., motor or wheel) will act according to the generated control command to bring the robot close to the next waypoint in the path strategy.

Illustratively, the present embodiment is provided with a robot or AGV whose position is represented by (x, y) and direction θ.

S301, inputting a path strategy: the path strategy is represented as a target path by a series of path points, each path point comprising a position coordinate (xi, yi) and a desired direction angle θi. The path policy is expressed as:；

s302, control command generation: the path tracking algorithm calculates the current robot position (x, y) and direction theta and the next path point in the path strategyThe error between them to generate the control command.

S3021, position error calculation: calculating the current position [ (]And->) Position error between the next waypoint (x and y)>And->：

；

S3022, direction error calculation: calculating the direction error between the current direction and the target direction, typically expressed using an angular difference, may be considered using a tangent functionTo calculate the angle error: />；

S303, using atan2 to ensure that the angle error is between-pi and pi:；

s304, executing control command generation by a PID controller algorithm: the controller uses the position errors delta x, delta y and the direction errors delta theta, and combines the control theory to generate a control command of the robot:

；

u (t) is the output of the controller at time t, typically denoted as a control command, such as speed or force.

dt represents the sampling time interval, i.e. the time interval during which the controller calculates the control command; if dt is 0.1, the controller calculates the control command every 0.1 seconds.

Kp, ki, kd are the three main parameters of the PID controller algorithm, representing the proportional gain, the integration time and the differentiation time, respectively. e (t) is an error at the current time, and is defined as a difference between the target value and the actual value. In the path tracking, e (t) represents the position error Δx, Δy, and the direction error Δθ, respectively.

S3041, position error calculation:；/>；/>；

s3042, control command generation:

；

where V represents the linear speed of the AGV wheel motor and ω represents the angular speed of the wheel motor.

Examples ten

The embodiment further discloses an internet of things storage AGV logistics management system:

(4) And the wireless receiving and transmitting module is used for: the AGV is arranged on each AGV and is in wireless communication with the processor for interactive transmission of information quality.

Through executing the program instructions, the processor can manage and control the behaviors of the storage AGVs of the Internet of things in real time.

Further, referring to fig. 8 to 9, program instructions stored in the registers in the present embodiment are shown; as shown in the figure, this embodiment only provides the c++ pseudo code form of the program to show its operation logic, and its operation principle includes:

(1) The path planning (Point start) function is responsible for performing path planning, connecting the start and end points in the warehouse by selecting the best path. The Dijkstra algorithm is used to find the best path, and the path between the start point and the end point is calculated, including path length and energy consumption.

(2) A trainSVM (vector < TrainingExample > trainingData) function for training a Support Vector Machine (SVM) model using a given training data set. The SVM learns how to relate path quality to features in order to make path quality predictions in S201.

(3) The pathcorrection control (Point currentPosition, point targetPosition, SVM model) function, which is the core of the control program, selects whether to correct the current path based on the path quality predicted by the SVM. The AGV is then guided to the target position using a path tracking algorithm based on the corrected path or the original path.

The current path policy is calculated by invoking path planning. The quality of the current path is predicted using svmModel. If the predicted path quality does not meet (is below the threshold), path correction is performed using correctPath. Finally, an exectep athtracking function is called to guide the AGV to move along the corrected or original path.

(4) executePathTracking (Path path) function which is responsible for actually guiding the AGV along a given path by controlling the mechanical transport to follow the path to reach the target position. Control commands for the AGV, such as position error Deltax, deltay and direction error Deltaθ, are calculated by a path tracking algorithm and then passed to the mechanical conveyance to achieve accurate path tracking.

All of the above examples merely represent embodiments of the invention which are described in more detail and are not to be construed as limiting the scope of the invention. It will be apparent to those skilled in the art that various modifications and improvements can be made without departing from the spirit of the invention.

Claims

1. The logistics management method for the storage AGVs of the Internet of things comprises RFID tags and information of each cargo in the storage Internet of things system, and is characterized in that: for each AGV controller, the following steps are performed:

s1, executing a flood filling algorithm: considering the warehouse area as a flood boundary; under the current time step, based on the position information provided by the RFID tag of the goods, path planning is executed for the current task of the AGV, and the method comprises the following steps:

s102, initializing: executing Dijkstra algorithm to generate an initial path strategy I1 as an initial solution; executing a flood filling algorithm and determining an initial water level L1 and a precipitation rate R;

s2, executing a support vector machine algorithm:

s201, in the alternation of the S103, for each generated neighbor solution R1, performing output correction optimal solution R3 by using a support vector machine model SVM, and taking the output correction optimal solution as an evaluation threshold T of candidate solutions; the evaluation threshold T metric performs the initial path strategy I1 or the final path strategy I2;

S202, in the precipitation function F2 of the S103, comparing the value of the best neighbor solution R2 generated by the precipitation function F2 in the previous time step with the error degree between the corrected best solution R3, and executing mapping operation to output a correction factor alpha; in the next time step, correcting the precipitation function F2 by using the correction factor α;

and S3, according to the measurement of S201, the controller of the AGV executes a final path strategy I2 to drive the AGV to move.

2. The internet of things storage AGV logistics management method of claim 1, wherein: in S101, the quality of the path is composed of a path length L and an energy consumption E;

the objective function F1 is:；

w1 and w2 are weights of path length and energy consumption, respectively;

the path length L is the distance between adjacent points on the path;

e represents energy consumption, which is energy consumption between adjacent points on the path;

the path P of the AGV is:；

in the step S101, the alternative implementation is as follows: the result of selecting the path P with the smallest F1 value is implemented.

3. The internet of things warehouse AGV logistics management method of claim 2, wherein: in the step S102, the Dijkstra algorithm includes:

s1023, selecting a path strategy I1 as a path corresponding to a node v with the minimum F1 (P):；

in S102, the initial water level L1 is set to a path length:；

in S102, the precipitation rate R is a regulating parameter, controlling the rate of water level drop.

4. The internet of things warehouse AGV logistics management method of claim 2, wherein: in S103, the precipitation function F2 includes:；

l2 represents the current water level, and t represents time;

in the step S103, the step of inner-circulating includes:

s1032, generating a neighbor solution R1 of the current path policy I1:；

s1033, selecting a best neighbor solution R2 from the neighbor solutions:；

f1 (R1) is the value of the objective function F1 on the neighbor solution R1;

S1034, comparing the objective function value of the best neighbor solution with the current solution, comprising:

s10342, calculating an objective function value of the current solution I1:；

s10343, execution:

；

s1035, repeatedly executing S1031-S1034.

5. The internet of things storage AGV logistics management method of claim 4, wherein: in the S1032, the generating function N (I1) is any combination of one or more of a mobile node operation and an add node operation; comprising the following steps:

the mobile node operates:

s10321, deleting node from path strategy I1：/>；

S10322 in the new positionAdd node->：/>；

Or/and the adding node operates:

s10321, adding nodes to path strategy I1：/>。

6. The internet of things storage AGV logistics management method of claim 4, wherein: in the step S2, a support vector machine SVM model is included, and the step of establishing the SVM model includes:

each example element Gi includes a known path policy and path quality information associated therewith;

s2002, feature engineering: converting the path planning problem P into defined features D of the SVM model, wherein the defined features comprise path length L, energy consumption E and target position O: ；

A feature vector representing an ith said example element Gi comprising a path length +.>Energy consumptionAnd target position->；

S2003, tag t preparation: providing each of the example elements Gi with a corresponding tagRepresenting the quality of the path plan;

s2004, SVM training: taking the definition feature D as input and the labelAs output:

；

t SVM is a mapping function responsible for performing the training process of the SVM model, causing the SVM model to learn how to input featuresMapped to the output label.

7. The internet of things warehouse AGV logistics management method of claim 6, wherein: in the S201, the step of outputting the corrected best solution R3 using the SVM model, and using the evaluation threshold T includes:

s2011, feature extraction: for the generated neighbor solution R1, feature extraction is performed, and a feature extraction function E (R1) maps the neighbor solution R1 to a feature vector FR1:；

s2012, using the SVM model to orient the featuresThe quantity FR1 performs output correction to obtain a corrected optimal solution:；

s2013, the measure of the evaluation threshold T:

；

8. The internet of things warehouse AGV logistics management method of claim 6, wherein: in the step S202, the step of calculating the correction factor α and correcting the precipitation function F2 includes:

ME (R2, R3) is a function of calculating the degree of error between R2 and R3;

s2022, using Sigmoid function to perform the mapping on the error degree, and outputting as one of the correction factors α:

；

the correction factor alpha is a value of a section between [0,1 ];

s2023, correcting the precipitation function F2: by increasing or decreasing the precipitation rate R, the convergence rate of the search is controlled:；

9. Thing networking storage AGV commodity circulation management system, its characterized in that: the internet of things storage AGV logistics management system comprises a processor and a register coupled with the processor, wherein a program instruction is stored in the register, and when the program instruction is executed by the processor, the processor is caused to execute the internet of things storage AGV logistics management method according to any one of claims 1-8.

10. A storage medium, characterized by: program instructions for executing the internet of things warehouse AGV logistics management method of any one of claims 1-8 are stored in the storage medium.