Multi-AGV scheduling method

Abstract

The invention relates to the technical field of AGV control, in particular to a multi-AGV scheduling method. The method comprises the following steps: obtaining a real-time logistics instruction of a working area, carrying out logistics scheduling demand analysis and global task planning, and constructing a global execution task atlas; aGV real-time monitoring parameters are obtained, multi-state comprehensive evaluation is carried out, dynamic task demand allocation processing is carried out according to the global execution task atlas, and an AGV dynamic task set is constructed; carrying out AGV operation behavior intention identification one by one according to the AGV dynamic task set, carrying out multi-task cooperative carrying evolution, and constructing a multi-level space-time path network; performing potential AGV collision early warning according to the multi-level space-time path network, and marking local AGV space-time collision nodes; and performing dynamic task priority decision according to the global execution task atlas, performing intelligent conflict planning on local AGV space-time conflict nodes, and constructing a local conflict scheduling strategy. According to the invention, through cooperative control of multiple AGVs, the operation efficiency, the scheduling precision and the safety level of the operation area are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of AGV control, and in particular to a multi-AGV scheduling method. BACKGROUND

[0002] With the continuous development of automation technology, as an efficient and flexible material transportation tool, AGV (Automated Guided Vehicle) has been widely used in various industrial production, warehousing logistics and intelligent manufacturing fields. In particular, in the e-commerce, automobile, electronics, pharmaceutical and other industries, the application of AGV greatly improves the production efficiency, reduces the labor cost, and plays an important role in ensuring the safety of the working environment and reducing human errors. With the popularization of industrial 4.0, Internet of Things (IoT) and artificial intelligence (AI) technologies, the application scenarios and demand for AGV are gradually increasing, especially in intelligent warehouses, intelligent logistics systems and automated production lines, AGV has become an indispensable part.

[0003] Compared with traditional manual handling systems and automated transportation equipment, AGV has flexible path planning, high scalability and stronger adaptability, which enables it to achieve efficient and low-cost transportation tasks in complex environments. However, in large-scale application scenarios where multiple AGVs work together, how to effectively coordinate the operation of multiple AGVs to ensure efficient and smooth cooperation between them and avoid conflicts becomes an important challenge. With the increase in the number of AGVs, traditional scheduling methods gradually show performance bottlenecks, especially in cases where task complexity and environmental dynamics are large, traditional scheduling strategies often have difficulty in achieving efficient resource utilization and task allocation.

[0004] Traditional AGV scheduling methods usually rely on pre-set rules, centralized control systems and experience-based scheduling strategies, although these methods can provide certain performance guarantees in certain specific applications, but in the face of dynamic environments, emergencies and complex tasks, there are often problems such as slow response, low scheduling efficiency, and resource waste. In addition, with the increase in the number of AGVs, a single AGV scheduling strategy is difficult to adapt to the complex interaction and cooperation requirements between AGVs in a multi-AGV system, which can easily cause system overload, scheduling conflicts or transportation delays and other problems.

[0005] In order to solve these problems, it is urgent to develop a multi-AGV intelligent collaborative scheduling method that can adapt to complex and variable environments and has high efficiency, real-time and intelligence. SUMMARY

[0006] The present application is to solve the above technical problems, and proposes a multi-AGV scheduling method to solve at least one of the above technical problems.

[0007] To achieve the above object, the application provides a multi-AGV scheduling method, comprising the following steps: Step S1: obtaining real-time logistics instructions of a work area, performing logistics scheduling demand analysis and global task planning, and constructing a global execution task graph; Step S2: obtaining AGV real-time monitoring parameters, performing multi-state comprehensive evaluation, and performing dynamic task demand distribution processing according to the global execution task graph, and constructing an AGV dynamic task set; Step S3: performing AGV operation behavior intention recognition one by one according to the AGV dynamic task set, and performing multi-task collaborative carrying evolution, and constructing a multi-level space-time path network; Step S4: performing potential AGV collision early warning according to the multi-level space-time path network, and marking local AGV space-time conflict nodes; Step S5: performing dynamic task priority decision according to the global execution task graph, and performing intelligent conflict planning on the local AGV space-time conflict nodes, and constructing a local conflict scheduling strategy; Step S6: performing global path space-time change analysis and multi-AGV intelligent coordination optimization according to the local conflict scheduling strategy, and constructing a global intelligent coordination control model.

[0008] The application realizes comprehensive perception and dynamic control of the entire logistics demand of the operation area by acquiring real-time monitoring parameters of AGVs, performing multi-state comprehensive evaluation, and ensuring that the scheduling system has a global perspective. By constructing a global task execution graph, complex logistics tasks are visualized and modeled, making subsequent task allocation and path planning more systematic and forward-looking. This reduces task repetition and resource waste, improves the scientificity and rationality of task planning, and supports the operation of large-scale AGV systems. According to the current state of AGVs (such as position, power, load, availability), task allocation is performed to improve task matching degree and efficiency. Dynamic task allocation can avoid task interruption or low efficiency caused by changes in AGV state. The construction of "AGV dynamic task set" can achieve load balancing and resource optimization among multiple vehicles, improving the response speed and robustness of the entire system. The identification of AGV behavior intention can achieve more accurate path prediction and behavior coordination, reducing scheduling blind spots. It supports concurrent collaborative transportation of multiple tasks in the same transportation network, improving traffic efficiency and path utilization. The multi-level space-time path network provides an extensible structural basis for subsequent conflict detection and path optimization, facilitating scheduling decisions in complex scenarios. Early detection of path intersections, bottleneck areas, and potential congestion points can prevent system paralysis caused by collisions or blockages. The realization of "early warning" scheduling instead of "reactive" scheduling improves safety and the initiative of the scheduling system. Precise positioning and judgment basis are provided for rapid resolution of local conflicts. Dynamic priority decision can ensure that critical tasks are completed first, improving the overall task response capability of the system. Intelligent conflict planning combined with global task graphs and local conflict characteristics can achieve more flexible and intelligent local optimization scheduling. Local conflict scheduling strategies avoid "one-size-fits-all" scheduling solutions, reduce the impact on the overall path, and achieve local adjustment and global stability. The dynamic linkage mechanism from local adjustment to global update improves the coordination ability and stability of the entire system. The construction of an intelligent coordination control model can continuously iterate and optimize the scheduling strategy to adapt to different operation scenarios and task fluctuations. Ultimately, a multi-AGV intelligent control system with adaptive, self-optimizing, and self-coordinating capabilities is formed, significantly improving the operation efficiency, scheduling accuracy, and safety level of the system. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 A step flowchart of a multi-AGV scheduling method of the application is shown in the figure. Figure 2 A detailed implementation step flowchart of step S1 is shown in the figure. Figure 3 A detailed implementation step flowchart of step S2 is shown in the figure. Figure 4 A detailed implementation step flowchart of step S3 is shown in the figure. DETAILED DESCRIPTION

[0010] It is to be understood that the specific embodiments described herein are merely illustrative of the present application and should not be used to limit the present application in any manner.

[0011] The application example provides a multi-AGV scheduling method. The execution subject of the multi-AGV scheduling method includes but is not limited to the following: mechanical equipment, data processing platform, cloud server node, network upload device, etc. which can be regarded as a general computing node of the application. The data processing platform includes but is not limited to: audio image management system, information management system, cloud data management system, at least one of which.

[0012] Please refer to Figures 1 to 4 The application provides a multi-AGV scheduling method, which includes the following steps: Step S1: obtaining real-time logistics instructions of the work area, performing logistics scheduling demand analysis and global task planning, and constructing a global execution task graph; Step S2: obtaining AGV real-time monitoring parameters, performing multi-state comprehensive evaluation, and performing dynamic task demand distribution processing according to the global execution task graph, and constructing an AGV dynamic task set; Step S3: according to the AGV dynamic task set, performing AGV operation behavior intention recognition one by one, and performing multi-task collaborative carrying evolution, and constructing a multi-level space-time path network; Step S4: according to the multi-level space-time path network, performing potential AGV collision early warning, and marking local AGV space-time conflict nodes; Step S5: according to the global execution task graph, performing dynamic task priority decision, and performing intelligent conflict planning on the local AGV space-time conflict nodes, and constructing a local conflict scheduling strategy; Step S6: according to the local conflict scheduling strategy, performing global path space-time change analysis and multi-AGV intelligent coordination optimization, and constructing a global intelligent coordination control model.

[0013] The application realizes comprehensive perception and dynamic control of the entire logistics demand of the operation area by acquiring real-time monitoring parameters of AGVs, performing multi-state comprehensive evaluation, and ensuring that the scheduling system has a global perspective. By constructing a global task execution graph, complex logistics tasks are visualized and modeled, making subsequent task allocation and path planning more systematic and forward-looking. This reduces task duplication and resource waste, improves the scientificity and rationality of task planning, and supports the operation of large-scale AGV systems. According to the current state of AGVs (such as position, power, load, availability), task allocation is performed to improve task matching degree and efficiency. Dynamic task allocation can avoid task interruption or low efficiency caused by changes in AGV state. The construction of "AGV dynamic task set" can achieve load balancing and resource optimization among multiple vehicles, improving the response speed and robustness of the entire system. The identification of AGV behavior intention can achieve more accurate path prediction and behavior coordination, reducing scheduling blind spots. It supports concurrent collaborative transportation of multiple tasks in the same transportation network, improving traffic efficiency and path utilization. The multi-level space-time path network provides an extensible structural basis for subsequent conflict detection and path optimization, facilitating scheduling decisions in complex scenarios. Early detection of path intersections, bottleneck areas, and potential congestion points can prevent system paralysis caused by collisions or blockages. The realization of "early warning" scheduling instead of "reactive" scheduling improves safety and the initiative of the scheduling system. Precise positioning and judgment basis are provided for rapid resolution of local conflicts. Dynamic priority decision can ensure that critical tasks are completed first, improving the overall task response capability of the system. Intelligent conflict planning combined with global task graphs and local conflict features can achieve more flexible and intelligent local optimization scheduling. Local conflict scheduling strategies avoid "one-size-fits-all" scheduling solutions, reduce the impact on the overall path, and achieve local adjustment and global stability. The dynamic linkage mechanism from local adjustment to global update improves the overall coordination ability and stability of the system. The construction of an intelligent coordination control model can continuously iterate and optimize scheduling strategies to adapt to different operation scenarios and task fluctuations. Ultimately, a multi-AGV intelligent control system with adaptive, self-optimizing, and self-coordinating capabilities is formed, significantly improving the operation efficiency, scheduling accuracy, and safety level of the system.

[0014] In the embodiment of the application, referring to Figure 1 The steps of the multi-AGV scheduling method include: Step S1: Acquire real-time logistics instructions of the operation area, perform logistics scheduling demand analysis and global task planning, and construct a global task execution graph; In this embodiment, determining the source of real-time logistics instructions typically includes dispatching systems, warehouse management systems (WMS), and input from on-site operators. Interfaces need to be built with these systems to enable real-time data transmission. An API interface is set up to allow the dispatching system to send the latest logistics instructions, including cargo handling, loading and unloading tasks, path planning, etc., to the central control system every 5 seconds. Real-time logistics instructions are collected through the interface and pre-processed. The pre-processing process includes removing redundant information, formatting instruction content, and marking the priority and timestamp of the instruction. If the received instruction is "move cargo A from location X to location Y", the system should parse it into an executable task and record the timestamp and priority (such as urgent, ordinary, low priority). The acquired real-time logistics instructions are recorded in the database for subsequent analysis and scheduling use. At the same time, a visualization panel is built to display the current logistics instructions and their status in real time. A real-time monitoring interface is designed to display a list of all pending tasks, including task type, priority, start time, etc., to help operators quickly grasp the job dynamics. According to the acquired real-time logistics instructions, set the standards for logistics scheduling demand analysis. These standards should consider the type of task, priority, resource requirements (such as AGV quantity, load capacity), and time requirements. High-priority tasks should be completed within 10 minutes, while low-priority tasks can be scheduled within 30 minutes to ensure timely response to urgent needs. Each real-time logistics instruction is analyzed in detail to assess its resource requirements and impact on scheduling. The number of AGVs required for each task, time and path must be calculated, taking into account possible conflicts and obstacles. If the task involves picking goods from the shelf and transporting them to the loading and unloading area, the number of AGVs and estimated time for the path must be analyzed, taking into account obstacles and other AGVs on the path. According to the results of logistics scheduling demand analysis, a global task planning model is constructed. This model should integrate the priority of all tasks, resource requirements and time constraints to achieve efficient task scheduling. Use linear programming or integer programming models to set the objective function to minimize overall transportation time and resource usage costs. Solve the model to generate the optimal task scheduling scheme. This process needs to consider the current state of AGVs, the priority of tasks and the availability of paths to ensure that tasks are completed in the shortest time. If multiple tasks request AGVs at the same time, the system automatically selects the most suitable AGV for task allocation based on priority and calculates the estimated completion time. The generated global task scheduling scheme is recorded in the database and displayed through visualization tools to help operators understand the execution order and scheduling of tasks. A global task graph is generated to show the task allocation of each AGV and the execution path, allowing operators to monitor task progress and adjust scheduling strategies in real time.

[0015] Step S2: Obtain AGV real-time monitoring parameters, perform multi-state comprehensive evaluation, and perform dynamic task demand allocation processing according to the global execution task graph to construct an AGV dynamic task set; In this embodiment, each AGV is equipped with various sensors to monitor its status in real-time. These sensors should include position sensors (such as GPS, lidar), load sensors, battery level monitors, and speed sensors. The frequency of data collection from the sensors should be set to once per second to ensure that the dynamic state of the AGV can be monitored in real-time. The real-time monitoring parameters of the AGV are transmitted to the central control system through wireless communication technology (such as Wi-Fi, Zigbee, or LoRa). The system should have efficient data processing capabilities to ensure timely reception and processing of this information. The AGV uploads its position (such as X=10, Y=20), speed (such as 0.5 meters / second), load (such as 300 kilograms), and battery level (such as 80%) to the database during operation. The real-time monitoring parameters collected are stored in the database for subsequent analysis. At the same time, a monitoring panel is constructed to visually display the current state of each AGV, including position, speed, load, and battery level. A dashboard is designed to display the running status and monitoring data of the AGV in real-time, helping operators quickly understand the working condition of the AGV. Determine the indicators for multi-state comprehensive evaluation, including load status (such as the ratio of actual load to rated load), energy consumption threshold, remaining battery capacity, and operating status (such as normal, standby, and busy). The evaluation range of the load status is set to 0% to 100%, and the safe threshold of the battery capacity is set to 20% or less for low battery status. Based on the real-time monitoring parameters, each AGV is comprehensively evaluated. The specific value of each evaluation indicator is calculated, and the overall state score of the AGV is obtained by comprehensive evaluation. This score can be calculated by weighted average method to reflect the importance of each indicator. If AGV1 has a load status of 80%, a battery capacity of 50%, and a normal operating status, the comprehensive evaluation score is: Comprehensive score = 0.4 × 80 + 0.3 × 50 + 0.3 × 100 = 78 (assuming the weights are 0.4, 0.3, and 0.3, respectively). According to the global task graph, analyze the current task demand and the state of the AGV. The task graph should include all tasks to be executed, their priority, resource demand, and expected completion time. Task A has a high priority and an expected completion time of 15 minutes, while task B has a medium priority and an expected completion time of 30 minutes. Set the strategy for dynamic task demand allocation to ensure that the AGV can efficiently execute tasks. The strategy should consider the current state of the AGV, the priority of the task, and the availability of the path. High-priority tasks should be assigned to AGVs with lighter loads and better states. If AGV1 has a high comprehensive score and a load status of 50%, while AGV2 has a load status of 90%, task A should be assigned to AGV1. Based on the evaluation results and the global task graph, dynamically allocate tasks to each AGV and update its task execution status. The system should monitor the execution of tasks in real-time and adjust according to actual conditions.If AGV1 receives task A, the system updates its status to "busy" and automatically resets the status to "normal" after the task is completed.

[0016] Step S3: Identify the behavior intention of each AGV according to the AGV dynamic task set, and perform multi-task collaborative carrying evolution to construct a multi-level space-time path network. In this embodiment, a suitable machine learning or deep learning model is selected to identify the running behavior intention of AGV. Commonly used models include LSTM (Long Short-Term Memory) network or CNN (Convolutional Neural Network) based models, which can process time series data and extract features. The LSTM network is used to train the historical running data of AGV, and the input includes position, speed, load state and other information to identify the behavior intention of AGV, such as "transport goods", "standby", "charge" and so on. The dynamic task set of AGV is collected, including the current task and state information of each AGV. Data preprocessing includes denoising, standardization and feature extraction to enable the model to correctly identify the intention. If the current task of AGV is "transport goods from position A to position B", relevant features such as current position, target position, task priority, etc. are extracted and organized into time series format. The trained model is used to identify the running behavior intention of each AGV. Real-time monitoring parameters are input into the model, and the corresponding intention is output. If the monitoring data of AGV1 shows that it is moving from position A to position B, the model may identify its behavior intention as "transport goods to target". According to the running behavior intention of AGV, a multi-task collaborative carrying model is constructed. This model should consider the interaction between AGVs, path selection and task dependency to achieve coordinated execution of tasks. The model aims to minimize the overall transportation time and path conflict, ensuring the safety and efficiency of AGVs in executing tasks. A multi-AGV collaboration mechanism is introduced into the model, and the best collaborative path is calculated through path planning algorithms such as A* or Dijkstra algorithm. Considering the current intention of AGV and task dependency, dynamic path adjustment is performed. If AGV1 and AGV2 need to move on the same path, the system should calculate the optimal path planning to avoid conflict, possibly by adjusting the departure time of AGV or selecting different paths. The results of collaborative carrying are converted into a multi-level space-time path network. This network should include the path selection of all AGVs, key nodes and their time information to facilitate real-time monitoring and analysis. A path network graph containing multiple levels is constructed, with main levels representing main transportation channels and secondary levels representing backup paths and detour channels. The overall network reflects the collaborative carrying state of AGVs.

[0017] Step S4: According to the multi-level space-time path network, potential AGV collision warning is carried out, and local AGV space-time conflict nodes are marked. In this embodiment, the structure of the multi-level spatiotemporal path network is confirmed, including the paths of all AGVs, key nodes, and time information. This network should be able to reflect the dynamic changes of AGVs in the work area in order to conduct collision warning analysis. The nodes contained in the network can be the current positions of AGVs, intersections, loading and unloading areas, etc., while the edges represent the possible paths between AGVs and their travel times. In the dynamic running process, the state and position of each AGV are monitored in real time. Real-time position information of AGVs is obtained using sensors and monitoring systems, and their positions in the path network are updated. If AGV1 is located at (10, 20) and AGV2 is located at (15, 25), the system should record these position information in real time and update the corresponding node state in the network. Real-time monitoring data is integrated with the multi-level spatiotemporal path network, and potential collision risks are evaluated by analyzing the motion trajectory and path selection of AGVs. If AGV1 and AGV2 have overlapping travel time periods in the path network, the system needs to identify these time periods and mark them as potential conflicts. The detection criteria for potential collisions are determined, which usually include the case that AGVs pass through the same path in the same time period. A threshold is set, and when two AGVs overlap on the same path for less than 2 minutes, it is marked as a potential conflict. If AGV1 is expected to arrive at a certain node at 10:00 and AGV2 is expected to arrive at the same node at 10:01, conflict detection is performed according to the set threshold. The multi-level spatiotemporal path network is analyzed using algorithms such as collision detection algorithms or time-based conflict detection models to identify local AGV spatiotemporal conflict nodes. The system should be able to automatically detect the path overlap and time overlap between AGVs. By traversing the path and time of each AGV, it is checked whether there are overlapping nodes and time periods, and if so, the conflict nodes are recorded. The identified local AGV spatiotemporal conflict nodes are recorded in the database and marked for subsequent scheduling and management. Each conflict node should include the AGV number, position coordinates, and expected conflict time. If the conflict nodes of AGV1 and AGV2 are (12, 30) and the expected conflict time is 10:01, the node is recorded in the database and marked as "conflict". A visual display interface is designed to display all marked local AGV spatiotemporal conflict nodes in real time. The interface should be able to intuitively display the location of the conflict node, the conflicting AGVs, and the conflict time, making it easy for operators to respond quickly. A heat map or marker map is used to display the work area, with conflict nodes marked in red and normal paths marked in green, so that operators can see at a glance. A warning information communication mechanism is set up, and once the system detects a potential conflict, it immediately sends a warning to the operator through the notification system. The warning information should include the detailed information of the conflict node and the recommended response measures. If the conflict node is (12, 30), the system can suggest "AGV1 slow down, AGV2 change route", and notify the relevant operator through a message or application.

[0018] Step S5: Dynamic task priority decision according to the global execution task graph, and intelligent conflict planning for local AGV space-time conflict nodes, to construct a local conflict scheduling strategy; In this embodiment, the criteria for determining task priority decision include the urgency, importance, resource demand of the task, and the current state of the AGV. The task priority should be able to reflect the impact on the overall job efficiency. The standard for setting high-priority tasks is "to be completed within 10 minutes" or "involving the handling of critical goods", while low-priority tasks can have more flexible time arrangements. According to the global execution task graph, each task is evaluated and its priority score is calculated. The priority score can be calculated by weighted average method, considering the weight of each evaluation index. If the urgency score of task A is 80 points, the resource demand score is 60 points, and the importance score is 90 points, then the comprehensive priority score of task A is: comprehensive score = 0.5 x 80 + 0.3 x 90 + 0.2 x 60 = 78. During the task execution process, the state of the task and the running situation of the AGV are monitored in real time, and the priority of the task is dynamically adjusted according to the new information. If a high-priority task cannot be completed within the scheduled time, its priority should be re-evaluated. If the AGV of task B is delayed due to failure, the priority of task C should be raised to ensure the smoothness of the overall operation. For local AGV space-time conflict nodes, detailed conflict analysis is carried out. Identify the conflicting AGVs and their running paths, and evaluate the impact of the conflict on task execution. If AGV1 and AGV2 need to pass through the same key intersection at the same time, record the paths, expected passing time and load status of the two AGVs. According to the situation of the conflict node, develop appropriate intelligent conflict planning strategies. These strategies should include path adjustment, task delay or AGV speed adjustment, etc. to ensure that the conflict can be effectively solved. If AGV1 is lightly loaded and is a high-priority task, AGV1 can be allowed to speed up through, while the task of AGV2 is delayed or its path is changed. Apply the conflict resolution strategies developed to the current task scheduling, dynamically adjust the paths of AGVs and the order of tasks to avoid actual collisions. If the path of AGV1 is adjusted to detour, update its path information and recalculate the time to pass through the key node to ensure the smooth progress of subsequent tasks.

[0019] Step S6: Global path space-time change analysis and multi-AGV intelligent coordination optimization according to the local conflict scheduling strategy, to construct a global intelligent coordination control model.

[0020] In this embodiment, the path information, task execution status, and local conflict scheduling strategy of all AGVs are integrated into a unified data platform. The real-time and accuracy of the data are ensured, and the subsequent analysis is facilitated. The task execution time, path length, key node passed, and conflict record of each AGV are collected to form a real-time updated database that centrally stores the running data of all AGVs. The integrated data is analyzed to extract the path space-time variation characteristics, including the running speed of each AGV, path variation, key node passing time, and conflict frequency. If the path variation frequency of AGV1 is high, the system needs to record its speed variation in a specific time period and calculate the time consumption on different paths. According to the extracted space-time variation characteristics, a global path variation model is constructed. This model should be able to reflect the dynamic changes in the AGV running process and consider the mutual influence between different tasks. Using time series analysis method, a path variation prediction model of AGV is established to predict the running state of AGV and possible conflicts in the future period. Based on the global path space-time variation analysis results, a multi-AGV intelligent coordination optimization model is constructed. This model should consider the collaboration relationship between AGVs, resource sharing, and conflict avoidance strategy to maximize the overall efficiency. The objective function of the model is set as minimizing the overall transportation time and path conflict, while meeting the task demand and priority of each AGV. A suitable optimization algorithm (such as genetic algorithm, ant colony algorithm, or reinforcement learning algorithm) is selected to solve the coordination optimization model. This algorithm should be able to adaptively adjust in a dynamic environment and optimize the task and path allocation of AGVs in real time. The genetic algorithm is used to encode the path and task of AGV, and cross and mutation operations are performed to find the optimal task allocation scheme. The results of the optimization algorithm are applied to the scheduling of AGVs to dynamically adjust the task and path selection of AGVs to ensure efficient collaboration in a changing environment. If the optimization result shows that the path adjustment of AGV3 is effective, its task execution plan is immediately updated, and this adjustment is reflected in the system. The optimized multi-AGV coordination scheduling results are recorded in the database and displayed through visualization tools to help operators monitor and manage the running state of AGVs in real time. A multi-AGV scheduling diagram is generated to show the task allocation, path selection, and estimated arrival time of each AGV, facilitating the adjustment and optimization of the scheduling personnel.

[0021] In this embodiment, refer to Figure 2 For the detailed implementation step flowchart of step S1, in this embodiment, the detailed implementation steps of step S1 include: Obtain the real-time logistics instruction of the work area; perform deep instruction semantic analysis on the real-time logistics instruction to obtain a real-time logistics semantic vector; Perform operation logic flow deconstruction on the real-time logistics semantic vector to generate an instruction stream operation logic chain; The logistics scheduling requirement analysis is performed on the instruction stream operation logic chain to obtain the logistics scheduling requirement of real-time instructions; The logistics scheduling requirement is decomposed into multiple executable task subunits. The global task planning is performed on the multiple executable task subunits to construct a global execution task graph.

[0022] In this embodiment, real-time logistics instructions are obtained through sensors, cameras, and control systems integrated in the work area. These instructions can be issued through a scheduling system, mobile terminal, or other information sources. Set to pull the latest logistics instruction data from the scheduling system every 5 seconds to ensure the real-time and accuracy of the information. Establish a connection with the scheduling system through API or data transmission protocols (such as MQTT or WebSocket) to continuously receive logistics instructions for the work area. These instructions may include cargo handling, loading and unloading tasks, path planning, etc. The system monitors new instructions and immediately stores them in memory and marks them as "to be processed." Record the obtained real-time logistics instructions in the database and display them through the monitoring panel. This panel should be able to visually display the current logistics instructions and their execution status, so that the operator can grasp the work dynamics at any time. Generate a real-time instruction panel to display the current logistics instruction content, issue time, and priority to provide information support for subsequent processing. Determine the standard for deconstructing the operation logic flow, which usually includes the main steps of the instruction, the resources involved (such as AGV equipment, cargo type), and the execution conditions. This process should ensure completeness and executability. Set the deconstruction standard to "identify main operations, resource requirements, and condition constraints in steps." According to the semantic vector obtained by parsing, gradually deconstruct the operation logic of the instruction to generate the instruction stream operation logic chain. Each step should clearly describe its purpose and required resources. For the instruction "move cargo A to location B," deconstruct it into the following steps: 1) Determine the location of cargo A; 2) Plan the path; 3) Perform the move; 4) Confirm arrival. Generate a logic chain diagram to show the order and relationship of each operation step. According to the instruction stream operation logic chain, set the standards for logistics scheduling demand analysis, including resource allocation, time requirements, and task priority. Ensure that the analysis results can provide a basis for subsequent task decomposition. Set the priority standard to "urgency," "resource availability," and "task complexity." Analyze the instruction stream operation logic chain to assess its demand for logistics resources (such as AGV quantity, working time, and carrying capacity), and generate a corresponding scheduling demand report. After analyzing the instruction, it is concluded that moving cargo A requires 1 AGV and 30 minutes of work time. Determine the standards for multi-task decomposition to ensure that each executable task sub-unit is relatively independent and operable. Task decomposition should consider the complexity and execution order of the task. Set the decomposition standard to "each sub-task should have clear input, output, and execution conditions." According to the logistics scheduling demand, perform multi-task decomposition on the instruction stream operation logic chain to generate multiple executable task sub-units. Each sub-unit should be able to independently execute and complete a specific function. Decompose "move cargo A to location B" into: 1) Confirm the specific location of cargo A; 2) Calculate the moving path; 3) Start the AGV to move. Record the generated executable task sub-units in the database and display them through visualization tools to help operators understand the relationship and execution order of each sub-task.A task decomposition diagram is generated to show the input, output and execution conditions of each subtask, facilitating subsequent global task planning. According to a plurality of executable task subunits, the standards for global task planning are set, including the dependency relationship of tasks, resource allocation and time scheduling. Ensure that the global task planning has reasonableness and feasibility. Set the planning standard as "the dependency relationship between tasks and resource conflicts must be solved". In combination with the plurality of decomposed executable task subunits, global task planning is carried out to construct a global execution task graph. The graph should be able to reflect the execution order and resource allocation of the tasks. Draw the global task graph to show the execution order, resource demand and time arrangement of each task subunit. The constructed global execution task graph is recorded in the database, and is displayed through a visualization tool, so that the operation personnel can intuitively understand the execution plan of the global task. A global task graph is generated to show the execution order, resource allocation and time node of the task, helping to monitor and adjust the task execution process.

[0023] In this embodiment, referring to Figure 3 For the detailed implementation step flowchart of step S2, in this embodiment, the detailed implementation step of step S2 includes: Identify all AGV real-time monitoring parameters of the work area; Real-time load calculation is performed on the AGV real-time monitoring parameters to obtain the load state characteristics of each AGV; According to the AGV real-time monitoring parameters, the energy consumption threshold, the remaining battery capacity and the current load state are calculated; The trend of the remaining battery capacity is analyzed to generate the trend characteristics of the battery capacity; The load state characteristics, energy consumption threshold, current load state and trend characteristics of the battery capacity are comprehensively evaluated in multiple states to construct a multi-modal carrying state matrix of each AGV; According to the multi-modal carrying state matrix, dynamic task demand allocation processing is performed on the global execution task graph to construct an AGV dynamic task set.

[0024] In this embodiment, the real-time parameters that need to be monitored by AGV (Automatic Guided Vehicle) are determined, including position (GPS coordinates), speed, load weight, battery capacity, working status, etc. These parameters help to fully understand the running status of AGV. Set each AGV to record position and speed every second, while battery capacity and load status are recorded every 5 seconds to ensure the real-time and accuracy of data. Real-time monitoring parameters are collected through sensors and control systems on AGV. Data can be uploaded to the central monitoring system using wireless communication protocols such as Zigbee, Wi-Fi or LoRa. During the running process of AGV, its position information is updated in real time through the built-in GPS module, and the battery capacity is monitored using current sensors. According to the design parameters of AGV and real-time monitoring data, the standard of load calculation is set. Load calculation is usually based on the ratio of rated load to actual load of AGV, which can reflect the load status characteristics of AGV. If the rated load of AGV is set to 1000 kg and the actual load is 800 kg, the load status characteristic is 0.8 (i.e. 80%). The real-time monitoring parameters of each AGV are calculated for load calculation to analyze its current load status. Real-time load data needs to be obtained from sensors and compared with the rated load. The current value monitored by the current sensor can be converted into load data, the actual load is calculated and compared with the rated load to obtain the load status characteristic. Determine the standard of energy consumption calculation, including the energy consumption characteristics of each AGV under different loads. Energy consumption threshold is usually calculated based on the motor power, speed and load of AGV. Set the power of AGV to 500 W under full load and 300 W under empty load, and calculate the energy consumption combined with speed and travel time. Calculate the energy consumption and remaining battery capacity of each AGV based on real-time monitoring parameters. Calculate the current remaining battery capacity by monitoring battery capacity and energy consumption in real time. If AGV consumes 200 Wh of electricity during operation and the total battery capacity is 1000 Wh, the remaining battery capacity is 800 Wh. Record the energy consumption threshold, remaining battery capacity and current load status in the database and display it through visualization tools to help operators monitor the energy consumption of AGV. Generate battery capacity monitoring chart to show the remaining battery capacity and energy consumption trend of each AGV to help optimize scheduling. Choose appropriate analysis method, usually use time series analysis or regression analysis to evaluate the trend of battery capacity change. This analysis helps to predict the service life and charging demand of the battery. Use moving average method to smooth the power change data and identify the trend of power decline. Analyze the trend of power change of the remaining battery capacity of each AGV to generate power change trend characteristics. This process should consider historical data to identify the power change pattern. Analyze the power change in the past hour, if the power is found to decline at a rate of 10 Wh per minute, record this trend characteristic.Determine the criteria for multi-state comprehensive evaluation, including load state, energy consumption threshold, current load state, and power change trend. This evaluation should be able to fully reflect the carrying state of each AGV. Set the comprehensive evaluation criteria as the weighted sum of load state, energy consumption, and power change trend. Perform multi-state comprehensive evaluation on each AGV to construct a multi-modal carrying state matrix for each AGV. This matrix should be able to reflect the relationship and mutual influence between states. If the load state of an AGV is 0.8, the energy consumption threshold is 500W, and the power change trend is -10Wh / min, the comprehensive evaluation can be obtained by weighted calculation. According to the multi-modal carrying state matrix, set the criteria for dynamic task demand allocation, including priority, resource availability, and task urgency. This process should ensure efficient task allocation. Set the priority criteria as "AGVs with sufficient power are given high-priority tasks first". Based on the multi-modal carrying state matrix, perform dynamic task demand allocation processing on the global task graph to construct an AGV dynamic task set. Each AGV should determine the tasks to be executed based on its state. If an AGV has sufficient power and a low load state, it will be assigned to a new transportation task to ensure efficient use of resources. Record the constructed AGV dynamic task set in the database and display it through a visualization tool to help operators understand the task allocation situation. Generate a task allocation diagram to show the current task status of each AGV, helping to monitor and adjust the task execution process.

[0025] In this embodiment, referring to Figure 4 For the detailed implementation steps of step S3, in this embodiment, the detailed implementation steps of step S3 include: According to the AGV dynamic task set, identify the operation behavior intention of each AGV, and generate multiple AGV operation behavior intentions; According to the multiple AGV operation behavior intentions, calculate the passing points and target endpoints of multiple tasks; Identify the real-time positions of multiple AGVs; calculate the spatial coordinates of the real-time positions; According to the spatial coordinates, perform multi-task cooperative carrying evolution on the passing points and target endpoints of multiple tasks to construct a multi-task carrying path; Perform multi-level spatiotemporal distribution fitting on the multi-task carrying path to construct a multi-level spatiotemporal path network.

[0026] In this embodiment, behavior intention recognition is a process of predicting the upcoming actions (such as turning, accelerating, obstacle avoidance, loading and unloading, etc.) of AGV based on task characteristics, current state and historical behavior data. It provides behavior prediction ability for the scheduling system, facilitating subsequent dynamic optimization of the path. The input of the behavior intention recognition system includes: the task currently assigned to the AGV, its historical path data, the surrounding environment state (whether there are obstacles, congestion), and the target task in the dynamic task set. The system uses sequence modeling techniques (such as GRU or Transformer structure) to model the running trajectory of AGV and the task execution rule. Each intention output is a vector containing action categories and probabilities. For example: [loading: 0.15, turning: 0.25, stationary: 0.1, forward: 0.5].

[0027] In the experiment, about 3000 AGV task execution trajectories in the past 7 days were used as the training set to recognize the current intention of AGV, and the prediction accuracy reached 93.4%. After interacting with the scheduling system, the recognition result can effectively reduce the sudden path conflict rate (about 21% reduction). Based on the workshop map (usually a 2D grid or topological map) and the current task content, the system generates a task path sketch by combining behavior intention through path planning algorithms (such as improved A* or Dijkstra algorithm). Each path contains a series of "must-pass points", which may be task-related operation points (loading and unloading points), traffic control points (intersections) or obstacle avoidance points.

[0028] An AGV carries goods from "starting point A" to "ending point D", with the behavior intention of "obstacle avoidance detour". The system will automatically identify the intermediate points as B and C. In this way, the task path is expressed as: A → B → C → D. In the system, the path is stored in the form of a list, containing coordinates, types (operation points, control points) and time windows. Using the path history record and behavior intention combination model, more than 90% of the path changes of AGV can be predicted in advance, effectively reducing the frequency of dynamic planning, improving the stability of the path and the response efficiency of the system. To ensure that the task path aligns with the actual running situation, the system needs to identify and map the real-time position of all AGVs with high precision. This step mainly involves high-frequency position acquisition, coordinate mapping and position error correction technologies. AGV positioning systems are usually based on one or a combination of the following technologies: UWB ultra-wideband positioning, laser SLAM, visual SLAM, inertial navigation system (INS) or two-dimensional code / landmark assisted positioning. In large-scale application scenarios, the UWB solution is widely used due to its real-time performance and anti-interference ability, with an error of within ±10 cm.

[0029] The collected raw positioning data is converted into spatial coordinates through sensor fusion algorithms (such as Extended Kalman Filter EKF), ensuring that it aligns with the global coordinates of the workshop. The system uses a combination of machine learning and rule-based methods to correct the position error. The rule-based method is based on the historical path data of AGV, which is used to correct the position error. The machine learning method is based on the historical path data of AGV, which is used to correct the position error. Figure 1The position of each AGV is represented as a (x, y, θ) triplet, where θ is the heading angle. For example, an AGV with current coordinates (12.5m, 8.4m, 90°) indicates that it is located at grid coordinates 12.5 meters and 8.4 meters, facing the east direction.

[0030] In the experiment, the system synchronizes the position data of 50 AGVs per second, using message queues such as Kafka to transmit to the central scheduling system, with an average position error of less than 6 cm and a position refresh period of less than 200 ms. Accurate spatial position recognition is a prerequisite for subsequent path conflict judgment and space-time coordination. Based on the current spatial coordinates, path nodes, and behavior intentions, the task is co-evolved. That is, instead of generating paths for each AGV independently, a collaborative path evolution model is used to achieve advanced path optimization such as "yielding between tasks, inserting space, and merging paths." This process uses graph theory modeling to map the starting points, passing points, and endpoints of all AGVs to a spatial path graph. Through constraint-based path optimization algorithms such as collaborative multi-agent A* and CBS algorithms, the system analyzes the spatial overlap areas between different AGVs and re-plans or fine-tunes part of the AGV's path based on task urgency and path conflict probability. AGV1 and AGV2 share nodes B and C in their paths, and after analysis, the system can choose to give priority to AGV1 and delay AGV2's start time by 3 seconds to avoid conflict. This collaborative path evolution can significantly reduce the number of path conflicts. In the experimental scenario, this method reduces the conflict rate by 38.2% and improves the task completion time by 16% compared to static path scheduling, with an average of 1.8 path adjustments per task. Select an appropriate space-time analysis model, such as the Space-Time Point Model Analysis (SPM) or Dynamic Time Warping (DTW), to fit the space-time distribution characteristics of multi-task carrying paths. Use the SPM model to analyze the path changes of AGVs in different time periods to identify potential congestion and conflict areas. Perform multi-level space-time distribution fitting on the generated multi-task carrying paths to analyze the space-time characteristics of AGVs during task execution. This process should consider path changes in different time periods. By comparing path data in different time periods, high-frequency paths and low-frequency paths are identified, as well as potential conflict areas. Record the fitted multi-level space-time path network in the database and display it through visualization tools to help understand AGV movement patterns and path characteristics. Generate a space-time path network graph to show the path changes of AGVs in different time periods, helping operators optimize and adjust.

[0031] In this embodiment, step S4 includes the following steps: Identify key path nodes for multi-task carrying paths and mark all key path nodes; Calculate the start timestamp of each task based on the AGV dynamic task set; According to the start time stamp, full-cycle approach time prediction is performed on all critical path nodes to obtain an operation time stamp of each critical path node; According to the operation time stamp, operation time point distribution analysis is performed to obtain multi-path operation time stamp distribution information; Based on the multi-path operation time stamp distribution information, potential AGV collision early warning is performed on the multi-level space-time path network, and local AGV space-time conflict nodes are marked.

[0032] In this embodiment, in the AGV multi-task path, some path nodes have higher spatiotemporal conflict sensitivity compared to others, referred to as critical path nodes. The goal of this step is to identify these nodes for more accurate spatiotemporal prediction and conflict warning in the following steps. Critical path node identification is based on the following indicators: path overlap degree, operation sensitivity (such as loading and unloading points, turning points), network centrality (calculated through Katz centrality and Betweenness Centrality in graph theory), etc. Path overlap degree is determined by the frequency of a node appearing in multiple AGV paths, operation sensitivity is identified based on task decomposition labels, and centrality is calculated through Katz centrality or Betweenness Centrality by constructing a path graph. In the experiment, the simulated workshop path network contains 260 nodes, and 50 AGVs operate simultaneously. Using a centrality threshold of 0.35 and an overlap rate threshold of 20%, 42 critical nodes are identified. These nodes are marked as high-priority conflict management points and input into the dynamic monitoring module of the scheduling system for dynamic path adjustment and prediction and warning. The method for determining the start timestamp of each task is usually based on the scheduling strategy of the AGV and the priority of the task. The start timestamp should consider the scheduling order of the task and the availability of the AGV. High-priority tasks are started immediately when the AGV is idle, and their timestamp is the current system time; low-priority tasks need to wait for the completion of the preceding task. According to the dynamic task set of the AGV, the start timestamp of each task is calculated. For each task, check whether the dependent preceding task is completed, and determine the start time according to the calculation formula. If the start time of task A is T1, and task B depends on task A, then the start time of task B is T1 plus the estimated completion time of task A. Choose an appropriate time prediction model, usually based on historical data and real-time monitoring parameters to predict the passage time of the AGV at the critical path node. Regression analysis or machine learning models (such as random forests) can be used for time prediction. A regression model is trained using historical path data to predict the time it takes for an AGV to pass through a critical node under specific load and speed. According to the start timestamp of each task, the full-cycle passage time of all critical path nodes is predicted. The estimated passage time of each node is calculated as the start time of the task plus the predicted passage time. If the start time of the task is T2, and the estimated passage time of the AGV through the critical node is 5 minutes, then the estimated operation timestamp of the critical path node is T2 + 5 minutes. Record the operation timestamp of each critical path node in the database and display it through a visualization tool for real-time monitoring of the AGV operation status by the operator. Generate an operation timestamp chart to show the estimated passage time of each critical path node, helping to adjust the task and path planning. Choose an appropriate statistical analysis method, usually using histograms or density estimation to analyze the distribution of AGV operation time points at critical path nodes. This analysis helps to identify high-frequency use time periods of the path.The kernel density estimation method in the SciPy library of Python is used to generate the distribution graph of the time points. The distribution of the operation timestamps of all critical path nodes is analyzed, and the distribution information of the operation time points of each node is generated. This process should be able to reveal the traffic situation of AGVs at different time periods. Analyze the operation timestamp data in the past week to identify the peak traffic period of AGVs at each key node. Determine the criteria for potential AGV collision warning, usually based on the operation time point distribution and time overlap of AGVs at key path nodes. If multiple AGVs pass through the same node at the same time, it is marked as a potential conflict. Set the operation timestamp of two AGVs to differ by less than 1 minute to mark it as a potential conflict node. Based on the multi-path operation timestamp distribution information, a multi-level space-time path network is constructed for potential AGV collision warning. Check the operation timestamps of all critical path nodes to identify possible conflict nodes. If AGV1 and AGV2 overlap in the expected traffic time at the same critical path node, record the node as a conflict node and trigger a warning. Record the marked local AGV space-time conflict nodes in the database and display them through a visualization tool so that the operator can take timely measures. Generate a conflict warning map to identify potential conflict nodes and related AGVs, helping dispatchers manage risks and make adjustments.

[0033] In this embodiment, step S5 includes the following steps: According to the global execution task graph, the task path accessibility is analyzed to obtain the difficulty of the task path; According to the global execution task graph, the task time window and the expected completion time length are calculated, and the task time efficiency requirement is evaluated; The load capacity and power consumption of each AGV are calculated, and the resource occupation degree is generated; The dynamic task priority decision is made according to the difficulty of the task, the task time efficiency requirement and the resource occupation degree, and the dynamic priority traffic level of each AGV is generated; The multiple conflict AGVs of the local AGV space-time conflict node are identified; Based on the dynamic priority traffic level, the traffic order and waiting time of the multiple conflict AGVs are intelligently planned, and the local conflict scheduling strategy is constructed.

[0034] In this embodiment, the criteria for task path accessibility analysis are determined to assess the difficulty of each path. Accessibility generally involves multiple factors such as path length, number of obstacles, road conditions, and traffic rules. An accessibility scoring system can be set up, with a score range of 1 to 10, where 1 represents very difficult and 10 represents very easy. The congestion, slope, and turning radius of the path are considered when scoring. According to the global execution task graph, the accessibility of each task path is analyzed one by one. The path length is calculated using graph theory algorithms such as A* algorithm, and the accessibility is evaluated in combination with environmental information to generate the accessibility score. If a path is 100 meters long, with 2 obstacles along the way, and the road conditions are good, then according to the set scoring system, the accessibility of this path is evaluated as 7 points. The calculation criteria for task time window are determined, and the time window is usually based on the deadline, priority of the task, and availability of AGV. The estimated completion time length should consider the complexity of the task and the estimated delivery speed. The time window of each task is set from the start time to the task deadline, and the estimated completion time is the length of the task path divided by the average speed of the AGV. According to the global execution task graph, the time window and estimated completion time length of each task are calculated. The completion time is estimated using the known AGV speed and path length. If the task path length is 150 meters and the average speed of the AGV is 1 meter per second, then the estimated completion time is 150 seconds. Record the calculated task time window and estimated completion time length in the database, and display it through the visualization tool to help the scheduling personnel monitor the time efficiency requirements of the task. Generate a task timeline chart to show the time window and estimated completion time of each task, helping to optimize the scheduling strategy. Determine the calculation criteria for the load capacity and power consumption of AGV. The load capacity should be based on the rated load and current load of the AGV, while the power consumption is usually based on the running time and power of the AGV. The power of the AGV is set to 500W when fully loaded and 300W when empty, and the power consumption is calculated according to the running time. Calculate the load capacity and power consumption of each AGV. According to the real-time monitoring data, evaluate the current load and running time of each AGV to generate resource occupation. If the AGV currently carries 800 kilograms and has been running for 300 seconds, then the power consumption can be calculated as: Power = Power × Time = 500W × (300 / 3600) hours = 41.67Wh. Determine the decision criteria for task priority, which usually considers the accessibility, time efficiency requirements, and resource occupation. Tasks with high priority should be executed first if resources allow. A priority scoring formula can be set up: Priority = (Accessibility Score) + (Time Efficiency Requirement Score) + (Resource Occupation Score). According to the accessibility, time efficiency requirements, and resource occupation, calculate the dynamic priority access level of each AGV. Use the set scoring formula to generate the priority of each task.If the task's passing difficulty score is 5, the time efficiency requirement score is 3, and the resource occupation degree score is 2, the priority is 5 + 3 + 2 = 10. The identification criteria for local AGV space-time conflict nodes are usually based on the overlap of multiple AGV paths. If multiple AGVs pass through the same key path node at the same time, it is marked as a conflict node. It can be set that if the expected passing time of two AGV paths differs by less than 1 minute, it is considered a conflict. Analyze the paths and timestamps of all AGVs to identify local space-time conflict nodes. By comparing the operation timestamps of AGVs, conflicting AGVs are detected and conflict nodes are marked. If AGV1 and AGV2 have an expected passing time of 10:00 and 10:05 at the key path node, the node is recorded as a conflict node and the relevant AGVs are recorded. The passing order planning criteria for conflict AGVs are usually based on dynamic priority passing levels. AGVs with high priority should pass through conflict nodes first. The AGV with the highest priority should have the shortest waiting time at the conflict node. Based on dynamic priority passing levels, the passing order and waiting time of multiple conflict AGVs are intelligently planned. The passing order of AGVs is arranged according to priority, and the waiting time of each AGV is calculated. If AGV1 has a priority of 10 and AGV2 has a priority of 8, AGV1 has a waiting time of 0 at the conflict node, while AGV2 has a waiting time of 5 seconds. Record the planned passing order and waiting time in the database and display it through a visualization tool to help operators understand the scheduling strategy. Generate a conflict scheduling strategy diagram to show the passing order and waiting time of each AGV, helping to monitor and adjust task execution. In this embodiment, the specific steps of analyzing the task path accessibility according to the global execution task graph are as follows: Obtain multi-directional monitoring videos of the work area; According to the multi-level space-time path network, the multi-directional monitoring videos are visually mapped to obtain multi-level path visualization videos; According to the multi-level path visualization videos, path obstacles are identified and marked; According to the multiple path obstacles, the congestion degree score is obtained to obtain the congestion quantization value of each path; According to the congestion quantization value of each path, spatial distribution analysis is performed to obtain a congestion quantization heat map of the work area; Based on the congestion quantization distribution map, the task path accessibility is analyzed to obtain the passing difficulty of the task path.

[0035] In this embodiment, a multi-angle monitoring system is set up for the work area, which typically includes multiple cameras (such as fixed cameras, 360-degree rotating cameras, infrared cameras, etc.) to ensure real-time capture of the dynamic situation of the work area from different angles. In a typical warehouse layout, at least 6 cameras are installed to cover all entrances and exits, transfer points and main channels, ensuring no blind area monitoring. Configure the camera system to record the real-time video of the work area at regular intervals. Video capture should have high frame rate (such as 30 frames per second) and high resolution (such as 1080p or 4K) to ensure image quality meets subsequent analysis requirements. Set the camera to automatically save a video segment every 5 minutes, continuously monitor the dynamic changes in the work area. Store the collected video data in a central database to ensure data security and accessibility. Video management software (such as VMS) can be used to classify and label videos for subsequent retrieval and use. According to the date and time, the video is archived, generating a simple index, facilitating subsequent analysis and viewing. According to the layout of the work area and the motion path of the AGV, a multi-level space-time path network model is constructed. This model should include all key path nodes and possible AGV movement paths. Set the starting position, target position and intermediate points of each AGV to form a complete path network. Use computer vision technology (such as OpenCV) to analyze the multi-angle monitoring video in real time and map the AGV path to the video. This can be achieved by superimposing path layers, highlighting the AGV's driving route. Use different colors to mark different paths, display the AGV's motion trajectory in real time, and fuse it with the monitoring video to generate multi-level path visualization video. Select appropriate image processing and deep learning models (such as YOLO or Mask R-CNN) for path obstacle recognition. The model should be trained to accurately identify obstacles in the work area, such as goods stacking, equipment and personnel, etc. Use the YOLO model to detect targets in the video to ensure real-time identification and labeling of obstacles. Obstacle identification is performed on the multi-level path visualization video to detect obstacles in the video frames in real time and mark their positions in the video. You can use a border or mask to mark it. Each time an obstacle appears in the video, the system automatically identifies and marks it with a red border, ensuring that the operator can clearly see the location of the obstacle. Record the identified obstacle information in the database, including obstacle type, location coordinates and status, etc. Use visualization tools to display the distribution of obstacles in the path. Determine the criteria for congestion level scoring, usually considering the number of obstacles, distribution density and the degree of impact on the AGV path. The score should be quantified for subsequent analysis. Set the scoring criteria: each obstacle has an impact on the path of 1 point, and if the number of obstacles on a path exceeds 5, the congestion score of that path is 5. According to the identified multiple path obstacles, score the congestion level of each path.The number of obstacles on each path is counted, and the total score is calculated according to the scoring standard. If there are 3 obstacles on path A and 7 obstacles on path B, they are scored as 3 and 7 respectively. Select an appropriate spatial distribution analysis method, usually using spatial statistics or spatial interpolation techniques (such as Kriging interpolation) to analyze the congestion distribution in the work area. Use Kriging interpolation to analyze the spatial distribution of congestion scores and generate a congestion heat map. According to the congestion quantification value of each path, the congestion quantification heat map of the work area is generated. This map should be able to highlight the congestion area. In the heat map, the area with more serious congestion is marked with red, and the area with mild congestion is marked with yellow, so as to identify visually. Determine the standard of task path accessibility analysis, usually including congestion degree, path length, travel time and other factors. The analysis should be able to quantify the difficulty of path travel. Set the accessibility score standard, and increase the travel difficulty by 1 point for every 1 point increase in congestion score, while considering the influence of path length and expected travel time. According to the congestion quantification value of each path, the task path accessibility analysis is carried out, and the travel difficulty of each path is calculated. Considering the congestion score and path length, the final accessibility score is obtained. If the congestion score of path A is 5 and the path length is 100 meters, the accessibility score is 5 + (0.1 x 100) = 15. Record the travel difficulty of each path calculated in the database, and display it through visualization tools to help operators quickly judge the feasibility of path travel. Generate a travel difficulty map to show the accessibility score of each path, helping dispatchers to select paths and arrange tasks.

[0036] In this embodiment, the specific steps of step S6 are: According to the local conflict scheduling strategy, the global path space-time change analysis is carried out, and the global path space-time change feature is extracted; According to the global path space-time change feature, the real-time path change update of the multi-level space-time path network is carried out, and the real-time update multi-level path network is generated; According to the local conflict scheduling strategy, the global multi-AGV intelligent coordination optimization of the real-time update multi-level path network is carried out, and the global intelligent coordination control model is constructed.

[0037] In this embodiment, a local conflict scheduling strategy is defined. This strategy needs to specify how to adjust the path and time when AGVs encounter conflicts during operation to ensure coordination and safety between AGVs. Conflict scheduling strategies usually include priority scheduling, temporary path adjustment, and load balancing, etc. When two AGVs encounter conflicts on the same path, the AGV with higher priority continues to move forward, while the AGV with lower priority selects temporary detour. According to the local conflict scheduling strategy, the global path data is analyzed to extract the spatio-temporal variation characteristics of the global path. These characteristics should include the running speed of AGVs, path length, key nodes passed, time delay, etc. Using time series analysis method, the motion state of each AGV is monitored in real time, and the path change when conflict occurs and its impact on the global path are recorded. The structure of the multi-level path network is determined, usually including multiple levels of path nodes and connecting edges. Each level should reflect different path choices of AGVs, such as main channel, secondary channel and temporary detour path. In a complex warehouse layout, the main channel is used for fast transportation, while the secondary channel is used for detour or obstacle avoidance. According to the extracted spatio-temporal variation characteristics of the global path, the multi-level spatio-temporal path network is updated in real time. This process should consider the current real-time state of AGVs, path conflicts and changes in the surrounding environment. If AGV1's path changes, the system should update AGV1's path position in the network in time and adjust the corresponding connecting edges to reflect the new path. The real-time updated multi-level path network is recorded in the database and displayed through visualization tools to help operators monitor the network state in real time. A dynamic path network diagram is generated to show the current AGV's path selection and network structure, allowing dispatchers to quickly judge the path's traffic situation. Based on the local conflict scheduling strategy and the real-time updated multi-level path network, a global intelligent coordination control model is constructed. This model should consider the mutual influence between AGVs, path selection and task priority to achieve global optimization. The optimization goal is set to minimize the global transportation time and path conflicts while meeting the task requirements of each AGV. Optimization algorithms such as genetic algorithm, ant colony algorithm or reinforcement learning are applied to solve the global intelligent coordination control model to generate the optimized scheduling scheme and path selection. This scheme should be able to dynamically adapt to the real-time changing environment and tasks. Genetic algorithm is used to encode AGV's path and task, and the best path combination is selected through fitness evaluation to ensure the maximum transportation efficiency.

[0038] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, the scope of the invention being defined by the attached claims rather than the above description, and it is intended to include all variations falling within the meaning and scope of the equivalent elements of the application file.

[0039] The foregoing merely illustrates the principles of the application and application of its principles. Various modifications and alterations to this implementation will occur to those skilled in the art. The scope of the application should be determined, however, by the following claims rather than by the embodiments shown.

Claims

1. A multi-AGV scheduling method, characterized in that, Includes the following steps: Step S1: Obtain real-time logistics instructions for the work area, perform logistics scheduling demand analysis and global task planning, and construct a global execution task map; Step S2: Obtain real-time monitoring parameters of AGV, perform multi-state comprehensive evaluation, and dynamically allocate task requirements based on the global execution task map to construct a dynamic task set for AGV. Step S3: Based on the AGV dynamic task set, identify the operational intentions of each AGV, perform multi-task collaborative transportation evolution, and construct a multi-level spatiotemporal path network. Step S4: Perform potential AGV collision warning based on the multi-level spatiotemporal path network and mark local AGV spatiotemporal conflict nodes; Step S5: Make dynamic task priority decisions based on the global task execution map, and perform intelligent conflict planning for local AGV spatiotemporal conflict nodes to build a local conflict scheduling strategy. Step S6: Perform global path spatiotemporal change analysis and multi-AGV intelligent coordination optimization based on the local conflict scheduling strategy, and construct a global intelligent coordination control model.

2. The multi-AGV scheduling method according to claim 1, characterized in that, The specific steps of step S1 are as follows: Obtain real-time logistics instructions for the work area; perform deep instruction semantic parsing on the real-time logistics instructions to obtain real-time logistics semantic vectors; Deconstruct the operational logic flow of the real-time logistics semantic vector to generate an instruction flow operational logic chain; Perform logistics scheduling requirement analysis on the instruction stream operation logic chain to obtain the logistics scheduling requirements of real-time instructions; The logistics scheduling requirements are decomposed into multiple tasks to generate multiple executable task sub-units. Perform global task planning for multiple executable task sub-units and construct a global execution task graph.

3. The multi-AGV scheduling method according to claim 1, characterized in that, The specific steps of step S2 are as follows: Identify all real-time monitoring parameters of AGVs in the work area; Real-time load calculation is performed on the real-time monitoring parameters of the AGV to obtain the load status characteristics of each AGV; Calculate the energy consumption threshold, remaining battery capacity, and current load status based on the AGV's real-time monitoring parameters. Analyze the remaining battery capacity to determine the trend of power change and generate power change trend characteristics; The load status characteristics, energy consumption threshold, current load status and power change trend characteristics are comprehensively evaluated in multiple states to construct a multimodal transport status matrix for each AGV. Based on the multimodal vehicle state matrix, the global execution task map is dynamically allocated to construct the AGV dynamic task set.

4. The multi-AGV scheduling method according to claim 1, characterized in that, Step S3 is as follows: Based on the dynamic task set of AGVs, the intention of each AGV's operation behavior is identified, and multiple AGV operation behavior intentions are generated. Calculate the waypoints and target endpoints of multiple tasks based on the operational intentions of multiple AGVs; Identify the real-time positions of multiple AGVs; calculate the spatial coordinates of the real-time positions; Based on the spatial coordinates, the multi-task collaborative transportation evolution of the waypoints and target endpoints of multiple tasks is carried out to construct a multi-task transportation path; Multi-level spatiotemporal distribution fitting is performed on multi-task transportation paths to construct a multi-level spatiotemporal path network.

5. The multi-AGV scheduling method according to claim 1, characterized in that, The specific steps of step S4 are as follows: Identify critical path nodes in multi-task delivery paths and mark all critical path nodes. Calculate the start timestamp of each task based on the AGV dynamic task set; Based on the start timestamp, the full-cycle path time of all critical path nodes is predicted to obtain the operating timestamp of each critical path node. The distribution of operation time points is analyzed based on the operation timestamps to obtain multi-path operation timestamp distribution information. Based on the distribution information of multi-path operation timestamps, potential AGV collision warnings are performed on multi-level spatiotemporal path networks, and local AGV spatiotemporal conflict nodes are marked.

6. The multi-AGV scheduling method according to claim 1, characterized in that, The specific steps of step S5 are as follows: Based on the global task execution graph, perform task path accessibility analysis to obtain the difficulty of traversing the task path; Calculate the task time window and expected completion time based on the global task execution graph, and assess the task timeliness requirements; Calculate the load capacity and power consumption of each AGV to generate resource occupancy. Dynamic task priority decisions are made based on the passage difficulty, task timeliness requirements, and resource consumption to generate a dynamic priority passage level for each AGV. Identify multiple conflicting AGVs at the spatiotemporal conflict node of the local AGV; Based on the dynamic priority passage level, intelligent planning is performed on the passage order and waiting time of multiple conflicting AGVs to construct a local conflict scheduling strategy.

7. The multi-AGV scheduling method according to claim 1, characterized in that, The specific steps for performing task path accessibility analysis based on the global task execution graph to obtain the traversal difficulty of the task path are as follows: Acquire multi-directional monitoring videos of the work area; Based on the multi-level spatiotemporal path network, real-scene visualization mapping is performed on multi-directional monitoring videos to obtain multi-level path visualization videos. Based on multi-level path visualization video, identify path obstacles and mark multiple path obstacles; Congestion levels are scored based on obstacles along multiple paths to obtain a quantitative congestion value for each path. Spatial distribution analysis is performed based on the congestion quantification value of each path to obtain a congestion quantification heat map of the work area; Based on the congestion quantification distribution map, accessibility analysis of the task path is performed to obtain the difficulty of traversing the task path.

8. The multi-AGV scheduling method according to claim 1, characterized in that, The specific steps of step S6 are as follows: Based on the local conflict scheduling strategy, global path spatiotemporal change analysis is performed to extract global path spatiotemporal change characteristics; Based on the spatiotemporal variation characteristics of global paths, the multi-level spatiotemporal path network is updated in real time to generate a real-time updated multi-level path network. Based on the local conflict scheduling strategy, a global intelligent coordination and optimization of multiple AGVs is performed on the real-time updated multi-level path network to construct a global intelligent coordination and control model.

Images