An automated terminal AGV intelligent scheduling method and system

Abstract

The present application relates to the field of intelligent scheduling technology, and discloses an automated terminal AGV intelligent scheduling method and system. The method comprises: collecting container task data through a terminal operation management system, performing multi-dimensional evaluation processing to obtain a task priority evaluation matrix, constructing an AGV state vector based on the matrix and predicting load capacity based on historical data, intelligently matching the task matrix and the AGV state vector through the Port‑MTCSA algorithm to obtain a full-process task chain, performing time series analysis based on the task chain to predict the terminal load distribution, and performing dynamic re-planning of AGV resources when the load exceeds a threshold. The present application solves the technical problem that existing automated terminal AGV scheduling methods lack dynamic priority evaluation, load prediction, task chain optimization, and forward-looking resource allocation in multi-task concurrent scenarios.

Description

Technical Field

[0001] The present application relates to the field of intelligent scheduling technology, and in particular to an automated terminal AGV intelligent scheduling method and system. Background Art

[0002] Existing automated vehicle (AGV) scheduling methods for automated terminals primarily utilize static path planning combined with simple conflict avoidance strategies to manage container transport operations at the terminal by precalculating AGV paths and task allocation. Traditional scheduling methods typically allocate tasks based on fixed priority rules and greedy algorithms, employing simple strategies such as first-come, first-served or shortest-distance-first to determine the order of AGV operations. Time window constraints and capacity limits are also employed to avoid resource conflicts. These methods can maintain basic scheduling functionality when terminal operations are relatively stable and task volumes are relatively small, providing technical support for the initial development of automated terminals.

[0003] However, existing technologies have significant deficiencies when dealing with multi-task concurrent scenarios, mainly reflected in the lack of intelligent evaluation and real-time adjustment mechanisms for dynamic task priorities, and the inability to reorder tasks according to the actual urgency and business value of terminal operations. Existing task allocation algorithms mostly use greedy or simple heuristic strategies, which are difficult to achieve global optimal allocation under multi-objective constraints. In particular, when considering multiple constraints such as time windows, AGV energy consumption, and equipment status, static scheduling schemes often lead to uneven resource utilization and low operating efficiency. In addition, traditional methods lack predictive scheduling capabilities based on historical operation data, and are unable to identify peak task congestion and resource bottlenecks in advance, resulting in passivity and lag in scheduling decisions, making it difficult to maintain optimal scheduling performance when the terminal's operating load changes dynamically.

[0004] Based on an in-depth analysis of the above technical deficiencies, we can further identify the fundamental problems of existing technologies in intelligent scheduling and decision-making: First, there is a lack of a dynamic assessment mechanism for task priorities that integrates multi-dimensional factors, making it impossible to make intelligent trade-offs between multiple dimensions such as the urgency of ship departure, container weight level, and customer priority; second, there is a lack of a load capacity prediction method based on the real-time status and historical operating mode of AGVs, making it impossible to accurately estimate the changes in the operating capacity of AGVs in future time windows; third, there is a lack of specialized task chain optimization algorithms for the three-layer operating structure of "quay crane-AGV-yard" at the terminal, making it impossible to achieve coordinated optimization of the entire loading, unloading, transportation, and storage process; fourth, there is a lack of a forward-looking resource allocation mechanism based on load forecasting, making it impossible to proactively and dynamically reallocate AGV resources before the load peak arrives. Summary of the Invention

[0005] The present application provides an automated terminal AGV intelligent scheduling method and system, which is used to solve the technical problems of existing automated terminal AGV scheduling methods lacking dynamic priority evaluation, load prediction, task chain optimization and forward-looking resource allocation in multi-task concurrent scenarios.

[0006] In the first aspect, the present application provides an automated terminal AGV intelligent scheduling method, which includes: collecting container task data through the terminal operation management system, performing multi-dimensional evaluation processing on the container task data according to the urgency of the ship's departure, the container weight level, the quay crane operation sequence, the yard distance coefficient and the customer priority, and obtaining a task priority evaluation matrix; constructing an AGV state vector according to the task priority evaluation matrix, and performing load capacity prediction processing on the AGV's current position coordinates, load status, remaining power and expected completion time through training with historical port operation data to obtain an AGV load prediction result; comparing the AGV load prediction result with the The task priority evaluation matrix is ​​matched, and the container task priority matrix and the AGV dynamic load state vector are intelligently matched through the terminal multi-task chain scheduling algorithm Port-MTCSA to obtain the full-process task chain of loading and unloading, transportation and storage; according to the full-process task chain of loading and unloading, transportation and storage, the terminal historical operation data is collected, and the load is predicted by time series analysis based on the ship arrival plan, weather factors and container type distribution to obtain the terminal operation load distribution; the terminal operation load distribution is compared with the preset load threshold. When the predicted load exceeds the threshold, the AGV resource allocation is dynamically re-planned to obtain the optimized scheduling instruction and execute the terminal AGV intelligent scheduling.

[0007] In a second aspect, the present application provides an automated terminal AGV intelligent scheduling system, the automated terminal AGV intelligent scheduling system comprising:

[0008] An evaluation module is used to collect container task data through the terminal operation management system, and perform multi-dimensional evaluation processing on the container task data based on the urgency of ship departure, container weight level, quay crane operation sequence, yard distance coefficient and customer priority to obtain a task priority evaluation matrix;

[0009] A prediction module is used to construct an AGV state vector based on the task priority evaluation matrix, and to perform load capacity prediction processing on the AGV's current position coordinates, load status, remaining power, and estimated completion time through training with historical port operation data to obtain an AGV load prediction result;

[0010] A matching module is used to match the AGV load prediction result with the task priority evaluation matrix, and intelligently match the container task priority matrix and the AGV dynamic load state vector through the terminal multi-task chain scheduling algorithm Port-MTCSA to obtain the full-process task chain of loading and unloading, transportation, and storage;

[0011] A collection module is used to collect historical terminal operation data based on the full-process task chain of loading and unloading, transportation, and storage, and to perform load forecasting processing based on ship arrival plans, weather factors, and container type distribution through time series analysis to obtain terminal operation load distribution;

[0012] The comparison module is used to compare the terminal operation load distribution with a preset load threshold, and when the predicted load exceeds the threshold, dynamically replan the AGV resource allocation, obtain the optimized scheduling instruction and execute the terminal AGV intelligent scheduling.

[0013] In a third aspect, an automated terminal AGV intelligent scheduling device is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the automated terminal AGV intelligent scheduling device executes the above-mentioned automated terminal AGV intelligent scheduling method.

[0014] In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, enables the computer to execute the above-mentioned automated terminal AGV intelligent scheduling method.

[0015] The technical solution provided in this application addresses the irrational resource allocation problem caused by static task allocation in existing technologies by establishing a multi-dimensional evaluation and processing mechanism based on the urgency of ship departure, container weight level, quay crane operation sequence, yard distance coefficient, and customer priority. The task priority evaluation matrix can dynamically adjust the execution order of tasks based on the actual terminal operations, avoiding the delay of urgent tasks caused by the traditional first-come, first-served policy. The construction of AGV state vectors and the prediction of load capacity based on historical port operation data overcome the shortcomings of existing technologies that ignore the dynamic state changes of AGVs. By comprehensively analyzing the AGV's current position coordinates, load status, remaining battery power, and expected completion time, it accurately estimates the AGV's future operating capacity, significantly improving the accuracy and executability of task allocation. The terminal multi-task chain scheduling algorithm, Port-MTCSA, achieves systematic optimization of the entire loading, unloading, transportation, and storage process through intelligent matching of the container task priority matrix and the AGV's dynamic load state vector. Compared with traditional segmented scheduling methods, the full-process task chain comprehensively considers the coordination of all operational links, reduces waiting time and resource idleness between operational links, and improves overall terminal operation efficiency.

[0016] The load forecasting process based on time series analysis establishes a forward-looking terminal load forecasting mechanism through a comprehensive analysis of ship arrival schedules, weather factors, and container type distribution. This addresses the existing problem of passively responding to load changes and enables AGV scheduling to proactively adjust resource allocation before peak loads arrive. The dynamic replanning process compares the terminal load distribution with a preset load threshold and automatically triggers the reallocation of AGV resources when the predicted load exceeds the threshold. This overcomes the lack of adaptive adjustment capabilities of traditional scheduling methods and achieves dynamic optimization of terminal AGV resources. Especially in the specific application field of intelligent AGV scheduling in automated terminals, the Port-MTCSA algorithm is specially designed for the particularities of the terminal's three-tiered operating structure: quay crane, AGV, and yard. The algorithm's three-tiered network modeling and dynamic programming recursive nature enable it to effectively handle the complex constraints of terminal operations. Compared with general vehicle routing problem-solving algorithms, Port-MTCSA significantly improves its computational efficiency and optimization effects in terminal environments. The introduction of time series analysis and load forecasting algorithms transforms scheduling decisions from passive response to active prevention. The overall solution achieves a technological leap from static scheduling to intelligent dynamic scheduling, providing key technical support for the efficient operation of modern automated terminals. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are some embodiments of the present invention. Those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0018] Figure 1 This is a schematic diagram of an embodiment of the method for intelligent scheduling of AGVs in automated terminals in an embodiment of the present application;

[0019] Figure 2 This is a schematic diagram of an embodiment of the automated terminal AGV intelligent dispatching system in the embodiment of the present application;

[0020] Figure 3 It is a schematic block diagram of the structure of the automated terminal AGV intelligent dispatching equipment in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] An embodiment of the present application provides an automated terminal AGV intelligent scheduling method and system. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products or devices.

[0022] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the intelligent scheduling method of the automated terminal AGV includes:

[0023] Step S101: Container task data is collected through the terminal operation management system, and a multi-dimensional evaluation process is performed on the container task data based on the urgency of the ship's departure, the container weight level, the quay crane operation sequence, the yard distance coefficient, and the customer priority to obtain a task priority evaluation matrix;

[0024] Step S102: construct an AGV state vector based on the task priority evaluation matrix, and perform load capacity prediction processing on the AGV's current position coordinates, load status, remaining power, and estimated completion time through training with historical port operation data to obtain an AGV load prediction result;

[0025] Step S103: Match the AGV load prediction result with the task priority evaluation matrix. Use the terminal multi-task chain scheduling algorithm Port-MTCSA to intelligently match the container task priority matrix and the AGV dynamic load state vector to obtain the full-process task chain of loading, unloading, transportation, and storage.

[0026] Step S104: historical terminal operation data is collected based on the entire process task chain of loading, unloading, transportation, and storage. Load forecasting is performed based on ship arrival plans, weather factors, and container type distribution through time series analysis to obtain terminal operation load distribution.

[0027] Step S105: Compare the terminal operation load distribution with the preset load threshold. When the predicted load exceeds the threshold, dynamically re-plan the AGV resource allocation, obtain the optimized scheduling instruction, and execute the terminal AGV intelligent scheduling.

[0028] It is understandable that the execution subject of this application can be the automated terminal AGV intelligent dispatching system, or it can be a terminal or a server, and the specific details are not limited here. The embodiment of this application is described by taking the server as the execution subject as an example.

[0029] Specifically, through the data interface with the port information platform, basic information such as container number, vessel voyage, quay crane operation plan, yard location, and estimated operation time is collected. This raw data is then structured and parsed, and data in different formats is uniformly converted into a standardized task dataset. The urgency coefficient is then calculated based on the difference between the ship's departure time and the current time. The specific calculation method is to convert the time difference into an urgency score, with the shorter the time, the higher the score. At the same time, the weight parameters in the container specification information are extracted, and 20-foot and 40-foot containers are classified and coded to form a weight grade coefficient matrix. Then, through a multi-dimensional weight allocation algorithm, the urgency score vector and weight grade coefficient matrix are weighted and fused with the quay crane operation sequence, yard distance coefficient, and customer priority to calculate the comprehensive priority value of each task and form a task priority assessment matrix.

[0030] Based on the task distribution in the task priority evaluation matrix, the real-time operating parameters of each AGV are collected, including the current position coordinates obtained by GPS positioning, the load status detected by the load sensor, the remaining power reported by the battery management system, and the estimated completion time calculated based on the current task. These four parameters are vectorized and encoded to form an AGV state vector. A correlation query is then performed with the terminal's historical operation database. AGV load change data from the past 30 days is filtered using a set time window. Feature extraction is performed on this historical data to analyze the load distribution patterns of AGVs in different yard areas. Load frequency calculation is used to calculate the load probability of 20-foot and 40-foot containers in each time period, generating a load probability distribution table. The current load status in the AGV state vector is matched with the load probability distribution table, and the difference between the AGV's remaining load capacity and the weight requirement of the task to be performed is analyzed to obtain the load matching coefficient. Finally, a correlation operation is performed with the AGV's estimated completion time, and load time series analysis is used to predict the load capacity trend of the AGV within the future time window.

[0031] The AGV load prediction results are fused with the task priority evaluation matrix to calculate the degree of match between the container task weight requirement and the AGV load capacity, generating an initial task-AGV matching table. Based on this matching table, a bipartite graph structure is constructed. The terminal multi-task chain scheduling algorithm, Port-MTCSA, decomposes terminal operations into three levels: quay crane loading and unloading nodes, AGV transportation nodes, and yard storage nodes. The connections between these three levels are modeled to form a network topology graph for the terminal operations. Node connections are extracted from the network topology graph and the complete container operation process from the quay crane to the yard is chain-orchestrated in the order of loading, unloading, transportation, and storage, generating multiple candidate task chains. Using a dynamic programming recursive method, the Port-MTCSA algorithm calculates the cumulative cost of each task chain from its starting state to its final state, including time cost, load cost, and distance cost. The optimal task chain score is determined by minimizing the path cost. The candidate task chains are ranked and screened based on the optimization score, and a greedy selection strategy is used to prioritize high-scoring task chains. Ultimately, the optimal matching between the container task priority matrix and the AGV dynamic load state vector is determined.

[0032] Based on the time node information in the entire loading, unloading, transportation, and storage process, data such as ship arrival times, operation completion times, and AGV operation trajectories for the past 90 days were extracted from the terminal's historical operation database. This historical data was arranged in a temporal sequence, and time series analysis was used to identify the cyclical and seasonal variations in ship arrival schedules, thereby establishing a ship arrival time series model. Weather forecast data and historical weather records were also collected to analyze the correlation between wind speed, rainfall, visibility, and terminal operation efficiency. A weather impact coefficient matrix was calculated through correlation analysis. Container transportation records were extracted from the historical operation dataset, and the distribution of 20-foot and 40-foot containers in different time periods was statistically analyzed to determine the container type distribution pattern. The ship arrival time series pattern, weather impact coefficient matrix, and container type distribution pattern were integrated into a model, and a load forecasting algorithm was used to calculate the task density distribution of each operation area within the future time window.

[0033] The load values ​​of each operating area in the terminal's load distribution are compared with the preset load thresholds. When the load value of a certain area exceeds the threshold, it is marked and identified to form a list of overloaded operating areas. Based on the location information and the degree of load excess of the overloaded areas, the current distribution status of AGV resources in each area is analyzed, the AGV quantity gap and surplus in each area are calculated, and an AGV resource reallocation demand matrix is ​​generated. The dynamic replanning algorithm calculates the optimal transfer path and number of AGVs from low-load areas to high-load areas based on the demand matrix, taking into account the AGV movement time cost and fuel consumption, and optimizes the resource adjustment plan. The AGV resource adjustment plan is converted into a specific scheduling instruction format that includes task allocation, path planning, and operation timing. The optimized scheduling instructions are issued to each AGV through the terminal's 5G communication network to perform intelligent scheduling.

[0034] For example, a terminal has three cargo ships docking simultaneously. Ship A's departure time is 6 hours later, Ship B's is 12 hours later, and Ship C's is 24 hours later. Based on the time urgency calculation, Ship A's tasks receive the highest priority. Currently, 15 AGVs are operating: 5 are unloaded in Yard 1, 8 are loading 40-foot containers in the quay crane area, and 2 are charging in Yard 2. Load prediction analysis indicates that unloaded AGVs are suitable for new transport tasks, while those currently loading must complete their current tasks first. The Port-MTCSA algorithm prioritizes Ship A's urgent tasks for the unloaded AGVs and simultaneously plans the shortest path from the quay crane to the designated yard. Time series analysis reveals that the afternoon is typically peak operating time. Combined with the weather forecast indicating no severe weather, the load in the quay crane area is predicted to reach the upper threshold within the next two hours. The dynamic replanning algorithm decided to deploy two fully charged AGVs to the quay crane area and pre-deploy one empty AGV to Yard No. 2 to prepare for the task. This predictive scheduling avoided the occurrence of operational bottlenecks.

[0035] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0036] The container number, vessel information, quay crane operation plan, yard location, and estimated operation time are collected through the data interface of the terminal operation management system. The container task data is then structured and parsed to obtain a standardized task data set.

[0037] The difference between the ship departure time in the standardized task dataset and the current time is calculated, and the urgency coefficient is quantified based on the urgency of the ship departure time to obtain the urgency score vector;

[0038] The weight parameters are extracted based on the container specification information in the standardized task dataset, and the load levels of 20-foot containers and 40-foot containers are classified and coded to obtain the weight level coefficient matrix.

[0039] The urgency score vector and the weight grade coefficient matrix are weighted and fused, and the quay crane operation sequence, yard distance coefficient and customer priority are comprehensively calculated and processed through a multi-dimensional weight allocation algorithm to obtain the task priority evaluation matrix.

[0040] Specifically, the terminal operations management system obtains raw data from the port information platform through a standardized data interface. This data includes multi-source heterogeneous data, including container numbers, vessel voyage information, quay crane operation plans, yard location coordinates, and estimated operation times. Structured parsing processing first converts the data in different formats, converting XML-formatted vessel information, JSON-formatted quay crane plans, and database-formatted yard data into a standardized data structure. The parsing process extracts key fields, converting container numbers into unique identifiers, parsing vessel information into attributes such as ship name, voyage number, and estimated departure time, parsing quay crane operation plans into operation sequence numbers, assigned quay crane numbers, and estimated loading and unloading times, and parsing yard location into specific yard area numbers and storage location coordinates. The standardized task dataset organizes all task information into unified fields, such as task ID, vessel ID, container specifications, starting location, destination location, and time window, forming a structured data table. Each row represents a specific container transport task, and each column represents an attribute dimension of that task.

[0041] The urgency coefficient quantification process relies on two core steps: time difference calculation and urgency mapping. First, the ship departure time corresponding to each task is extracted from the standardized task dataset. The difference between the departure time and the current time is calculated to obtain the remaining time. The remaining time is then quantified in hours. The urgency coefficient is then assigned using a piecewise function. Tasks with a remaining time between 0 and 6 hours are assigned an urgency coefficient between 0.9 and 1.0, between 6 and 12 hours to 0.7 and 0.9, between 12 and 24 hours to 0.5 and 0.7, and over 24 hours to 0.3 and 0.5. The quantification is performed using linear interpolation, calculating the precise urgency coefficient for each time period based on the remaining time. Linear interpolation determines the upper and lower bounds of the time period and the corresponding coefficient range, then calculates the exact coefficient value in a proportional manner. The urgency score vector arranges the urgency coefficients of all tasks in order, forming a one-dimensional array structure. Each element in the vector corresponds to the urgency coefficient value of a task, and the length of the vector equals the total number of tasks.

[0042] Weight parameter extraction begins with the container specification information field in the standardized task dataset, which contains container dimensions and load information. Classification and coding processes categorize containers into two primary categories: 20-foot and 40-foot containers. Each category is further subdivided based on load capacity. 20-foot containers are divided into three weight classes: light, medium, and heavy, corresponding to load ranges of 0 to 10 tons, 10 to 15 tons, and 15 to 24 tons, with code values ​​of 1, 2, and 3, respectively. 40-foot containers are similarly divided into three weight classes: 0 to 15 tons, 15 to 25 tons, and 25 to 30 tons, with code values ​​of 4, 5, and 6, respectively. The weight class coefficient matrix is ​​a two-dimensional matrix, with rows representing task numbers and columns representing weight class attributes. The matrix values ​​represent the corresponding weight class codes. During matrix construction, the weight class code for each task is determined based on the container specification information, and the code value is entered into the corresponding position in the matrix. The number of rows in the matrix equals the total number of tasks, and the number of columns is fixed to one to store the weight class codes.

[0043] The multi-dimensional weight allocation algorithm performs a weighted fusion calculation on the urgency score vector, the weight grade coefficient matrix, the quay crane operation sequence, the yard distance coefficient, and the customer priority. The quay crane operation sequence is numbered according to the order of the quay crane's operation plan. The task sequence with the earlier number has a higher weight. The sequence weight is calculated using an inverse relationship. The sequence weight of the first task is 1.0, the second is 0.5, the third is 0.33, and so on. The yard distance coefficient is obtained by calculating the straight-line distance from the quay crane position to the target yard position. The closer the distance, the higher the weight. The distance weight is calculated using an inverse proportional relationship, specifically 1000 divided by the distance in meters and then divided by 1000 to obtain a standardized weight value. Customer priority is comprehensively assessed based on factors such as the customer's VIP level, years of cooperation, and business volume, and is divided into three levels: high, medium, and low, with corresponding weight coefficients of 1.0, 0.7, and 0.4, respectively. The weight fusion calculation uses a linear weighted summation method and is calculated using the following formula:

[0044] M=G×0.3+H×0.2+J×0.2+K×0.15+L×0.15+O×0.15, where M is the comprehensive priority score, G is the urgency coefficient, H is the weight level coefficient, J is the job sequence coefficient, K is the distance coefficient, and O is the customer priority coefficient. Based on the comprehensive priority score of each task, the task priority evaluation matrix organizes the comprehensive priority scores of all tasks in matrix form, with rows representing tasks and columns representing priority scores. The values ​​in the matrix reflect the importance of each task in scheduling.

[0045] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0046] According to the task distribution in the task priority evaluation matrix, the real-time operating parameters of the AGV are collected, and the current position coordinates, load status, remaining power and estimated completion time of the AGV are vectorized and encoded to obtain the AGV state vector;

[0047] The AGV state vector is correlated with the terminal's historical operation database. The AGV load change data over the past 30 days is filtered through a time window to extract features and obtain a historical load feature sample set.

[0048] Based on a historical load characteristic sample set, the load distribution of AGVs in different yard areas was statistically analyzed. The load probability of 20-foot and 40-foot containers was quantified by load frequency calculation to obtain a load probability distribution table.

[0049] According to the matching calculation between the current load state in the AGV state vector and the load probability distribution table, the difference analysis of the remaining load capacity of the AGV and the weight requirement of the task to be executed is performed to obtain the load matching coefficient;

[0050] The load matching coefficient is correlated with the estimated completion time of the AGV, and the load capacity change trend of the AGV in the future time window is predicted through load time series analysis to obtain the AGV load prediction result.

[0051] Specifically, the real-time operating parameters of the AGV are collected based on the task distribution in the task priority evaluation matrix. The real-time status information of each AGV is obtained through the communication interface with the AGV's onboard control system. The current position coordinates are obtained through the fusion positioning of the GPS positioning module and the terminal's internal positioning system. The load status is detected in real time by the onboard weighing sensor to detect the weight of the currently carried container. The remaining battery power percentage is read by the battery management system. The estimated completion time is calculated based on the remaining path of the currently executed task and the AGV's standard driving speed. Vectorized encoding processing converts these four real-time parameters into numerical vector form. The position coordinates are represented by the X-axis and Y-axis values ​​of the terminal coordinate system. The load status is represented by the ratio of the current load to the AGV's maximum load capacity. The remaining battery power is expressed as a decimal between 0 and 1. The estimated completion time is expressed as an integer in minutes. The AGV state vector arranges these four values ​​in a fixed order to form a four-dimensional vector, with each component of the vector corresponding to a state parameter.

[0052] The AGV state vector is associated with the terminal's historical operation database for query. A time window filtering mechanism is used to extract AGV load change records from the historical database over the past 30 days. This time window filtering performs range queries based on the timestamp field of the data records, with the filtering condition being that the record time is greater than the current time minus 30 days and less than the current time. Feature extraction extracts key features such as AGV number, timestamp, location information, load status, and task type from the filtered historical records. Feature extraction uses attribute selection methods from data mining and determines the most predictive feature dimensions based on correlation analysis of load prediction. The historical load feature sample set organizes the extracted feature data into a two-dimensional sample-feature matrix. Each row represents an AGV state sample at a historical moment, and each column represents a feature dimension. The size of the sample set depends on the total number of data records and the number of feature dimensions within 30 days.

[0053] Based on a historical load feature sample set, a statistical analysis of the load distribution of AGVs in different yard areas was conducted. The statistical analysis used a grouping statistical method, grouping historical samples by yard area. Each yard area contains the historical load records of all AGVs in that area. The load frequency calculation counts the number of occurrences and load distribution of 20-foot and 40-foot containers in each yard area. During the frequency calculation process, 20-foot containers were classified and counted according to three levels: light load, medium load, and heavy load. 40-foot containers were also classified according to three load levels. The load probability quantification process divides the number of occurrences of each level by the total number of load records in the area to obtain a probability value. The load probability distribution table uses a three-dimensional table structure. The first dimension is the yard area, the second dimension is the container type, and the third dimension is the load level. The values ​​in the table are the corresponding probability values.

[0054] A matching calculation is performed based on the current load state in the AGV state vector and the load probability distribution table. The matching calculation first determines the yard area where the AGV is located based on its current position. The load probability distribution of this area is then queried from the load probability distribution table. Difference analysis is performed to calculate the difference between the AGV's remaining load capacity and the weight requirement of the task to be performed. The remaining load capacity is equal to the AGV's maximum load capacity minus the current load capacity. The weight requirement of the task to be performed is obtained by extracting the container weight information of the corresponding task from the task priority evaluation matrix. The load matching coefficient is obtained by normalizing the ratio of the remaining load capacity to the task weight requirement. When the remaining load capacity is greater than the task weight requirement, the matching coefficient is close to 1. When the remaining load capacity is insufficient, the matching coefficient is close to 0. The matching calculation uses a sigmoid function for nonlinear mapping to convert the difference into a matching coefficient between 0 and 1.

[0055] The load matching coefficient is correlated with the AGV's expected completion time. This correlation establishes a composite evaluation model for load matching and time factors. Load timing analysis uses a time series prediction method to analyze how the AGV's load capacity changes over time. Time series analysis is based on time series data from a historical load feature sample set, and uses a moving average method to identify trends and cyclical patterns in load changes. Prediction processing calculates the load state of the AGV after completing the current task within a future time window. The prediction calculation considers load changes during task execution, including increases when loading containers and decreases when unloading containers. The AGV load prediction results include the expected load state and load capacity availability at each future time point. The prediction results are stored in a two-dimensional array of time-load state, providing load capacity prediction information for subsequent task allocation decisions.

[0056] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0057] The AGV load prediction results are fused with the task priority evaluation matrix, and the matching degree between the container task weight requirement and the AGV load capacity is calculated to obtain the task-AGV initial matching table.

[0058] Based on the task-AGV initial matching table, a bipartite graph structure is constructed. The terminal multi-task chain scheduling algorithm Port-MTCSA is used to perform three-layer network modeling of the quay crane loading and unloading nodes, AGV transportation nodes, and yard storage nodes to obtain the terminal operation network topology.

[0059] The task chain path is extracted based on the node connection relationship in the terminal operation network topology diagram. The container loading, unloading, transportation and storage operation sequence from the quay crane to the yard is chain-orchestrated to obtain a set of candidate task chains.

[0060] The time constraints and load constraints in the candidate task chain set are cumulatively calculated, and the path cost from the starting state to the ending state of the task chain is minimized through dynamic programming recursion to obtain the task chain optimization score;

[0061] The candidate task chains are sorted and screened based on the task chain optimization score, and the high-scoring task chains are prioritized through a greedy selection strategy to obtain the matching result between the container task priority matrix and the AGV dynamic load state vector.

[0062] The matching results of the container task priority matrix and the AGV dynamic load state vector are verified, and the load conflicts and time conflicts in the task chain are detected and processed to obtain the full-process task chain of loading, unloading, transportation and storage.

[0063] Specifically, the AGV load prediction results are fused with the task priority evaluation matrix. The data fusion process first correlates the AGV load capacity data in the load prediction results with the task weight requirement data in the task priority evaluation matrix. The fusion algorithm uses matrix operations to perform a Cartesian product between the AGV load capacity matrix and the task weight requirement matrix to generate all possible AGV-task combinations. The matching degree calculation process evaluates the load compatibility of each AGV-task combination by dividing the AGV's remaining load capacity by the task weight requirement to obtain a load matching ratio. A ratio greater than or equal to 1 indicates sufficient AGV load capacity, while a ratio less than 1 indicates insufficient load capacity. The matching degree is converted to a value between 0 and 1 by performing a logarithmic transformation and normalization on the ratio. The initial task-AGV matching table is constructed in a two-dimensional table format, with rows representing task numbers and columns representing AGV numbers. The values ​​in the table represent the corresponding matching degree scores. During the table construction process, all calculated matching degree scores are entered into the corresponding positions according to the corresponding relationship between tasks and AGVs.

[0064] Based on the initial task-AGV matching table, a bipartite graph is constructed. A bipartite graph is a special graph structure that divides nodes into two disjoint sets: one set contains all task nodes, and the other contains all AGV nodes. Edges connect only nodes between the two sets. The terminal multi-task chain scheduling algorithm, Port-MTCSA, uses a three-layer network model for quay crane loading and unloading nodes, AGV transport nodes, and yard storage nodes. This three-layer network structure decomposes the terminal operation process into three levels: the quay crane loading and unloading node layer contains all quay crane loading and unloading operation locations, the AGV transport node layer contains all AGV transport path nodes, and the yard storage node layer contains all yard storage locations. The network modeling process constructs a network topology by defining node attributes and connection relationships. Node attributes include parameters such as location coordinates, capacity constraints, and operation time. Connection relationships are determined by calculating the distance and connectivity between nodes. The terminal operation network topology diagram graphically represents all nodes and connection relationships in the three-layer network. Each node in the diagram represents a work location, and each edge represents a feasible transportation path.

[0065] Task chain paths are extracted based on the node connectivity in the terminal operation network topology. Path extraction uses a graph traversal algorithm to search for all possible paths from the quay crane node to the yard node. The path search considers node connectivity and capacity constraints, eliminating unreachable path combinations. Chain orchestration arranges the container operation process from the quay crane to the yard according to the loading, unloading, transportation, and storage time sequence. The loading and unloading phase includes operations such as unloading containers from the ship to the quay crane. The transportation phase involves the AGV transporting containers from the quay crane to the yard. The storage phase involves storing containers at designated yard locations. The candidate task chain set contains all task chain paths that meet the requirements of the operation process. Each task chain includes a sequence of operation nodes and a corresponding AGV allocation plan. The size of the set depends on the number of feasible paths and the availability of AGVs.

[0066] The time constraints and load constraints in the candidate task chain set are cumulatively calculated. The time constraints include the operation time of each operation node and the movement time of the AGV between nodes. The load constraints include the load status and load capacity limit of the AGV at each node. The cumulative cost calculation adopts the path cost accumulation method. Starting from the starting node of the task chain, the operation cost of each node and the movement cost between nodes are calculated one by one, and all costs are accumulated to obtain the total cost of the task chain. Dynamic programming recursion is an optimization algorithm that solves complex problems by decomposing them into sub-problems. During the recursive process, a state transition table is maintained to record the optimal solution of each sub-problem. The state transition equation defines the cost calculation rule from one state to another. The minimization process selects the path with the lowest cost as the optimal solution by comparing the cumulative costs of all candidate paths. The task chain optimization score is expressed in the form of the inverse or negative number of the cost. The lower the cost, the higher the score.

[0067] Candidate task chains are sorted and screened based on their optimized scores. The sorting algorithm uses efficient sorting methods such as quick sort or merge sort, arranging task chains in descending order of score. A greedy selection strategy is a heuristic optimization method that selects the task chain with the highest score in the current state for each assignment. The core concept of the greedy strategy is that local optimal selection can lead to a global optimal solution. Prioritized allocation assigns high-scoring task chains based on the sorting results. The allocation process checks AGV availability and task time window constraints to ensure the feasibility of the allocation scheme. The matching results of the container task priority matrix and the AGV dynamic load state vector record the final assignment relationship and corresponding score information for each task and AGV. The matching results of the container task priority matrix and the AGV dynamic load state vector are then verified to check the rationality and feasibility of the allocation scheme. Load conflict detection verifies whether the AGV's load capacity meets the weight requirements of the assigned task. Time conflict detection checks whether the task schedules overlap or violate time window constraints. Conflict detection and processing adopts the constraint satisfaction problem solving method to identify potential conflicts by checking the satisfaction of all constraints. When a conflict is detected, the reallocation mechanism is triggered to adjust the task allocation plan. The full-process task chain of loading and unloading-transportation-storage is the verified and optimized final scheduling plan, which includes the complete operation process of each container task and the corresponding AGV allocation plan.

[0068] In a specific embodiment, the process of performing three-layer network modeling processing on the quay crane loading and unloading nodes, AGV transport nodes, and yard storage nodes using the terminal multi-task chain scheduling algorithm Port-MTCSA can specifically include the following steps:

[0069] Node information is extracted based on the task allocation relationship in the task-AGV initial matching table, and spatial coordinate mapping is performed on the quay crane loading and unloading positions, AGV transportation paths, and yard storage locations to obtain a three-layer node coordinate set.

[0070] The three-layer node coordinate set is divided into layers according to the terminal operation process. The node connection relationship of the quay crane layer, transportation layer and yard layer is topologically constructed and processed using the terminal multi-task chain scheduling algorithm Port-MTCSA to obtain a hierarchical network structure.

[0071] A connection weight matrix is ​​constructed based on the node attribute information in the hierarchical network structure. The distance weights from the quay crane loading and unloading node to the AGV transport node and the distance weights from the AGV transport node to the yard storage node are quantified and calculated to obtain a node connection weight table.

[0072] The node connection weight table is correlated with the container transport capacity constraint. The transport path weights of 20-foot and 40-foot containers are assigned differently using the Port-MTCSA algorithm to obtain the capacity constraint weight matrix.

[0073] The three-layer network structure is finally constructed based on the capacity constraint weight matrix. The connectivity of the quay crane loading and unloading nodes, AGV transportation nodes and yard storage nodes is verified through the graph theory algorithm to obtain the terminal operation network topology.

[0074] Specifically, node information is extracted based on the task assignment relationships in the initial task-AGV matching table. The node information extraction process identifies all involved quay crane locations, key points on the AGV transport path, and yard storage locations from the matching table. The extraction algorithm traverses each task-AGV combination in the matching table, obtaining the starting quay crane location and target yard location from the task information, and the current location and reachable path nodes from the AGV information. Spatial coordinate mapping converts the actual locations in the terminal's physical layout into a digital coordinate system. The quay crane loading and unloading locations are assigned coordinates based on the physical distribution of the terminal's shoreline. Each quay crane position corresponds to a two-dimensional coordinate point. The AGV transport path contains all feasible path nodes connecting the quay crane and the yard. The path node coordinates are determined based on the actual layout of the terminal road network. The yard storage locations are mapped based on the yard area division and container storage locations. The three-layer node coordinate set divides all extracted nodes into three sets according to their functional types. The quay crane layer set contains the coordinates of all quay crane loading and unloading positions, the transport layer set contains the coordinates of all AGV path nodes, and the yard layer set contains the coordinates of all yard storage locations. Each set is arranged in order according to the node number, and the sets are associated through hierarchical relationships.

[0075] The three-layer node coordinate set is hierarchically divided according to the terminal operation process. This hierarchical division follows the natural process sequence of container operations. The quay crane layer, as the first layer, is responsible for container loading and unloading operations; the transport layer, as the middle layer, is responsible for container transportation operations; and the yard layer, as the final layer, is responsible for container storage operations. The terminal multi-task chain scheduling algorithm, Port-MTCSA, constructs a topological connection relationship between the nodes in the three layers. This topology construction process defines the relationships between nodes within a layer and the connection rules between nodes in the layers. Nodes within the quay crane layer are connected based on the adjacency of quay cranes, nodes within the transport layer are connected based on the connectivity of paths, and nodes within the yard layer are connected based on the adjacency of yard areas. Inter-layer connections are determined by analyzing the reachability of the operation process. Connections between quay crane nodes and transport nodes are based on the reachable paths of AGVs from the quay crane locations, and connections between transport nodes and yard nodes are based on the feasible paths of AGVs to the yard locations. The hierarchical network structure is represented by a graph data structure. The node set contains all nodes in the three layers, and the edge set contains all intra-layer and inter-layer connection relationships. Each node in the network structure carries attribute information such as location coordinates, capacity constraints, and operation time.

[0076] A connection weight matrix is ​​constructed based on node attribute information in the hierarchical network structure. The weight matrix is ​​a two-dimensional array structure with rows and columns corresponding to nodes in the network. Matrix elements represent the connection weights between corresponding nodes, reflecting the cost or priority of the connection between nodes. Distance weights are quantified using the Euclidean distance formula to calculate the spatial distance between nodes. The distance from a quay crane loading and unloading node to an AGV transport node is calculated as the straight-line distance between the two coordinates. The distance from an AGV transport node to a yard storage node is also determined using coordinate calculations. Shorter distances result in larger weights, indicating a higher connection priority. The weight calculation considers the physical constraints and operational limitations of the nodes. The weight calculation for quay crane nodes incorporates the crane's operational capacity and current load status; the weight calculation for AGV transport nodes incorporates the path's capacity and traffic conditions; and the weight calculation for yard nodes incorporates storage capacity and current occupancy. A node connection weight table organizes all calculated weights according to node correspondence. This table format facilitates querying and modifying weight information. The weight table update mechanism dynamically adjusts weights based on the terminal's real-time operational status.

[0077] The node connection weight table is correlated with container transport capacity constraints, including AGV load capacity limits and route capacity limits. This correlation analysis identifies dependencies between weight values ​​and capacity constraints, adjusting weight values ​​accordingly when capacity constraints change. The Port-MTCSA algorithm assigns differentiated weights to transport routes for 20-foot and 40-foot containers. This differentiation is based on the weight difference and transport difficulty of the two types of containers. 20-foot containers are lighter and easier to transport, resulting in a higher path weight, while 40-foot containers are heavier and more difficult to transport, resulting in a lower path weight. This differentiation is achieved by multiplying the base weight by a container type coefficient. The type coefficient for 20-foot containers is greater than 1, while the type coefficient for 40-foot containers is less than 1. The specific value of the coefficient is determined by the weight ratio and transportation cost ratio of the two containers. The capacity constraint weight matrix incorporates the influence of capacity constraints and container type on the original weight matrix. The matrix element calculation incorporates distance weight, capacity weight, and type weight. The matrix dimension corresponds to the number of nodes, and the matrix symmetry is determined by the directionality of the connections between nodes. The three-layer network structure is finally constructed based on the capacity constraint weight matrix. Weight information is integrated into the network's edge attributes during network construction. The weight value of each edge is derived from the capacity constraint weight matrix. Once the network is constructed, a terminal operation network model is formed. Graph theory algorithms verify the network's connectivity. Connectivity verification checks whether a reachable path exists between any two nodes. The verification algorithm uses a depth-first search or breadth-first search to traverse all nodes and edges, confirming the network's integrity and consistency. Verification includes both strong and weak connectivity checks. Strong connectivity requires the existence of a bidirectional reachable path between any two nodes, while weak connectivity requires a connected path between any two nodes by treating the directed graph as an undirected graph. Appropriate connectivity requirements are selected based on the characteristics of the terminal operation process. The terminal operation network topology is the final, verified network structure. The graph contains the spatial layout and connectivity of all operation nodes. Node attributes include location coordinates, operational capabilities, and capacity constraints, while edge attributes include distance weights, capacity weights, and type weights. The topology provides the network foundation for subsequent path planning and task allocation.

[0078] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0079] The terminal's historical operation data is collected based on the time node information in the entire process of loading, unloading, transportation, and storage. The ship arrival time, operation completion time, and AGV operation trajectory in the past 90 days are extracted and processed to obtain a historical operation data set.

[0080] The historical operation data set is arranged in sequence according to the time dimension. The periodic laws and seasonal changes of ship arrival plans are analyzed through time series analysis to obtain the ship arrival time series pattern.

[0081] Weather forecast data and historical weather records were collected based on the ship arrival time series model. Correlation analysis was performed on the correlation between wind speed, rainfall, visibility and terminal operation efficiency to obtain the weather impact coefficient matrix.

[0082] Extract type distribution information based on container transportation records in historical operation data sets, perform statistical calculations on the quantity distribution of 20-foot and 40-foot containers in different time periods, and obtain the container type distribution pattern;

[0083] The ship arrival time series pattern, weather impact coefficient matrix and container type distribution law are comprehensively modeled. The task density of each operation area in the future time window is predicted and calculated through the load prediction algorithm to obtain the terminal operation load distribution.

[0084] Specifically, historical terminal operation data is collected based on time node information within the entire loading, unloading, transportation, and storage process. This time node information includes the start, end, and duration of each operation. The data collection process queries the terminal operation management database for all operation records within the past 90 days. The query criteria are range-filtered based on the timestamp field to ensure data integrity and timeliness. Data extraction is performed based on three key dimensions: ship arrival time, operation completion time, and AGV trajectory. Ship arrival time is derived from port scheduling records, representing the actual arrival timestamp of each ship. Operation completion time is derived from the quay crane and yard operation logs, representing the completion time of each container task. AGV trajectory information is obtained from the historical records of the vehicle's GPS system, representing the movement path and time of each AGV. The historical operation dataset is organized in a structured data table format. Each row represents a complete operation event and includes fields such as timestamp, operation type, location information, equipment number, and task status. The size of the dataset depends on the frequency of operations within the 90-day period and the level of record detail.

[0085] The historical job dataset is sequenced along the time dimension. The sequence arrangement process sorts all job records in chronological order by timestamp, forming a time series data structure. Time series data is a collection of data points arranged in chronological order, with each data point containing a timestamp and corresponding numerical or status information. Time series analysis uses statistical methods to identify regular patterns in the data. The analysis process includes trend analysis, periodicity analysis, and seasonality analysis. Trend analysis identifies the long-term direction of data change, periodicity analysis identifies repetitive patterns in the data, and seasonality analysis identifies the periodic fluctuation characteristics of the data. Pattern recognition processing uses autocorrelation analysis and spectral analysis to detect the regularity of ship arrival times. Autocorrelation analysis calculates the correlation of time series at different time intervals, and spectral analysis uses Fourier transforms to identify periodic components in the data. Ship arrival time series patterns include characteristic parameters such as cycle length, fluctuation amplitude, and peak time. These parameters reflect the temporal regularity and changing trends of ship arrivals.

[0086] Weather forecast data and historical weather records are collected based on a time-series model of ship arrivals. Weather data collection uses a meteorological data interface to obtain daily weather information for the past 90 days and a weekly forecast. This information includes meteorological parameters such as wind speed, rainfall, visibility, temperature, and humidity. Correlation analysis uses the Pearson correlation coefficient method to analyze the linear relationship between weather factors and terminal operating efficiency. The correlation coefficient is calculated based on paired samples of historical weather data and operating efficiency data. The calculation uses weather parameters as independent variables and operating efficiency indicators as dependent variables, and the correlation coefficient is calculated through statistical calculation. Correlation analysis identifies which weather factors have the most significant impact on operating efficiency. The degree of impact is determined by the absolute value of the correlation coefficient. A correlation coefficient close to 1 indicates a strong positive correlation, close to -1 indicates a strong negative correlation, and close to 0 indicates no correlation. A weather impact coefficient matrix quantifies the impact of various weather factors on different operating steps. The rows of the matrix represent the type of weather factor, the columns represent the type of operating step, and the matrix elements are the corresponding impact coefficient values. The coefficient values ​​reflect the magnitude of the change in operating efficiency under specific weather conditions.

[0087] Type distribution information is extracted from container transport records in historical operation datasets. The type distribution information extraction process identifies the specification type of each container from the operation records, including two main categories: 20-foot and 40-foot containers. The extraction algorithm traverses all transport records and counts the number of containers of different types. The statistical calculation process groups the extracted type information by time period, using different granularities such as hours, days, and weeks. Statistics include indicators such as the number, proportion, and trend of each container type within each time period. Quantity distribution calculation uses counting to count the absolute number of 20-foot and 40-foot containers in each time period. Ratio distribution calculation uses division to determine the percentage of each container type in the total quantity. Trend calculation compares quantity differences between adjacent time periods to identify increases or decreases. Container type distribution patterns include temporal distribution characteristics, quantity ratio characteristics, and periodicity characteristics. Temporal distribution characteristics reflect the distribution patterns of different container types within a day or week. Quantity ratio characteristics reflect the relative quantity relationship between two container types. Periodicity characteristics reflect the cyclical changes in container type distribution. The ship arrival time series pattern, weather impact coefficient matrix, and container type distribution pattern are integrated into a model. The integrated modeling process integrates these three different data patterns to establish a multi-factor prediction model. The load prediction algorithm is a prediction method based on historical data and influencing factors. The algorithm inputs include time series pattern parameters, weather impact coefficients, and type distribution pattern parameters. The algorithm outputs a load forecast for a future time window. The prediction calculation uses a weighted summation method, linearly combining the time series forecast, weather impact correction, and type distribution forecast. The weight coefficients are set based on the historical forecast accuracy of each factor. The calculation formula is: the predicted load equals the time series forecast multiplied by the time weight plus the weather correction multiplied by the weather weight plus the type forecast multiplied by the type weight. Task density prediction is calculated separately for different terminal operation areas. The forecast value for each area is calculated independently based on the historical load data and influencing factors. The terminal operation load distribution includes the expected task density values ​​for each operation area within the future time window. The load distribution data provides load forecast information for AGV intelligent scheduling decisions, addressing the technical problem of the lack of predictive scheduling capabilities in existing technologies.

[0088] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0089] Compare the load value of each operating area in the terminal operation load distribution with the preset load threshold one by one, mark and identify the areas that exceed the threshold, and obtain a list of overloaded operating areas;

[0090] Based on the area location information and load excess degree in the overloaded operation area list, the current AGV resource allocation status in each area is analyzed and evaluated to obtain the AGV resource reallocation demand matrix;

[0091] According to the AGV resource reallocation demand matrix, the dynamic replanning algorithm is called to optimize the transfer path and transfer quantity of AGVs from low-load areas to high-load areas, and the AGV resource adjustment plan is obtained.

[0092] The AGV resource adjustment plan is converted into a scheduling instruction format, and the task allocation, path planning and operation sequence of each AGV are re-issued through the terminal communication network to obtain optimized scheduling instructions and execute terminal AGV intelligent scheduling.

[0093] Specifically, the load value of each operating area in the terminal's load distribution is compared against a preset load threshold. This load comparison process uses a numerical comparison algorithm to compare the current load value of each operating area with the corresponding threshold. This comparison is performed by traversing each element in the load distribution array, checking each element individually. When the load value exceeds the threshold, an overload flag is triggered. The preset load threshold is a critical value determined based on the terminal's designed capacity and historical operational data. The threshold setting takes into account factors such as quay crane operating capacity, yard storage capacity, and AGV traffic capacity. Different operating areas have different threshold parameters. The threshold for quay crane areas is based on the maximum operating efficiency of the quay cranes, the threshold for yard areas is based on storage capacity and stacking height restrictions, and the threshold for transport lanes is based on AGV traffic density and traffic flow. A flag recognition process marks areas where the comparison result is true. The flag information includes attributes such as area number, load excess magnitude, and load excess time. A list of overloaded operating areas uses a list data structure to store all marked areas. Each entry in the list contains the area ID, load excess degree, and urgency level. A dynamic update mechanism for the list adds or removes area entries based on real-time load changes.

[0094] Based on the regional location information and load excess degree in the overloaded operation area list, the analysis and evaluation process first extracts the geographic location coordinates and load excess value of each overloaded area from the list. The location information is used to calculate the distance relationship and connectivity between areas, and the load excess degree is used to determine the priority and allocation quantity of resource allocation. The current AGV resource allocation status analysis obtains the real-time location, task status and load status of each AGV by querying the AGV management database. The analysis algorithm counts the number of AGVs distributed in each operation area, the number of idle AGVs and the number of loaded AGVs, and calculates the AGV resource density and utilization rate of each area. The evaluation process uses a load balancing analysis method to calculate the load difference between each area and the degree of AGV uneven distribution. The load difference is quantified by the standard deviation calculation method to quantify the discrete degree of load distribution between areas. The degree of AGV uneven distribution is evaluated by comparing the deviation of the AGV density in each area from the ideal density. The AGV resource reallocation demand matrix uses a two-dimensional matrix structure to represent the resource allocation demand between regions. The rows of the matrix represent the source region, and the columns represent the target region. The matrix element values ​​represent the number of AGVs required to be allocated from the source region to the target region. The demand calculation is based on the load gap of the target region and the resource redundancy of the source region.

[0095] The dynamic replanning algorithm is invoked based on the AGV resource reallocation demand matrix. This optimization algorithm decomposes the complex resource allocation problem into multiple subproblems. The core concept of the algorithm is to construct the optimal solution to the original problem using the optimal solutions to the subproblems. The algorithm inputs include the demand matrix, the current AGV status information, and terminal path network data. The algorithm outputs the optimal AGV deployment plan, including the deployment path and deployment sequence for each AGV. The optimization calculation uses a shortest path algorithm to calculate the optimal transfer path for AGVs from low-load areas to high-load areas. Path optimization considers factors such as distance cost, time cost, and traffic congestion. Transfer quantity optimization uses integer programming to determine the optimal number of AGVs for each deployment direction, ensuring a balanced load distribution across all areas after deployment. Path calculation uses the Dijkstra algorithm or the A-star algorithm to search for the shortest path from the source node to the destination node. The algorithm maintains a distance table recording the shortest distance to each node and continuously updates the distance value through relaxation operations until the optimal path is found. The AGV resource adjustment plan contains specific deployment instructions for each deployed AGV. The instructions include the target area, transfer path, estimated arrival time and new task assignment. The feasibility verification of the plan checks the capacity and time window constraints of the deployment path.

[0096] The AGV resource adjustment plan is converted into a dispatch instruction format. The instruction format conversion process converts the abstract dispatch information in the adjustment plan into specific operational instructions that the AGV can execute. This conversion process includes path coordinate conversion, speed parameter setting, and task sequence arrangement. The dispatch instruction format uses a standardized data structure, including fields such as instruction type, target location, execution time, and priority. The instruction encoding follows the format standard specified by the terminal communication protocol to ensure accurate transmission and parsing of instructions. The terminal communication network is the data transmission network connecting the central dispatch center and each AGV. The communication network uses wireless communication technology to support real-time data transmission and two-way communication. The network protocol uses TCP / IP or a dedicated industrial communication protocol to ensure reliable and real-time data transmission. The re-issuance process sends new dispatch instructions to each AGV via the communication network. The issuance process includes steps such as instruction packaging, network transmission, and instruction receipt confirmation. Instructions are issued using broadcast or point-to-point communication, depending on the urgency and coverage of the instructions. Task assignments are re-assigned to assign new container transport tasks to the relocated AGVs. Path planning is re-assigned to calculate new driving routes and navigation instructions for each AGV. Operational scheduling is re-assigned to adjust the start and completion times of each AGV's operation to ensure coordinated operations after relocation. The optimized scheduling instructions are the final set of instructions optimized by the dynamic replanning algorithm. These instructions include the relocation instructions and new task instructions for all relevant AGVs. The instruction execution monitoring mechanism tracks the execution status and completion status of each AGV.

[0097] The above describes the intelligent dispatching method of the automated terminal AGV in the embodiment of the present application. The following describes the intelligent dispatching system of the automated terminal AGV in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the automated terminal AGV intelligent dispatching system includes:

[0098] An evaluation module is used to collect container task data through the terminal operation management system, and perform multi-dimensional evaluation processing on the container task data based on the urgency of ship departure, container weight level, quay crane operation sequence, yard distance coefficient and customer priority to obtain a task priority evaluation matrix;

[0099] A prediction module is used to construct an AGV state vector based on the task priority evaluation matrix, and to perform load capacity prediction processing on the AGV's current position coordinates, load status, remaining power, and estimated completion time through training with historical port operation data to obtain an AGV load prediction result;

[0100] A matching module is used to match the AGV load prediction result with the task priority evaluation matrix, and intelligently match the container task priority matrix and the AGV dynamic load state vector through the terminal multi-task chain scheduling algorithm Port-MTCSA to obtain the full-process task chain of loading and unloading, transportation, and storage;

[0101] A collection module is used to collect historical terminal operation data based on the full-process task chain of loading and unloading, transportation, and storage, and to perform load forecasting processing based on ship arrival plans, weather factors, and container type distribution through time series analysis to obtain terminal operation load distribution;

[0102] The comparison module is used to compare the terminal operation load distribution with a preset load threshold, and when the predicted load exceeds the threshold, dynamically replan the AGV resource allocation, obtain the optimized scheduling instruction and execute the terminal AGV intelligent scheduling.

[0103] above Figure 2 The automated terminal AGV intelligent scheduling system in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The automated terminal AGV intelligent scheduling device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0104] Reference Figure 3 In the embodiment of the present invention, there is also provided an automated terminal AGV intelligent dispatching device, which can be a server, and its internal structure can be as follows: Figure 3 As shown. The automated terminal AGV intelligent scheduling device includes a processor, a memory, a display screen, an input device, a network interface and a database connected via a system bus. Among them, the computer-designed processor is used to provide computing and control capabilities. The memory of the automated terminal AGV intelligent scheduling device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the automated terminal AGV intelligent scheduling device is used to store the corresponding data in this embodiment. The network interface of the automated terminal AGV intelligent scheduling device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0105] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the automated terminal AGV intelligent scheduling equipment to which the solution of the present invention is applied.

[0106] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the steps of the automated terminal AGV intelligent scheduling method.

[0107] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0108] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the existing technology, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling an automated terminal AGV intelligent scheduling device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program code.

[0109] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An automated terminal AGV intelligent scheduling method, characterized in that: The method comprises: Collecting container task data through the terminal operation management system, performing multi-dimensional evaluation processing on the container task data based on the urgency of ship departure, container weight level, quay crane operation sequence, yard distance coefficient and customer priority, and obtaining a task priority evaluation matrix; According to the task distribution in the task priority evaluation matrix, the real-time operating parameters of the AGV are collected, the current position coordinates, load status, remaining power and expected completion time of the AGV are vectorized and encoded to obtain the AGV state vector, and the load capacity of the AGV is predicted based on the current position coordinates, load status, remaining power and expected completion time of the AGV through training with historical port operation data to obtain the AGV load prediction result; The AGV load prediction result is matched with the task priority evaluation matrix. The container task priority matrix and the AGV dynamic load state vector are intelligently matched using the terminal multi-task chain scheduling algorithm Port-MTCSA to obtain the full-process task chain of loading and unloading, transportation, and storage. Collect historical terminal operation data based on the time node information in the entire loading, unloading, transportation and storage process task chain, and perform load forecasting based on ship arrival plans, weather factors and container type distribution through time series analysis to obtain the terminal operation load distribution; Compare the terminal operation load distribution with a preset load threshold, and dynamically re-plan the AGV resource allocation when the predicted load exceeds the threshold, obtain optimized scheduling instructions, and execute terminal AGV intelligent scheduling; The AGV load prediction result is matched with the task priority evaluation matrix, and the container task priority matrix and the AGV dynamic load state vector are intelligently matched by the terminal multi-task chain scheduling algorithm Port-MTCSA to obtain the full-process task chain of loading and unloading, transportation and storage, including: The AGV load prediction result is data-fused with the task priority evaluation matrix, and the matching degree between the container task weight requirement and the AGV load capacity is calculated to obtain the task-AGV initial matching table; Based on the task-AGV initial matching table, a bipartite graph structure is constructed. The terminal multi-task chain scheduling algorithm Port-MTCSA is used to perform three-layer network modeling on the quay crane loading and unloading nodes, AGV transportation nodes, and yard storage nodes to obtain the terminal operation network topology diagram. Extracting task chain paths based on the node connection relationships in the terminal operation network topology graph, and chain-orchestrating the container loading, unloading, transportation, and storage operation sequence from the quay crane to the container yard to obtain a set of candidate task chains; The time constraints and load constraints in the candidate task chain set are cumulatively calculated, and the path cost of the task chain from the starting state to the ending state is minimized through dynamic programming recursion to obtain the task chain optimization score; Based on the task chain optimization score, candidate task chains are sorted and screened, and high-scoring task chains are prioritized through a greedy selection strategy to obtain a matching result between the container task priority matrix and the AGV dynamic load state vector; Verify the matching result of the container task priority matrix and the AGV dynamic load state vector, perform conflict detection on the load conflict and time conflict in the task chain, and obtain the full-process task chain of loading, unloading, transportation and storage; The bipartite graph structure is constructed based on the task-AGV initial matching table, and the three-layer network modeling processing of the quay crane loading and unloading nodes, AGV transportation nodes and yard storage nodes is performed through the terminal multi-task chain scheduling algorithm Port-MTCSA to obtain the terminal operation network topology diagram, including: Extract node information based on the task allocation relationship in the task-AGV initial matching table, perform spatial coordinate mapping on the quay crane loading and unloading position, AGV transportation path, and yard storage position to obtain a three-layer node coordinate set; The three-layer node coordinate set is hierarchically divided according to the terminal operation process, and the node connection relationship of the quay crane layer, transportation layer and yard layer is topologically constructed and processed using the terminal multi-task chain scheduling algorithm Port-MTCSA to obtain a hierarchical network structure; A connection weight matrix is ​​constructed based on the node attribute information in the hierarchical network structure, and the distance weights from the quay crane loading and unloading node to the AGV transportation node and the distance weights from the AGV transportation node to the yard storage node are quantified and calculated to obtain a node connection weight table; The node connection weight table is associated with the container transportation capacity constraint and analyzed, and the transport path weights of 20-foot and 40-foot containers are differentiated by the Port-MTCSA algorithm to obtain a capacity constraint weight matrix. The three-layer network structure is finally constructed based on the capacity constraint weight matrix. The connectivity of the quay crane loading and unloading nodes, AGV transportation nodes and yard storage nodes is verified through the graph theory algorithm to obtain the terminal operation network topology diagram.

2. The automated terminal AGV intelligent scheduling method according to claim 1 is characterized in that: The container task data is collected through the terminal operation management system, and the container task data is evaluated and processed in multiple dimensions according to the urgency of ship departure, container weight level, quay crane operation sequence, yard distance coefficient and customer priority to obtain a task priority evaluation matrix, including: Collect container numbers, vessel information, quay crane operation plans, yard locations, and estimated operation times through the data interface of the terminal operation management system, and perform structured parsing on the container task data to obtain a standardized task data set; Calculate the difference between the ship departure time in the standardized task data set and the current time, and quantify the urgency coefficient based on the urgency of the ship departure time to obtain an urgency score vector; Extracting weight parameters based on the container specification information in the standardized task dataset, classifying and coding the load levels of 20-foot containers and 40-foot containers to obtain a weight level coefficient matrix; The urgency score vector is weightedly fused with the weight grade coefficient matrix, and the quay crane operation sequence, yard distance coefficient and customer priority are comprehensively calculated and processed through a multi-dimensional weight allocation algorithm to obtain a task priority evaluation matrix.

3. The automated terminal AGV intelligent scheduling method according to claim 1 is characterized in that: The real-time operating parameters of the AGV are collected according to the task distribution in the task priority evaluation matrix, and the current position coordinates, load status, remaining power and expected completion time of the AGV are vectorized and encoded to obtain the AGV state vector. The load capacity prediction processing of the current position coordinates, load status, remaining power and expected completion time of the AGV is performed through training with historical port operation data to obtain the AGV load prediction result, including: The AGV state vector is correlated with the terminal historical operation database, and feature extraction processing is performed on the AGV load change data in the past 30 days through time window screening to obtain a historical load feature sample set; Based on the historical load feature sample set, a statistical analysis is performed on the load distribution of AGVs in different yard areas, and the load probability of 20-foot and 40-foot containers is quantified by load frequency calculation to obtain a load probability distribution table; According to the current load state in the AGV state vector and the load probability distribution table, a matching calculation is performed, and the difference analysis processing is performed on the remaining load capacity of the AGV and the weight requirement of the task to be performed to obtain the load matching coefficient; The load matching coefficient is correlated with the estimated completion time of the AGV, and the load capacity change trend of the AGV in the future time window is predicted through load time series analysis to obtain the AGV load prediction result.

4. The automated terminal AGV intelligent scheduling method according to claim 1 is characterized in that: The terminal operation load distribution is obtained by collecting historical terminal operation data based on the time node information in the loading, unloading, transportation and storage process task chain, and performing load forecasting processing on the ship arrival plan, weather factors and container type distribution through time series analysis, including: The ship arrival time, operation completion time and AGV operation trajectory of the past 90 days were extracted and processed to obtain a historical operation data set; The historical operation data set is arranged in sequence according to the time dimension, and pattern recognition processing is performed on the periodic laws and seasonal changes of the ship arrival plan through time series analysis to obtain the ship arrival time series pattern; Based on the ship arrival time series model, weather forecast data and historical weather records are collected, and correlation analysis is performed on the correlation between wind speed, rainfall and visibility and terminal operation efficiency to obtain a weather impact coefficient matrix; Extracting type distribution information based on the container transportation records in the historical operation data set, performing statistical calculations on the quantity distribution of 20-foot containers and 40-foot containers in different time periods, and obtaining a container type distribution pattern; The ship arrival time series pattern, the weather impact coefficient matrix and the container type distribution law are comprehensively modeled, and the task density of each operation area in the future time window is predicted and calculated by the load prediction algorithm to obtain the terminal operation load distribution.

5. The method for intelligent dispatching of AGVs at automated terminals according to claim 1, characterized in that: The terminal operation load distribution is compared with a preset load threshold, and when the predicted load exceeds the threshold, the AGV resource allocation is dynamically replanned to obtain an optimized scheduling instruction and execute the terminal AGV intelligent scheduling, including: Comparing the load value of each operating area in the terminal operation load distribution with a preset load threshold one by one, marking and identifying areas exceeding the threshold, and obtaining a list of overloaded operating areas; Based on the area location information and load excess degree in the overloaded operation area list, the current allocation status of AGV resources in each area is analyzed and evaluated to obtain an AGV resource reallocation demand matrix; According to the AGV resource reallocation demand matrix, a dynamic replanning algorithm is called to optimize the transfer path and transfer quantity of AGVs from low-load areas to high-load areas, thereby obtaining an AGV resource adjustment plan; The AGV resource adjustment plan is converted into a scheduling instruction format, and the task allocation, path planning and operation timing of each AGV are re-issued through the terminal communication network to obtain optimized scheduling instructions and execute terminal AGV intelligent scheduling.

6. An automated terminal AGV intelligent dispatching system, characterized in that: Used to implement the automated terminal AGV intelligent scheduling method according to any one of claims 1 to 5, the automated terminal AGV intelligent scheduling system includes: An evaluation module is used to collect container task data through the terminal operation management system, and perform multi-dimensional evaluation processing on the container task data based on the urgency of ship departure, container weight level, quay crane operation sequence, yard distance coefficient and customer priority to obtain a task priority evaluation matrix; A prediction module is used to construct an AGV state vector based on the task priority evaluation matrix, and to perform load capacity prediction processing on the AGV's current position coordinates, load status, remaining power, and estimated completion time through training with historical port operation data to obtain an AGV load prediction result; A matching module is used to match the AGV load prediction result with the task priority evaluation matrix, and intelligently match the container task priority matrix and the AGV dynamic load state vector through the terminal multi-task chain scheduling algorithm Port-MTCSA to obtain the full-process task chain of loading and unloading, transportation, and storage; A collection module is used to collect historical terminal operation data based on the full-process task chain of loading and unloading, transportation, and storage, and to perform load forecasting processing based on ship arrival plans, weather factors, and container type distribution through time series analysis to obtain terminal operation load distribution; The comparison module is used to compare the terminal operation load distribution with a preset load threshold, and when the predicted load exceeds the threshold, dynamically replan the AGV resource allocation, obtain the optimized scheduling instruction and execute the terminal AGV intelligent scheduling.

7. An automated terminal AGV intelligent dispatching device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the method for intelligent scheduling of automated terminal AGVs according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor is enabled to execute the automated terminal AGV intelligent scheduling method according to any one of claims 1 to 5.

Images