An internet container management deployment method based on an AI large model

Abstract

The application relates to the technical field of Internet and discloses an Internet container management deployment method based on an AI large model, which comprises the following steps: collecting real-time freight data and empty container position information of a port through a sensor network, generating a cargo flow distribution map, calculating the empty container supply-demand difference between ports, and determining an imbalance index; if the imbalance index exceeds a preset threshold, judging the demand fluctuation type of the current port, constructing a reinforcement learning model according to the demand fluctuation type, generating a candidate operation scheme between ports, and obtaining an optimized path set; screening a feasible path subset from the optimized path set based on congestion data of the current port, if the feasible path subset is empty, obtaining a replacement path from a backup route database, and determining a final deployment path; converting the final deployment path into an operation instruction and sending the operation instruction to a container scheduling system, receiving an execution confirmation feedback returned by the system, and correcting the cargo flow distribution map. The application improves the empty container operation efficiency and decision accuracy.

Description

Technical Field

[0001] This application relates to the field of Internet technology, and in particular to an Internet container management and deployment method based on an AI large model. Background Technology

[0002] With the continuous development of global trade and shipping, container shipping has become a major mode of international logistics. Due to regional differences in cargo flow, some ports often experience empty container accumulation, while others suffer from empty container shortages. To alleviate this imbalance, operators typically need to allocate empty containers to meet transportation demand. In existing technologies, container allocation mainly relies on manual experience or prediction methods based on static data. On the one hand, manual allocation is inefficient and struggles to cope with the complexity and real-time nature of global shipping routes; on the other hand, traditional methods rely heavily on historical statistical data, failing to accurately reflect real-time cargo flow and port congestion, easily leading to allocation delays or resource waste. Furthermore, existing allocation schemes lack in-depth analysis of demand fluctuation patterns. In practical applications, container demand is often affected by both cyclical factors (such as seasonal peaks) and unforeseen events (such as port strikes and extreme weather). If different types of demand fluctuations are not distinguished, allocation models struggle to adapt quickly, leading to decision-making biases. Moreover, current allocation route selections are mostly fixed schemes, lacking dynamic optimization mechanisms. Especially in situations such as port congestion and shipping route disruptions, existing methods lack the ability to automatically select and switch alternative routes, resulting in insufficient flexibility and robustness in allocation. Therefore, this application proposes an internet-based container management deployment method based on an AI-powered large-scale model. This method combines real-time data collection, intelligent algorithm optimization, and execution feedback updates to achieve dynamic and intelligent management of global empty container allocation, thereby improving allocation efficiency, reducing transportation costs, and enhancing system robustness. Summary of the Invention

[0003] To address the aforementioned technical issues, this application provides an internet-based container management deployment method based on an AI large model, which aims to improve the efficiency and accuracy of empty container relocation.

[0004] Firstly, this application provides a method for managing and deploying internet-based container systems based on an AI-powered large-scale model, the method comprising:

[0005] Step S1: Collect real-time freight data and empty container location information of multiple ports within the target area through a sensor network, generate a cargo flow distribution map, calculate the supply and demand difference of empty containers between ports based on the cargo flow distribution map, and determine the imbalance index.

[0006] Step S2: If the imbalance index exceeds a preset threshold, the current port demand fluctuation type is determined based on historical data. The demand fluctuation type includes periodic fluctuations and sudden fluctuations. A reinforcement learning model is constructed based on the demand fluctuation type to generate candidate transportation schemes between ports and obtain an optimized path set.

[0007] Step S3: Obtain the current port congestion data, and based on the congestion data, filter out a subset of feasible paths from the optimized path set. If the subset of feasible paths is empty, obtain alternative paths from the backup route database and determine the final dispatch path.

[0008] Step S4: Convert the final allocation path into a dispatch instruction and send it to the container dispatch system. Receive the execution confirmation feedback returned by the system and revise the cargo flow distribution map based on the execution confirmation feedback.

[0009] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0010] This application discloses an internet-based container management deployment method based on an AI big data model. Through comprehensive analysis of real-time port freight data, empty container location information, and historical dispatch records, it achieves intelligent management of the entire empty container dispatch process. First, real-time freight data and empty container information collected through sensor networks can accurately generate cargo flow distribution maps, and based on this, calculate the differences and imbalances in empty container supply and demand between ports. This enables dynamic evaluation of the priority of empty container allocation at ports, effectively avoiding resource waste and transportation delays caused by shortages or surpluses of empty containers. Second, based on historical data, the system can determine the type of port demand fluctuations, distinguishing between periodic and sudden fluctuations. Reinforcement learning models are then built for different fluctuation types, ensuring that the generated candidate transportation plans balance transportation costs and timeliness requirements, improving the adaptability and robustness of the transportation strategy. Third, by combining current port congestion data with the selection of feasible path subsets from the optimized path set, the system can avoid congested ports in real time, ensuring the feasibility of transportation routes. When the feasible path subset is empty, alternative routes are obtained from the backup route database, and transportation timeliness and cost are evaluated to form preferred routes and determine the final allocation route, ensuring the continuity and efficiency of the empty container allocation plan.

[0011] Meanwhile, this application updates the empty container inventory and cargo flow data of each port in a timely manner by receiving execution confirmation feedback from the container scheduling system, and regenerates the cargo flow distribution map based on the updated data, thereby providing an accurate initial state for the next round of allocation, realizing closed-loop scheduling and dynamic optimization. Overall, this application significantly improves the efficiency and accuracy of empty container allocation, reduces logistics costs, and improves the utilization rate of port operating resources through the collaborative work of multiple links such as real-time data collection, historical data analysis, reinforcement learning optimization, feasible path screening and feedback correction. At the same time, it enhances the system's ability to cope with emergencies and port congestion, and significantly optimizes the intelligent scheduling level of the global container logistics network. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram of an embodiment of an internet container management deployment method based on an AI large model, as described in this application.

[0014] Figure 2 This is a flowchart illustrating the implementation of the optimized path set in the embodiments of this application. Detailed Implementation

[0015] This application provides a deployment method for internet container management based on an AI large model. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion; 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 explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0016] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the Internet container management deployment method based on AI large model in this application includes:

[0017] Step S1: Collect real-time freight data and empty container location information of multiple ports within the target area through a sensor network, generate a cargo flow distribution map, calculate the supply and demand difference of empty containers between ports based on the cargo flow distribution map, and determine the imbalance index.

[0018] Specifically, in order to promptly grasp the empty container inventory and demand at various ports and avoid allocation decision errors due to information lag, it is necessary to monitor and model the real-time operational status of ports and shipping routes. This application collects real-time freight data and empty container location information from multiple ports within the target area through a sensor network. For example, it obtains empty container location information through GPS positioning devices installed on containers, obtains the number of empty containers entering and leaving the port through RFID scanning equipment at the port, and obtains vessel navigation dynamics through the Automatic Identification System (AIS). This data is then aggregated to a data processing platform, and based on the above information... A cargo flow distribution map, using ports as nodes and inter-port cargo flow as edges, represents the cargo flow direction and empty container distribution at each port. Based on this map, the supply-demand difference for empty containers at each port is calculated. This difference refers to the discrepancy between the available empty containers and the demand at each port, used to determine whether a port is a supply or demand port. Finally, by comparing the supply-demand difference between ports with the corresponding cargo flow, an imbalance index is obtained. This index quantifies the urgency of empty container allocation between ports; for example, a larger imbalance index indicates a more urgent need for empty container allocation at that port. Dynamic monitoring and quantitative analysis of the empty container supply and demand status between ports ensures the real-time nature and accuracy of subsequent allocation decisions, providing fundamental data support for the system's intelligent optimization of allocation paths.

[0019] Step S2: If the imbalance index exceeds the preset threshold, the current port demand fluctuation type is determined based on historical data. The demand fluctuation type includes periodic fluctuations and sudden fluctuations. A reinforcement learning model is constructed based on the demand fluctuation type to generate candidate transportation schemes between ports and obtain an optimized path set.

[0020] Specifically, when an imbalance indicator exceeds a preset threshold, historical empty container allocation records are first retrieved, and the distance between the current port scenario and the corresponding indicator in the historical records is calculated to measure their similarity. If the distance is below the set threshold, the corresponding historical records are used as matching samples and aggregated into a subset. Subsequently, the freight volume data in the subset is decomposed into time series components, including trend components, seasonal components, and residual components. These components are compared with preset thresholds to determine the type of demand fluctuation. The type of demand fluctuation refers to the changing pattern of port empty container demand over time, used to distinguish whether there is regular periodic demand or abnormal demand driven by sudden factors. Reinforcement learning models are constructed to adapt to different types of demand fluctuation. When the demand fluctuation type is periodic, transportation cost is used as the main weight factor of the reward function, and historical successful paths are used as training samples, causing the model to tend to select low-cost solutions during iterative training. When the demand fluctuation type is sudden, transportation timeliness is used as the main weight factor of the reward function, and a perturbation factor is introduced during training to simulate allocation delays or abnormal situations under sudden scenarios, thereby improving the robustness of the model. Multiple candidate transport plans are generated using the above method. Each plan includes specific transport routes and quantities. These plans are then comprehensively evaluated in terms of cost and timeliness. Finally, a set of plans that meet preset conditions is selected, resulting in an optimized route set. The optimized route set refers to a set of candidate transport routes that achieve the best balance between cost and timeliness under the current port demand environment, after comprehensive evaluation. This method allows for flexible adjustment of allocation strategies under different demand fluctuation scenarios, comprehensively considering historical patterns and unforeseen events. It not only improves the adaptability and robustness of candidate transport plans but also provides high-quality input for subsequent route selection, ensuring a more intelligent and efficient allocation process throughout the system.

[0021] Step S3: Obtain the current port congestion data, and based on the congestion data, filter out a subset of feasible paths from the optimized path set. If the subset of feasible paths is empty, obtain alternative paths from the backup route database and determine the final dispatch path.

[0022] Specifically, to ensure that the transportation plan is not only optimal in terms of cost and timeliness but also feasible in actual operation, it is necessary to screen based on the real-time operation status of the port. First, congestion data of the target port and its related transshipment ports are obtained through the real-time data interface of the port management system. Indicators such as average waiting time of ships, berth occupancy rate, and yard utilization rate are obtained and stored as a congestion dataset with timestamps. Then, the congestion data is mapped to each path in the optimized path set. The congestion index is calculated for each of the origin port, transshipment port, and destination port involved in the path. Based on the congestion index of all ports on the path, the overall capacity of the path is calculated. The capacity reflects the smoothness of the path under the current conditions. The higher the value, the more feasible the path is. If the path contains multiple routes, the capacity of each segment is weighted and summed to obtain the total capacity of the path. After obtaining the capacity of each route, it is compared with preset traffic conditions. Routes with capacity exceeding a threshold are selected to form a subset of feasible routes. If the subset of feasible routes is not empty, the routes are sorted in descending order of capacity and used as candidate inputs for the next scheduling stage. If the subset of feasible routes is empty, meaning all solutions in the current optimized route set cannot be executed due to congestion, the backup route database is accessed to extract alternative route information related to the current port. These backup routes are then evaluated for transportation timeliness and cost. Following a priority order of timeliness followed by cost, one or more alternative routes are ultimately determined, and the optimal solution is selected as the final dispatch route. This step performs a secondary screening of the optimized route set based on actual port operating conditions, ensuring that the dispatch routes meet both economic and timeliness requirements while also possessing practical feasibility, thereby effectively reducing the risk of dispatch failure due to port congestion or unforeseen events.

[0023] Step S4: Convert the final allocation route into a dispatch instruction and send it to the container dispatch system. Receive the execution confirmation feedback returned by the system and revise the cargo flow distribution map based on the execution confirmation feedback.

[0024] Specifically, the final allocation route is converted into an executable transport instruction, including information such as the origin port, destination port, number of empty containers to be transported, and estimated shipping time. This instruction is then sent to the container dispatching system. Upon receiving the instruction, the dispatching system arranges specific execution actions in the vessel plan and loading plan, such as determining the specific vessel, loading / unloading berths, and estimated voyage time. Subsequently, the dispatching system returns execution confirmation feedback, including whether the instruction was successfully received, the execution progress status, and the estimated arrival time. Based on this execution confirmation feedback, the relevant information in the cargo flow distribution map is updated, specifically including reducing the corresponding number of empty containers from the origin port, adding an in-transit transport record at the destination port, and marking the estimated arrival time, thereby forming a new empty container supply and demand status. Thus, the revised cargo flow distribution map can reflect the latest empty container distribution and flow in real time, providing accurate input data for the next round of imbalance index calculation and allocation decisions. This step achieves closed-loop management from scheme planning to execution feedback and data correction, ensuring that the allocation results can be truly implemented and dynamically reflected in the system model. This not only improves the reliability of allocation execution, but also enhances the adaptability and continuous optimization capability for subsequent scheduling.

[0025] In one specific embodiment, generating a cargo flow distribution map includes the following steps:

[0026] Real-time freight data and empty container location information are cleaned and integrated to generate a structured dataset;

[0027] Based on the structured dataset, a cargo flow distribution map is generated using a data visualization algorithm. The cargo flow distribution map represents the cargo flow and empty container distribution between ports.

[0028] Specifically, real-time freight data and empty container location information acquired by the sensor network are cleaned and integrated to generate a structured dataset. Sensor data may contain noise, missing, or duplicate records; therefore, outlier detection algorithms can be used to remove invalid data during the cleaning process, such as using the isolated forest method to identify and remove outliers. In the data integration stage, the cleaned data is categorized according to port identifiers and timestamps to obtain a tabular dataset containing freight flow, empty container quantity, and geographic coordinates. Based on the structured dataset, a freight flow distribution map is generated using data visualization algorithms. The map uses ports as nodes and freight flow between ports as edges. A heatmap method is used to map port freight flow as a color gradient, and point density is used to represent the difference in empty container inventory, thus highlighting high-demand or high-redundancy areas. In another implementation, a Sankey diagram can be further incorporated to depict the flow paths between ports as directed edges, with edge width representing flow magnitude and arrow direction representing supply and demand, thus more intuitively representing imbalances. For example, in a specific application scenario, freight data from port A to port B is extracted from a structured dataset. If the daily freight volume is 1500 TEUs, and port A has a surplus of 300 empty TEUs while port B has a shortage of 200 empty TEUs, a heatmap can highlight the path from A to B with warm colors, and a Sankey diagram can connect the two nodes with a wide side, marking their supply-demand difference. This allows users to intuitively identify the potential allocation needs for this path. This step achieves the cleaning, integration, and visualization of real-time logistics data, ensuring data quality and visually presenting the freight relationships and empty container supply-demand distribution between ports. This provides an accurate and intuitive foundation for subsequent imbalance index calculations and allocation path optimization.

[0029] In one specific embodiment, calculating the supply and demand difference of empty containers between ports and determining the imbalance index specifically includes the following steps:

[0030] Extract the empty container inventory and demand of each port from the cargo flow distribution map, and build a directed graph with each port as a node and the cargo flow between ports as edges.

[0031] Based on the empty container inventory and demand, the supply and demand nodes in the directed graph are determined, and the maximum flow algorithm is used to calculate the empty container supply and demand difference between the supply and demand nodes.

[0032] The ratio of the supply-demand difference of empty containers to the total freight volume among all ports is used as the imbalance indicator for each port, and the priority of empty container allocation at each port is determined based on the magnitude of the imbalance indicator.

[0033] Specifically, to avoid long-term idleness of empty containers in some ports while severe shortages occur in others, it is necessary to quantitatively analyze the supply and demand status of empty containers at each port and calculate allocation priorities using mathematical models. This provides an accurate basis for generating subsequent transportation plans. Specifically, the empty container inventory and demand for each port are extracted from the cargo flow distribution map. Empty container inventory refers to the number of available empty containers currently held by the port, while demand is the future empty container usage predicted based on cargo flow. Based on the extracted inventory and demand, supply and demand nodes are determined in the graph: when inventory exceeds demand, the port is defined as a supply node; otherwise, it is a demand node. Then, a maximum flow algorithm (e.g., the Ford-Fulkerson method) is used on this directed graph to calculate the supply-demand difference between supply and demand nodes. The maximum flow algorithm continuously searches for augmenting paths from supply nodes to demand nodes in the directed graph, increasing flow along the path until no feasible path remains. The final maximum flow value is the difference in empty containers that can be allocated from the supply side to the demand side, reflecting the degree of supply-demand balance between ports. After obtaining the supply-demand difference value for each port pair, it is calculated as a ratio to the total freight flow between all ports to obtain the imbalance index for each port. The imbalance index is the normalized result of the empty container supply-demand difference value and the total freight flow, used to characterize the degree of supply-demand imbalance of the port in the overall network. Based on the magnitude of the imbalance index, the priority of empty container allocation for ports can be ranked. Port pairs with higher indices indicate greater supply-demand differences and should be allocated first. For example, in a network containing ten ports, if port A has an empty container inventory of 5000 and a demand of 3000, then port A is a supply node; port B has an inventory of 2000 and a demand of 3500, then port B is a demand node. The freight flow between the two ports is 2000. Calculated using the maximum flow algorithm, the final available allocation difference value is 1500. If the total freight flow of all ports is 10000, then the imbalance index for the port pair A and B is 0.15, indicating that this path has a higher allocation priority. Conversely, for Port C, if the inventory is 4000 and the demand is 4500, the supply-demand difference is only -500, which accounts for a small proportion of the total flow. Therefore, its imbalance index is only 0.05, indicating a low allocation priority. This step achieves precise calculation based on a graph theory model, allowing the supply-demand difference of empty containers to be expressed intuitively through quantitative indicators. This not only ensures the scientific and objective nature of allocation priority judgment but also adapts to port networks of different sizes, from a few ports within a region to hundreds of ports globally, thus significantly improving the decision-making efficiency and accuracy of empty container allocation.

[0034] In one specific embodiment, determining the current port demand fluctuation type based on historical data includes the following steps:

[0035] If the similarity value is lower than the preset threshold, the corresponding historical empty box allocation records will be used as matching samples to form a subset of data.

[0036] Time series decomposition was performed on the freight volume data of the subset to obtain the trend component, seasonal component, and residual variance;

[0037] If the seasonal component exceeds the preset threshold and the residual variance is lower than the preset threshold, then the current port demand fluctuation type will be set to cyclical fluctuation.

[0038] If the residual variance exceeds the preset threshold and there is no obvious seasonal component, the current port demand fluctuation type will be set to sudden fluctuation.

[0039] Specifically, after detecting that the imbalance index of a port exceeds a preset threshold, the similarity between the current imbalance index and the corresponding index in historical empty container allocation records is first calculated, using Euclidean distance as the metric. If the similarity is lower than the preset threshold (e.g., 0.1), it indicates that the current scenario has a strong similarity to some historical scenarios, and this historical record is used as a match. All matching results are then aggregated to form a subset. For example, during the holiday transportation peak, the imbalance index of Port D is 0.25, matching 15 similar historical records from around Christmas in the past three years. Subsequently, the freight volume data in the subset is decomposed into time series data. The solution results include trend components, seasonal components, and residual components. The trend component represents long-term trends, the seasonal component reflects periodic patterns, and the residual component reflects random fluctuations. When analyzing the decomposition results, if the amplitude of the seasonal component exceeds a preset threshold while the variance of the residual component is below a preset threshold, it indicates that the port's empty container demand is mainly affected by periodic patterns, and therefore its demand fluctuation type is set to periodic fluctuation. Conversely, if the residual variance exceeds a preset threshold and does not exhibit obvious periodic characteristics, it indicates that the port's demand is mainly driven by sudden events, and therefore its demand fluctuation type is set to sudden fluctuation. The demand fluctuation type refers to the pattern of change in the port's empty container demand over time, used to distinguish between regular seasonal changes and abnormal demand driven by occasional factors. For example, when analyzing empty container data on the Asia-Europe route, if a subset of the dataset shows a clear sinusoidal waveform in cargo volume over the past three months, with a seasonality of 25% and low residual variance, then the port's demand can be judged to be cyclical. In another scenario, if the port's recent cargo volume experiences abnormally large fluctuations, with residual variance increasing by more than 50% of the historical average and no cyclical pattern observed, then it is judged to be a sudden fluctuation. Using historical data to classify and identify current port demand changes not only ensures the rationality and accuracy of the model input but also allows for flexible adjustments to subsequent allocation plans in both cyclical and sudden scenarios.

[0040] In one specific embodiment, constructing a reinforcement learning model based on the type of demand fluctuation includes the following steps:

[0041] When the demand fluctuation is cyclical, the transportation cost of each transportation route is used as the main weighting factor of the reward function, and historical successful routes are used as training samples for the reinforcement learning model.

[0042] When demand fluctuations are sudden, transportation timeliness is used as the main weighting factor of the reward function, and a perturbation factor is introduced to simulate sudden scenarios during the training process of the reinforcement learning model.

[0043] Specifically, the first step is to determine the type of demand fluctuation. If the detection results indicate that the current port demand fluctuation is cyclical, then the transportation cost of each candidate transport route is used as the main weighting factor of the reward function. This means the model will prioritize routes with lower transportation costs during training. Simultaneously, historically successful transport routes are used as training samples for the reinforcement learning model. The training data also includes the current cargo flow distribution map and port imbalance indicators, which will be explained in detail later. This makes it easier for the model to reuse proven effective route solutions during iterative convergence. If the detection results indicate that the current port demand fluctuation is sudden, then when constructing the reinforcement learning model, transportation timeliness is used as the main weighting factor of the reward function. This prioritizes minimizing transportation time and ensuring rapid response capabilities. Furthermore, a disturbance factor is introduced during model training to simulate sudden scenarios, such as sudden port strikes, temporary shutdowns caused by typhoons, or ship delays. The disturbance factor artificially increases environmental uncertainty, allowing the model to learn how to select routes under abnormal conditions during training, thereby enhancing its robustness to unforeseen events. By combining demand fluctuation types with the core parameters of reinforcement learning models, transportation costs can be effectively reduced in cyclical fluctuation scenarios, while timeliness requirements can be quickly responded to in sudden fluctuation scenarios. Its beneficial effect is to improve the model's adaptability to different logistics scenarios and enhance the practicality and stability of dispatching solutions.

[0044] In one specific embodiment, obtaining the optimized path set includes the following steps:

[0045] The current cargo flow distribution map and imbalance index are input into the reinforcement learning model for iterative training to generate multiple candidate transportation schemes. Each candidate transportation scheme includes the transportation route and the transportation quantity.

[0046] Cost and timeliness assessments are conducted on candidate transportation schemes, and candidate schemes that meet preset conditions are selected based on the assessment results to form an optimized path set.

[0047] Specifically, the current cargo flow distribution map and imbalance index are input into the reinforcement learning model for iterative training. The cargo flow distribution map represents the distribution of cargo traffic and empty container supply and demand among ports, while the imbalance index represents the degree of difference between the supply and demand of empty containers among ports. This input serves as the state space vector of the model, which can comprehensively reflect the real-time operating status of the system. Secondly, during the iterative training of the model, the action space is defined as different combinations of transport paths and their corresponding transport quantities. The action space is explored and utilized through an epsilon greedy strategy, that is, the historically optimal path is selected with a certain probability to maintain convergence and stability, while new paths are explored with a certain probability to improve the diversity of the schemes. Each round of training generates a set of candidate transport schemes, in which each scheme contains a complete transport path sequence (i.e., a combination of origin port, stopover port and destination port) and the number of empty containers allocated on the path.

[0048] After generating multiple candidate shipping plans, each plan undergoes a dual evaluation of cost and timeliness. Cost evaluation includes fuel costs, loading and unloading costs, and transshipment costs. Fuel costs are calculated by multiplying the total route distance by the unit fuel cost. Loading and unloading costs are calculated based on the shipping quantity and the port's unit loading and unloading price. Transshipment costs are estimated based on a fixed fee added to each stopover port. Timeliness evaluation includes sailing time and transshipment delay. Sailing time is obtained by dividing the total route distance by the average ship speed. Transshipment delay is weighted and summed based on historical port congestion data. For example, if the total route distance is 8,000 nautical miles and the ship speed is 15 knots, the sailing time is approximately 22 days, and with a 2-day transshipment delay, the timeliness is 24 days. By comparing the evaluation results, if a plan's overall performance meets preset conditions, such as a cost below a threshold of $50,000 and a timeliness of less than 10 days, then the plan is selected.

[0049] Suppose there is a regional port network where the supply and demand of empty containers at each port are significantly unbalanced, as follows: High-supply ports (excess empty containers): Port A (severe empty container backlog, high inventory costs, urgent need for shipment), Port B (ample empty container supply, significant shipping capacity); High-demand ports (empty container shortage): Port F (imminent large export orders, urgent need for empty containers to ensure loading, highest shortage level), Port C (some demand, but slightly less urgency than Port F); Transit port: Port E (an important logistics hub with dense shipping routes, but its own supply and demand are balanced). A relocation plan needs to be developed to move empty containers from Ports A and B to Ports F and C, aiming to achieve the optimal balance between controlling transportation costs and ensuring timely relocation, in order to address the urgent need at Port F and optimize overall operational efficiency. Figure 2The diagram shows the implementation flowchart of the optimized route set. The supply and demand data (cargo flow distribution map) and the degree of difference (imbalance index) of the aforementioned ports are input into the reinforcement learning model. The model uses the current state (which port has more, which port has less, and by how much) as the basis for decision-making. The model explores using an epsilon greedy strategy, selecting known, cost-effective historical routes (e.g., direct routes) with a high probability, and randomly trying new routes (e.g., routes via transit port E) with a certain probability to discover potential better solutions. Each iteration generates a candidate transport plan, which includes a specific route and transport volume (e.g., "transport 200 empty containers from port A to port E, and then from port E to port F"). Each candidate plan undergoes cost and timeliness evaluation. Finally, all plans that meet the conditions are included in the optimized route set. This set contains multiple transport routes with advantages in both cost and timeliness, along with corresponding transport volumes, providing high-quality input for subsequent port congestion analysis and route decision-making. This improves the overall efficiency of empty container transport while balancing economy and timeliness.

[0050] In one specific embodiment, selecting a subset of feasible paths from the optimized path set based on congestion data specifically includes the following steps:

[0051] The port management system is used to obtain data on vessel waiting times and berth occupancy rates at ports involved in the route, calculate the congestion index for each port, and calculate the overall traffic capacity of the route based on the congestion index.

[0052] The overall traffic capacity of the route is compared with the preset traffic conditions. Optimized routes with traffic capacity higher than the preset threshold are selected. The optimized routes that meet the conditions are sorted in descending order of traffic capacity to form a subset of feasible routes.

[0053] Specifically, the port management system obtains data on vessel waiting times and berth occupancy rates at ports involved in the route. This system updates data in real time using sensors and log systems deployed on berths, wharves, and vessels. For example, in port A, the system returns an average vessel waiting time of 4 hours and a berth occupancy rate of 85%. In port B, the system returns a waiting time of 2 hours and an occupancy rate of 75%. Based on the obtained waiting times and occupancy rates, the congestion index for each port is calculated using the formula: Congestion Index = Waiting Time / Standard Waiting Time + Occupancy Rate / Standard Occupancy Rate. The standard waiting time and standard occupancy rate are set based on historical averages. For example, if the standard waiting time is 2 hours and the standard occupancy rate is 70%, then the congestion index for port A is 4 / 2 + 85 / 70 ≈ 2.21, and the congestion index for port B is 2 / 2 + 75 / 70 ≈ 2.07. Subsequently, the calculated port congestion indices are mapped to the set of ports involved in the route. The overall capacity of the route is calculated based on the average congestion index of all ports on the route, using the formula: Capacity = 1 - Average Congestion Index. If a route includes ports A and B, the overall capacity of the route is 1 - (2.21 + 2.07) / 2 ≈ -1.14. The overall capacity of the route is compared with a preset capacity threshold, for example, a threshold of 0.7. If the capacity is lower than the threshold, the route is considered not to meet the transportation conditions and is excluded from the feasible route subset; conversely, routes with a capacity higher than the threshold are marked as meeting the conditions. Then, all routes meeting the conditions are sorted in descending order of capacity to generate the final feasible route subset. This subset is output to the subsequent transportation decision module for actual transportation route selection and scheduling optimization, thereby ensuring that route selection considers both the actual port congestion situation and improves the efficiency of empty container transportation and the reliability of the overall logistics system.

[0054] In one specific embodiment, obtaining alternative routes from the backup route database and determining the final allocation route specifically includes the following steps:

[0055] Extract route information related to the current port from the backup route database to form a candidate route set;

[0056] The candidate route set is evaluated for transportation timeliness and total cost, and backup routes that meet the preset conditions are selected and sorted according to the rule of timeliness first and cost second best.

[0057] If multiple routes meet the conditions simultaneously, a weighted score is calculated to determine the priority order of the routes, and the preferred route in the ranking results is determined as the final allocation route.

[0058] Specifically, by querying the backup route database, information related to the current port is extracted, including the origin port, destination port, sailing distance, estimated call time, and historical transport records, forming a candidate route set. Simultaneously, based on the current port's geographical location and empty container relocation needs, irrelevant routes are filtered out, retaining only those matching the target port. Subsequently, each route in the candidate route set is evaluated for transport timeliness and total cost. Transport timeliness is calculated by dividing the sailing distance by the average ship speed and adding the estimated call time. Total cost includes fuel costs, port fees, and time value loss. If the route involves transshipment, the timeliness and cost of each segment are accumulated to form a total evaluation value. This evaluation value is then compared with preset conditions to select backup routes that meet both timeliness and cost requirements. Then, the selected routes are sorted according to the rule of timeliness first and cost second. If multiple routes meet the conditions at the same time, a weighted score is further calculated (e.g., timeliness weight 0.6, cost weight 0.4) to determine the priority order. The route with the highest weighted score is selected as the final dispatch route. At the same time, the detailed information of the route is updated, including the origin, destination and estimated arrival time, so as to ensure that a reasonable and efficient alternative can still be provided for empty container dispatch even when the subset of feasible routes is empty.

[0059] In one specific embodiment, the evaluation of transportation timeliness and total cost for the candidate route set includes the following steps:

[0060] Input the candidate route set into the timeliness assessment module, calculate the transportation timeliness based on the sailing distance and average ship speed, and accumulate the stop delay time of each segment to obtain the total transportation timeliness of each alternative route;

[0061] The candidate route set is simultaneously input into the cost assessment module to calculate fuel costs, port fees, and transshipment costs, and the total cost evaluation result for each alternative route is formed by combining the number of empty containers.

[0062] Specifically, the candidate route set is input into the timeliness assessment module. For each route, its sailing distance, the number of expected ports of call, and the call time are obtained. The transport timeliness is calculated using the formula: Transport Timeliness = Sailing Distance / Average Ship Speed ​​+ Call Delay Time. The average ship speed is set based on historical shipping data, for example, 15 knots. For routes involving multiple segments, the sailing timeliness of each segment is calculated separately and then summed to obtain the total transport timeliness, thereby quantifying the expected transport time of each alternative route. Simultaneously, the candidate route set is input into the cost assessment module. For each route, fuel costs, port fees, and transshipment costs are calculated. Fuel costs are calculated by multiplying the cost per nautical mile by the sailing distance. Port fees are accumulated according to a fixed standard. Transshipment costs are calculated based on the number of transshipment ports and handling fees. These costs are then combined with the number of empty containers transported to calculate the unit cost, forming the total cost evaluation result for each alternative route. For example, a route from A to B, 2000 nautical miles away, with a ship speed of 15 knots, is expected to stop once, with a total transit time of approximately 5.5 days, fuel costs of $1000, port fees of $1000, transshipment fees of $500, and a total cost of $2500. Through the above steps, the transit time and cost quantification data of candidate routes can be obtained simultaneously, providing an accurate basis for subsequent screening and ranking.

[0063] In one specific embodiment, correcting the cargo flow distribution map based on execution confirmation feedback includes the following steps:

[0064] Based on the execution confirmation feedback, update the empty container inventory and cargo flow data of each port. Using the updated data, regenerate the cargo flow distribution map. Based on the regenerated cargo flow distribution map, determine the initial state of the next round of allocation.

[0065] Specifically, the system receives feedback information from the container scheduling system, including the number of empty containers moved, timestamps, and related freight adjustment data. Then, it analyzes the empty container relocation details based on the feedback, extracting the number of empty containers reduced at the source port and increased at the destination port, and adjusts the empty container inventory at each port accordingly. Simultaneously, it updates the inter-port transport flow data based on changes in freight flow in the feedback to ensure data real-time performance and accuracy. Next, using the updated empty container inventory and freight flow data as input, it regenerates the freight flow distribution map. It calculates the flow vectors between ports, designating ports as nodes and flow directions as directed edges, and labels the flow values ​​on the edges. Visualization tools can be used to render the distribution map, such as generating heatmaps to display high-flow areas, to intuitively reflect changes in freight patterns. Finally, based on the regenerated freight flow distribution map, it analyzes unbalanced nodes and sets the states of ports with excess or insufficient inventory as the initial state for the next round of allocation, used for subsequent calculations and allocation, thereby ensuring the continuity of the allocation process and the efficiency of iterative optimization.

[0066] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0067] If the integrated unit is implemented as 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 this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0068] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for deploying internet-based container management based on an AI large-scale model, characterized in that, The method includes: Step S1: Collect real-time freight data and empty container location information of multiple ports within the target area through a sensor network, generate a cargo flow distribution map, calculate the supply and demand difference of empty containers between ports based on the cargo flow distribution map, and determine the imbalance index. Step S2: If the imbalance index exceeds a preset threshold, the current port demand fluctuation type is determined based on historical data. The demand fluctuation type includes periodic fluctuations and sudden fluctuations. A reinforcement learning model is constructed based on the demand fluctuation type to generate candidate transportation schemes between ports and obtain an optimized path set. Determining the current port demand fluctuation type based on historical data includes: obtaining historical empty container allocation records for each port; calculating the similarity value between the port's current imbalance index and the corresponding index in the historical empty container allocation records; if the similarity value is lower than a preset threshold, the corresponding historical empty container allocation records are used as matching samples to form a subset; the freight volume data of the subset is decomposed into a time series to obtain trend components, seasonal components, and residual variance; if the seasonal component exceeds a preset threshold and the residual variance is lower than a preset threshold, the current port demand fluctuation type is set to periodic fluctuation; if the residual variance exceeds a preset threshold and there is no obvious seasonal component, the current port demand fluctuation type is set to sudden fluctuation. The reinforcement learning model constructed according to the demand fluctuation type includes: when the demand fluctuation type is periodic fluctuation, the transportation cost of each transportation route is used as the main weight factor of the reward function, and historical successful routes are used as training samples for the reinforcement learning model; when the demand fluctuation type is sudden fluctuation, the transportation timeliness is used as the main weight factor of the reward function, and a disturbance factor is introduced to simulate sudden scenarios during the training process of the reinforcement learning model. Step S3: Obtain the current port congestion data, and based on the congestion data, filter out a subset of feasible paths from the optimized path set. If the subset of feasible paths is empty, obtain alternative paths from the backup route database and determine the final dispatch path. Step S4: Convert the final allocation path into a dispatch instruction and send it to the container dispatch system. Receive the execution confirmation feedback returned by the system and revise the cargo flow distribution map based on the execution confirmation feedback.

2. The method according to claim 1, characterized in that, Generating a cargo flow distribution map includes: The real-time freight data and empty container location information are cleaned and integrated to generate a structured dataset; Based on the structured dataset, a cargo flow distribution map is generated using a data visualization algorithm. The cargo flow distribution map represents the cargo flow and empty container distribution between ports.

3. The method according to claim 2, characterized in that, Calculate the supply and demand differences of empty containers between ports and determine the imbalance indicators, including: Extract the empty container inventory and demand of each port from the cargo flow distribution map, and build a directed graph with each port as a node and the cargo flow between ports as edges. Based on the empty container inventory and demand, the supply nodes and demand nodes in the directed graph are determined, and the maximum flow algorithm is used to calculate the empty container supply and demand difference value between the supply nodes and the demand nodes. The ratio of the empty container supply-demand difference to the total freight volume among all ports is used as the imbalance index for each port, and the priority of empty container allocation for each port is determined based on the magnitude of the imbalance index.

4. The method according to claim 1, characterized in that, The optimized path set includes: The current cargo flow distribution map and the imbalance index are input into the reinforcement learning model for iterative training to generate multiple candidate transportation schemes. Each candidate transportation scheme includes a transportation route and a transportation quantity. The candidate transportation schemes are evaluated for cost and timeliness, and candidate schemes that meet preset conditions are selected based on the evaluation results to form an optimized path set.

5. The method according to claim 1, characterized in that, Based on the congestion data, a subset of feasible paths is selected from the optimized path set, including: The port management system is used to obtain data on vessel waiting time and berth occupancy rate at ports involved in the route, calculate the congestion index of each port, and calculate the overall traffic capacity of the route based on the congestion index. The overall traffic capacity of the path is compared with the preset traffic conditions, and optimized paths with traffic capacity higher than the preset threshold are selected. The optimized paths that meet the conditions are sorted in descending order of traffic capacity to form a subset of feasible paths.

6. The method according to claim 1, characterized in that, Obtain alternative routes from the backup route database to determine the final allocation route, including: Extract route information related to the current port from the backup route database to form a candidate route set; The candidate route set is evaluated for transportation timeliness and total cost, and backup routes that meet preset conditions are selected and sorted according to the rule of timeliness first and cost second best. If multiple routes meet the conditions simultaneously, a weighted score is calculated to determine the priority order of the routes, and the preferred route in the ranking results is determined as the final allocation route.

7. The method according to claim 6, characterized in that, The evaluation of the transport time and total cost of the candidate route set includes: The candidate route set is input into the timeliness assessment module, and the transportation timeliness is calculated based on the sailing distance and average ship speed. The delay time of each stop is added up to obtain the total transportation timeliness of each alternative route. The candidate route set is simultaneously input into the cost assessment module to calculate fuel costs, port fees, and transshipment costs, and the total cost assessment result for each alternative route is formed by combining the number of empty containers.

8. The method according to claim 1, characterized in that, The cargo flow distribution map is revised based on the execution confirmation feedback, including: Based on the execution confirmation feedback, update the empty container inventory and cargo flow data of each port. Using the updated data, regenerate the cargo flow distribution map. Based on the regenerated cargo flow distribution map, determine the initial state of the next round of allocation.

Images