Container logistics ecological chain large model system based on joint model

Abstract

The invention discloses a container logistics ecological chain large model system based on a joint model, and relates to the technical field of container logistics, and the system comprises real equipment, a Tos system, a real-time data concentrator, a virtual port, a database, and an LLM. The real equipment is key operation objects of a port site and comprises quay cranes, wheel cranes, inner container trucks, storage yards, ships and the like, and each object continuously outputs real-time operation state data; and the Tos system is responsible for receiving and processing real-time data from real equipment. According to the container logistics ecological chain large model system based on the joint model, a scheduling strategy with sensing, decision-making and self-adaption capabilities is formed by training an agent in a virtual port environment, and real-time deployment and application are performed in an actual port, so that multi-type equipment and multi-task concurrent scheduling can be supported; efficient, real-time response and dynamic rescheduling capability are realized, and the overall intelligent level, loading and unloading efficiency and handling capacity of port operation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of container logistics, and in particular to a large model system of a container logistics ecological chain based on a joint model. Background Art

[0002] Container logistics refers to a logistics model that utilizes containers as standardized transport units to achieve efficient, safe, and convenient transportation of goods via various modes of transport, including sea, land, and air. Within the container logistics ecosystem, each link is closely interconnected, forming a complex and efficient logistics system. Ocean shipping, with its numerous advantages, including low cost, large transport volumes, environmental protection and energy conservation, high safety, and global coverage, has become the dominant mode of container logistics.

[0003] However, current container terminals generally face challenges such as complex operational processes, a wide variety of equipment, dispersed operational nodes, and high task concurrency. Frequent changes in ship arrival schedules, uncertain container inflows and outflows, and real-time changes in equipment status make traditional scheduling methods difficult to meet practical needs.

[0004] Most ports still rely on manual experience or static rule systems for scheduling, lacking the ability to quickly respond to emergencies and globally optimize the overall system. This can easily lead to scheduling conflicts, unbalanced resource allocation, low equipment utilization, limited overall operational efficiency, and hindered port throughput capacity. Summary of the Invention

[0005] The purpose of the present invention is to provide a large model system of container logistics ecological chain based on a joint model to solve the problems raised in the above background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solutions: a large model system for container logistics ecological chain based on a joint model, comprising:

[0007] Real equipment: Key operational objects at the port site, including quay cranes, wheel cranes, container trucks, storage yards, and ships. Each object continuously generates real-time operational status data, such as location, status, load, and operational progress.

[0008] Tos system: The port's original dispatching and management system, responsible for receiving and processing real-time data from real equipment;

[0009] Real-time data aggregator: responsible for aggregating real-time data from all devices and scene objects to form a digital scene status snapshot, which serves as the input source for the digital port system;

[0010] Virtual Port: The system's intelligent decision-making hub, responsible for building a digital port simulation environment that simulates container trucks, yard machinery, and yard operations. It includes two core intelligent modules: a scheduling agent and a loading map agent. The scheduling agent makes real-time decisions through reinforcement learning / multi-agent systems, while the loading map agent optimizes the order in which containers are embarked and disembarked.

[0011] Database: Stores all historical and real-time data, including equipment status, dispatch results, loading diagrams, instructions, etc., provides data query and background support for LLM, and supports model training data accumulation and playback verification. The database serves as the system's data storage center and interacts with other parts of the system through SQL or other data access interfaces.

[0012] LLM (Large Language Model): Receives natural language input from human-computer interaction, converts it into structured instructions or policy rules, interacts with the database, assists the scheduling agent in understanding task intent, generating rules, and helps explain and generate emergency strategies in abnormal situations (such as how to schedule in the event of ship delays or mechanical failures).

[0013] Furthermore, the real-time operating status data produced by the real equipment is collected in real time through sensors, GPS, IoT devices, etc., and the collected raw data is standardized and structured, and uploaded to the Tos system through various interface protocols (such as OPC UA, API, IoT) to achieve status fusion, and then transmitted to the real-time data aggregator by the Tos system.

[0014] Furthermore, the data standardization abstracts the raw data of each device into a unified scheduling language (such as "waiting for loading" and "in transit"). The specific operations are: establishing a status tag library (such as "machine status = {idle, operating, faulty, standby}"), and using rule engines / event stream processing technology to generate semantic event streams (such as "a box is about to leave the port"), and finally unifying time synchronization and spatial positioning (all objects use the same port coordinate system).

[0015] Furthermore, the real-time collection objects include quay cranes, wheel cranes, container trucks, storage yards, ships, and other equipment data. The key data items of each object are as follows:

[0016] Quay crane: crane number, operation status, location information, target vessel, spreader height, hook raising and lowering events, etc.

[0017] Wheel crane: current yard location, operation instructions, moving direction, spreader status, and container number to be operated;

[0018] Internal container truck: location information, mission ID, cargo status, driving route, speed, fuel / power level, etc.

[0019] Yard: yard layout, container occupancy, channel congestion, and job queue information;

[0020] Ship: sailing schedule, berthing plan, loading and unloading tasks, cargo priority, container number list, and loading plan;

[0021] Other equipment data: maintenance status, power system, meteorology, hydrology, RFID, differential, GPS, etc.

[0022] Furthermore, the Tos system acts as a data transfer intermediary, transferring data from real devices to a real-time data aggregator, while interacting with other parts of the system (such as the control execution layer) to execute scheduling instructions.

[0023] Furthermore, the real-time data collector includes the following key components:

[0024] Protocol adaptation layer: supports multiple communication protocols such as OPCUA, Modbus, CAN, MQTT, HTTP, WebSocket, etc., and is used to achieve unified access of data streams of heterogeneous devices and systems;

[0025] Device driver management: Provide lightweight driver adapter plug-ins for each type of device, which can be dynamically hot-swapped;

[0026] Data buffering and priority control: processing high-frequency equipment data (such as container truck / machine movement trajectory) and low-frequency data (such as asynchronous processing of planned instructions).

[0027] Furthermore, the virtual port is built on the Unity 3D engine platform to map the equipment, scenes, and operation processes of the real port, and serves as a training platform, testing platform, and human-machine collaborative interface for AI scheduling strategies. The construction content and composition of the virtual port are as follows:

[0028] Port entity 3D model: Entirely adopt commercial models, avoid 3D modeling as much as possible, speed up R&D, incorporate physical engine components to simulate the actual motion trajectory and collision behavior of equipment, and model objects include quay cranes, wheel cranes, container trucks, container yards, ships, berths, channels, and crossings;

[0029] Operation process and behavior modeling: Recreate the real container transport operation chain, such as: unloading → container truck → yard → gantry crane / stacking. Specifically, the task chain modeler defines the input / output status and conversion rules of each type of equipment, and uses Python / C# scripts to control the unloading sequence and container truck allocation logic. It also simulates congestion, failures, command delays, temporary task insertion, etc.

[0030] Virtual-reality synchronization mechanism: access real-time data aggregator, realize the synchronization mapping of real device state, virtual device can receive real scheduling instructions, and feedback to the scheduling agent, can switch between "simulation running" and "real following" two modes, support unified platform for training and actual deployment;

[0031] Scheduling agent and stowage plan agent access:

[0032] Scheduling agent receives current environment state (device position, task queue), generates scheduling decision (scheduling which truck / machinery / yard area, etc.), feedback job efficiency index and bottleneck analysis, etc.

[0033] Stowage plan agent automatically generates stowage scheme according to ship stowage plan, cabin structure, cargo priority, etc., supports joint optimization with scheduling strategy.

[0034] Further, the agent training: using Unity ML-Agents platform, based on port simulation environment to build multi-agent reinforcement learning model, train scheduling agent and stowage plan agent, so that they have the ability to realize efficient task allocation, device coordination and scheduling, and intelligent use of resources in complex dynamic environment, so as to maximize port throughput efficiency and reduce energy consumption and congestion.

[0035] Further, the agent is as follows:

[0036] Scheduling agent: through reinforcement learning / multi-agent system of Agents or custom algorithm to make real-time decision (truck allocation, device sequence, task flow optimization), real-time allocate tasks to shore cranes, trucks, wheel cranes, etc., realize multi-task scheduling, can respond to scheduling according to port dynamic situation (such as congestion, delay);

[0037] Stowage plan agent: optimize container loading / unloading sequence, combine cabin, center of gravity, cargo type and other restrictions to generate reasonable stowage plan for the ship, and affect the scheduling strategy to balance efficiency and cabin structure;

[0038] Collaborative agent (optional): learn the cooperation mechanism among multiple devices to improve the coherence and collaboration of the whole system.

[0039] Further, the LLM accesses structured database through SQL Agent, Pandas Agent or custom retrieval interface, accesses data types including real-time scheduling state (device job state, task queue, truck position, etc.), historical job record (past scheduling instructions, exception handling record, scheduling result), environment data (weather information, tide, berth state, etc.), so that LLM can make factual response based on real-time data and historical record in port scheduling system, improve the rationality of strategy, the accuracy of answer and traceability.

[0040] The present invention provides a large-scale model system for container logistics ecological chain based on a joint model, which has the following beneficial effects:

[0041] The present invention trains embodied intelligent agents in a virtual port environment to form a scheduling strategy with perception, decision-making and self-adaptation capabilities, and deploys and applies it in real time in an actual port. It can support the concurrent scheduling of multiple types of equipment (such as bridge cranes, wheel cranes, AGVs, etc.) and multiple tasks, achieve efficient, real-time response and dynamic rescheduling capabilities, and improve the overall intelligence level, loading and unloading efficiency and throughput capacity of port operations.

[0042] 1. This system uses a real-time data collector to collect and synchronize port equipment status within seconds, ensuring the high real-time performance of the dispatching system. At the same time, the dispatching agent works in conjunction with the loading diagram agent to dynamically optimize task allocation and loading and unloading plans, and quickly respond to dispatch based on port dynamics (such as congestion and delays), significantly improving overall throughput efficiency and operational coordination.

[0043] 2. This system integrates intelligent agents, reinforcement learning models, optimization algorithms, and a rule interpreter driven by a large language model (LLM). It supports multi-objective task scheduling and multi-device collaborative control. It can execute strategies and perform real-time rescheduling in complex environments, ensuring efficient and flexible operation of the system. It also has the ability to quickly respond to emergencies and achieve global optimal control of the entire system. It can automatically trigger the rescheduling process in abnormal situations such as ship delays, equipment failures, and missing containers.

[0044] 3. This system builds a highly realistic and interactive virtual port system, serving as a training platform, testing platform, and human-machine collaborative interface for AI scheduling strategies. The integration of the virtual port and real-world data creates a digital platform that integrates simulation and real-world scenarios. This platform can be used for algorithm verification, scheduling drills, and pre-deployment testing to ensure the effectiveness and reliability of scheduling strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a logical architecture diagram of a large model system of container logistics ecological chain based on a joint model of the present invention;

[0046] Figure 2 This is a data flow chart of the container logistics ecological chain large model system based on the joint model of the present invention. DETAILED DESCRIPTION

[0047] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0048] like Figure 1-Figure 2As shown, a container logistics ecosystem chain large model system based on a joint model includes:

[0049] Real devices: key operation objects in the port site, including quayside cranes, ship-to-shore cranes, yard trucks, yards, ships, etc., each object continuously outputs real-time running state data such as position, state, load, operation progress, etc.

[0050] In this embodiment, the real-time running state data output by the real device is collected in real time through sensors, GPS, IoT devices, etc., and the raw data collected is standardized and structured, and then uploaded to the Tos system through various interface protocols (such as OPCUA, API, IoT), to realize state fusion, and then transmitted to the real-time data collector by the Tos system.

[0051] Data standardization abstracts the raw data of each device into a unified scheduling language (such as "waiting for loading" and "in transit"), and the specific operation is: establishing a state label library (such as "machine state = {idle, working, fault, standby}"), and using rule engine / event stream processing technology to generate semantic event stream (such as "a container is about to leave the port"), and finally time synchronization and space positioning are unified (all objects use the same port area coordinate system).

[0052] In this embodiment, the real-time collected objects include quayside cranes, ship-to-shore cranes, yard trucks, yards, ships, and other device data, and the key data items of each object are as follows:

[0053] Quayside crane: crane number, operation state, position information, target ship, spreader height, hooking event, etc.

[0054] Ship-to-shore crane: current yard position, operation instruction, moving direction, spreader state, to-be-operated container number;

[0055] Yard truck: position information, task ID, load state, driving path, speed, fuel / electricity, etc.

[0056] Yard: yard layout, container position occupation, channel congestion degree, operation queuing information;

[0057] Ship: ship period, berthing plan, loading and unloading task, cargo priority, container number list, stowage plan;

[0058] Other device data: maintenance state, power system, meteorology, hydrology, RFID, differential, GPS, etc.

[0059] Tos system: the original scheduling management system of the port, responsible for receiving and processing real-time data from real devices. The Tos system acts as an intermediary for data transmission, transmitting data from real devices to the real-time data collector, and interacting with other parts of the system (such as the control execution layer) to execute scheduling instructions.

[0060] Real-time data aggregator: This is responsible for aggregating real-time data from all devices and scene objects, forming a digital scene status snapshot that serves as the input source for the digital port system. In this embodiment, the real-time data aggregator includes the following key components:

[0061] Protocol adaptation layer: supports multiple communication protocols such as OPCUA, Modbus, CAN, MQTT, HTTP, WebSocket, etc., and is used to achieve unified access of data streams of heterogeneous devices and systems;

[0062] Device driver management: Provide lightweight driver adapter plug-ins for each type of device, which can be dynamically hot-swapped;

[0063] Data buffering and priority control: processing high-frequency equipment data (such as container truck / machine movement trajectory) and low-frequency data (such as asynchronous processing of planned instructions).

[0064] Virtual Port: The system's intelligent decision-making center is responsible for building a simulated environment for the digital port, simulating container trucks, yard machinery, and yard operations. It includes two core intelligent modules: the scheduling agent and the loading map agent. The scheduling agent makes real-time decisions through reinforcement learning / multi-agent systems, and the loading map agent optimizes the order in which containers are boarded and disembarked.

[0065] In this embodiment, the virtual port is built on the Unity 3D engine platform to mirror the equipment, scenes, and operational processes of a real port. It also serves as a training platform, testing platform, and human-machine collaborative interface for AI scheduling strategies. The construction content and composition of the virtual port are as follows:

[0066] Port entity 3D model: Entirely adopt commercial models, avoid 3D modeling as much as possible, speed up R&D, incorporate physical engine components to simulate the actual motion trajectory and collision behavior of equipment, and model objects include quay cranes, wheel cranes, container trucks, container yards, ships, berths, channels, and crossings;

[0067] Operation process and behavior modeling: Recreate the real container transport operation chain, such as: unloading → container truck → yard → gantry crane / stacking. Specifically, the task chain modeler defines the input / output status and conversion rules of each type of equipment, and uses Python / C# scripts to control the unloading sequence and container truck allocation logic. It also simulates congestion, failures, command delays, temporary task insertion, etc.

[0068] Virtual-reality synchronization mechanism: Connects to a real-time data aggregator to achieve synchronous mapping of real-world device states. Virtual devices can receive real-world scheduling instructions and feed them back to the scheduling agent. It can switch between "simulation operation" and "reality follow-up" modes, supporting a unified platform for training and actual deployment.

[0069] The dispatching agent and the loading diagram agent are connected:

[0070] The dispatching agent receives the current environment status (equipment location, task queue), generates dispatching decisions (which truck / machine / yard area to dispatch, etc.), and provides feedback on operational efficiency indicators and bottleneck analysis.

[0071] The loading diagram agent automatically generates loading plans based on the ship loading plan, cabin structure, cargo priority, etc., and supports joint optimization with the scheduling strategy.

[0072] In this embodiment, agent training: using the Unity ML-Agents platform, a multi-agent reinforcement learning model is built based on the port simulation environment to train the scheduling agent and the loading map agent. This enables them to achieve efficient task allocation, coordinated equipment scheduling, and intelligent resource utilization in a complex dynamic environment, thereby maximizing port throughput efficiency and reducing energy consumption and congestion.

[0073] Scheduling Agent: Through agents or customized algorithms, reinforcement learning / multi-agent systems make real-time decisions (container truck allocation, equipment sequence, task flow optimization), assign tasks to quay cranes, container trucks, wheel cranes and other equipment in real time, realize multi-task scheduling, and respond to port dynamics (such as congestion and delays) in scheduling.

[0074] Loading diagram agent: Optimizes the order of container embarkation and disembarkation, generates a reasonable loading diagram for the ship based on constraints such as space, center of gravity, and cargo type, and influences scheduling strategies to balance efficiency and cabin structure.

[0075] Collaborative Agent (optional): Learn the collaboration mechanism between multiple devices to improve the overall system coherence and coordination.

[0076] Database: Stores all historical and real-time data, including equipment status, dispatch results, loading diagrams, and instructions. It provides data query and background support for LLM, and supports the accumulation and playback of model training data. As the system's data storage center, the database interacts with other parts of the system through SQL or other data access interfaces.

[0077] LLM (Large Language Model): Receives natural language input from human-computer interactions, converts it into structured instructions or policy rules, interacts with the database, and assists the dispatching agent in understanding task intent, generating rules, and helping to explain and generate emergency strategies in abnormal situations (such as scheduling in the event of ship delays or mechanical failures). LLM uses SQL Agent, Pandas Agent, or customized search interfaces to implement natural language access to structured databases. Accessible data types include real-time dispatch status (equipment operating status, task queues, container truck locations, etc.), historical operation records (previous dispatch instructions, exception handling records, dispatch results), and environmental data (meteorological information, tides, berth status, etc.). This enables LLM to make fact-based responses based on real-time data and historical records in the port dispatching system, improving the rationality of strategies, the accuracy of answers, and traceability.

[0078] The embodiments of the present invention are presented for purposes of illustration and description and are not intended to be exhaustive or to limit the invention to the disclosed forms. Many modifications and variations will be apparent to those skilled in the art. The embodiments are chosen and described in order to better illustrate the principles of the invention and its practical application and to enable those skilled in the art to understand the invention and design various embodiments with various modifications as suited for specific applications.

Claims

1. A large model system for container logistics ecological chain based on a joint model, characterized by: Including real equipment, Tos system, real-time data collector, virtual port, database, LLM; Real equipment: Key operational objects at the port site, including quay cranes, wheel cranes, container trucks, storage yards, and ships, each of which continuously generates real-time operational status data. Tos system: The port's original dispatching and management system, responsible for receiving and processing real-time data from real equipment; Real-time data aggregator: responsible for aggregating real-time data from all devices and scene objects to form a digital scene status snapshot, which serves as the input source for the digital port system; Virtual Port: The system's intelligent decision-making center, responsible for building a digital port simulation environment, simulating container trucks, yard machinery, and yard operations. It includes two core intelligent modules: the scheduling agent and the loading diagram agent. Database: stores all historical and real-time data, including equipment status, dispatch results, loading diagrams, and instructions, provides data query and background support for LLM, and supports model training data accumulation and playback verification; LLM: Receives natural language input from human-computer interaction, converts it into structured instructions or policy rules, interacts with the database, assists the scheduling agent in understanding task intent, generates rules, and helps explain and generate emergency strategies in abnormal situations.

2. A large model system for container logistics ecological chain based on a joint model according to claim 1, characterized in that: The real-time operating status data produced by the real equipment is collected in real time through sensors, GPS, and IoT devices. The collected raw data is standardized and structured, uploaded to the Tos system through various interface protocols, and then transmitted to the real-time data aggregator by the Tos system.

3. The large model system of container logistics ecological chain based on joint model according to claim 2 is characterized in that: The data standardization abstracts the original data of each device into a unified scheduling language. The specific operations are: establishing a state tag library, and using the rule engine / event stream processing technology to generate a semantic event stream, and finally unifying time synchronization and spatial positioning.

4. The large model system of container logistics ecological chain based on joint model according to claim 2 is characterized in that: The real-time collection objects include quay cranes, wheel cranes, container trucks, storage yards, ships, and other equipment data. The key data items of each object are as follows: Quay crane: crane number, operation status, location information, target vessel, spreader height, hook raising and lowering events; Wheel crane: current yard location, operation instructions, moving direction, spreader status, and container number to be operated; Internal container truck: location information, mission ID, cargo status, driving route, speed, fuel / power level; Yard: yard layout, container occupancy, channel congestion, and job queue information; Ship: sailing schedule, berthing plan, loading and unloading tasks, cargo priority, container number list, and loading plan; Other equipment data: maintenance status, power system, meteorology, hydrology, RFID, differential, GPS.

5. The large-scale model system of container logistics ecological chain based on the joint model according to claim 1 is characterized in that: The Tos system transmits data from real devices to a real-time data aggregator, while interacting with other parts of the system to execute scheduling instructions.

6. The large model system of container logistics ecological chain based on joint model according to claim 1 is characterized in that: The real-time data collector includes the following key components: Protocol adaptation layer: supports multiple communication protocols to achieve unified access of data streams from heterogeneous devices and systems; Device driver management: Provide lightweight driver adapter plug-ins for each type of device, which can be dynamically hot-swapped; Data buffering and priority control: processing high-frequency device data and low-frequency data.

7. The large model system of container logistics ecological chain based on joint model according to claim 1 is characterized in that: The virtual port is built on the Unity 3D engine platform to mirror the equipment, scenes, and operational processes of a real port. It also serves as a training platform, testing platform, and human-machine collaborative interface for AI scheduling strategies. The construction content and composition of the virtual port are as follows: Port entity 3D model: This model fully adopts commercial models and incorporates physical engine components to simulate the actual motion trajectory and collision behavior of equipment. Modeling objects include quay cranes, wheel cranes, container trucks, container yards, ships, berths, passages, and crossings. Operation process and behavior modeling: Restore the real container transportation operation chain. Specifically, the task chain modeler defines the input / output status and conversion rules of each type of equipment, and uses Python / C# scripts to control the unloading sequence and container truck allocation logic. At the same time, congestion, failures, command delays, and temporary task insertion are simulated; Virtual-reality synchronization mechanism: Connect to the real-time data aggregator to achieve synchronous mapping of the real-world device status. Virtual devices can receive real-world scheduling instructions and feed them back to the scheduling agent. The dispatching agent and the loading diagram agent are connected: The scheduling agent receives the current environment status, generates scheduling decisions, and provides feedback on job efficiency indicators and bottleneck analysis; The loading diagram agent automatically generates loading plans based on the ship loading plan, cabin structure, and cargo priority, and supports joint optimization with the scheduling strategy.

8. The large model system of container logistics ecological chain based on joint model according to claim 7 is characterized in that: The agent training described above uses the Unity ML-Agents platform to build a multi-agent reinforcement learning model based on a port simulation environment, and trains the scheduling agent and the loading map agent to enable them to achieve efficient task allocation, coordinated equipment scheduling, and intelligent resource utilization in a complex dynamic environment.

9. The large-scale model system of container logistics ecological chain based on the joint model according to claim 7 is characterized in that: The intelligent agent is specifically as follows: Scheduling Agent: Through agents or custom algorithms, reinforcement learning / multi-agent systems make real-time decisions, assign tasks to equipment in real time, realize multi-task scheduling, and respond to port dynamics. Loading diagram agent: Optimizes the order of container embarkation and disembarkation, generates a reasonable loading diagram for the ship based on space, center of gravity, and cargo type restrictions, and influences the scheduling strategy to balance efficiency and cabin structure.

10. The large-scale model system of container logistics ecological chain based on the joint model according to claim 1 is characterized in that: The LLM implements natural language access to the database through SQL Agent, Pandas Agent or customized retrieval interface, and the access data types include real-time scheduling status, historical operation records, and environmental data.

Images