An international logistics whole-process graphical operation system and method based on deep integration of finite state machine and AI agent

Abstract

This invention discloses a graphical operating system and method for the entire international logistics process, based on the deep integration of finite state machines and AI agents. The system includes: a graphical state machine editor for constructing a visualized state diagram representing the entire lifecycle of logistics processes using state nodes and event-driven processes using directed edges; a multi-role permission engine for binding the sales, operations, customs declaration, warehousing, and transportation roles of freight forwarding companies to the operational permissions of nodes; an AI agent factory for injecting corresponding AI agents into preset key state nodes, wherein the AI ​​agents have autonomous operation modes and assisted operation modes; and a dual-engine drive module, including a state machine execution engine and an AI scheduling engine. When the process reaches a target node, the AI ​​scheduling engine wakes up the corresponding AI agent to perform tasks such as quoting, order creation, or anomaly prediction, and drives the state machine to automatically or semi-automatically flow to the next node along the directed edges based on the execution results. This invention achieves a three-in-one integration of "graphical collaboration + multi-role participation + AI autonomous execution," greatly reducing the manual operation load and decision-making threshold of international logistics.

Description

Technical Field

[0001] This invention relates to the field of supply chain digitization and process control technology, specifically to a graphical operating system and method for the entire international logistics process based on the deep integration of finite state machines and AI agents. Background Technology

[0002] International logistics (especially LCL / LCL services) involves multiple participants, including freight forwarders, customs brokers, warehouses, trucking companies, shipping companies, and destination port agents. The business process covers more than ten stages, including inquiry / quotation, booking, warehousing, customs declaration, release, LCL / LCL loading, loading onto the ship, departure, arrival at the port, customs clearance, and cargo collection. For a long time, this industry has faced the following technical challenges:

[0003] First, there is a serious information silo problem. Different roles (sales, operations, customs brokers, warehouse managers, truck drivers) use their own independent systems or tools (Excel, email, instant messaging) to transmit information, resulting in the inability to synchronize the status of the entire process in real time. The flow information of a shipment is scattered in multiple places, and operators need to frequently switch systems and communicate repeatedly to obtain the complete status.

[0004] Second, blurred lines of responsibility lead to low operational efficiency. In traditional processes, there is a lack of refined system control over the operational authority of each role in the current stage, often resulting in unauthorized operations or shirking of responsibility. For example, a warehouse manager may load and consolidate containers without confirming the customs clearance status, leading to the entire container being seized during subsequent inspection and creating the risk of "joint liability."

[0005] Third, manual operations are labor-intensive and have high decision-making thresholds. The inquiry / quotation process requires operators to manually search for freight rates from multiple shipping companies, compare delivery times, and generate a quotation; the order creation process requires manual extraction of key fields from historical shipping orders and emails to fill in each field. These repetitive and rule-based operations consume a lot of manpower and are prone to errors.

[0006] In the existing technology, although some supply chain management systems have introduced workflow engines or state machines for process control, they generally have the following defects: (1) The state transition logic is fixed and cannot be flexibly arranged according to different business scenarios; (2) There is a lack of fine mapping of multi-role permissions, and it is impossible to achieve three-dimensional control of "role-node-event"; (3) AI capabilities exist in the form of independent plug-ins or add-ons, and are not deeply integrated with the state machine, so it is impossible to achieve a closed loop of "process reaches node → AI automatically executes → state automatically transitions". Summary of the Invention

[0007] Purpose of the invention

[0008] To address the aforementioned deficiencies in existing technologies, this invention provides a graphical operating system and method for the entire international logistics process, based on the deep integration of finite state machines and AI agents. The aim is to achieve a three-in-one architecture of "graphical collaboration + multi-role participation + AI autonomous execution," thereby reducing the manual operation load and decision-making threshold in international logistics.

[0009] Technical solution

[0010] The present invention achieves the above objectives through the following technical solutions:

[0011] In a first aspect, the present invention provides a graphical operating system for the entire international logistics process based on the deep integration of finite state machines and AI agents, characterized in that it includes: a graphical state machine editing module, a role and permission mapping module, an AI agent factory, and a state machine execution engine.

[0012] The graphical state machine editing module is configured to provide a visual editing interface and generate a state diagram of the entire international logistics process in response to user operations. The state diagram includes multiple state nodes and multiple directed edges. The state nodes are used to represent the business stages of international logistics during the flow process, and the directed edges are used to represent the state transition paths triggered by events. The state nodes include at least inquiry / quotation nodes and order creation nodes.

[0013] The role permission mapping module is configured to store a role permission configuration table, which is used to establish a mapping relationship between multiple preset roles in the freight forwarding company's organizational structure and the operation permissions of the status nodes and the directed edges.

[0014] The AI ​​agent factory is configured to store and manage multiple preset AI agents, each AI agent corresponding to at least one state node, and each AI agent includes at least two operating modes: autonomous execution mode and auxiliary suggestion mode.

[0015] The state machine execution engine is communicatively connected to the graphical state machine editing module, the role and permission mapping module, and the AI ​​intelligent agent factory. The state machine execution engine includes: a node trigger listening unit, an AI scheduling unit, a state-driven unit, and a synchronization push unit.

[0016] The node triggering listening unit is used to listen to the current state node of the state graph, and generate an agent invocation signal when the current state node is determined to be the target triggering node.

[0017] The AI ​​scheduling unit, in response to the agent call signal, wakes up from the AI ​​agent factory and executes the target AI agent corresponding to the current state node.

[0018] The target AI agent is configured to: automatically complete the business operations of the node and generate flow decisions in the autonomous execution mode based on the context data of the current international logistics task, or generate operation suggestions and provide them to the user terminal for manual confirmation in the auxiliary suggestion mode.

[0019] The state driving unit is used to drive the state graph to transition from the current state node to the next state node along the corresponding target directed edge, based on the execution result of the target AI agent or the confirmation instruction fed back by the user terminal.

[0020] The synchronous push unit is used to push the state transition results and the execution log of the AI ​​agent to all related role terminals in real time.

[0021] Secondly, this invention provides a graphical operation method for the entire international logistics process based on the deep integration of finite state machines and AI agents, comprising the following steps:

[0022] S1: Generate a state diagram of the entire international logistics process through a graphical state machine editing module. The state diagram includes multiple state nodes and multiple directed edges.

[0023] S2: Establish a mapping relationship between multiple preset roles and the operation permissions of the state node and the directed edge through the role permission mapping module;

[0024] S3: The node triggering listening unit of the state machine execution engine listens to the current state node of the state diagram, and generates an agent calling signal when the current state node is determined to be the target triggering node;

[0025] S4: In response to the agent's call signal, the AI ​​scheduling unit wakes up the AI ​​agent factory and executes the target AI agent corresponding to the current state node;

[0026] S5: The target AI agent automatically completes the business operation of the node and generates a flow decision in autonomous execution mode based on the context data of the current international logistics task, or generates operation suggestions in auxiliary suggestion mode and provides them to the user terminal for manual confirmation.

[0027] S6: The state driving unit drives the state graph to move from the current state node to the next state node along the corresponding target directed edge based on the execution result of the target AI agent or the confirmation instruction fed back by the user terminal.

[0028] S7: The state transition results and the execution log of the AI ​​agent are synchronously pushed to all related role terminals in real time through the synchronous push unit.

[0029] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0030] Beneficial effects

[0031] Process visualization and standardization: The complex multi-role collaborative operations in international logistics are abstracted into a graphical state machine model, realizing visualized control of "one ticket to the end".

[0032] Fine-grained control of role permissions: By binding "status nodes + roles" through a mechanism, business permissions are solidified into the graphical topology structure, solving the problem of unclear responsibility boundaries in traditional operations.

[0033] Deep integration of AI and state machines: By treating AI agents as embedded attributes of state nodes rather than external plugins, a closed loop of "process arrival → AI automatic execution → automatic state transition" is achieved, significantly reducing the workload of manual operations.

[0034] Dual-engine arbitration mechanism: Through the difference comparison arbitration unit, the decision-making problem when AI suggestions conflict with human intentions is solved, and two-way feedback between humans and machines is realized. Attached Figure Description

[0035] Figure 1 This is a block diagram of the overall architecture of a graphical operating system for the entire international logistics process based on the deep integration of finite state machines and AI agents, provided for embodiments of the present invention.

[0036] Figure 2 This is a schematic diagram of the entire process of international logistics sea freight full container load (FCL) shipments provided in an embodiment of the present invention.

[0037] Figure 3 This is a sequence diagram of the fully automated quotation closed loop process according to Embodiment 1 of the present invention.

[0038] Figure 4 This is a sequence diagram of the interaction between two people working together to create an order, according to an embodiment of the present invention.

[0039] Figure 5 This is a sequence diagram of the process interaction of multi-channel intelligent task handover and notification based on role switching perception in Embodiment 3 of the present invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0041] System Overall Architecture

[0042] like Figure 1 As shown, the graphical operating system for the entire international logistics process, based on the deep integration of finite state machines and AI agents, provided by this invention, consists of four core modules: a graphical state machine editing module, a role and permission mapping module, an AI agent factory, and a state machine execution engine. The state machine execution engine, as the core scheduling hub of the system, internally includes a node trigger listening unit, an AI scheduling unit, a state-driven unit, a synchronization push unit, and an intelligent task handover and multi-channel notification module. The three external modules (graphical state machine editing module, role and permission mapping module, and AI agent factory) interact with the execution engine through triggering and callback mechanisms: the graphical state machine editing module provides visualized state diagram definition capabilities; the role and permission mapping module maintains a three-dimensional permission table of "role-node-event"; and the AI ​​agent factory is responsible for the registration, storage, and dual-mode (autonomous execution mode / assisted suggestion mode) management of AI agents. These modules work together to achieve a closed loop of "process arrival at node → AI automatic execution → automatic state transition → real-time synchronization of multiple roles."

[0043] International Logistics Full Process Status Diagram

[0044] like Figure 2 As shown, the international logistics full-process state diagram defined in this invention (taking full container load (FCL) business as an example) includes the following state nodes and directed edges:

[0045] Main process phase one (business access): `① Inquiry / Quote` → `② Pending Confirmation` → `③ Confirmed` → `④ Booking` → `⑤ Booked`. The trigger event for the `Inquiry / Quote` node is "Submit Quote", the trigger event for the `Pending Confirmation` node is "Customer Confirmation", the trigger event for the `Confirmed` node is "Initiate Booking", and the trigger event for the `Booking` node is "SO Confirmation".

[0046] FCL Operation Sub-State Machine (`⑥ In Operation`): The FCL sub-state machine is entered via the "Start Operation" event from node `⑤ Booked`. This sub-state machine contains ten sub-state nodes: `Pending Container Pickup` → `Container Pickup in Progress` → `Container Loading in Progress` → `Container Returned` → `Pending Material Replenishment` → `Material Replenishment Completed` → `Pending Customs Declaration` → `Customs Declaration in Progress` → `Released` → `Loaded onto Ship`. The corresponding triggering events are, in sequence: vehicle dispatch, arrival at the factory, return of loaded container, operation completed, SI submission, port entry confirmation, declaration, customs release, and VGM confirmation for loading onto ship. This sub-state machine covers the entire terminal operation process from vehicle dispatch and container pickup to completion of loading onto ship.

[0047] Main process phase two (transportation and destination port processing): Starting from the `loaded on board` node, the process proceeds through the "departure" event to `⑦ sailed` → `⑧ arrived at port` → `⑨ destination port processing` → `⑩ cargo picked up`. The corresponding triggering events are, in order: arrival at port, destination port operation, and confirmation of receipt of POD / MBL.

[0048] Financial Settlement (Parallel Process): The financial settlement process includes three nodes: Invoicing → Settlement in Progress → Closing Account. This process runs in parallel with the transportation and destination port processing stages ⑦-⑩, and can be executed at any time during these stages. It is connected to the main process by a dotted line to indicate its independent parallel nature.

[0049] Example 1: Fully Automated Quotation Closed Loop (AI Agent Autonomous Execution Mode)

[0050] This embodiment takes the inquiry / quotation stage in the entire international logistics process as an example to explain in detail the complete workflow of the AI ​​agent in autonomous execution mode. Figure 3 This is a sequence diagram of the process interaction in this embodiment, showing the complete message interaction sequence between the user terminal, the state machine execution engine, the quotation agent, and the external data source.

[0051] Scene setting:

[0052] Suppose that a sales user named "Manager Zhang" from a freight forwarding company needs to provide a sea freight quote to a customer for a new shipment of export goods (product name: electronic components, quantity: 5 pallets, weight: 850kg, volume: 3.2CBM, port of loading: Shanghai, port of destination: Los Angeles).

[0053] Step 1: Loading the State Graph and Listening to Nodes

[0054] After the system starts, the graphical state machine editing module loads the pre-generated international logistics full-process state diagram (such as...). Figure 2 (As shown). This state diagram uses state nodes to represent logistics links and directed edges to represent state transition paths. The state nodes include the following in sequence: `① Inquiry / Quote` → `② Pending Confirmation` → `③ Confirmed` → `④ Booking` → `⑤ Booked` → (Entering the FCL operation sub-state machine: `Pending Container Pickup` → `Container Pickup in Progress` → `Container Loading in Progress` → `Container Returned` → `Pending Material Replenishment` → `Material Replenishment` → `Pending Customs Declaration` → `Customs Declaration in Progress` → `Released` → `Loaded onto Ship`) → `⑦ Ship Departed` → `⑧ Arrived at Port` → `⑨ Destination Port Processing` → `⑩ Cargo Pickup`. In addition, the financial settlement process (`Invoicing` → `Settlement in Progress` → `Closing Account`) runs parallel to the `⑦-⑩` transportation and destination port processing stages, and is represented by dashed lines to indicate independent parallel relationships.

[0055] In this embodiment, the current state node is the `inquiry / quotation` node. The node triggering monitoring unit of the state machine execution engine continuously monitors the current state node. When it detects that the system has entered the `inquiry / quotation` node, it determines that the node is the target triggering node and then generates an agent invocation signal.

[0056] Step 2: AI Agent Wake-up and Context Loading

[0057] In response to the agent's call signal, the AI ​​scheduling unit queries the AI ​​agent factory for the AI ​​agent identifier bound to the "inquiry / quotation" node. The query reveals that the agent bound to the "inquiry / quotation" node is a quoting agent (e.g.,...). Figure 3 As shown, the state machine execution engine wakes up the quoting agent and loads the context data.

[0058] The AI ​​scheduling unit wakes up the pricing agent and loads the context data of the current international logistics task into the agent's execution environment. This context data includes, but is not limited to:

[0059] Goods information: Product name, quantity, weight, volume, packaging method;

[0060] Shipping information: Port of loading, port of destination, estimated shipping time;

[0061] Customer information: Customer ID, credit rating, historical quotes;

[0062] Market data: Current real-time freight rates, route transit times, and shipping schedules of various shipping companies.

[0063] Step 3: Autonomous Execution by the AI ​​Agent

[0064] After being awakened, the bidding agent operates in autonomous execution mode. This agent has a pre-built multi-objective optimization algorithm and automatically completes the bid generation according to the following sub-steps:

[0065] Sub-step 3.1 (Intent Recognition): The quoting agent first parses the cargo and transportation information in the context data to identify the intent type of this inquiry. In this embodiment, the recognition result is "general cargo sea freight export quotation".

[0066] Sub-step 3.2 (Multi-source data retrieval and fusion): The pricing agent invokes multiple data sources in parallel based on the identified intent type:

[0067] Call the shipping company's freight rate interface to obtain real-time ocean freight rates from multiple shipping companies (such as COSCO Shipping, Maersk, etc.) on the Shanghai to Los Angeles route;

[0068] Access the local port miscellaneous charges database to obtain additional charges such as terminal handling fees, document fees, and EU entry summary customs declarations for Shanghai Port;

[0069] Access the customs brokerage fee rate database for the port of origin and port of destination to obtain the customs declaration and inspection service fees;

[0070] Call the trucking company's API to obtain a land transport quote from the Shanghai warehouse to Shanghai Port.

[0071] Sub-step 3.3 (Structured Calculation and Scheme Generation): Based on the retrieved multi-source data, the pricing agent generates comparative pricing schemes from multiple shipping companies according to preset pricing calculation rules (occupancy rate × chargeable tonnes + various surcharges + service fees). Each scheme includes: shipping company name, voyage number, estimated departure date, estimated arrival date, ocean freight unit price, total freight, details of various surcharges, total price, and validity period.

[0072] Sub-step 3.4 (Output Validation): The pricing agent encapsulates the generated pricing scheme according to a pre-defined output data scheme. This scheme specifies the fields, data types, and format requirements that the quotation must include. If the output data does not meet the schema requirements (e.g., missing the required field "Estimated Departure Date" or an incorrect format of the "Total Quotation" field), the pricing agent automatically triggers recalculation until the output conforms to the schema specifications.

[0073] Step 4: State-Driven and Automatic Flow

[0074] After the pricing agent generates a price quote, it returns the structured pricing scheme as the execution result to the state-driven unit. The state-driven unit determines that the execution result meets the preset output data schema requirements and does not require manual confirmation (in autonomous execution mode). It then automatically drives the state graph from the `Inquiry / Quote` node along the `Quote Completed` directed edge to the `② Pending Confirmation` node (e.g., ...). Figure 2 Main process phase one shown).

[0075] Step 5: Real-time synchronization push

[0076] After the state transition is complete, the synchronous push unit will push the following information to all associated role terminals in real time:

[0077] Status transition notification: `Inquiry / Quote → Pending Confirmation`;

[0078] The execution log of the bidding agent includes data retrieval time, the number of data sources retrieved, and the number of solutions generated.

[0079] The complete quotation plan is displayed in the form of a graphical card on the terminal interface of the sales role "Manager Zhang", and is also copied to the operations role and the finance role.

[0080] Technical effects of this embodiment:

[0081] The entire quoting process, from state node triggering to automatic state transition, requires no manual intervention and takes approximately 3-5 seconds (compared to 30-60 minutes for traditional manual quoting). The quoting agent automatically aggregates and structures freight rates, surcharges, and service fees by concurrently accessing multi-source data, avoiding repetitive tasks such as manual queries, calculations, and itemized data entry. Simultaneously, the automatic state machine transitions ensure process continuity and real-time data synchronization.

[0082] Example 2: Human-Machine Collaborative Order Creation (AI Agent Assisted Suggestion Mode)

[0083] This embodiment takes the order creation stage in the entire international logistics process as an example to explain in detail the complete workflow of the AI ​​agent in the assisted suggestion mode, as well as the operation of the difference comparison arbitration mechanism. Figure 4 This is a sequence diagram of the process interaction in this embodiment, showing the complete message interaction sequence between the operator, the state machine engine, the order creation agent, and the difference comparison arbitration unit.

[0084] Scene setting:

[0085] Following Example 1, the state diagram has automatically transitioned to the `Order Creation` node. On the terminal interface of the user "Operator Li", the `Order Creation` node is highlighted, indicating that an order creation operation is required.

[0086] Step 1: State Node Triggering and AI Wake-up

[0087] The node trigger monitoring unit detects that the current state node is the "Order Creation" node, determines that this node is the target trigger node, and generates an agent invocation signal. The AI ​​scheduling unit queries the AI ​​agent factory for the AI ​​agent identifier bound to the "Order Creation" node. The query confirms that the agent bound to the "Order Creation" node is the order creation agent.

[0088] Unlike Embodiment 1, in this embodiment, the order-setting agent is configured to operate in auxiliary suggestion mode. This mode configuration is preset by the system administrator in the AI ​​agent factory and can be flexibly switched according to business scenarios.

[0089] Step 2: Context Data Loading and Semantic Analysis

[0090] The AI ​​scheduling unit loads the context data of the current international logistics task into the execution environment of the order-creating intelligent agent. The context data includes:

[0091] Cargo information and transportation information inherited from Example 1;

[0092] PDF files of historical entrustment documents sent by the customer via email;

[0093] This customer's historical order database records;

[0094] The communication records (instant chat text) between the sales role "Manager Zhang" and the customer during the inquiry stage.

[0095] After the order creation agent starts, the following preprocessing sub-steps are executed first:

[0096] Sub-step 2.1 (Email Semantic Analysis): The order creation agent calls the built-in natural language processing module to perform semantic analysis on the customer's email body and attachment (order form PDF), and extract the following key fields: shipper's name and address, consignee's name and address, notifying party, product name, HS code, number of pieces, weight, volume, container type requirement, estimated shipping date, and special instructions (such as "urgent request" or "refrigerated container required").

[0097] Sub-step 2.2 (Historical Data Matching): The order creation agent retrieves the customer's historical order database based on the extracted customer information, matches historical orders with similar product names, similar routes, and similar container types, and extracts commonly used field values ​​from historical orders (such as commonly used customs brokers, commonly used trucking companies, commonly used insurance rates, etc.) as candidate filling values.

[0098] Sub-step 2.3 (Auto-filling fields): The order creation agent extracts and matches the results from the above two steps, and fills them in according to the preset order data model to generate an order draft. This order draft has approximately 80% of its fields filled (such as shipper / consignee information, product name, HS code, weight and dimensions, container type, estimated shipping date, etc.). The remaining approximately 20% of fields (such as "booking agent selection" and "special remarks") are left blank due to insufficient data or ambiguity.

[0099] Step 3: Outputting auxiliary suggestions and obtaining manual confirmation

[0100] After the order creation agent generates the order draft, it pushes the order draft and the following auxiliary information to the user terminal of the "Operator Li" role in the auxiliary suggestion mode:

[0101] Populated fields are highlighted: Fields that have been automatically populated by AI are marked with a green background and accompanied by data source labels (such as "Source: Customer Email" "Source: Historical Order #2025-0342").

[0102] Fields to be filled: Fields marked in yellow that require manual entry are left blank, and AI suggestions are provided (e.g., "Suggested booking agent: XX International Logistics, matching degree 92%, reason: this agent has advantageous freight rates on the Los Angeles route").

[0103] Confidence rating: Each autofill field comes with an AI confidence score (0-100%) to help humans determine whether a review is needed.

[0104] Step 4: Manual Review and Conflict Triggering

[0105] Operator Li reviewed the draft order generated by the order creation agent on the terminal. During the review, Operator Li noticed that the order creation agent recommended "A International Logistics" as the agent in the "Booking Agent" field. However, based on personal experience, Operator Li judged that due to the special nature of the goods (electronic components, which have high time requirements), he preferred to use the direct flight service of "B Shipping Company".

[0106] Operator Li changed the "Booking Agent" field from "A International Logistics" to "B Shipping Company" on the user terminal and clicked the "Submit" button.

[0107] At this point, the difference comparison arbitration unit is triggered.

[0108] Step 5: Comparison and Arbitration of Differences

[0109] The discrepancy comparison arbitration unit detected a conflict between the operation suggestion generated by the booking agent (Option 1: Booking Agent A International Logistics) and the manual instruction entered by the user (Option 2: Booking Agent B Shipping Company), and immediately initiated the arbitration process (e.g., Figure 4 As shown, the difference comparison is used to calculate the difference between different schemes in the arbitration unit, visualize the comparison, and assist in decision-making.

[0110] Sub-step 5.1 (Calculation of Scheme Differences): The difference comparison arbitration unit calculates the complete logistics paths for Scheme 1 and Scheme 2 respectively, including:

[0111] Option 1 (A International Logistics): Shanghai → (A Company's own barge) → Ningbo → (A Company's contracted MSC mainline vessel) → Los Angeles, with an estimated total travel time of 18 days and a total freight cost of USD 2,850;

[0112] Option 2 (Shipping Company B): Shanghai → (Shipping Company B's direct mainline vessel) → Los Angeles, estimated total time 14 days, total freight USD 3,200.

[0113] Sub-step 5.2 (Visualization of Differences): The difference comparison arbitration unit displays the comparison results of the two schemes in a visual manner by highlighting them on the directed edges of the state diagram.

[0114] A comparison card pops up on the directed edge from the `Order Creation` node to the `Booking` node;

[0115] The comparison card displays the key differences between Option 1 and Option 2 in a left-right column format: time difference (14 days vs. 18 days, Option 2 is 4 days faster), cost difference (USD 3,200 vs. USD 2,850, Option 1 saves USD 350), shipping company reputation score, etc.

[0116] Each difference indicator is accompanied by a color code: green indicates that the solution is superior in that indicator, and red indicates that it is inferior.

[0117] Sub-step 5.3 (Assisted Decision Making): While displaying the difference comparison, the difference comparison arbitration unit also generates an auxiliary decision-making suggestion based on the context of the current task (the product name is electronic components, and the value is relatively high): "It is recommended to choose Option 2 (direct flight by Shipping Company B). Reason: Electronic components have a high value, timeliness takes precedence over cost, and the 4-day time difference may affect the customer's production plan."

[0118] Step Six: Final Decision and State Transition

[0119] After reviewing the comparison card and decision support suggestions generated by the difference comparison arbitration unit, Operator Li confirmed the selection of Option 2 (direct flight by Shipping Company B) and clicked the "Confirm Option 2" button on the terminal.

[0120] After receiving the confirmation instruction from the user terminal, the state-driven unit completes the remaining fields in the order draft according to the booking agent information in Scheme 2, and drives the state graph to move from the "Order Creation" node along the "Order Creation Completed" directed edge to the "Booking" node.

[0121] Step 7: Real-time synchronization push

[0122] The synchronous push unit will push the following information to all associated role terminals in real time:

[0123] Status transition notification: `Order creation → Booking`;

[0124] Complete order information (including the final adopted option two);

[0125] Difference Comparison Arbitration Log: Includes complete comparison data between Option 1 and Option 2, AI-assisted decision-making suggestions, and the user's final selection record.

[0126] Technical effects of this embodiment:

[0127] This embodiment demonstrates the core value of the AI ​​agent in the assisted suggestion mode—"AI does 80%, human review 20%." The order-creating agent automatically fills in 80% of the fields in the order draft through email semantic parsing and historical data matching, freeing operators from tedious data entry. When AI suggestions conflict with human experience, the difference comparison arbitration unit transforms "human-machine conflict" into "data-driven rational decision-making" through multi-dimensional solution comparison and visualization, preserving the final decision-making power of humans while ensuring that the decision-making process is fully supported by data.

[0128] Example 3: Multi-channel intelligent task handover and notification based on role switching awareness

[0129] This embodiment follows Embodiment 2, taking the scenario of "order creation completed and status transitioning from operator to booking operator" as an example, to explain in detail the complete working mechanism of the system in the graphical process of sensing role switching and executing multi-channel intelligent notifications. Figure 5 This is a sequence diagram of the process interaction in this embodiment, showing the complete message interaction sequence between the state-driven unit, the intelligent task handover and multi-channel notification module, the multi-channel notification gateway, and the target user.

[0130] Scene setting:

[0131] Following Example 2, the state-driven unit has moved the state graph from the `Order Creation` node along the `Order Completion` directed edge to the `Booking` node. At this point, the dominant role in the business process needs to switch from "Operation Role (Operator Li)" to "Booking Role (Booking Agent Wang)". This role switch triggers the intelligent task handover notification process.

[0132] Step 1: Character Switching Perception and Target Analysis

[0133] After the state-driven unit completes the state transition, it synchronously calls the intelligent task handover and multi-channel notification module.

[0134] First, the role switching detection unit obtains the set of source roles (i.e., "operation roles") bound to the source status node (`order creation` node) of this transfer, and the set of target roles (i.e., "booking roles") bound to the target status node (`booking` node).

[0135] The role switching detection unit compares the two sets and determines that the source role set and the target role set are inconsistent, indicating that a role change event has occurred. The role switching detection unit then generates a role switching signal and transmits the signal, along with the node identifier of the target state node, to the receiving dynamic parsing unit.

[0136] After receiving the role switching signal, the receiving dynamic parsing unit queries the role permission configuration table in the role permission mapping module based on the node identifier (`booking` node) of the target status node, and dynamically parses out all active users corresponding to the "booking role". Assume the query result is: the currently active booking roles include "Wang Booking Agent" (primary responsible) and "Zhao Booking Agent" (alternate).

[0137] The receiving dynamic parsing unit further determines the primary recipient of this notification as "Wang Booking Agent" from the above users based on the preset load balancing strategy (such as polling by employee number or prioritizing the least current task volume), and extracts the contact information stored in the system: user terminal device ID (used for in-application push), work email (wang@freight.com), and work mobile phone number (138XXXX1234).

[0138] Step 2: Semantic encapsulation of structured notification payloads

[0139] The context semantic encapsulation unit generates a structured notification payload based on the current state transition event. This payload is encapsulated in JSON format and contains the following core fields:

[0140] {

[0141] "notification_id": "NOTIFY-20260707-003",

[0142] "task_id": "SHP-2026-0707-0088",

[0143] "source_state": "Order creation",

[0144] "target_state": "Booking",

[0145] "transfer_time": "2026-07-07T14:32:18+08:00",

[0146] "trigger_summary": "AI-assisted order creation has been completed. Operator Li has confirmed the adoption of Option 2 (direct flight with Shipping Company B). Complete order data has been generated."

[0147] "urgent_level": "normal",

[0148] "graphical_snapshot": {

[0149] "current_node_coordinates": "(x: 320, y: 150)",

[0150] "highlighted_path": "Order Creation → Booking",

[0151] "view_link": "https: / / platform.freight.com / state-machine / task / SHP-2026-0707-0088"

[0152] },

[0153] "pending_actions": [

[0154] Please confirm the booking agent for this shipment (already pre-selected as Shipping Company B).

[0155] Please enter the booking number into the system and upload the SO (Booking Confirmation Form)

[0156] ],

[0157] "attachments": [

[0158] "Link to full order details page"

[0159] AI-generated booking suggestion report ]

[0161] }

[0162] The context semantic encapsulation unit also captures the topology of the current state graph, generates a visual topology snapshot with the 'Booking' node highlighted at the center and the 'Order Creation → Booking' directed edge marked with a flashing arrow, and generates a temporary access link (valid for 7 days).

[0163] Step 3: Multi-channel parallel distribution and form adaptation

[0164] After receiving the encapsulated structured notification payload and the visual snapshot link, the multi-channel distribution execution unit performs differentiated channel distribution (e.g., based on the three contact methods provided by the recipient's dynamic parsing unit) according to the three methods provided by the recipient's dynamic parsing unit. Figure 5 As shown, distribution is carried out in parallel through three channels: in-app push, email, and SMS gateway.

[0165] In-application push channel: The multi-channel distribution execution unit renders the core fields (task ID, source status, target status, and to-do items) in the structured notification payload as in-application system message cards, and pushes them in real time to the user terminal currently logged in by "Wang Booking Agent" via a WebSocket long connection. After Wang Booking Agent logs into the system, a red message badge will pop up in the upper right corner of the status graph. Clicking it will expand the message card, and the "Booking" node will be accompanied by a breathing light effect for strong visual reminder.

[0166] Email Channel: The multi-channel distribution execution unit invokes a preset email template engine to render the structured notification payload into a formatted HTML email. A visual topology snapshot is inserted at the top of the email body, task details are displayed in a table format in the middle, and a deep link button for "one-click jump to this status node" is attached at the bottom. The email is sent to wang@freight.com via an SMTP gateway.

[0167] SMS gateway channel: The multi-channel distribution execution unit compresses the notification payload into a plain text digest of no more than 140 characters (example: "[XX Freight Forwarder] You have a booking pending: Task SHP-2026-0707-0088 has been transferred from order creation to booking. Please log in to the system to confirm the booking agent. Link: https: / / dwz.cn / xxx"), and sends it to 138XXXX1234 through the operator's SMS gateway.

[0168] Step 4: Delivery Confirmation and Timeout Escalation Mechanism

[0169] After the multi-channel distribution execution unit completes the sending, it writes the record of this notification to the notification log table and starts the delivery confirmation and timeout retransmission unit.

[0170] The delivery confirmation and timeout retransmission unit starts a timer with a preset timeout threshold of 2 hours. This unit continuously monitors whether "Wang Booking Agent" has performed a "read confirmation" action on any notification from any channel (e.g., clicking a deep link in an email, opening an in-app message card, or directly logging into the system and entering the "booking" node).

[0171] Scenario A (Normal Confirmation): Assume that 5 minutes after booking agent Wang receives the SMS notification on his phone, he logs into the system and clicks the in-app message card. The client immediately sends a "read" confirmation signal back to the delivery confirmation and timeout retransmission unit. Upon receiving the signal, the unit stops the timer, marks the notification status as "read," and stops subsequent retransmissions. Simultaneously, the synchronization push unit sends the "read" status back to operator Li's terminal. Operator Li sees the notification "Booking agent Wang has received the task," completing the closed-loop handover.

[0172] Scenario B (Timeout Without Confirmation - Elevation Strategy): Assume that the booking agent, surnamed Wang, was on leave that day and did not see the message. The 2-hour timer expires without a confirmation signal. The delivery confirmation and timeout retransmission unit automatically executes the retransmission and elevation strategy:

[0173] First retransmission (30-minute interval): Send a duplicate notification to Wang, the booking agent, through three channels, and add the "[Urgent]" tag to the message header.

[0174] Second retransmission (60-minute interval): Still not confirmed. The delivery confirmation and timeout retransmission unit calls the receiver's dynamic parsing unit to parse out the contact information of the superior user "Manager Liu" based on the superior role (i.e., "Operations Supervisor") of the "Booking Role" in the role permission configuration table.

[0175] Upgrade Notice: The multi-channel distribution execution unit sent an upgrade notice to "Supervisor Liu" stating: "Subordinate booking agent Wang has not processed the emergency booking task (SHP-2026-0707-0088), which has timed out by 3 hours. Please intervene." At the same time, the system changed the border color of the 'Booking' node on the status graph from yellow (pending processing) to red (timeout warning).

[0176] Technical effects of this embodiment:

[0177] This embodiment demonstrates the core innovative value of this invention in the "multi-role collaborative handover" stage. Traditional notification systems merely "broadcast messages," while this invention achieves "targeted routing notifications based on state diagram topology awareness." The system doesn't send spam; instead, it accurately knows "who should do it next" and allows the next role to "understand the current progress at a glance" through a visual snapshot. Simultaneously, through a "timeout upgrade mechanism," management rules are solidified into system logic, avoiding the persistent problem in the freight forwarding industry of "QQ / WeChat messages being overwhelmed, leading to cargo delays."

[0178] Industrial applicability

[0179] This invention provides a graphical operating system and method for the entire international logistics process, deeply integrated with finite state machines and AI agents. This system can be widely applied to various freight forwarding companies, international logistics supply chain management platforms, and cross-border e-commerce logistics service platforms. By abstracting the complex multi-role collaborative operations of international logistics into a graphical state machine model and introducing AI agents to automate and semi-automate key nodes, this invention significantly improves the operational efficiency of the entire international logistics process, reduces labor costs, and minimizes operational errors. It possesses significant industrial practical value and broad market application prospects.

[0180] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A graphical operating system for the entire international logistics process based on the deep integration of finite state machines and AI agents, characterized in that, include: The graphical state machine editing module is configured to provide a visual editing interface and generate a state diagram of the entire international logistics process in response to user operations. The state diagram includes multiple state nodes and multiple directed edges. The state nodes are used to represent the business stages of international logistics during the flow process, and the directed edges are used to represent the state transition paths triggered by events. The state nodes include at least an inquiry / quotation node and an order creation node. The role and permission mapping module is configured to store a role and permission configuration table, which is used to establish a mapping relationship between multiple preset roles in the organizational structure of a freight forwarding company and the operation permissions of the status node and the directed edge. The AI ​​agent factory is configured to store and manage multiple preset AI agents, each AI agent corresponding to at least one state node, and each AI agent includes at least two operating modes: autonomous execution mode and auxiliary suggestion mode. The state machine execution engine is communicatively connected to the graphical state machine editing module, the role and permission mapping module, and the AI ​​intelligent agent factory, respectively. The state machine execution engine includes: A node-triggered listening unit is used to listen to the current state node of the state graph and generate an agent invocation signal when the current state node is determined to be a target trigger node. The AI ​​scheduling unit, in response to the agent call signal, wakes up from the AI ​​agent factory and executes the target AI agent corresponding to the current state node; The target AI agent is configured to: automatically complete the business operation of the node and generate a flow decision in the autonomous execution mode based on the context data of the current international logistics task, or generate operation suggestions and provide them to the user terminal for manual confirmation in the auxiliary suggestion mode; The state driving unit is used to drive the state graph to transition from the current state node to the next state node along the corresponding target directed edge, based on the execution result of the target AI agent or the confirmation instruction fed back by the user terminal. The synchronous push unit is used to push the state transition results and the execution log of the AI ​​agent to all related role terminals in real time.

2. The graphical operating system for the entire international logistics process based on the deep integration of finite state machines and AI agents as described in claim 1, characterized in that, The state node also includes at least one of the following: Booking milestones, warehousing milestones, reporting milestones, customs release milestones, LCL / packing milestones, loading milestones, sailing milestones, arrival milestones, clearance milestones, and cargo collection milestones.

3. The graphical operating system for the entire international logistics process based on the deep integration of finite state machines and AI agents as described in claim 1, characterized in that, Each AI agent in the AI ​​agent factory contains the following attributes: The agent identifier is used to uniquely identify the AI ​​agent. The bound state node identifier is used to specify the state node served by the AI ​​agent; A set of trigger conditions, used to define the conditions under which the AI ​​agent is awakened; The operating mode identifier indicates whether the current mode is autonomous execution or auxiliary suggestion mode. Output data schema, used to constrain the structured format of the AI ​​agent's output data; The timeout threshold is used to specify the maximum waiting time for the AI ​​agent to execute.

4. The graphical operating system for the entire international logistics process based on the deep integration of finite state machines and AI agents as described in claim 3, is characterized in that... The state-driven unit is further configured to: refuse to execute the state transition and return a recalculation instruction to the target AI agent when the output data of the target AI agent does not meet the structured requirements of the output data schema.

5. The graphical operating system for the entire international logistics process based on the deep integration of finite state machines and AI agents as described in claim 1, characterized in that, The AI ​​agents preset in the AI ​​agent factory include at least: a quotation agent, which is bound to the inquiry / quotation node and configured to automatically generate a quotation based on cargo information, port of origin, port of destination and current market freight rate data; and an order creation agent, which is bound to the order creation node and configured to automatically fill in order fields and generate an order draft based on historical consignment data and email semantic analysis.

6. The graphical operating system for the entire international logistics process based on the deep integration of finite state machines and AI agents as described in claim 1, characterized in that, The state machine execution engine further includes a difference comparison arbitration unit: the difference comparison arbitration unit is configured to calculate the difference degree between the first solution corresponding to the operation suggestion and the second solution corresponding to the manual instruction when the operation suggestion generated by the target AI agent in the assisted suggestion mode conflicts with the manual instruction input by the user through the user terminal, and to display the difference degree in a visual manner on the directed edge of the state graph for the user to make a final decision.

7. The graphical operating system for the entire international logistics process based on the deep integration of finite state machines and AI agents as described in claim 1, characterized in that, The graphical state machine editing module is further configured to: in response to a user's drag-and-drop operation on the state diagram, add state nodes or adjust the connection relationships of directed edges in the state diagram, and automatically update the role permission configuration table in the role permission mapping module according to the added state nodes or adjusted directed edges.

8. The graphical operating system for the entire international logistics process based on the deep integration of finite state machines and AI agents as described in claim 1, characterized in that, The state machine execution engine also includes an intelligent task handover and multi-channel notification module: The intelligent task handover and multi-channel notification module is communicatively connected to the state-driven unit, the role and permission mapping module and the synchronous push unit, respectively. The intelligent task handover and multi-channel notification module includes: The role switching detection unit is used to detect the target role set bound to the target state node after the transition when the state driving unit performs the state transition, and to detect whether the target role set has changed from the source role set bound to the source state node before the transition. The receiver dynamic parsing unit is used to dynamically parse the user terminal address, email address and mobile communication number corresponding to each target role in the target role set according to the role permission configuration table in the role permission mapping module when the role switching detection unit determines that a role has changed. The context semantic encapsulation unit is used to generate a structured notification payload based on the current state transition event. The structured notification payload includes at least: the unique identifier of the current international logistics task, the name of the source state node before the transition, the name of the target state node after the transition, the execution summary of the AI ​​agent that triggered this transition or the user operation summary, and a graphical topological location screenshot / link of the target state node in the state diagram. A multi-channel distribution execution unit is configured to send the structured notification payload to the corresponding user terminal in parallel or asynchronously through at least one notification channel based on the address information parsed by the receiver dynamic parsing unit; the notification channel includes at least: an in-application push channel, an email channel, and an SMS gateway channel; The delivery confirmation and timeout retransmission unit is used to monitor the target role's read confirmation signal for the structured notification payload. If the read confirmation signal is not received within a preset timeout threshold, the multi-channel distribution execution unit is re-executed according to a preset retransmission strategy (including interval retransmission and / or escalation notification to the superior role).

9. A graphical operation method for the entire international logistics process based on the deep integration of finite state machines and AI agents, applied to the system described in any one of claims 1 to 8, characterized in that, Includes the following steps: S1: Generate a state diagram of the entire international logistics process through a graphical state machine editing module. The state diagram includes multiple state nodes and multiple directed edges. S2: Establish a mapping relationship between multiple preset roles and the operation permissions of the state node and the directed edge through the role permission mapping module; S3: The node triggering listening unit of the state machine execution engine listens to the current state node of the state diagram, and generates an agent calling signal when the current state node is determined to be the target triggering node; S4: In response to the agent call signal, the AI ​​scheduling unit wakes up the AI ​​agent factory and executes the target AI agent corresponding to the current state node; S5: The target AI agent automatically completes the business operation of the node and generates a flow decision in autonomous execution mode based on the context data of the current international logistics task, or generates operation suggestions in auxiliary suggestion mode and provides them to the user terminal for manual confirmation. S6: The state driving unit drives the state graph to move from the current state node to the next state node along the corresponding target directed edge based on the execution result of the target AI agent or the confirmation instruction fed back by the user terminal. S7: The state transition results and the execution log of the AI ​​agent are synchronously pushed to all related role terminals in real time through the synchronous push unit.

10. The graphical operation method for the entire international logistics process based on the deep integration of finite state machine and AI intelligent agent as described in claim 9, characterized in that, Before waking up and executing the target AI agent in step S4, a permission verification step is also included: Obtain the user role that initiated the current interactive operation, and verify whether the user role has the permission to trigger state transition under the current state node according to the role permission configuration table; If the verification passes, the target AI agent is woken up and executed; if the verification fails, execution is refused and an insufficient permission prompt is returned to the user terminal.

11. The graphical operation method for the entire international logistics process based on the deep integration of finite state machine and AI intelligent agent as described in claim 9, characterized in that, After the target AI agent generates operation suggestions in the assisted suggestion mode in step S5, the method further includes a difference comparison arbitration step: When the operation suggestion generated by the target AI agent conflicts with the human instruction input by the user, the difference between the first solution corresponding to the operation suggestion and the second solution corresponding to the human instruction is calculated, and the difference is displayed in a visual manner on the directed edge of the state graph for the user to make a final decision.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 9 to 11.

Images