A shipping AI office visual display method and system based on multiple agents

Abstract

The application discloses a shipping AI office visual display method and system based on multiple agents, which comprises the following steps: receiving the input content of a user on a client; using a large language model to perform deep semantic analysis on all available information input by the user to obtain a structured inquiry intention; according to the structured inquiry intention, splitting the processing of the inquiry of the user into multiple subtasks, wherein each intention corresponds to a subtask, each subtask corresponds to a special agent and a plurality of shipping-related multiple-agent cooperation protocols; for each subtask, calling the special agent corresponding to the subtask to perform shipping data acquisition processing, finally integrating all multi-source information to generate a shipping analysis result; calling the file agent corresponding to each file format to generate a multi-modal target format file; and realizing shipping AI office visual display through file online preview technology, and providing online preview, download and link sharing functions for the multi-modal target format file.

Description

Technical Field

[0001] This invention relates to the technical field of deep integration of artificial intelligence and shipping, and in particular to a shipping AI office visualization method and system based on multi-agent intelligence. Background Technology

[0002] With the rapid development of artificial intelligence technology, its penetration into the shipping industry continues to deepen, and intelligent shipping has become an industry trend. However, current mainstream shipping software (shipping AI office software) still focuses on the development of mapping functions, such as optimizing the accuracy of electronic charts, improving route planning logic, and strengthening the ability to display ship positioning and port information. This has led to a series of industry pain points that urgently need to be addressed:

[0003] 1. Dispersed functional architecture and cumbersome operation process. The software's functional modules are fragmented, making it difficult for users to find the shipping-related content they want at a glance. The operation process is cumbersome, requiring multiple steps to complete the target task, which significantly reduces work efficiency.

[0004] Second, the content output format is too simplistic, and the data value mining is insufficient. The data presentation method is mainly based on basic tables, which can only realize the export function of raw data. It lacks in-depth integration, analysis and visualization of shipping business data, and cannot provide strong support for users' decision-making.

[0005] Third, the software suffers from weak intelligent interaction capabilities and a lack of support for autonomous decision-making. It only possesses basic data query and export functions, lacks a deep understanding of shipping business scenarios, and is unable to perform intelligent analysis, reasoning, and decision-making suggestions based on input data. The output is limited to a mere listing of raw data.

[0006] Fourth, the interaction methods are limited, restricting collaboration efficiency. The interaction methods are mainly based on traditional command input and result query, and natural language conversational interaction has not yet been realized, making it difficult to achieve efficient collaboration through natural language conversation.

[0007] Currently, mainstream shipping software on the market is essentially still a map display tool for port, vessel, and shipping route information, failing to build an office-oriented functional system adapted to shipping operations. Users need to frequently switch between multiple applications and input complex commands to complete basic operations when handling shipping office tasks, resulting in a high barrier to entry and low business processing efficiency. Furthermore, the content generated by existing software does not fully meet the needs of office scenarios, requiring extensive manual modification and adjustment before it can be applied to formal office settings, significantly increasing the user's workload.

[0008] Furthermore, shipping office scenarios involve the processing of multimodal content such as text, tables, images, and charts, which often requires professional handling. Existing shipping software struggles to intelligently edit and generate documents while maintaining the original format. Common office document formats such as mixed text and images, complex tables, and multi-level lists are prone to errors and loss during AI processing, requiring users to spend a significant amount of time on secondary formatting, further complicating the workflow.

[0009] Therefore, there is an urgent need for a shipping AI office solution that can be deeply integrated with shipping office scenarios, integrate multimodal processing capabilities, and support natural language interaction. Summary of the Invention

[0010] To address the shortcomings of existing shipping AI office software, such as fragmented functions, limited interaction methods, insufficient multimodal content processing capabilities, and the need for extensive manual modification of generated content, which makes it unsuitable for direct adaptation to office scenarios, this invention proposes a multi-agent-based shipping AI office visualization method. Based on user-input natural language queries, this method utilizes multi-agent intelligence to intelligently process the entire process from demand understanding, content analysis, and generation, achieving a one-stop intelligent service for generating shipping-related multimodal content. This invention also relates to a multi-agent-based shipping AI office visualization system.

[0011] The technical solution of the present invention is as follows:

[0012] A shipping AI-based visual office display method based on multi-agent technology includes the following steps:

[0013] Shipping office needs interaction and input steps: Receive user input on the client, which includes mandatory natural language description of shipping office needs, as well as optional uploaded auxiliary files, associated shipping knowledge bases, and selected online search options;

[0014] Intent recognition and structuring steps: Utilizing a large language model that integrates a shipping domain knowledge base and a multi-agent collaboration protocol, deep semantic parsing is performed on all available information input by the user. All available information includes the shipping office requirements described in natural language, as well as auxiliary files uploaded by the user and / or associated shipping domain knowledge bases. After parsing, a structured query intent containing the analysis object, time range, target dimension, and file format requirements is output.

[0015] Task decomposition and scheduling steps: Based on the structured query intent, the processing of shipping office needs is divided into multiple sub-tasks, where each intent in the structured query intent corresponds to a sub-task, and each sub-task corresponds to a dedicated intelligent agent and several shipping-related multi-agent collaboration protocols; the dedicated intelligent agent is used to execute the sub-task corresponding to the intent and obtain data support or functional assistance based on the multi-agent collaboration protocols; the dedicated intelligent agent includes: a search intelligent agent, a document parsing intelligent agent, and a knowledge base retrieval intelligent agent;

[0016] Multi-agent collaborative execution steps: For each subtask, the dedicated agent corresponding to that subtask is invoked to perform shipping data acquisition and processing: The search agent generates multi-round search keywords based on the analysis direction in the structured query intent, sequentially searches the network for shipping data related to the search keywords, and finally outputs the network search information obtained from the multi-round search; The document parsing agent parses the auxiliary files uploaded by the user, extracts and outputs file parsing information related to the structured query intent; The knowledge base retrieval agent extracts and outputs knowledge base retrieval information related to the structured query intent from the shipping domain knowledge base using RAG technology; Finally, the network retrieval information, file parsing information, and knowledge base retrieval information output by all invoked dedicated agents are integrated, and in-depth analysis and reasoning are performed to generate shipping analysis results corresponding to the shipping office needs;

[0017] Multimodal content generation steps: Based on the shipping analysis results, generate a shipping analysis result file in a preset format; based on the target file format in the structured query intent, use multimodal processing technology to call the file agent corresponding to each target file format to generate a multimodal target format file containing text files, presentation files, audio files and interactive web page links;

[0018] Visualization and distribution steps: The shipping AI office visualization is realized through online file preview technology, providing online preview, download and link sharing functions for the generated multimodal target format files.

[0019] Preferably, in the shipping office needs interaction and input step, the user's input content also includes the selected in-depth thinking option;

[0020] The task decomposition and scheduling steps further include: determining the complexity of the shipping office requirements based on the structured query intent; if it is a simple shipping office question that the large language model can directly output an accurate answer without external data support, and the user's input does not include deep thinking options, then the large language model is directly called to output the shipping analysis results corresponding to the shipping office requirements; if it is a complex shipping office question that the large language model cannot directly answer, or the user's input includes deep thinking options, then according to the structured query intent, the processing of the shipping office requirements is divided into multiple sub-tasks, each sub-task corresponding to a dedicated intelligent agent and several shipping-related multi-agent cooperation protocols.

[0021] Preferably, the shipping office needs interaction and input steps further include: the user exploring the shipping office needs in depth through multi-round natural language dialogue in the client, and gradually clarifying the needs;

[0022] The intent recognition and structuring steps further include: when the user's input on the client involves historical dialogue content, combining the corresponding historical dialogue input content, historical shipping analysis results, and the input content of the current dialogue, using the large language model that integrates a shipping domain knowledge base and a multi-agent collaboration protocol to perform intent parsing of the shipping office needs of the current dialogue, and outputting a structured query intent.

[0023] Preferably, in the shipping office requirements interaction and input step, the uploaded auxiliary files include text files, code files, and images;

[0024] In the multimodal content generation step, the file agent includes PDF agent, PPT agent, HTML agent, podcast agent, Excel agent, and CSV agent.

[0025] Preferably, in the intent recognition and structuring step, the structured query intent is JSON format data, including whether to generate a report, whether to send an email, the target file format for output, and whether to persist it to a database;

[0026] In the multimodal content generation step, the text files of the multimodal target format files include markdown files, excel files, and csv files; the presentation files include pdf files and ppt files; the audio files are podcast files; and the interactive web page links are html files.

[0027] Preferably, in the intent recognition and structuring step, if the shipping office needs propose an analysis requirement but do not specify a concrete analysis target, then at least one of the following—market dynamics, green shipping, smart shipping, operating costs, and next trend—is taken as the analysis target and written into the structured query intent.

[0028] Preferably, in the task decomposition and scheduling steps, the shipping-related multi-agent cooperation protocol includes ship data MCP, typhoon warning MCP, email assistant MCP, web page deployment MCP, and port dynamics MCP;

[0029] In the multi-agent collaborative execution step, the search agent generates multi-round search keywords based on the analysis direction in the structured query intent, and sequentially retrieves shipping data related to the search keywords in the network through the shipping data retrieval function based on the ship data MCP, the meteorological data retrieval function based on the typhoon warning MCP, and / or the port information retrieval function based on the port dynamics MCP, and finally outputs all the results obtained from the multi-round retrieval.

[0030] In the multimodal content generation step, if the target file format includes HTML, then based on the shipping analysis results, using an HTML agent and combining the webpage generation and deployment function based on the webpage deployment MCP, an interactive dynamic webpage is generated. The dynamic webpage can be previewed and shared via webpage links.

[0031] Preferably, in the visualization and distribution step, the online file preview technology includes kkfileview, OnlyOffice, and Collabora Online;

[0032] View the generation process and progress of each target format file and the shipping analysis results through the client;

[0033] If the structured query intent includes the intent to send an email, then the email agent generates a notification email based on the preset email template and all target files of all formats generated in the multimodal content generation step, using the email sending function of the email assistant MCP, and automatically sends it to the email receiving address in the structured query intent.

[0034] A shipping AI-based office visualization system, based on multi-agent technology, includes a shipping office demand interaction and input module, an intent recognition and structuring module, a task decomposition and scheduling module, a multi-agent collaborative execution module, a multimodal content generation module, and a visualization display and distribution module, connected sequentially.

[0035] The shipping office needs interaction and input module is used to receive user input on the client. The input includes mandatory natural language descriptions of shipping office needs, as well as optional uploaded auxiliary files, associated shipping knowledge bases, and selected online search options.

[0036] The intent recognition module utilizes a large language model that integrates a shipping domain knowledge base and a multi-agent collaboration protocol to perform deep semantic analysis on all available information input by the user. This available information includes the shipping office requirements described in natural language, as well as auxiliary files uploaded by the user and / or associated shipping domain knowledge bases. After analysis, it outputs a structured query intent that includes the analysis object, time range, target dimension, and file format requirements.

[0037] The task decomposition and scheduling module is used to break down the processing of shipping office needs into multiple sub-tasks based on the structured query intent. Each intent in the structured query intent corresponds to a sub-task, and each sub-task corresponds to a dedicated intelligent agent and several shipping-related multi-agent collaboration protocols. The dedicated intelligent agent is used to execute the sub-task corresponding to the intent and obtain data support or functional assistance based on the multi-agent collaboration protocols. The dedicated intelligent agents include: a search intelligent agent, a document parsing intelligent agent, and a knowledge base retrieval intelligent agent.

[0038] The multi-agent collaborative execution module: For each subtask, it calls the dedicated agent corresponding to that subtask to perform shipping data acquisition and processing: The search agent generates multi-round search keywords based on the analysis direction in the structured query intent, sequentially searches the network for shipping data related to the search keywords, and finally outputs the network search information obtained from the multi-round search; The document parsing agent parses the auxiliary files uploaded by the user, extracts and outputs file parsing information related to the structured query intent; The knowledge base retrieval agent extracts and outputs knowledge base retrieval information related to the structured query intent from the shipping domain knowledge base using RAG technology; Finally, it integrates the network retrieval information, file parsing information, and knowledge base retrieval information output by all the called dedicated agents, performs in-depth analysis and reasoning, and generates shipping analysis results corresponding to the shipping office needs;

[0039] The multimodal content generation module is used to generate a shipping analysis result file in a preset format based on the shipping analysis results; and to generate a multimodal target format file containing text files, presentation files, audio files, and interactive web page links by calling the file agents corresponding to each target file format respectively through multimodal processing technology based on the target file format in the structured query intent.

[0040] The visualization and distribution module is used to realize the visualization of shipping AI office through online file preview technology, and provides online preview, download and link sharing functions for the generated multimodal target format files.

[0041] Preferably, in the shipping office demand interaction and input module, the user's input content also includes the selected in-depth thinking option;

[0042] The task decomposition and scheduling module further includes: determining the complexity of the shipping office requirements based on the structured query intent; if it is a simple shipping office question that the large language model can directly output an accurate answer without external data support, and the user's input does not include deep thinking options, then the large language model is directly called to output the shipping analysis results corresponding to the shipping office requirements; if it is a complex shipping office question that the large language model cannot directly answer, or the user's input includes deep thinking options, then according to the structured query intent, the processing of the shipping office requirements is divided into multiple sub-tasks, each sub-task corresponding to a dedicated intelligent agent and several shipping-related multi-agent cooperation protocols.

[0043] The beneficial effects of this invention are as follows:

[0044] This invention provides a shipping AI office visualization method based on multi-agent technology, also known as a shipping AI office visualization method based on multi-agent content generation. This method supports user interaction on the client side using natural language as the core interaction method, supporting flexible input of "mandatory input items (shipping office needs described in natural language) + optional auxiliary items (uploaded auxiliary files, associated shipping domain knowledge bases, and selected online search options)," lowering the barrier to entry and eliminating the need for users to write complex commands or repeatedly switch applications. It solves the problems of single interaction methods and cumbersome operation in existing shipping software, achieving human-computer dialogue-style office work. Furthermore, it supports multi-turn dialogue to deeply explore needs and gradually clarify requirements, significantly improving user office efficiency. Relying on a large language model that integrates a shipping domain knowledge base and a multi-agent collaboration protocol (MCP), it achieves deep semantic analysis of shipping office needs, auxiliary files, and the full range of information in the shipping domain knowledge base, outputting information including the analysis object, time range, target dimension, and file format. The system seeks standardized, structured query intents to ensure unambiguous intents and complete elements, providing clear guidance for subsequent task decomposition. Especially when users provide auxiliary documents and / or knowledge bases, the system can deeply understand the needs of complex office scenarios through multi-source information fusion and analysis. This allows for planning the direction of content generation, adjusting content bias and richness, and reducing intent misjudgment. It is particularly suitable for complex office scenarios, making up for the shortcomings of existing technologies in terms of insufficient understanding of office needs and lack of autonomous decision-making capabilities. Based on the structured query intents, complex shipping office needs are broken down into standardized sub-tasks, achieving precise matching of needs, dedicated intelligent agents, and multi-agent collaboration protocols. The dedicated intelligent agents include search intelligent agents, document parsing intelligent agents, and knowledge base retrieval intelligent agents. This allows each dedicated intelligent agent to focus on its corresponding sub-task, and the multi-agent collaboration protocol provides targeted data / functional support, achieving centralized functions and efficient execution. At the same time, the task decomposition mode can achieve parallel processing of sub-tasks through scheduling, improving overall processing efficiency.For each subtask, a dedicated intelligent agent corresponding to that subtask is invoked to perform shipping data acquisition and processing. Finally, the multi-source information output by all invoked dedicated intelligent agents (including network retrieval information obtained through multiple rounds of searching by the search agent, file parsing information extracted by the document parsing agent, and knowledge base retrieval information obtained by the knowledge base retrieval agent through RAG technology) is integrated, and deep analysis and reasoning are performed to generate shipping analysis results corresponding to the shipping office requirements. In this way, through the division of labor and collaboration among multiple dedicated intelligent agents, efficient acquisition and in-depth processing of multi-source shipping data can be achieved. The multi-round retrieval by the search agent ensures the comprehensiveness of the searched data; the document parsing agent accurately extracts file parsing information related to the structured query intent from auxiliary files; and the knowledge base retrieval agent, through RAG technology, can quickly obtain the knowledge base retrieval information most relevant to the structured query intent in the shipping field knowledge base. Then, through multi-source information integration and in-depth analysis and reasoning, shipping analysis results with decision-making reference value can be generated, rather than the raw data pile output by existing shipping software. This ensures both the comprehensiveness of the shipping analysis results and... The shipping analysis results are concise, streamlined, and more practical. Based on these results, pre-formatted (such as the widely used Markdown format) shipping analysis result files are generated. This allows for output of files viewable and editable on most systems, even when the user doesn't specify an output file format, facilitating user use and distribution. For different file formats, multimodal processing technology is used, employing the corresponding file intelligence agent to process the files, ensuring professionalism and structural integrity. This prevents damage to elements such as mixed text and images, complex tables, and multi-level lists, eliminating the need for users to spend significant time on secondary formatting and improving user satisfaction. Online file preview technology enables visualized display of shipping AI office work, allowing for real-time visualization of multimodal files. This provides an intuitive presentation of file content without relying on external tools, facilitating quick access to file content and quality verification. Download and link sharing functions are also provided, enabling users to quickly distribute office work results and improving the continuity, efficiency, and convenience of shipping office processes. This invention, through interaction and input of shipping office needs, intent recognition and structuring, task decomposition and scheduling, multi-agent collaborative execution, multimodal content generation, and visualization and distribution, combined with RAG technology, dedicated intelligent agents, multi-agent collaboration protocols, and online file preview technology, enables users to query shipping office needs using only natural language. It achieves intelligent processing throughout the entire process from need understanding, content analysis, and generation, ensuring the efficiency and accuracy of the generated shipping analysis results, the structural integrity and usability of the generated target format files, and realizing a one-stop intelligent service for generating multimodal content in shipping-related office scenarios such as document creation, presentation generation, webpage generation, information processing, and audio playback.

[0045] This invention provides users with an optional deep thinking option, allowing them to choose whether to perform in-depth analysis based on the complexity of the shipping office problem. For simple shipping office problems, not using the deep thinking option improves response efficiency (i.e., faster generation of shipping analysis results), while using the deep thinking option for complex shipping office problems makes the analysis results more accurate. If the large language model can directly output an accurate answer without external data support, and the user's input does not include the deep thinking option, then the large language model is directly invoked to output the shipping analysis result corresponding to the shipping office requirement. If the large language model cannot directly answer a complex shipping office problem, or if the user's input includes the deep thinking option, then the processing of the shipping office requirement is broken down into multiple sub-tasks before further processing. The dual judgment conditions of the large language model and the deep thinking option can more accurately distinguish the complexity of shipping office problems, preventing situations where users forget to select the deep thinking option for complex shipping office problems, resulting in a simplistic analysis and ensuring the accuracy of the analysis results. Meanwhile, simple shipping office problems are directly output using large language models, significantly shortening response time and reducing redundant system resource consumption. For complex shipping office problems or scenarios where users explicitly require in-depth analysis, a refined processing flow involving multi-agent collaboration and multi-agent cooperation protocols is initiated to ensure comprehensive data acquisition and in-depth analytical reasoning, avoiding the problem of incomplete analysis caused by simplified processes. This design balances efficient response to simple queries with guaranteed processing quality for complex shipping office problems or in-depth analysis needs, achieving a dual optimization of "efficiency and depth," and adapting to diverse scenarios in shipping office work, from rapid queries to in-depth analysis.

[0046] This invention's human-computer interaction client supports multi-turn natural language dialogue, aligning with users' progressive thinking habits in actual office work. It allows for a deep understanding of a shipping issue from scratch (e.g., first asking "Analyze the fuel consumption of a certain vessel in the last quarter," then supplementing with "Analyze whether there is room for optimization in the fuel consumption of a certain vessel in the last quarter"). Adding new questions based on historical dialogues can correct and guide users to obtain the desired shipping research report. The intent recognition and structuring steps incorporate the correlation between historical dialogue content and past shipping analysis results. When a user's new request involves context, it can automatically link historical dialogue input and historical shipping analysis results, eliminating the need for users to repeat themselves. This improves the accuracy of intent parsing, avoids misjudgment due to contextual fragmentation, and ensures the continuity of requests and the consistency of analysis. This context-linked parsing method allows multi-turn dialogues to form a closed loop of "refining requests—updating intent—deepening analysis," ensuring that the final generated shipping analysis results continuously align with the user's dynamically adjusted core needs. This overcomes the shortcomings of existing technologies that lack continuity and cannot adapt to dynamic user needs.

[0047] The auxiliary files uploaded by this invention include text files, code files, and images, precisely matching the actual data carrier needs of shipping office scenarios. Text files can carry structured information such as shipping contracts and voyage logs; code files allow users to import custom data processing logic or voyage data analysis algorithms; and images can include visual materials such as ship drawings, port layout maps, and voyage trajectory screenshots. This design achieves comprehensive coverage of multiple types of auxiliary data, allowing users to fully utilize diverse resources to support shipping office needs analysis. It avoids problems such as the inability to upload key information or insufficient analysis basis due to file type incompatibility, further improving the comprehensiveness of data support and the richness of analysis reports.

[0048] This invention's document intelligent agents include PDF, PPT, HTML, podcast, Excel, and CSV agents, achieving professional processing with "exclusive agents corresponding to exclusive formats." The PDF agent can generate standardized PDF files, a commonly used format in daily work, suitable for formal office archiving scenarios; the PPT agent can automatically extract core data and conclusions and complete the layout, generating PPT files that meet presentation needs; the HTML agent can generate interactive HTML web pages and links, transforming dry text descriptions into actionable web pages, improving user experience and facilitating sharing via links; the podcast agent can convert text to speech, catering to users' fragmented listening needs; and the Excel and CSV agents can accurately extract structured data to generate tables, suitable for data statistics and secondary analysis scenarios. Compared to existing technologies that rely on general tools for multimodal conversion, are prone to formatting errors (such as distorted tables and messy text and image layouts), and cannot meet the needs of specific office scenarios, this design ensures the conversion quality of each target format through segmented intelligent agents, avoids secondary formatting costs, and comprehensively covers high-frequency scenarios in shipping offices such as archiving, reporting, collaboration, listening, and data statistics. This allows users to obtain files suitable for different purposes without the need for additional external tools, greatly improving office convenience and the efficiency of reusing results.

[0049] This invention addresses the situation where analytical needs arise in shipping office work, but no specific analytical objectives are clearly defined. It identifies at least one of the following as analytical objectives: market dynamics, green shipping, smart shipping, operating costs, and future trends. These objectives are then incorporated into the structured query intent. Even if the user is unfamiliar with the scenario or lacks professional knowledge of shipping analysis frameworks, the system can automatically provide core dimensions such as market dynamics, green shipping, smart shipping, operating costs, and future trends, generating corresponding reports. This avoids situations where users are unsure how to break down their needs or miss key analytical points, resulting in reports lacking practicality. It allows non-professional users to quickly obtain comprehensive and professional analytical reports.

[0050] This invention's multi-agent collaboration protocol is precisely integrated with core shipping office scenarios. The ship data MCP provides core data such as ship positioning and navigation trajectory, the typhoon warning MCP outputs real-time and historical meteorological risk information, and the port dynamics MCP synchronizes key data such as port throughput and berthing efficiency, avoiding the problems of MCP generalization and weak correlation with shipping needs. The search agent accurately calls the ship data MCP (to obtain navigation trajectory), the typhoon warning MCP (to obtain weather information for the sea area along the route), and the port dynamics MCP (to obtain port berthing restrictions) based on the analysis direction in the structured query intent (such as "ship navigation risk analysis"), significantly improving the targeting and efficiency of data retrieval, reducing irrelevant data interference, and ensuring that the acquired shipping data is highly consistent with the analysis objectives. By linking HTML intelligent agents with web-based deployment MCP, the generated web pages are no longer ordinary static pages, but dynamic web pages that support interactive operations (such as directory navigation, data chart interaction, content annotation, etc.), perfectly adapting to the needs of team collaborative analysis and online discussions in shipping offices. At the same time, web-based deployment MCP supports web page generation and deployment, and links can be shared directly. Users do not need to perform complex operations such as additional server deployment and domain name configuration, and can quickly share dynamic web pages with collaborators, improving the efficiency of disseminating office results.

[0051] This invention also relates to a shipping AI office visualization system based on multi-agent technology. This system corresponds to the aforementioned shipping AI office visualization method based on multi-agent technology and can be understood as a system that implements the aforementioned method. It includes a shipping office demand interaction and input module, an intent recognition and structuring module, a task decomposition and scheduling module, a multi-agent collaborative execution module, a multimodal content generation module, and a visualization display and distribution module. These modules work collaboratively, with natural language interaction as the core, and are compatible with multiple types of auxiliary files and knowledge base calls, eliminating the need for users to switch applications. Users can input complex commands to achieve interactive human-computer interaction, significantly lowering the barrier to entry. During this process, users can gain a deep understanding of a shipping issue from scratch through multiple rounds of dialogue, ultimately obtaining the desired report results. Leveraging multi-agent collaboration with a shipping-specific MCP (Multi-Channel Programming Platform), the system achieves accurate acquisition and in-depth analysis of multi-source data, generating shipping analysis results that are both professional and valuable for decision-making, avoiding the need for raw data compilation and manual secondary processing. It supports multi-format file generation and professional online preview, adapting to diverse office scenarios such as archiving, reporting, and collaboration. Combined with convenient download and sharing functions, it improves the efficiency of results flow. This system completely solves the pain points of existing shipping software, such as fragmented functions, complex interactions, insufficient analytical depth, and poor multi-modal format compatibility. It comprehensively optimizes office convenience, data accuracy, and collaborative efficiency, significantly reducing labor costs and adapting to the needs of all shipping office scenarios. Attached Figure Description

[0052] Figure 1 This is a flowchart of the shipping AI office visualization method based on multi-agent technology according to the present invention.

[0053] Figure 2 This is an example diagram illustrating the HTML webpage generation process of the present invention.

[0054] Figure 3 This is an example image of an HTML webpage for this invention.

[0055] Figure 4 This is an example diagram illustrating the shipping analysis report generation process of the present invention.

[0056] Figure 5 and Figure 6 This is an example diagram of the shipping analysis report results of the present invention.

[0057] Figure 7 This is a structural block diagram of the shipping AI office visualization display system based on multi-agent systems according to the present invention. Detailed Implementation

[0058] The present invention will now be described with reference to the accompanying drawings.

[0059] This invention discloses a shipping AI-based visual display method for office work based on multi-agent systems, such as... Figure 1 As shown, the specific steps include:

[0060] I. Shipping Office Needs Interaction and Input Steps: Receive user input on the client side. The input includes mandatory natural language descriptions of shipping office needs, as well as optional uploaded auxiliary files (including text files, code files, and images, etc.), associated shipping knowledge bases, selected online search options, and selected in-depth thinking options.

[0061] After users input their shipping office needs described in natural language on the client, they can choose whether to upload supporting files (i.e., shipping-related documents used to assist in the analysis of the inquiry; users can upload text files, code files, and images locally to provide file support for their shipping questions), whether to link to shipping knowledge bases (including personal and team knowledge bases to provide data support for their shipping questions and to plan the direction of content generation), whether to use the online search function, and whether to use the deep thinking function, depending on the difficulty of the inquiry (i.e., shipping office needs).

[0062] For example, when a user enters "Analysis of COSCO Shipping Lotus's shipping trajectory over the past month" in the client, because the query is a broad analytical question, the voyage log and shipping contract documents of the Lotus vessel are uploaded to assist in the analysis.

[0063] This invention also provides external memory and state management functions, allowing users to delve deeper into shipping office needs through multi-round natural language dialogue within the client, gradually clarifying the requirements. In other words, users can raise new questions based on previous queries without needing to repeat themselves. For example, if a user previously asked about "COSCO Shipping Lotus's shipping trajectory analysis for the past month," after receiving the analysis results, they can further ask, "Optimize the shipping trajectory from a fuel economy perspective," without needing to specify which trajectory.

[0064] Furthermore, the shipping knowledge base in this embodiment of the invention can be divided into a personal knowledge base and a team knowledge base. The personal knowledge base is accessible only to the individual user, while the team knowledge base is accessible to all members of the user's team. The construction of the knowledge base can be a long-term process. In the daily work of users or team members, data files (such as ship navigation logs, shipping contracts, port operation specifications, cargo manifests, etc.) and images (such as ship structure diagrams, port layout maps, route trajectory screenshots, etc.) related to shipping office work can be added to the knowledge base in batches at irregular intervals, continuously expanding the data scale of the knowledge base.

[0065] It is worth noting that data files and images need to undergo quality screening before being added to the knowledge base, removing damaged files, files with weak content relevance, and files with low image clarity to ensure the accuracy of the knowledge base.

[0066] II. Intent Recognition and Structured Process: Utilizing a large language model that integrates a shipping domain knowledge base and a multi-agent collaboration protocol (hereinafter referred to as the large language model), deep semantic parsing is performed on all available information input by the user. This available information includes the shipping office requirements described in natural language, as well as auxiliary files uploaded by the user and / or associated shipping domain knowledge bases. The parsed output includes a structured query intent containing the analysis object, time frame, target dimensions (including whether to send an email, generate a report, persist data to a database, conduct an online search, or engage in deep thinking), file format requirements (also known as the target output file format), the address for receiving emails, and whether to display results only on the current interface. The structured query intent can be in JSON format.

[0067] Specifically, if the user's input includes auxiliary documents and / or a shipping knowledge base, then a large language model is used to perform deep semantic analysis on the user's input shipping office requirements, auxiliary documents, and / or shipping knowledge base; if the user's input does not include auxiliary documents and a shipping knowledge base, then a large language model is used to perform deep semantic analysis on the user's input shipping office requirements.

[0068] Furthermore, when the user's input on the client involves historical dialogue content, the system combines the corresponding historical dialogue input, historical shipping analysis results, and the input content of the current dialogue. It then utilizes the large language model, which integrates a shipping domain knowledge base and a multi-agent collaboration protocol, to perform intent parsing on the shipping office needs of the current dialogue, outputting a structured query intent. Taking "COSCO Shipping Lotus's shipping trajectory analysis for the past month" as an example, after obtaining the corresponding shipping analysis results, if the user further asks "Optimize the above trajectory from a fuel economy perspective," this embodiment of the invention will utilize the large language model, combining the input content corresponding to the query "COSCO Shipping Lotus's shipping trajectory analysis for the past month" and the shipping analysis results, as well as the input content corresponding to the query "Optimize the above trajectory from a fuel economy perspective," to perform intent parsing, ultimately outputting a structured query intent, making the output structured query intent more accurate.

[0069] Because different users have varying abilities to define the analytical goals in their inquiries, if an analytical requirement is mentioned in the shipping office requirements but no specific analytical goal is clearly defined, at least one of the following—market dynamics, green shipping, smart shipping, operating costs, and future trends—will be used as the analytical goal and written into the structured query intent. This embodiment of the invention constructs a requirement-analysis goal mapping table based on historical high-frequency analytical goals in shipping office work and their corresponding shipping office requirements. Thus, when users do not specify a concrete analytical goal, they can use this mapping table to select the most relevant analytical goal for their shipping office requirements, thereby improving the effectiveness of shipping analysis results and user satisfaction.

[0070] III. Task Decomposition and Scheduling Steps: Based on the structured query intent, determine the complexity of the shipping office requirements. If it is a simple shipping office question that a large language model can directly output an accurate answer without external data support, and the user's input does not include deep thinking options, then directly call the large language model to output the shipping analysis results corresponding to the shipping office requirements. If it is a complex shipping office question that a large language model cannot directly answer, or the user's input includes deep thinking options, then according to the structured query intent, the processing of the shipping office requirements is divided into multiple sub-tasks. Each intent in the structured query intent corresponds to a sub-task, and each sub-task corresponds to a dedicated intelligent agent and several shipping-related multi-agent collaboration protocols (including ship data MCP, typhoon warning MCP, email assistant MCP, webpage deployment MCP, port dynamics MCP, etc.). The dedicated intelligent agent is used to execute the sub-task corresponding to the intent and obtain data support or functional assistance based on the multi-agent collaboration protocols. The dedicated intelligent agent includes: a search intelligent agent, a document parsing intelligent agent, and a knowledge base retrieval intelligent agent.

[0071] For example, when the user inputs a query of "COSCO Shipping Lotus's shipping trajectory analysis in the past month", the sub-tasks include "information extraction", "network search", "content generation", "format processing", and "risk warning".

[0072] In this embodiment of the invention, the results are output in the form of shipping analysis results by default. If the user explicitly wants to generate a report, that is, if the "generate report" item in the structured query intent is "generate report", then the task decomposition and scheduling steps will further generate a shipping analysis report based on the shipping analysis results.

[0073] IV. Multi-Agent Collaborative Execution Steps: For each subtask, the dedicated agent corresponding to that subtask is invoked to perform shipping data acquisition and processing: The search agent generates multi-round search keywords based on the analysis direction in the structured query intent, sequentially searches the network for shipping data related to the search keywords, and finally outputs the network search information obtained from the multi-round search; The document parsing agent parses the auxiliary files uploaded by the user, extracts and outputs file parsing information related to the structured query intent; The knowledge base retrieval agent extracts and outputs knowledge base retrieval information related to the structured query intent from the shipping domain knowledge base using RAG technology; Finally, the network retrieval information, file parsing information, and knowledge base retrieval information output by all invoked dedicated agents are integrated, and in-depth analysis and reasoning are performed to generate shipping analysis results corresponding to the shipping office needs.

[0074] Specifically, the search agent generates multi-round search keywords based on the analysis direction in the structured query intent, and sequentially retrieves shipping data related to the search keywords in the network through the shipping data retrieval function based on the ship data MCP, the meteorological data retrieval function based on the typhoon warning MCP, and / or the port information retrieval function based on the port dynamics MCP, and finally outputs all the results obtained from the multi-round search.

[0075] The document parsing agent outputs information related to structured query intent through three steps: decomposing structured query intent and generating keywords, parsing multiple types of auxiliary documents, and filtering based on association with structured query intent.

[0076] The structured query intent decomposition and keyword generation process includes: First, parsing the JSON-formatted structured query intent to extract core elements: analysis direction (e.g., "ship fuel cost analysis"), key entities (e.g., ship name, time period, port name), and demand type (e.g., data statistics, specification query, fault analysis). Then, using the TF-IDF algorithm and a shipping industry dictionary (including ship terminology, port names, and business scenario vocabulary), core keywords (e.g., "COSCO Shipping Lotus," "Q3 2024," "fuel consumption," "port berthing fees") are selected to generate an intent keyword library.

[0077] Multi-type auxiliary file parsing includes: using different methods to parse different file formats. For example, for PDF files, PyPDF2 is used to extract plain text content; for scanned PDFs (image-based text), Tesseract OCR is used for text recognition, while pdfplumber is used to extract table data (preserving row and column structure) and image position information; for Word files, python-docx is used to parse the document structure, extract the main text, tables, and image annotations, and preserve chapter hierarchy; for Excel files, pandas is used to read table data, standardize table headers (e.g., unify "fuel consumption" and "fuel usage" as "fuel consumption"), and identify data dimensions (e.g., time, ships, numerical values); for TXT files, the text content is read directly, and redundant spaces and line breaks are removed using regular expressions to optimize the text structure.

[0078] The association filtering with structured query intents includes: first, comparing the parsed structured content (such as text entities, table fields, image tags, and code variables) with the intent keyword library, retaining information with a high degree of matching; then, using the Sentence-BERT model to calculate the semantic similarity between the parsed content and the analysis direction (such as the semantic similarity between the analysis direction "fuel cost analysis" and "2024 Q3 ship fuel replenishment record" in the file), retaining information with a similarity ≥ 0.7; then, using the SimHash algorithm to remove duplicate information; finally, integrating the filtered structured query intent-related information into a unified format and outputting the results.

[0079] The knowledge base retrieval agent uses RAG technology to search and locate information related to the structured query intent from the knowledge base, and sorts the searched information based on its relevance to the structured query intent. The agent then selects a preset number of information entries with the highest relevance as the information related to the structured query intent.

[0080] It is worth noting that if the user's input does not include online search options, the search agent will not be invoked; if the user's input does not include auxiliary files and knowledge bases, the document parsing agent will not be invoked; if the complexity judgment result of the shipping office needs is a simple shipping office problem, then the multi-agent collaborative execution step does not need to be executed.

[0081] V. Multimodal Content Generation Steps: Based on the shipping analysis results, generate a shipping analysis result file in a preset format (set as needed, generally in Markdown format); based on the target file format in the structured query intent, use multimodal processing technology to call the corresponding file agents (including PDF agent, PPT agent, HTML agent, podcast agent, Excel agent, CSV agent, etc.) for each target file format to generate a multimodal target format file containing text files, presentation files, audio files, and interactive web page links.

[0082] Among them, the text files of the multimodal target format files include Markdown files, Excel files, CSV files, etc.; the presentation files include PDF files, PPT files, etc.; the audio files are podcast files; and the interactive web page links are HTML files.

[0083] If the target file format contains HTML, then based on the shipping analysis results, using an HTML agent and combining the webpage generation and deployment function based on the webpage deployment MCP, an interactive dynamic webpage is generated. The dynamic webpage can be previewed and shared via a webpage link.

[0084] VI. Visualization and Distribution Steps: The shipping AI office visualization is realized through online file preview technology (such as kkfileview, OnlyOffice, Collabora Online), providing online preview, download and link sharing functions for the generated multimodal target format files.

[0085] In embodiments of the present invention, the generation process and progress of various target format files, shipping analysis results, and shipping analysis reports can be viewed through a client, such as... Figure 2 As shown, this demonstrates the HTML page generation process for the query "COSCO Shipping Lotus Shipping Track Analysis in the Past Month". The middle section displays the HTML code generation process, clearly showing the complete code generation process including tag writing, style definition, and data embedding. The bottom displays the percentage of generation progress (e.g., ...). Figure 2 (Displaying 1.6%) allows users to monitor the file generation status in real time. The final generated result is as follows: Figure 3As shown, the left side is a content navigation bar, including an introduction, ship information and background, detailed analysis of shipping routes, relevant data and context, risk factors, optimization suggestions, and conclusions. Users can jump to the corresponding location on the HTML webpage by clicking on the target item in the navigation bar, which is convenient for users to browse quickly. The generated results can be shared via URL, which is convenient for office collaboration and dissemination.

[0086] Figure 4 The process of generating the shipping analysis report for the inquiry "Analysis of COSCO Shipping Lotus's Shipping Track in the Past Month" is clear, fully visualized, and uses vibrant, changing colors to highlight the report's progress. The final shipping analysis report includes a detailed analysis of COSCO Shipping Lotus's shipping track (such as...). Figure 5 As shown), risk factors (also known as influencing factors, such as...) Figure 6 (as shown) and optimization suggestions (also known as strategic suggestions, such as...) Figure 6 As shown), where, Figure 5 The detailed analysis of the COSCO Shipping Lotus shipping trajectory shown includes port call records (such as the estimated / actual arrival and departure times and deviation analysis of Pusan ​​Port, Shanghai Port, Ningbo Port, and Singapore Port), timeline and delay analysis (such as the definition of high-frequency activity periods and low-activity periods and the reasons for delays, such as the Red Sea crisis and the impact of port congestion), and regional activity patterns (such as the regional distribution characteristics of 60% of calls being made at ports in the Yangtze River Delta and 20% at Singapore Port, with East Asia as the core and Southeast Asia as the expansion point). Figure 6 The risk factors identified include geopolitical disturbances, cost pressures, and data limitations, while optimization recommendations include full-chain resource integration, digital and intelligent upgrades, and low-carbon transformation.

[0087] In this embodiment of the invention, if the structured query intent includes the intent to send an email, then the email agent generates a notification email based on a preset email template and all target files of various formats generated in the multimodal content generation step, using the email sending function of the email assistant MCP, and automatically sends it to the email receiving address in the structured query intent.

[0088] Based on the same inventive concept, one or more embodiments of this specification also provide a shipping AI office visualization display system based on multi-agent technology. Since the working principle of the shipping AI office visualization display system based on multi-agent technology is the same as that of the aforementioned shipping AI office visualization display method based on multi-agent technology, the implementation of the shipping AI office visualization display system based on multi-agent technology can refer to the aforementioned implementation of the shipping AI office visualization display method based on multi-agent technology, and the repeated parts will not be described again.

[0089] Figure 7This specification provides a structural block diagram of a shipping AI-based office visualization system, which is used in one or more embodiments. Figure 7 As shown, the system includes, in sequence, a shipping office demand interaction and input module 101, an intent recognition and structuring module 102, a task decomposition and scheduling module 103, a multi-agent collaborative execution module 104, a multimodal content generation module 105, and a visualization display and distribution module 106. Among these,

[0090] The shipping office needs interaction and input module 101 is used to receive user input content on the client. The input content includes mandatory shipping office needs described in natural language, as well as optional uploaded auxiliary files, associated shipping field knowledge bases, selected online search options, and selected in-depth thinking options.

[0091] The intent recognition module 102 utilizes a large language model that integrates a shipping domain knowledge base and a multi-agent collaboration protocol to perform deep semantic analysis on all available information input by the user. The available information includes the shipping office requirements described in the natural language, as well as the auxiliary files uploaded by the user and / or the associated shipping domain knowledge base. After analysis, the module outputs a structured query intent that includes the analysis object, time range, target dimension, and file format requirements.

[0092] The task decomposition and scheduling module 103 is used to decompose the processing of shipping office needs into multiple sub-tasks according to the structured query intent. Each intent in the structured query intent corresponds to a sub-task, and each sub-task corresponds to a dedicated intelligent agent and several shipping-related multi-agent cooperation protocols. The dedicated intelligent agent is used to execute the sub-task corresponding to the intent and obtain data support or functional assistance based on the multi-agent cooperation protocols. The dedicated intelligent agent includes: a search intelligent agent, a document parsing intelligent agent, and a knowledge base retrieval intelligent agent.

[0093] The multi-agent collaborative execution module 104, for each subtask, calls the dedicated agent corresponding to that subtask to perform shipping data acquisition and processing: the search agent generates multi-round search keywords based on the analysis direction in the structured query intent, sequentially searches the network for shipping data related to the search keywords, and finally outputs the network search information obtained from the multi-round search; the document parsing agent parses the auxiliary files uploaded by the user, extracts and outputs file parsing information related to the structured query intent; the knowledge base retrieval agent extracts and outputs knowledge base retrieval information related to the structured query intent from the shipping domain knowledge base using RAG technology; finally, it integrates the network retrieval information, file parsing information, and knowledge base retrieval information output by all the called dedicated agents, performs in-depth analysis and reasoning, and generates shipping analysis results corresponding to the shipping office needs;

[0094] The multimodal content generation module 105 is used to generate a shipping analysis result file in a preset format based on the shipping analysis results; based on the target file format in the structured query intent, it uses multimodal processing technology to call the file agent corresponding to each target file format to generate a multimodal target format file containing text files, presentation files, audio files and interactive web page links.

[0095] The visualization and distribution module 106 is used to realize the visualization of shipping AI office through online file preview technology, and provides online preview, download and link sharing functions for the generated multimodal target format files.

[0096] Furthermore, the task decomposition and scheduling module 103 also includes: determining the complexity of the shipping office requirements based on the structured query intent; if it is a simple shipping office question that the large language model can directly output an accurate answer without external data support, and the user's input does not include deep thinking options, then the large language model is directly called to output the shipping analysis results corresponding to the shipping office requirements; if it is a complex shipping office question that the large language model cannot directly answer, or the user's input includes deep thinking options, then according to the structured query intent, the processing of the shipping office requirements is divided into multiple sub-tasks, each sub-task corresponding to a dedicated intelligent agent and several shipping-related multi-agent cooperation protocols.

[0097] The system provided in this invention, based on user-submitted shipping office needs, uploaded auxiliary files, associated shipping knowledge bases, and shipping-related multi-agent collaboration protocols, can generate and display highly relevant shipping research reports. The system includes multi-round question interaction, file upload, knowledge base access, network search, report generation, multi-type report preview and display, and related system operation and interaction logic.

[0098] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail with reference to the accompanying drawings and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention. In short, all technical solutions and improvements that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention patent.

Claims

1. A shipping AI-based visual office display method based on multi-agent technology, characterized in that, Includes the following steps: Shipping office needs interaction and input steps: Receive user input on the client, which includes mandatory natural language description of shipping office needs, as well as optional uploaded auxiliary files, associated shipping knowledge bases, and selected online search options; Intent recognition and structuring steps: Utilizing a large language model that integrates a shipping domain knowledge base and a multi-agent collaboration protocol, deep semantic parsing is performed on all available information input by the user. All available information includes the shipping office requirements described in natural language, as well as auxiliary files uploaded by the user and / or associated shipping domain knowledge bases. After parsing, a structured query intent containing the analysis object, time range, target dimension, and file format requirements is output. Task decomposition and scheduling steps: Based on the structured query intent, the processing of shipping office needs is divided into multiple sub-tasks, where each intent in the structured query intent corresponds to a sub-task, and each sub-task corresponds to a dedicated intelligent agent and several shipping-related multi-agent collaboration protocols; the dedicated intelligent agent is used to execute the sub-task corresponding to the intent and obtain data support or functional assistance based on the multi-agent collaboration protocols; the dedicated intelligent agent includes: a search intelligent agent, a document parsing intelligent agent, and a knowledge base retrieval intelligent agent; Multi-agent collaborative execution steps: For each subtask, the dedicated agent corresponding to that subtask is invoked to perform shipping data acquisition and processing: The search agent generates multi-round search keywords based on the analysis direction in the structured query intent, sequentially searches the network for shipping data related to the search keywords, and finally outputs the network search information obtained from the multi-round search; The document parsing agent parses the auxiliary files uploaded by the user, extracts and outputs file parsing information related to the structured query intent; The knowledge base retrieval agent extracts and outputs knowledge base retrieval information related to the structured query intent from the shipping domain knowledge base using RAG technology; Finally, the network retrieval information, file parsing information, and knowledge base retrieval information output by all invoked dedicated agents are integrated, and in-depth analysis and reasoning are performed to generate shipping analysis results corresponding to the shipping office needs; Multimodal content generation steps: Based on the shipping analysis results, generate a shipping analysis result file in a preset format; based on the target file format in the structured query intent, use multimodal processing technology to call the file agent corresponding to each target file format to generate a multimodal target format file containing text files, presentation files, audio files and interactive web page links; Visualization and distribution steps: The shipping AI office visualization is realized through online file preview technology, providing online preview, download and link sharing functions for the generated multimodal target format files.

2. The method according to claim 1, characterized in that, In the aforementioned shipping office needs interaction and input steps, the user's input also includes the selected in-depth thinking options; The task decomposition and scheduling steps further include: determining the complexity of the shipping office requirements based on the structured query intent; if it is a simple shipping office question that the large language model can directly output an accurate answer without external data support, and the user's input does not include deep thinking options, then the large language model is directly called to output the shipping analysis results corresponding to the shipping office requirements; if it is a complex shipping office question that the large language model cannot directly answer, or the user's input includes deep thinking options, then according to the structured query intent, the processing of the shipping office requirements is divided into multiple sub-tasks, each sub-task corresponding to a dedicated intelligent agent and several shipping-related multi-agent cooperation protocols.

3. The method according to claim 1 or 2, characterized in that, The interaction and input steps for shipping office needs also include: users exploring shipping office needs in depth through multi-round natural language dialogue in the client, gradually clarifying the needs; The intent recognition and structuring steps further include: when the user's input on the client involves historical dialogue content, combining the corresponding historical dialogue input content, historical shipping analysis results, and the input content of the current dialogue, using the large language model that integrates a shipping domain knowledge base and a multi-agent collaboration protocol to perform intent parsing of the shipping office needs of the current dialogue, and outputting a structured query intent.

4. The method according to claim 1 or 2, characterized in that, In the aforementioned shipping office requirements interaction and input steps, the uploaded auxiliary files include text files, code files, and images; In the multimodal content generation step, the file agent includes PDF agent, PPT agent, HTML agent, podcast agent, Excel agent, and CSV agent.

5. The method according to claim 1 or 2, characterized in that, In the intent recognition and structuring step, the structured query intent is JSON format data, including whether to generate a report, whether to send an email, the target file format for output, and whether to persist it to a database; In the multimodal content generation step, the text files of the multimodal target format files include markdown files, excel files, and csv files; the presentation files include pdf files and ppt files; the audio files are podcast files; and the interactive web page links are html files.

6. The method according to claim 1 or 2, characterized in that, In the intent identification and structuring step, if the shipping office requirements raise an analysis requirement but do not specify a concrete analysis target, then at least one of the following—market dynamics, green shipping, smart shipping, operating costs, and next trend—will be taken as the analysis target and written into the structured query intent.

7. The method according to claim 1 or 2, characterized in that, In the task decomposition and scheduling steps, the shipping-related multi-agent cooperation protocols include ship data MCP, typhoon warning MCP, email assistant MCP, web page deployment MCP, and port dynamics MCP. In the multi-agent collaborative execution step, the search agent generates multi-round search keywords based on the analysis direction in the structured query intent, and sequentially retrieves shipping data related to the search keywords in the network through the shipping data retrieval function based on the ship data MCP, the meteorological data retrieval function based on the typhoon warning MCP, and / or the port information retrieval function based on the port dynamics MCP, and finally outputs all the results obtained from the multi-round retrieval. In the multimodal content generation step, if the target file format includes HTML, then based on the shipping analysis results, using an HTML agent and combining the webpage generation and deployment function based on the webpage deployment MCP, an interactive dynamic webpage is generated. The dynamic webpage can be previewed and shared via webpage links.

8. The method according to claim 7, characterized in that, In the visualization and distribution steps, the online file preview technologies include kkfileview, OnlyOffice, and Collabora Online; View the generation process and progress of each target format file and the shipping analysis results through the client; If the structured query intent includes the intent to send an email, then the email agent generates a notification email based on the preset email template and all target files of all formats generated in the multimodal content generation step, using the email sending function of the email assistant MCP, and automatically sends it to the email receiving address in the structured query intent.

9. A shipping AI-based office visualization system based on multi-agent technology, characterized in that, This includes, in sequence, a shipping office needs interaction and input module, an intent recognition and structuring module, a task decomposition and scheduling module, a multi-agent collaborative execution module, a multimodal content generation module, and a visualization display and distribution module. The shipping office needs interaction and input module is used to receive user input on the client. The input includes mandatory natural language descriptions of shipping office needs, as well as optional uploaded auxiliary files, associated shipping knowledge bases, and selected online search options. The intent recognition module utilizes a large language model that integrates a shipping domain knowledge base and a multi-agent collaboration protocol to perform deep semantic analysis on all available information input by the user. This available information includes the shipping office requirements described in natural language, as well as auxiliary files uploaded by the user and / or associated shipping domain knowledge bases. After analysis, it outputs a structured query intent that includes the analysis object, time range, target dimension, and file format requirements. The task decomposition and scheduling module is used to break down the processing of shipping office needs into multiple sub-tasks based on the structured query intent. Each intent in the structured query intent corresponds to a sub-task, and each sub-task corresponds to a dedicated intelligent agent and several shipping-related multi-agent collaboration protocols. The dedicated intelligent agent is used to execute the sub-task corresponding to the intent and obtain data support or functional assistance based on the multi-agent collaboration protocols. The dedicated intelligent agents include: a search intelligent agent, a document parsing intelligent agent, and a knowledge base retrieval intelligent agent. The multi-agent collaborative execution module: For each subtask, it calls the dedicated agent corresponding to that subtask to perform shipping data acquisition and processing: The search agent generates multi-round search keywords based on the analysis direction in the structured query intent, sequentially searches the network for shipping data related to the search keywords, and finally outputs the network search information obtained from the multi-round search; The document parsing agent parses the auxiliary files uploaded by the user, extracts and outputs file parsing information related to the structured query intent; The knowledge base retrieval agent extracts and outputs knowledge base retrieval information related to the structured query intent from the shipping domain knowledge base using RAG technology; Finally, it integrates the network retrieval information, file parsing information, and knowledge base retrieval information output by all the called dedicated agents, performs in-depth analysis and reasoning, and generates shipping analysis results corresponding to the shipping office needs; The multimodal content generation module is used to generate a shipping analysis result file in a preset format based on the shipping analysis results; and to generate a multimodal target format file containing text files, presentation files, audio files, and interactive web page links by calling the file agents corresponding to each target file format respectively through multimodal processing technology based on the target file format in the structured query intent. The visualization and distribution module is used to realize the visualization of shipping AI office through online file preview technology, and provides online preview, download and link sharing functions for the generated multimodal target format files.

10. The system according to claim 9, characterized in that, In the shipping office needs interaction and input module, the user's input also includes the selected in-depth thinking options; The task decomposition and scheduling module further includes: determining the complexity of the shipping office requirements based on the structured query intent; if it is a simple shipping office question that the large language model can directly output an accurate answer without external data support, and the user's input does not include deep thinking options, then the large language model is directly called to output the shipping analysis results corresponding to the shipping office requirements; if it is a complex shipping office question that the large language model cannot directly answer, or the user's input includes deep thinking options, then according to the structured query intent, the processing of the shipping office requirements is divided into multiple sub-tasks, each sub-task corresponding to a dedicated intelligent agent and several shipping-related multi-agent cooperation protocols.

Images