A method and apparatus for visual flow field analysis driven by a large language model

Abstract

This invention relates to the field of visual analytics technology, and discloses a method and apparatus for flow field visual analysis driven by a large language model. The method includes: acquiring a flow field dataset to be analyzed and user commands; performing semantic analysis on the user commands using a large language model to obtain structured commands; based on the structured commands, retrieving corresponding target data from the flow field dataset using the large language model, and generating descriptive text corresponding to the target data; running a visualization agent based on the target data and the descriptive text to obtain visualization results; and displaying the target data according to the data display method indicated by the user commands. By using a large language model to achieve natural language-based interaction, user operation commands or text commands are converted into structured commands, thereby presenting visualization results and explanations in real time according to user needs, greatly improving the user experience.

Description

Technical Field

[0001] This invention relates to the field of visual analytics technology, specifically to a method and apparatus for flow field visual analytics driven by a large language model. Background Technology

[0002] Flow field data has wide applications in the field of visualization. The complexity of flow field data research mainly lies in its high dimensionality and multivariate characteristics. Currently popular flow field visualization analysis tools still suffer from poor usability and high learning costs in practical use. These tools usually rely on complex query statements or parameter settings and cannot process users' natural language. Users not only need to have professional knowledge in related fields, but also must master how to operate specific systems.

[0003] Therefore, there is an urgent need for a flow field visual analysis method that can interact with natural language, transforming users' natural language commands into specific visualization tasks, which greatly improves the user experience. Summary of the Invention

[0004] In view of this, this application provides a method and apparatus for flow field visual analysis driven by a large language model, which can transform the user's natural language instructions into specific visualization tasks, greatly improving the user experience. The technical solution is as follows.

[0005] In a first aspect, the present invention provides a large language model-driven flow field visual analysis method, the method comprising:

[0006] Obtain the data set of the flow field to be analyzed and the user commands;

[0007] The user's command is semantically analyzed using a large language model to obtain a structured command.

[0008] Based on this structured instruction, the corresponding target data is obtained from the flow field dataset to be analyzed through the large language model, and the corresponding descriptive text of the target data is generated.

[0009] Based on the target data and the descriptive text, run the visualization agent to obtain the visualization results;

[0010] Display the target data according to the data display method specified by the user instruction.

[0011] In one alternative implementation, the user instruction is semantically analyzed using a large language model to obtain structured instructions, including:

[0012] The user command is semantically analyzed using a large language model to obtain the corresponding task identifier. The task identifier is then parameterized using predefined parsing rules and transformed into a structured command.

[0013] In one alternative implementation, the user instructions include natural language, interface button operation instructions, or mouse gesture instructions.

[0014] In one optional implementation, the acquisition of corresponding target data from the flow field dataset to be analyzed via the large language model includes:

[0015] Extract the corresponding instruction task from the structured instructions; based on the instruction task, identify the corresponding proxy function in the large language model proxy layer; execute the instruction task through the proxy function to obtain the corresponding target data from the flow field dataset to be analyzed.

[0016] In one alternative implementation, the descriptive text corresponding to the target data includes:

[0017] The target data is analyzed using this large language model to generate corresponding descriptive text. This descriptive text includes: variable descriptions, dimensions, units of measurement, or value range information.

[0018] In one optional implementation, the target data is displayed according to the data display method indicated by the user instruction, including:

[0019] Extract the corresponding data display method from the user instruction; the data display method includes charts, renderings, models, text descriptions, or video animations; display the target data according to the data display method.

[0020] The present invention provides a flow field visual analysis method driven by a large language model, which has the following advantages.

[0021] The flow field visualization analysis method driven by a large language model of this invention first requires acquiring the flow field dataset to be analyzed and specific user commands. These user commands can be generated by the user through natural language input, clicking interface buttons, or performing specific mouse gestures. Then, a pre-trained large language model performs semantic analysis on the user commands to obtain the corresponding task identifier. Parameters are extracted from the task identifier using preset parsing rules, and a structured transformation is performed to obtain the structured command corresponding to the user command. This structured command is input into the proxy layer of the large language model, and the corresponding proxy function in the proxy layer executes the command task corresponding to the structured command, thereby retrieving the corresponding target data from the flow field dataset to be analyzed. Simultaneously, the large language model generates descriptive text messages corresponding to the target data. This descriptive text information includes variable descriptions, dimensions, units of measurement, and value range information to help users better understand the background and meaning of the target data. Based on the target data and the corresponding descriptive text, visualization results are generated. Before displaying the visualization results, the visualization display method needs to be obtained from the user commands. This display method includes charts, rendered images, models, text descriptions, or video animations. The system displays target data according to the user's instructions. By using a large language model to recognize the user's specific instructions, it enables natural language-based interaction. This allows the system to convert user commands or text commands into structured instructions, thereby obtaining the corresponding target data. The large language model then generates corresponding descriptive explanations, presenting the target data and their accompanying descriptions in real time. This helps users better understand the background and meaning of the target data, significantly enhancing the user experience.

[0022] Secondly, the present invention provides a large language model-driven flow field visual analysis device, the device comprising:

[0023] The acquisition module is used to acquire the data set of the flow field to be analyzed and user commands;

[0024] The instruction conversion module is used to perform semantic analysis on the user's instruction using a large language model to obtain structured instructions;

[0025] The target data acquisition module is used to acquire the corresponding target data from the flow field dataset to be analyzed based on the structured instructions and the large language model, and generate the corresponding descriptive text of the target data.

[0026] The results generation module is used to run a visualization agent based on the target data and the descriptive text to obtain visualization results;

[0027] The results display module is used to display the target data according to the data display method indicated by the user instruction.

[0028] In one alternative implementation, the instruction conversion module is specifically used for:

[0029] The user command is semantically analyzed using a large language model to obtain the corresponding task identifier. The task identifier is then parameterized using predefined parsing rules and transformed into a structured command.

[0030] In one optional implementation, the target data acquisition module is specifically used for:

[0031] Extract the corresponding instruction task from the structured instructions; based on the instruction task, identify the corresponding proxy function in the large language model proxy layer; execute the instruction task through the proxy function to obtain the corresponding target data from the flow field dataset to be analyzed.

[0032] In one optional implementation, the target data acquisition module is further configured to:

[0033] The target data is analyzed using this large language model to generate corresponding descriptive text. This descriptive text includes: variable descriptions, dimensions, units of measurement, or value range information.

[0034] In one alternative implementation, the results display module is specifically used for:

[0035] Extract the corresponding data display method from the user instruction; the data display method includes charts, renderings, models, text descriptions, or video animations; display the target data according to the data display method.

[0036] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the large language model-driven flow field visual analysis method described in the first aspect or any corresponding embodiment thereof.

[0037] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the large language model-driven flow field visual analysis method described in the first aspect or any corresponding embodiment thereof.

[0038] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the large language model-driven flow field visual analysis method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0039] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0040] Figure 1 This is a flowchart of a flow field visual analysis method driven by a large language model provided in an embodiment of the present invention.

[0041] Figure 2 This is a schematic diagram of the front-end interface of a progressive visual analytics system based on a large language model, according to an exemplary embodiment.

[0042] Figure 3 This is a schematic diagram of the visual analysis process of a progressive visual analysis system based on a large language model, according to an exemplary embodiment.

[0043] Figure 4 This is a schematic diagram illustrating the backend user command processing flow of a progressive visual analytics system based on a large language model, according to an exemplary embodiment.

[0044] Figure 5 This is a schematic diagram of a plug-in visualization design for a progressive visual analysis system based on a large language model, according to an exemplary embodiment.

[0045] Figure 6 This is a schematic diagram illustrating the visualization semantic enhancement process of a progressive visual analytics system based on a large language model, according to an exemplary embodiment.

[0046] Figure 7 This is a schematic diagram illustrating the global streamline distribution within a dataset according to an exemplary embodiment.

[0047] Figure 8 This is a schematic diagram illustrating the global velocity distribution of a dataset according to an exemplary embodiment.

[0048] Figure 9 This is a schematic diagram illustrating the velocity distribution in a local region according to an exemplary embodiment.

[0049] Figure 10 The system, as illustrated in an exemplary embodiment, generates a heat map and automatically describes a schematic diagram.

[0050] Figure 11 This is a schematic diagram illustrating the meridional and latitudinal velocity distribution near a relevant sea area according to an exemplary embodiment.

[0051] Figure 12This is a schematic diagram illustrating the underwater velocity distribution according to an exemplary embodiment.

[0052] Figure 13 This is a schematic diagram illustrating a comparison of temperature and salinity distribution in a local area according to an exemplary embodiment.

[0053] Figure 14 This is a schematic diagram illustrating the trend of regional average temperature and salinity changes according to an exemplary embodiment.

[0054] Figure 15 This is a schematic diagram of the structure of a flow field visual analysis device driven by a large language model, provided in an embodiment of this application.

[0055] Figure 16 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0057] It should be understood that the term "instruction" mentioned in the embodiments of this application can be a direct instruction, an indirect instruction, or an indication of a relationship. For example, A instructing B can mean that A directly instructs B, such as B being able to obtain information through A; it can also mean that A indirectly instructs B, such as A instructing C, so B can obtain information through C; or it can mean that there is a relationship between A and B.

[0058] In the description of the embodiments of this application, the term "correspondence" may indicate that there is a direct or indirect correspondence between two things, or that there is an association between two things, or that there is a relationship of instruction and being instructed, configuration and being configured, etc.

[0059] In the embodiments of this application, "predefined" can be achieved by pre-storing corresponding codes, tables or other means that can be used to indicate relevant information in the device (e.g., including terminal devices and network devices). This application does not limit the specific implementation method.

[0060] Real-time analysis of large-scale flow field data faces challenges such as inefficiency and complex interactions. Streamline-based visualization provides a technical means for studying flow field data. As the time span and variable dimensions increase, the scale of ensemble simulation data expands exponentially. The traditional computation-then-analysis approach involves lengthy waiting times and lacks the ability to intervene midway, while the large number of intermediate results also puts pressure on storage and I / O.

[0061] The complexity of flow field data research lies primarily in its high dimensionality and multivariate characteristics. Flow field data often originates from long-term observations or high-precision numerical simulations, covering not only multidimensional attributes such as velocity, temperature, and salinity, but also timescales ranging from weeks to years. In terms of spatial capacity, intermediate data generated during computation and visualization place severe demands on data storage and I / O, especially the results from volume rendering, which further exacerbate the data expansion. In practical visualization applications, the size of the loaded data is typically limited to the GB level or lower. While directly expanding storage capacity may seem like a solution, it fails to address the core issue. Traditional visualization workflows generally employ a post-processing model, requiring all experiments or simulations to complete and generate all results before inputting them into the system for visualization and analysis. This results in significant delays in analytical feedback during large-scale or complex calculations, as analysis cannot begin until all data processing is complete. This not only slows down research progress but may also cause missed opportunities to adjust experimental strategies, limiting the ability to dynamically explore data and promptly identify potential problems.

[0062] With the increasing demand for scientific data visualization, interactive operations enable users to analyze and explore flow field data more deeply, greatly facilitating the understanding and interpretation of complex datasets. However, currently popular flow field visualization and analysis tools still suffer from poor usability in practical use. These interfaces often rely on complex query statements or parameter settings, posing a significant challenge for users without a technical background. Users not only need professional knowledge in the relevant field but also must master how to operate the specific system, which undoubtedly increases the difficulty of use.

[0063] In recent years, natural language processing (NLP) technology has experienced rapid development, especially with the emergence of pre-trained models and large language models, which have provided strong technical support for understanding and processing natural language. Integrating NLP and understanding technologies into visual interactive interfaces is an innovative development direction.

[0064] Most current visualization systems employ a fixed graphical user interface (GUI), lacking flexibility in component layout and adjustment. When numerous visualization components are integrated onto a single interface, users often feel overwhelmed, increasing the burden of operation and exploration. Furthermore, the fixed visualization design hinders users from reviewing their operational history. When users move on to new cases, previous operations and configurations are overwritten and reset by the new operations, restricting their ability to switch flexibly between different datasets or cases, impacting the coherence and efficiency of their research. Finally, adjustments to system components (such as adding or removing chart support) affect the layout of other components, indicating a need for improved scalability.

[0065] Therefore, this invention provides a flow field visual analysis method driven by a large language model. By recognizing the user's specific instructions through a large language model, it achieves interaction based on natural language. It can convert the user's operation instructions or text instructions into structured instructions, thereby obtaining the corresponding target data. It can also generate corresponding descriptive descriptions through the large language model and present the target data and corresponding descriptive descriptions in real time to help users better understand the background and meaning of the target data, greatly improving the user experience.

[0066] The large language model-driven flow field visualization analysis method provided in this embodiment of the invention is as follows: Figure 1 As shown, it includes the following steps.

[0067] S101. Obtain the data set of the flow field to be analyzed and the user instructions.

[0068] Optionally, in the above steps, the flow field dataset can involve multiple technical fields, such as airflow data in the aerospace field, water flow data in the environmental engineering field, or ocean data. User instructions can be generated by the user through natural language input, clicking interface buttons, or performing specific mouse gestures.

[0069] S102. Perform semantic analysis on the user's instruction using a large language model to obtain structured instructions.

[0070] Specifically, in step S102, the user instruction is semantically analyzed using a large language model to obtain the task identifier corresponding to the user instruction; the task identifier is then parameterized using preset parsing rules and structurally transformed to obtain a structured instruction.

[0071] S103. Based on the structured instructions, the corresponding target data is obtained from the flow field dataset to be analyzed through the large language model, and the corresponding descriptive text of the target data is generated.

[0072] Specifically, in step S103, the corresponding instruction task is extracted from the structured instructions; based on the instruction task, the corresponding proxy function in the large language model proxy layer is identified; and the instruction task is executed through the proxy function to obtain the corresponding target data from the flow field dataset to be analyzed.

[0073] Optionally, in the above steps, the target data is analyzed using the large language model to generate corresponding descriptive text for the target data; the descriptive text includes: variable descriptions, dimensions, units of measurement, or value range information.

[0074] S104. Based on the target data and the descriptive text, run the visualization agent to obtain the visualization results.

[0075] Specifically, in step S104, the visualization results include the target data and the corresponding descriptive text, which can help users quickly understand the background and meaning of the target data.

[0076] S105. Display the target data according to the data display method indicated by the user instruction.

[0077] Optionally, in the above steps, the corresponding data display method is extracted from the user instruction; the data display method includes charts, renderings, models, text descriptions, or video animations; and the target data is displayed according to the data display method.

[0078] Based on the above embodiments, this invention also provides a progressive visual analytics system based on a large language model. This system combines backend processing with a frontend interface, including a visualization interface, a large language model, and a dataset. The frontend interface is as follows: Figure 2 As shown, Figure 2 In the diagram, a represents the management view; b1 represents the conversation view content area; b2 represents the conversation view input area; and c represents the gallery view.

[0079] Visual analytics workflow as follows Figure 3 As shown, the user first initializes the analysis session in the visualization interface, selecting the flow field dataset to be studied. Then, the user can generate specific instructions through natural language input, clicking interface buttons, or performing specific mouse gestures. These instructions are transmitted to the system backend via a WebSocket connection established between the frontend and backend. This connection establishes a persistent, bidirectional communication channel, allowing for real-time, continuous data exchange.

[0080] Figure 2In the interface, the conversation management view is located on the left side. Users can create new conversations using the buttons at the top of the screen. Each conversation is automatically assigned a default name and a unique identifier. Users can double-click the conversation tag to edit the name to distinguish different discussion topics. Conversations not only record user actions but also save the context of the dialogue, facilitating understanding and continuation of the conversation flow. For multi-task analysis, the system allows switching between different conversations by clicking the conversation tag. When a conversation is selected, other conversations are blocked, but their context information is preserved. This allows users to focus on the currently active conversation while easily returning to other conversations. Below the management view are "Load Chat," "Save Chat," and "Delete Chat" buttons. Users can load previously saved conversation records to review past conversations or continue unfinished tasks, save current conversation records for future viewing or discussion, or delete unwanted conversations to keep the list tidy.

[0081] The conversation view input area integrates data file upload and parsing functions. Users can select one or more data files to upload using the "Select File" button in the input area, supporting multiple file selection for batch data processing. After selecting a file, users need to click the "Upload" button to complete the data upload process. Once the data file is successfully uploaded, the system automatically performs loading and parsing operations. This process includes recognizing and reading the file content, analyzing the data structure, and extracting and displaying basic descriptive information of the data (dimensions, fields, data range, etc.), allowing users to quickly grasp the basic characteristics of the dataset. The input box in the center of the area serves as the main channel for user interaction with the system. Users can input commands or queries using natural language in this input box and send these commands by clicking the arrow buttons.

[0082] The conversation view content area provides a highly interactive and feature-rich general display space. This area can display various forms of visualization results, including images, dynamic canvases, models, and various charts. These visualization results are generated by different backend agents, and the frontend selects appropriate components for rendering and display based on requirements. The system adds interactive tools to each visualization component in the content area for assistance. Taking the map canvas as an example, users can select areas of interest on the map to generate flow charts. For chart displays, various statistical chart styles are supported, and users can freely adjust the displayed items in the charts, such as the visibility of data columns and the display range. As for the 3D view, users can observe the structure of the model through rotation, zoom, and other operations. Furthermore, during user interaction, if domain-related questions arise, users can simply ask questions via text input, and the system provides reference suggestions based on a large language model. This design greatly reduces the complexity of user operation and the learning curve, enabling even non-professional users to easily master and utilize the system for data analysis.

[0083] The Gallery view is used to save and record visualizations of interest from the Session view. It allows users to pin important charts or images during data analysis and exploration via a "pin" button within the Session content area. Simply click this button to place the selected visualization at an appropriate scale into the gallery for easy review later. To remove a pinned item, users can click the "unpin" button. Furthermore, the Gallery view supports saving visualizations from different sessions. This feature facilitates comparative analysis of results across different case studies or datasets, helping users gain multi-faceted insights into the data.

[0084] In the backend, a classifier based on a large language model is responsible for scheduling user commands, accurately classifying and forwarding them to the processing queue, and converting them into a structured input format. Next, the structured commands are fed into an LLM (Large Language Model) proxy layer composed of numerous proxy functions, each responsible for executing its corresponding task.

[0085] Once the commands are categorized, parsed, and executed, the agent initiates the visualization generation process. This outputs various forms of visualization, including images, charts, and 3D models, providing a comprehensive view and analysis of the flow field data. This visualization data is transmitted in real-time to the user interface via a WebSocket connection for viewing and exploration. Simultaneously, based on a LargeVision-LanguageModel (LVLM), the system automatically understands and interprets the visualizations, generating corresponding descriptive text. Users can submit further query commands based on the visualizations, and the system saves and updates contextual information relevant to the current session to ensure the continuity of the user's thought process. Furthermore, the system provides a "gallery" function, allowing users to store visualizations generated from different sessions for convenient subsequent analysis.

[0086] The backend large language model performs user command processing as follows: Figure 4 As shown.

[0087] Upon receiving a user's original command, the system first performs intent recognition. Using a large language model, such as GPT-3.5, it understands the semantics of the command and obtains the corresponding task identifier. Then, based on the task identifier, specific parsing rules are selected to extract key parameters from the command, resulting in a JSON-formatted string. Next, the JSON string is automatically converted into a structured object using the Pydantic specification. Throughout this process, based on prompt engineering, the system fully leverages the natural language understanding capabilities of the large language model. By assigning roles to the model in the prompts, providing task descriptions, data format specifications, and input / output examples, the accuracy and standardization of the large language model's output are ensured. Furthermore, the system includes a proxy layer, serving as the core of the backend processing logic. The proxy layer comprises multiple independent, modular proxy functions (or "services"), each responsible for executing a specific set of tasks. Upon receiving structured user command information, the system automatically selects the appropriate proxy to complete the entire workflow from data processing to final visualization output. These proxy functions are independent of each other, allowing for individual updates and optimizations without affecting other parts of the system, ensuring maintainability and scalability. In addition, the system provides an "adaptation layer" to enhance its versatility. If a new dataset needs to be adapted, the adaptation layer can be easily adjusted without modifying other details of the system.

[0088] Meanwhile, to reduce reliance on storage resources, the system employs a proxy function that transmits results to the front end in real time after completing data processing and visualization tasks, while automatically releasing intermediate results generated during the process. This mechanism avoids the long-term storage requirements of large amounts of intermediate data, optimizes the use of storage resources, and accelerates data processing and result presentation. Taking streamline and trajectory rendering tasks as an example, this process typically includes multiple steps such as parameter setting, velocity field generation, curve integral calculation, and saving and loading the model or snapshot. The proxy layer design shields the complex details behind the scenes; users only need to issue relevant commands, such as the selected region, the number of initial seed points, and the curve integral step size. The system automatically parses these commands into standardized inputs, triggers the corresponding proxy to execute the complete processing flow, and ultimately obtains a 2D image or 3D model of the streamline, which is then transmitted to the front end for display via data stream.

[0089] When visualization results need to be displayed, different analysis tasks usually require different forms of visualization, such as statistical charts, 2D renderings, 3D models, text descriptions, video animations, etc.

[0090] Therefore, the system features plug-in visualization, providing a general visualization display area that supports multiple components. This area can dynamically select and load the corresponding visualization components based on the type of visualization result, such as... Figure 5 As shown. For example, if it's a streamline drawing task, the display area will load an image rendering component; if it's data statistical analysis, it will switch to a specific type of statistical chart component. This design allows the same area to flexibly display different visualizations. Another advantage is that adding new components or deleting existing components will not affect other modules, improving the system's scalability and maintainability.

[0091] The core of this system lies in visual semantic enhancement based on a large language model. Semantic enhancement requires collecting the meaning of user commands, the described visualization results, and the configuration information of the visualization components supporting those results. The visualization results images are losslessly converted to text format using Base64 encoding. Information such as the command theme, chart type, and original data embedded in the charts are extracted from the backend's returned data to determine the relevant configuration parameters of the visualization components. A prompt word generation template is then developed for each visualization component. For example... Figure 6 As shown, chart type, theme, data, and preset configuration can be integrated into a prompt text. Simultaneously, the Base64 encoding of the visual image is sent as additional information to the visual language model Gemini. By combining the image content and contextual information in the prompt, Gemini can generate descriptive text related to the instructions and images.

[0092] The following specific example illustrates the progressive visual analytics system based on a large language model described in the above embodiments, using an ocean dataset as an example. A high-precision ocean dataset generated by the Earth numerical simulation device "Huan" serves as the basis for the analysis.

[0093] The dataset description includes 4 dimensions and 7 variables. The dimensions are: lon (longitude, 780 data points); lat (latitude, 480 data points); depth (depth, 33 data points); and time (time, 1 time point). The variable information includes: lon (longitude, in degrees east, ranging from 0 to 360 degrees); Lat (latitude, in degrees north, ranging from -90 to 90 degrees); depth (depth, in meters, representing different depth levels); time (time, representing the time of data recording); ssh (sea surface height, in meters); t (temperature, in degrees Celsius); s (salinity, in salinity units (PSU); u (meridional velocity, in meters per second); v (zonal velocity, in meters per second).

[0094] This dataset covers daily average ocean data over many years, providing a detailed record of key physical parameters of the global oceans. It includes latitude, longitude, and depth information for each geographic location, corresponding temperature and salinity data, and water flow velocities in the longitudinal and latitudinal directions, covering important ocean regions of the Indian and Pacific Oceans.

[0095] In the visual analysis interface, manually select a region and input the specified number of seed points to generate streamlines, from 32°S to 52°N latitude and 30°E to 180°W longitude, or sprinkle 2000 points across the entire region with a step size of 0.1. Based on the feedback from the dialog area, the system can correctly identify and parse each command and draw a global streamline distribution map as expected, such as... Figure 7 As shown.

[0096] right Figure 7Analysis revealed that the density of streamlines varied across different regions. For example, in the Philippine Sea and near the North Pacific, streamlines were relatively dense and nearly parallel. While slender streamlines also existed in the northern Indian Ocean, their distribution was uneven. Based on this preliminary analysis, a question was posed to the progressive visualization analysis system: "Why are the streamlines in the western North Pacific and related sea areas denser and more parallel, while the streamline distribution in the northern Indian Ocean appears sparse and uneven?" The system's feedback on the streamline visualization results is as follows.

[0097] The denser and more parallel streamlines in the western North Pacific and surrounding waters are primarily due to the following factors: Monsoon influence: This region is significantly affected by monsoons, which create regular wind fields on the ocean surface, forming relatively parallel ocean currents. These wind fields push the seawater to form streamlines. Ocean current influence: This region is a convergence point for several important ocean currents, such as the Japan Current and the Kuroshio Current. The direction and intensity of these currents affect the density and arrangement of local ocean currents. Topographic influence: The complex topography of this region, with numerous islands and seamounts, also influences ocean circulation and current arrangement. In contrast, the sparser and less uniform streamlines in the northern Indian Ocean may be due to: Monsoon and equatorial currents: The influence of the monsoon is relatively greater in the northern Indian Ocean, and the variations in monsoon and the influence of equatorial currents lead to uneven current distribution. Topography and seafloor features: The topography and seafloor features of the northern Indian Ocean also affect the stability of ocean circulation and the density of streamlines. Climate change has an impact on the northern Indian Ocean as well. Factors such as temperature variations and instability in ocean circulation systems affect the density and arrangement of ocean currents. In summary, the differences in streamline density and arrangement between the western North Pacific and the northern Indian Ocean are mainly due to a combination of factors, including monsoon influences, ocean current effects, and topographical and geomorphological factors.

[0098] The system's response indicates that the North Pacific is significantly influenced by monsoons, with regular wind patterns driving parallel ocean currents. Furthermore, multiple ocean currents converge in the vicinity of this area, and the Philippine coast has complex topography, including mountains and islands. This aligns with the actual situation, namely the presence of straits, sea peaks, and capes along the coast of this region, and the existence of a circulation pattern in the North Pacific. In contrast, the northern Indian Ocean, besides being influenced by monsoon currents, may also be affected by its open topography and climate change.

[0099] In marine data research, seawater velocity and thermohaline distribution are crucial to marine dynamic systems, and the interaction of these elements affects local climate, marine ecological balance, and sustainable use of resources.

[0100] The above analysis also revealed spiral streamlines in regions such as the Indian Ocean and the North Pacific, with these vortex streamlines being relatively dense. Since the density of streamlines is related to flow velocity, the velocity distribution across all regions of the dataset was examined. Figure 8 As shown, areas with eddies typically have fast currents, especially near the Mozambique Channel and the Comoros Islands. Furthermore, narrow strips of water with relatively fast currents also exist along the Philippine coast, in the East China Sea, and in southern Japan. Zooming in on these areas using the zoom tool reveals... Figure 9 As shown, the left side is near the Mozambique Channel, and the right side is near the East China Sea and the Philippine Sea.

[0101] Based on the above analysis, further questions were raised about the system, and relevant reference opinions were obtained. The question to the system was: "Spiral streamlines were also found, considered to be vortices, and the streamlines of these vortices are mostly quite dense. It was found that the flow velocity is relatively fast in areas with vortices, especially near Mozambique and the Comoros Islands. In addition, long, narrow areas with relatively fast seawater flow were found along the Philippine coast, the East China Sea, and southern Japan. Please explain this situation." The system's explanation was as follows: The appearance of spiral streamlines usually indicates the presence of strong oceanic vortices in the area. An oceanic vortex is a large-scale oceanic circulation structure formed by rotating seawater. The seawater inside the vortex has a relatively high velocity and strong horizontal and vertical motion. The reason why the streamlines of the vortex are mostly dense may be due to the very active seawater movement inside the vortex. The following is an explanation for this situation: The vortex areas near Mozambique and the Comoros Islands are likely to have marine environments that easily form stable vortex structures. These vortices may be influenced by factors such as topography and ocean currents. In eddy regions, the increased seawater velocity is due to the rotational motion within the eddy driving the seawater movement, resulting in high-speed flow. The rapid seawater flow along the Philippine coast, in the East China Sea, and in southern Japan may be related to local topography, climate, and circulation systems. For example, the area near the Philippines is located on the edge of the Western Pacific Warm Pool, where strong ocean currents are formed due to the influence of monsoons; the East China Sea is influenced by warm and cold currents, forming a complex ocean circulation system; and southern Japan is located at the confluence of the Western Pacific Warm Current and the Black Lake, resulting in relatively fast seawater flow. In summary, the rapid seawater flow in eddy regions and certain specific areas may be influenced by a combination of factors, including topography, ocean currents, and climate, leading to faster flow rates in the marine environment.

[0102] The seawater within the eddy exhibits strong horizontal and vertical motion, resulting in active ocean currents. Stable eddy structures tend to form in the Mozambique region. The eastern coast of the Philippines lies on the edge of the Western Pacific Warm Pool, influenced by monsoons, resulting in strong ocean currents. The East China Sea is influenced by warm and cold currents, forming a complex ocean circulation system. Southern Japan is located at the confluence of the Northwest Pacific Warm Current and the Kuroshio Current, resulting in faster seawater speeds. These complex geographical and climatic conditions may all contribute to faster seawater currents in these areas. Based on this, the velocity distribution within a local area was examined. An area near the Philippine Sea and the Pacific Ocean was delineated, and an instruction to view the attributes was issued to the system. The system automatically generated a heat map of the velocity distribution in this area and provided descriptive text based on the map's visualization elements, such as numerical ranges and color mappings. Figure 10 As shown, the color mapping method of the heatmap can be seen, that is, the change from blue to white and then to red represents the speed from low to high.

[0103] The above analysis reveals a prominent red area along the southwestern coast of the relevant sea area. Zooming in on the specific values ​​shows that the peak current velocity in this area, measured in units, is several times higher than that in the eastern Pacific Ocean (light yellow area). For ease of description, we will refer to the aforementioned Pacific Ocean as Area A and the relevant sea area as Area B. Reviewing the streamline diagram, the current direction in Area A is primarily latitudinal, while the current direction in Area B is primarily longitudinal. Therefore, further investigation into the longitudinal and latitudinal velocity distribution is needed.

[0104] like Figure 11 As shown, region B not only exhibits a high latitudinal velocity, but its meridional velocity is also significant (the red and blue blocks in region B represent opposite meridional velocities). In contrast, the meridional velocity in region A is almost zero. This directly results in a much higher overall velocity in region B compared to region A. Subsequently, the velocity distribution at different depths was examined. Velocity images for the 5th, 10th, and 15th underwater layers (corresponding to depths of 100 meters, 300 meters, and 800 meters, respectively) were plotted as follows: Figure 12 As shown. By analyzing... Figure 12Analysis shows that region B maintains a relatively high flow velocity at layer 10, while at layer 15, the entire region exhibits only a low velocity value. With the aid of a large model, it was understood that the slowdown in seawater flow velocity is actually an inevitable result of the interaction of various physical properties of seawater (such as friction and viscosity), energy distribution, temperature and salinity stratification, seafloor topography, and the Coriolis effect. These complex factors work together to shape the unique dynamic characteristics of deep-sea currents. Next, the performance of temperature and salinity at different times was explored. Data files from two time points with a six-month interval, such as January 1, 2014, and July 1, 2014, were selected, and two independent sessions were established to avoid data interference during the analysis process. The visualization results were observed and compared in the gallery view. Figure 13 As shown, the left side represents January 1st, and the right side represents July 1st. The sea temperature values ​​on January 1st are generally lower than those on July 1st, and this difference is more pronounced around 20 degrees North latitude. However, their salinity differences are small; despite narrowing the numerical boundaries of the color mapping, their salinity distributions remain similar.

[0105] At this point, examine the trend of attribute changes in this sea area over the entire time period. For example... Figure 14 As shown, the horizontal axis represents the time step, and the vertical axis represents the average attribute value of the region. January 1st indicates the initial time step. The study found that the average temperature difference in this sea area reached 3 degrees Celsius, and the system's descriptive text indicates that the average temperature was lowest at time step 50 (around late February) and reached its highest near July. The average salinity, however, showed relatively small fluctuations and remained relatively stable over the six months.

[0106] In summary, the flow field visualization analysis method driven by a large language model provided in this invention first requires acquiring the flow field dataset to be analyzed and specific user commands. These user commands can be generated by the user through natural language input, clicking interface buttons, or performing specific mouse gestures. Then, a pre-trained large language model performs semantic analysis on the user commands to obtain the corresponding task identifier. Parameters are extracted from the task identifier using preset parsing rules, and a structured transformation is performed to obtain the structured command corresponding to the user command. This structured command is input into the proxy layer of the large language model, and the corresponding proxy function in the proxy layer executes the command task corresponding to the structured command, thereby retrieving the corresponding target data from the flow field dataset to be analyzed. Simultaneously, the large language model generates descriptive text messages corresponding to the target data. These descriptive text messages include variable descriptions, dimensions, units of measurement, and value range information to help users better understand the background and meaning of the target data. Based on the target data and the corresponding descriptive text, visualization results are generated. Before displaying the visualization results, the visualization display method needs to be obtained from the user commands. This display method includes charts, rendered images, models, text descriptions, or video animations. The system displays target data according to the user's instructions. By using a large language model to recognize the user's specific instructions, it enables natural language-based interaction. This allows the system to convert user commands or text commands into structured instructions, thereby obtaining the corresponding target data. The large language model then generates corresponding descriptive explanations, presenting the target data and their accompanying descriptions in real time. This helps users better understand the background and meaning of the target data, significantly enhancing the user experience.

[0107] This application also provides a large language model-driven flow field visual analysis device, which is used to implement the above embodiments and preferred embodiments, and will not be repeated hereafter. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0108] This application provides a flow field visual analysis device driven by a large language model. Figure 15 This is a schematic diagram of a large language model-driven flow field visual analysis device provided in an embodiment of this application. The device includes:

[0109] The acquisition module 1501 is used to acquire the data set of the flow field to be analyzed and user commands;

[0110] The instruction conversion module 1502 is used to perform semantic analysis on the user instruction using a large language model to obtain a structured instruction.

[0111] The target data acquisition module 1503 is used to acquire the corresponding target data from the flow field dataset to be analyzed based on the structured instructions and the large language model, and generate the corresponding descriptive text of the target data.

[0112] The result generation module 1504 is used to run a visualization agent based on the target data and the descriptive text to obtain visualization results;

[0113] The result display module 1505 is used to display the target data according to the data display method indicated by the user instruction.

[0114] In an optional implementation, the instruction conversion module 1502 is specifically used for:

[0115] The user command is semantically analyzed using a large language model to obtain the corresponding task identifier. The task identifier is then parameterized using predefined parsing rules and transformed into a structured command.

[0116] In one optional implementation, the target data acquisition module 1503 is specifically used for:

[0117] Extract the corresponding instruction task from the structured instructions; based on the instruction task, identify the corresponding proxy function in the large language model proxy layer; execute the instruction task through the proxy function to obtain the corresponding target data from the flow field dataset to be analyzed.

[0118] In an optional implementation, the target data acquisition module 1503 is further configured to:

[0119] The target data is analyzed using this large language model to generate corresponding descriptive text. This descriptive text includes: variable descriptions, dimensions, units of measurement, or value range information.

[0120] In one alternative implementation, the result display module 1505 is specifically used for:

[0121] Extract the corresponding data display method from the user instruction; the data display method includes charts, renderings, models, text descriptions, or video animations; display the target data according to the data display method.

[0122] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0123] In this embodiment, the large language model-driven flow field visual analysis device is presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0124] This invention also provides a computer device having the above-described features. Figure 15 The large language model-driven flow field visualization analysis device is shown.

[0125] Please see Figure 16 , Figure 16 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 16 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information in a graphical user interface on an external input / output device (such as a display device coupled to the interface). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 16 Take a processor 10 as an example.

[0126] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0127] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0128] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0129] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0130] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 16 Taking the example of a connection between China and Israel via a bus.

[0131] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0132] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0133] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and all such modifications and variations fall within the scope defined by the appended claims.

Claims

1. A flow field visualization analysis method driven by a large language model, characterized in that, The method includes: Obtain the data set of the flow field to be analyzed and the user commands; The user instructions are semantically analyzed using a large language model to obtain structured instructions; Based on the structured instructions, the corresponding target data is obtained from the flow field dataset to be analyzed through the large language model, and the corresponding descriptive text of the target data is generated. Based on the target data and the descriptive text, run the visualization agent to obtain the visualization results; The target data is displayed according to the data display method indicated by the user instruction; The step of performing semantic analysis on the user instructions using a large language model to obtain structured instructions includes: The user instructions are semantically analyzed using a large language model to obtain the task identifier corresponding to the user instructions; The task identifier is parameterized by a preset parsing rule and then converted into a structured form to obtain a structured instruction. The step of obtaining the corresponding target data from the flow field dataset to be analyzed through the large language model includes: Extract the corresponding instruction task from the structured instructions; Based on the instruction task, identify the corresponding proxy function in the proxy layer of the large language model; The proxy function executes the instruction task to obtain the corresponding target data and visualization results from the flow field dataset to be analyzed; The generation of descriptive text corresponding to the target data includes: The visualization result image is losslessly converted into text format using Base64 encoding. The instruction theme, chart type, and original data information embedded in the chart are extracted from the backend returned data. This allows us to obtain the relevant configuration parameters of the visualization component. The chart type, theme, data, and preset configuration are then integrated into a prompt text. At the same time, the Base64 encoding of the visualization image is sent as additional information to the visual language model. By combining the image content and the contextual information in the prompt, descriptive text related to the instruction and image is generated.

2. The method according to claim 1, characterized in that, The user instructions include natural language, interface button operation instructions, or mouse gesture instructions.

3. The method according to claim 1, characterized in that, The generation of descriptive text corresponding to the target data includes: The target data is analyzed using the large language model to generate corresponding descriptive text for the target data; The descriptive text includes: variable descriptions, dimensions, units of measurement, or range of values.

4. The method according to claim 3, characterized in that, The step of displaying the target data according to the data display method indicated by the user instruction includes: The corresponding data display method is extracted from the user command; the data display method includes charts, rendered images, models, text descriptions, or video animations; The target data will be displayed according to the data display method described above.

5. A flow field visual analysis device driven by a large language model, characterized in that, The device includes: The acquisition module is used to acquire the data set of the flow field to be analyzed and user commands; The instruction conversion module is used to perform semantic analysis on the user instructions using a large language model to obtain structured instructions. The target data acquisition module is used to acquire corresponding target data from the flow field dataset to be analyzed based on the structured instructions and through the large language model, and generate corresponding descriptive text for the target data; The result generation module is used to run a visualization agent based on the target data and the descriptive text to obtain visualization results; The results display module is used to display the target data according to the data display method indicated by the user instruction; Specifically, the instruction conversion module is used for: The user instructions are semantically analyzed using a large language model to obtain the task identifier corresponding to the user instructions; The task identifier is parameterized by a preset parsing rule and then converted into a structured form to obtain a structured instruction. The target data acquisition module is specifically used for: Extract the corresponding instruction task from the structured instructions; Based on the instruction task, identify the corresponding proxy function in the proxy layer of the large language model; The proxy function executes the instruction task to obtain the corresponding target data and visualization results from the flow field dataset to be analyzed; The target data acquisition module is also used for: The visualization result image is losslessly converted into text format using Base64 encoding. The instruction theme, chart type, and original data information embedded in the chart are extracted from the backend returned data. This allows us to obtain the relevant configuration parameters of the visualization component. The chart type, theme, data, and preset configuration are then integrated into a prompt text. At the same time, the Base64 encoding of the visualization image is sent as additional information to the visual language model. By combining the image content and the contextual information in the prompt, descriptive text related to the instruction and image is generated.

6. The apparatus according to claim 5, characterized in that, The results display module is specifically used for: The corresponding data display method is extracted from the user command; the data display method includes charts, rendered images, models, text descriptions, or video animations; The target data will be displayed according to the data display method described above.

Images