Three-dimensional geographic information system control method and device, storage medium and electronic device

Abstract

The invention provides a control method and device of a three-dimensional geographic information system, a storage medium and an electronic device, and relates to the technical field of virtual three-dimensional geography, the method is applied to an unreal engine, and the method comprises the steps that a text instruction is obtained, the text instruction is used for instructing execution of target operation on the three-dimensional geographic information system, and the target operation is executed on the three-dimensional geographic information system; the text instruction is an instruction obtained by converting an obtained voice instruction or an instruction obtained through an interactive interface; determining an operation instruction corresponding to the text instruction through a large language model; and executing the target operation on the three-dimensional geographic information system according to the operation instruction. By adopting the method, the problem that the three-dimensional geographic information system cannot be efficiently controlled is solved, and the control efficiency is improved.

Description

Technical Field

[0001] This specification relates to the field of virtual 3D geographic technology, and in particular to control methods and devices, storage media, and electronic devices for 3D geographic information systems. Background Technology

[0002] In recent years, with the booming development of Virtual Reality (VR), Augmented Reality (AR), and the gaming industry, the application scenarios of Three-Dimensional Geographic Information Systems (GIS) have become increasingly widespread. From urban planning and architectural design to environmental monitoring, 3D GIS has become a powerful tool for visualizing and analyzing information about the Earth's surface. However, existing 3D GIS control methods are mainly based on traditional graphical user interface (GUI) operations, where users manipulate objects and viewpoints in a 3D scene through mouse clicks, drags, and keyboard input. While this method provides an intuitive control experience to some extent, it presents several significant problems for both professional and non-professional users: High operational complexity: In order to perform relatively complex tasks, such as transformations between multi-level coordinate systems, rapid positioning of specific locations, and precise spatial operations, users need to master a series of cumbersome operation steps and shortcut key combinations, which poses a significant learning obstacle for beginners.

[0003] Limitations of the interactive interface: Although GUI operation is intuitive, in some scenarios, such as when the user is outdoors or in special equipment (such as VR headset), direct operation of the device may be inconvenient or even impossible. This limits the scenario applicability and user experience of 3D GIS.

[0004] There is currently no effective solution to the problem of inefficient control of 3D geographic information systems in related technologies.

[0005] Therefore, it is necessary to improve the relevant technology to overcome the aforementioned defects. Summary of the Invention

[0006] To overcome the problem of inefficient control of three-dimensional geographic information systems in related technologies, this specification provides control methods and devices, storage media, and electronic devices for three-dimensional geographic information systems.

[0007] According to a first aspect of the embodiments of this specification, a control method for a three-dimensional geographic information system is provided, applied to Unreal Engine, comprising: acquiring a text instruction, wherein the text instruction is used to instruct the three-dimensional geographic information system to perform a target operation, the text instruction being an instruction converted from an acquired voice instruction or an instruction acquired through an interactive interface; determining an operation instruction corresponding to the text instruction through a large language model; and performing the target operation on the three-dimensional geographic information system according to the operation instruction.

[0008] Optionally, determining the operation instruction corresponding to the text instruction through a large language model includes: sending the text instruction to the large language model, wherein, after receiving the text instruction, the large language model performs semantic parsing on the text instruction to obtain a semantic parsing result; the large language model generates a structured invocation instruction conforming to the MCP tool invocation specification based on the semantic parsing result, and sends the structured invocation instruction to the MCP server; obtaining the operation instruction sent by the MCP server, wherein, after obtaining the structured invocation instruction, the MCP server encapsulates the structured invocation instruction into a standardized operation instruction according to the transmission protocol format between the MCP server and the Unreal Engine; the operation instruction is a JSON instruction package executable by the Unreal Engine.

[0009] Optionally, after performing the target operation on the three-dimensional geographic information system according to the operation instructions, the method further includes: broadcasting the execution result of the target operation via voice; and / or displaying the execution result through the interactive interface; and / or sending the execution result of the target operation to the MCP server, wherein the MCP server, after obtaining the execution result, sends the execution result to the large language model, and the large language model performs result analysis on the execution result.

[0010] Optionally, the semantic parsing result includes: intent classification result and parameter values ​​for performing the target operation; the large language model generates a structured invocation instruction conforming to the MCP tool invocation specification based on the semantic parsing result, including: determining the system prompt words and tool list provided by the MCP server, wherein the system prompt words are used to indicate the output format and output constraints of the large language model, and the tool list contains the registration information of each tool in the MCP server; generating the structured invocation instruction based on the system prompt words and tool list, wherein the structured invocation instruction includes target information, the target information including: the tool name pointing to the tool registered in the MCP server, the function name of the function that the tool is allowed to call, the parameter values ​​for performing the target operation, and a first identifier; the first identifier is used to identify a function call.

[0011] Optionally, when the target operation includes multiple sub-operations, the structured invocation instruction includes: target information and a second identifier corresponding to each sub-operation; wherein the second identifier corresponding to the sub-operation is used to point to the first identifier corresponding to the remaining sub-operations, and is used to control the execution order of the sub-operations.

[0012] Optionally, when the large language model generates a structured invocation instruction conforming to the MCP tool invocation specification based on the semantic parsing result, if the parameter value for executing the target operation cannot be quantitatively determined, it performs parameter value completion processing according to preset rules, and records the completion processing process in the structured invocation instruction through the field value of the target field.

[0013] Optionally, after performing the target operation on the three-dimensional geographic information system according to the operation instruction, the method further includes: if the target operation includes multiple sub-operations and only some of the multiple sub-operations are successfully executed on the three-dimensional geographic information system, reverting the three-dimensional geographic information system to the state before the target operation was executed.

[0014] According to a second aspect of the embodiments of this specification, a control device for a three-dimensional geographic information system, applied in Unreal Engine, is also provided, comprising: an acquisition module for acquiring text instructions, wherein the text instructions are used to instruct the three-dimensional geographic information system to perform a target operation, and the text instructions are instructions obtained by converting acquired voice instructions or instructions obtained through an interactive interface; a determination module for determining the operation instruction corresponding to the text instructions through a large language model; and an execution module for executing the target operation on the three-dimensional geographic information system according to the operation instructions.

[0015] Optionally, the determining module is further configured to send the text instruction to the large language model, wherein, after receiving the text instruction, the large language model performs semantic parsing on the text instruction to obtain a semantic parsing result; the large language model generates a structured invocation instruction conforming to the MCP tool invocation specification based on the semantic parsing result, and sends the structured invocation instruction to the MCP server; and obtains the operation instruction sent by the MCP server, wherein, after obtaining the structured invocation instruction, the MCP server encapsulates the structured invocation instruction into a standardized operation instruction according to the transmission protocol format between the MCP server and the Unreal Engine; the operation instruction is a JSON instruction package executable by the Unreal Engine.

[0016] Optionally, the control device of the 3D geographic information system further includes a processing module, used to: verbally announce the execution result of the target operation after the target operation is performed on the 3D geographic information system according to the operation instruction; and / or display the execution result through the interactive interface; and / or send the execution result of the target operation to the MCP server, wherein the MCP server, after obtaining the execution result, sends the execution result to the large language model, and the large language model performs result analysis on the execution result.

[0017] Optionally, the semantic parsing result includes: intent classification result and parameter values ​​for performing the target operation; the control device of the three-dimensional geographic information system is further configured to determine the system prompt words and tool list provided by the MCP server through the large language model, wherein the system prompt words are used to indicate the output format and output constraints of the large language model, and the tool list contains registration information for each tool in the MCP server; generate the structured invocation instruction based on the system prompt words and tool list, wherein the structured invocation instruction includes target information, the target information including: tool name pointing to the tool registered in the MCP server, function name of the function that the tool is allowed to call, parameter values ​​for performing the target operation, and a first identifier; the first identifier is used to identify a function call.

[0018] Optionally, when the target operation includes multiple sub-operations, the structured invocation instruction includes: target information and a second identifier corresponding to each sub-operation; wherein the second identifier corresponding to the sub-operation is used to point to the first identifier corresponding to the remaining sub-operations, and is used to control the execution order of the sub-operations.

[0019] Optionally, the control device of the three-dimensional geographic information system is further configured to, in the process of generating a structured invocation instruction conforming to the MCP tool invocation specification based on the semantic parsing result through the large language model, perform parameter value completion processing according to preset rules when the parameter value for executing the target operation cannot be quantitatively determined, and record the completion processing process in the structured invocation instruction through the field value of the target field.

[0020] Optionally, the control device for the three-dimensional geographic information system further includes: a rollback module, used to roll back the three-dimensional geographic information system to the state before the execution of the target operation after the target operation is executed on the three-dimensional geographic information system according to the operation instruction, in the case that the target operation includes multiple sub-operations and only some of the sub-operations are successfully executed on the three-dimensional geographic information system.

[0021] According to a third aspect of the embodiments of this specification, a computer-readable storage medium is also provided, wherein a computer program is stored in the computer program, which is configured to execute the control method of the three-dimensional geographic information system described above when running.

[0022] According to a fourth aspect of the embodiments of this specification, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the control method of the three-dimensional geographic information system through the computer program.

[0023] In the embodiments described in this specification, the traditional operation mode of 3D GIS is revolutionized by leveraging the intelligent parsing of natural speech and a large language model. Users issue natural language commands, and Unreal Engine then calls the LLM to determine the corresponding operation instructions and executes the corresponding operations on the 3D GIS. This process avoids lengthy manual operations and complex interface navigation, not only improving control efficiency but also making operation more intuitive and convenient, greatly enhancing the user's interactive experience. Especially in emergency or professional scenarios, it can quickly respond to user needs and achieve efficient 3D spatial control.

[0024] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this specification. Attached Figure Description

[0025] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this specification and, together with the description, serve to explain the principles of this specification.

[0026] Figure 1 This is a flowchart illustrating a control method for a three-dimensional geographic information system according to an exemplary embodiment.

[0027] Figure 2 This is an overall architecture diagram of a control method for a three-dimensional geographic information system illustrated in this specification according to an exemplary embodiment.

[0028] Figure 3 This is a timing diagram illustrating a control method for a three-dimensional geographic information system according to an exemplary embodiment of this specification.

[0029] Figure 4 This is a hardware structure diagram of a computer device containing the control device of the three-dimensional geographic information system in the embodiments of this specification.

[0030] Figure 5 This is a block diagram illustrating a control device for a three-dimensional geographic information system according to an exemplary embodiment. Detailed Implementation

[0031] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this specification as detailed in the appended claims.

[0032] The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of this specification. The singular forms “a,” “the,” and “the” as used in this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0033] It should be understood that although the terms first, second, third, etc., may be used in this specification to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this specification, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0034] The embodiments described in this specification will now be described in detail.

[0035] like Figure 1 As shown, Figure 1 This is a flowchart illustrating a control method for a three-dimensional geographic information system according to an exemplary embodiment of this specification. The method is applied to Unreal Engine and includes the following steps 101-103: Step 101: Obtain text instructions, wherein the text instructions are used to instruct the execution of target operations on the three-dimensional geographic information system, and the text instructions are instructions obtained by converting acquired voice instructions or instructions obtained through an interactive interface; Optionally, Unreal Engine can use a microphone or other audio input device to capture the user's voice input, or directly obtain the user's text commands through the interactive interface. To ensure high-quality reception of voice commands, noise reduction techniques, echo cancellation, and audio compression can be employed to reduce background noise interference and improve the accuracy of speech recognition.

[0036] Optionally, Unreal Engine can convert speech to text using a speech-to-text plugin, and supports multiple languages.

[0037] Optionally, the converted text commands are "Turn the camera left", "Switch to Mercator coordinate system", "Zoom in to the building entrance", etc.

[0038] Step 102: Determine the operation instruction corresponding to the text instruction using a large language model; Step 103: Perform the target operation on the three-dimensional geographic information system according to the operation instructions.

[0039] The above steps, leveraging the intelligent parsing of natural language speech and the Large Language Model (LLM), revolutionize the traditional operation mode of 3D GIS. Users send out natural language commands, and Unreal Engine then invokes the LLM to determine the corresponding operation instructions and executes the appropriate actions on the 3D GIS. This process avoids lengthy manual operations and complex interface navigation, not only improving control efficiency but also making operation more intuitive and convenient, greatly enhancing the user experience. Especially in emergency or professional scenarios, it can quickly respond to user needs and achieve efficient 3D spatial control.

[0040] In an exemplary embodiment, determining the operation instruction corresponding to the text instruction using a large language model includes the following steps S11-S12: Step S11: Send the text instruction to the large language model, wherein, after receiving the text instruction, the large language model performs semantic parsing on the text instruction to obtain a semantic parsing result; the large language model generates a structured invocation instruction that conforms to the Modular Control Protocol (MCP) tool invocation specification based on the semantic parsing result, and sends the structured invocation instruction to the MCP server; Step S12: Obtain the operation instruction sent by the MCP server, wherein, after obtaining the structured call instruction, the MCP server encapsulates the structured call instruction into a standardized operation instruction according to the transmission protocol format between the MCP server and the Unreal Engine; the operation instruction is a JSON instruction package executable by the Unreal Engine.

[0041] It should be noted that, in this embodiment, the text command, as the direct carrier of the user's intent, is transmitted to the Large Language Model (LLM). The LLM parses the text command to obtain the semantic parsing result. After the parsing is completed, the LLM generates a structured invocation command that conforms to the MCP tool invocation specification based on the parsed semantic result. Subsequently, the LLM sends this structured command to the MCP server as a blueprint for executing the command.

[0042] In this embodiment, the MCP server acts as both a translator and scheduler of instructions. Upon receiving a structured invocation instruction from the LLM, the MCP server standardizes and encapsulates the instruction according to the Unreal Engine (UE5)'s transport protocol format, such as the JSON encapsulation standard under TCP transmission. This encapsulation process involves converting key information such as tool names, function names, parameter values, and identifiers into JSON instruction packages that UE5 can directly parse and execute. This function of the MCP server not only ensures that complex instructions generated by the LLM can be accurately recognized by UE5, but also optimizes data exchange efficiency, providing clear instruction guidance for subsequent 3D GIS operations. Ultimately, UE5 executes the specific operations indicated by the user based on these standardized operation instructions, achieving seamless integration with the 3D scene.

[0043] The steps S11-S12 above, through the semantic parsing capabilities of LLM and the instruction encapsulation function of the MCP server, construct an efficient and accurate execution link from natural language instructions to 3D GIS operations, greatly improving the convenience and intelligence of user interaction with 3D GIS.

[0044] In an exemplary embodiment, after performing the target operation on the three-dimensional geographic information system according to the operation instruction, the method further includes: broadcasting the execution result of the target operation via voice; and / or displaying the execution result through the interactive interface; and / or sending the execution result of the target operation to the MCP server, wherein, after obtaining the execution result, the MCP server sends the execution result to the large language model, and the large language model performs result analysis on the execution result.

[0045] In this embodiment, the execution result of the target operation is announced by voice and / or displayed through the interactive interface, allowing the user to clearly understand whether Unreal Engine has correctly executed the instructions, thus increasing the user-friendliness of Unreal Engine.

[0046] It's important to note that the execution results will be transmitted back to the Large Language Model (LLM) via the MCP server for analysis. The MCP server, acting as an intermediary coordinator, is responsible for collecting state data after the UE5 operation, such as the new viewpoint position, coordinate system changes, and zoom level. This data is sent to the LLM in a structured format. Upon receiving this data, the LLM performs in-depth analysis to evaluate the success rate and effectiveness of the operation, and can even identify and correct potential errors, such as speculating on the most suitable parameter settings when user instructions are ambiguous. This cycle not only enables the LLM to understand user intent more accurately but also allows it to continuously learn and optimize the parsing and execution of similar commands in the future.

[0047] In an exemplary embodiment, the semantic parsing result includes: intent classification result and parameter values ​​for performing the target operation; The large language model generates a structured invocation instruction conforming to the MCP tool invocation specification based on the semantic parsing result, including: determining the system prompt words and tool list provided by the MCP server, wherein the system prompt words are used to indicate the output format and output constraints of the large language model, and the tool list contains the registration information of each tool in the MCP server; generating the structured invocation instruction based on the system prompt words and tool list, wherein the structured invocation instruction includes target information, the target information including: the tool name pointing to the tool registered in the MCP server, the function name of the function that the tool is allowed to call, the parameter value for performing the target operation, and a first identifier; the first identifier is used to identify a function call.

[0048] It should be noted that in this embodiment, when the LLM receives a text command, it refers to the system prompts and comprehensive tool list provided by the MCP server. This information is the cornerstone of communication between the LLM and the MCP server, ensuring that the format of the output command conforms to the MCP protocol requirements and that the parameters and functions of the command are within the scope of 3D GIS operations. The system prompts clarify the LLM's task: how to correctly parse the user's text command and output the results in a standardized format. They also set output constraints, such as parameter types, units, and ranges, to prevent the generation of erroneous or incompatible commands.

[0049] Based on the above information, LLM begins deep semantic parsing, transforming text commands into clear operational intentions. Next, according to the format constraints defined in the tool list and system prompts of the MCP server, LLM generates structured call commands conforming to the MCP tool call specifications. This command is no longer a simple text response, but a precise instruction containing target information, including the tool name, function name, parameter values, and a first identifier. The tool name and function name directly point to the specific functional module registered in the MCP server, ensuring the command is correctly identified and executed; the parameter values ​​detail the specific operations, such as rotation angle, coordinate system type, or zoom level—information indispensable for execution; and the first identifier, as a unique identifier for a function call, is used to track and record the execution process, facilitating subsequent querying, auditing, and error handling, ensuring the clarity and manageability of each call.

[0050] In an exemplary embodiment, when the target operation includes multiple sub-operations, the structured invocation instruction includes: target information and a second identifier corresponding to each sub-operation; wherein, the second identifier corresponding to the sub-operation is used to point to the first identifier corresponding to the remaining sub-operations, and is used to control the execution order of the sub-operations.

[0051] It should be noted that when the target operation involves a series of complex and interdependent sub-operations, the above-described structured instruction design can be used to ensure the consistency and accuracy of the operation. Specifically: When a user issues a voice command containing multiple sub-operations, such as "First, switch the viewpoint to the Mercator coordinate system, then pan to the center of the map of China, and finally zoom in to display the Great Wall section," the Large Language Model (LLM) parses this command by breaking it down into independent but ordered sub-operation commands. For each sub-operation, the structured call command generated by the LLM not only includes the tool name, function name, parameter values, and first identifier (e.g., the value of the callback_id field below), but also adds a second identifier (e.g., the value of the wait_for field below) to clarify the dependencies and execution order between sub-operations.

[0052] The second identifier works in conjunction with the first identifier, acting like a "clue" in the operation instructions. It points to the first identifier corresponding to the remaining sub-operations, forming a clear execution chain. For example, in the example above, the second identifier of the operation instruction to pan to the center of the map of China will point to the first identifier of the coordinate system switching instruction, ensuring that the system has switched to the Mercator coordinate system before performing the pan. Similarly, the second identifier of the operation instruction to zoom in on the Great Wall section will point to the first identifier of the pan instruction, ensuring that the system performs the zoom operation only after the pan operation is completed. This design ensures that the execution order of the operation instructions aligns with the user's original intent. Figure 1 This avoids system state chaos caused by improper sequence and achieves accurate parsing and orderly execution of multi-step text instructions.

[0053] In an exemplary embodiment, when the large language model generates a structured invocation instruction conforming to the MCP tool invocation specification based on the semantic parsing result, if the parameter value for executing the target operation cannot be quantitatively determined, it performs parameter value completion processing according to preset rules, and records the completion processing process in the structured invocation instruction through the field value of the target field.

[0054] It should be noted that in the operation and control of a 3D geographic information system, users may not provide completely specific parameter values, especially when describing operational details. They may use more abstract or generalized expressions, such as "zoom in a little" or "move north a bit." While these natural language descriptions are intuitive and easy to understand, they lack precise numerical information, posing a challenge to the system's direct execution. Therefore, this embodiment introduces an intelligent parameter completion mechanism to ensure that even with insufficient parameter information, structured invocation instructions conforming to the MCP tool invocation specification can be generated, enabling the system to accurately execute user intentions. Specifically: When a Large Language Model (LLM) performs semantic parsing on text instructions, if it encounters a target operation parameter that cannot be directly quantified, such as "slightly" in "zoom in slightly," the LLM will complete the parameter value according to preset rules. These preset rules are usually based on default settings or statistical analysis of historical operation data and can automatically fill in appropriate parameter values ​​according to the context. For example, "zoom in slightly" might be completed as "zoom in on the lens, magnification is 1.2x."

[0055] It's important to note that the parameter completion process is not arbitrary; it involves logical judgment and data analysis to ensure that the completed parameter values ​​are reasonable and align with the user's underlying intent. Furthermore, to enhance transparency and traceability, the LLM records the parameter completion process in the generated structured invocation instructions through specific target fields (such as the assumptions field described below). This includes the reason for the completion, the source of the rules, and the specific values ​​after completion. For example, the instruction might include a "Parameter Completion" field, detailing how "slightly magnify" is interpreted as "magnification ratio of 1.2x".

[0056] It's important to note that this mechanism not only solves the problem of insufficient parameter information but also builds a bridge between users and the system, enabling users to interact with the 3D GIS system in a more natural and intuitive way, without worrying about operational failures or errors due to imprecise expressions. Simultaneously, by recording the parameter completion process, LLM can gradually learn user preferences and habits, continuously improving the accuracy and response speed of natural language command parsing, thereby providing a more personalized and efficient voice control experience.

[0057] In an exemplary embodiment, after performing the target operation on the three-dimensional geographic information system according to the operation instruction, the method further includes: if the target operation includes multiple sub-operations and only some of the multiple sub-operations are successfully executed on the three-dimensional geographic information system, reverting the three-dimensional geographic information system to the state before the target operation was executed.

[0058] It's important to note that in 3D Geographic Information Systems (GIS), user commands delivered via natural language often involve a series of complex sub-operations, and the execution status of each sub-operation can affect the integrity of the entire operation chain. For example, a command might be: "Rotate the viewpoint 45 degrees to the left, then zoom in to area xx, and finally switch to Mercator coordinates." In such multi-step commands, each sub-operation must execute successfully to ensure the user's intent is fully realized. However, in actual operation, a step might fail; for instance, "zoom in to area xx" might fail due to network latency preventing the timely loading of the relevant area map.

[0059] To address the aforementioned issues and ensure robustness and consistency of operations, a transaction rollback mechanism was designed. When executing multi-step operations, if any sub-operation fails, Unreal Engine will proactively roll back the 3D GIS system to the state before the entire target operation was executed—that is, the initial state before the user command began execution.

[0060] In an exemplary embodiment, the method further includes: recording and reporting the failed sub-operation and the reason for the operation failure when a rollback operation is performed, wherein the reason for the operation failure includes, but is not limited to, network problems and resource loading timeouts; broadcasting voice prompts, wherein the voice prompts are used to indicate the reason for the operation failure; performing corresponding optimization operations based on the reason for the operation failure, wherein the optimization operations include, but are not limited to, network optimization and resource optimization; and, when a new text instruction is obtained, determining a new operation instruction corresponding to the new text instruction through a large language model, and performing corresponding operations on the three-dimensional geographic information system based on the new operation instruction, wherein the new text instruction is a text instruction re-determined by the user based on the voice prompts.

[0061] In one exemplary embodiment, a dynamic caching and prediction mechanism is established between the Large Language Model (LLM) and the Modular Control Protocol (MCP) server to accelerate the parsing and execution of operation instructions. Specifically, the MCP server creates a dynamic cache database based on historical user instructions and operation patterns. Whenever a text instruction is obtained, the LLM retrieves a structured instruction from this dynamic cache database that matches the semantic parsing result. If a matching instruction is found, it is used directly to perform the operation, avoiding repeated parsing and parameter extraction processes and significantly improving response speed. Simultaneously, the LLM can also predict the next possible sub-operation based on the user's frequently used instructions and preferences, preparing corresponding structured instructions in advance. This allows multi-step operations to respond quickly before the user issues the next instruction, further optimizing the user experience.

[0062] Obviously, the embodiments described above are merely some embodiments of the present invention, and not all embodiments. To better understand the above method, the following description, in conjunction with embodiments, illustrates the process, but is not intended to limit the technical solutions of the embodiments of the present invention: Figure 2 This specification illustrates an overall architecture diagram of a control method for a three-dimensional geographic information system according to an exemplary embodiment, wherein... 1) The main function of the semantic understanding layer is: Natural Language Parsing: Receives natural language input from users (e.g., "Add a tree to the map") and extracts semantic entities, operational intentions, and parameter ranges.

[0063] Entity Recognition: Identifying "map" and "tree" as target objects and object types, respectively.

[0064] Intent Detection: Identifies the action type corresponding to "add".

[0065] Parameter extraction: such as position, scaling, rotation and other auxiliary conditions.

[0066] Function mapping: Mapping the extracted intents and parameters to the function interfaces registered in the Function Service Layer (MCP). For example, mapping "add a tree" to the AddObject function, with the object type being Tree.

[0067] Generate structured instructions: Based on the MCP interface specification, convert the parsed results into structured instructions (such as JSON format or other interactive intermediate formats).

[0068] The instruction consists of three parts: Function identifier (Action / Method): such as AddObject; Target / Entity: such as Tree; Parameter information: such as { "location": [100, 50, 0], "scale": 1.0}; 2) Functional Service Layer (MCP): The Model Context Protocol (MCP) acts as an intermediate bridge layer, responsible for transforming the abstract instructions of the LLM into executable function packages. Its specific functions include: Interface registration and management: MCP maintains a functional registry, for example: AddObject(type, location, scale); DeleteObject(id); MoveObject(id, new_location); This registry can be accessed by the LLM so that the LLM can call the appropriate interface after semantic parsing.

[0069] Command Validation and Encapsulation: After receiving structured commands from the LLM, the MCP performs validity checks (such as whether the target object exists and whether the parameters are within the specified range). Valid commands are then encapsulated into a standardized JSON command package, containing: action: Operation type (e.g., "AddObject"); entity: object type or ID (e.g., "Tree"); params: Detailed parameters (such as position, scaling value).

[0070] Execution: MCP sends the generated JSON instruction package to the UE5 client.

[0071] 3) Function Execution Layer (UE5): UE5, as the execution environment layer, primarily includes the following functions: Command parsing: The UE5 MCP client receives JSON command packets, calls the local command parsing module, and converts them into specific blueprints or C++ API calls.

[0072] Object operation execution: Calling UE5's World API to perform operations in the virtual scene: adding objects (trees, buildings, etc.); modifying object properties (position, size, rotation, etc.); deleting objects; Result feedback: UE5 returns the operation result (success / failure status, object ID, etc.) to MCP, which then passes it to LLM for subsequent semantic dialogue closure.

[0073] To better understand, a complete application example is given below, demonstrating the three-layer interaction logic: User input: "Place a big tree in the center of the scene"; Step 1: LLM parsing (semantic understanding layer): Entity recognition: Object = Tree; Intent recognition: Add Object; Parameter explanation: Position = center of scene (mapped to coordinates [0,0,0]); Size = large (mapped to scale=2.0) Generate structured instructions: { "action": "AddObject", "entity": "Tree", "params": { "location": [0, 0, 0], "scale": 2.0 } }; Step 2: MCP processing (functional service layer): Verification: Check if AddObject(Tree) is in the registry → valid.

[0074] Encapsulated as a JSON command package: { "command_id": "12345", "action": "AddObject", "entity": "Tree", "params": { "location": [0, 0, 0], "scale": 2.0 }, "timestamp": "2025-09-02T15:30:00Z" }; Step 3: UE5 Execution (Function Execution Layer): UE5 client receives JSON packet → calls UE5 API: World->SpawnActor(TreeClass, FVector(0,0,0), FRotator::ZeroRotator,Scale=2.0); Generate an enlarged tree object in the center of the scene.

[0075] Return execution result: { "command_id": "12345", "status": "success", "object_id": "Tree_001 }

[0076] Figure 3This is a timing diagram illustrating a control method for a three-dimensional geographic information system according to an exemplary embodiment of this specification. For better understanding, the following explanation is provided with specific examples: Step 1: The user inputs voice commands through the microphone (such as "Turn the camera left, switch to Mercator coordinate system"), and UE5 captures the real-time voice stream through the audio interface.

[0077] Step 2: The speech-to-text plugin converts speech into text and supports multiple languages.

[0078] Step 3: Send the text command to the LLM large model via HTTP protocol for intent classification and parameter extraction. Specifically: 1. Input preparation: The speech-to-text module outputs a plain text command, such as "Turn the camera left", "Switch to Mercator coordinate system", "Zoom in to the building entrance". The system will encapsulate this text into a request body and send it to the LLM server (such as the OpenAI API or a self-built LLM service) via an HTTP POST request.

[0079] 2. Request content design: System Prompt: Predefines the AI's role and output format. For example: { "role": "system", "content": "You are an instruction parser. Your task is to extract intent and parameters from user input and output structured instructions in JSON format." }; User Prompt: The actual text input by the user.

[0080] { "role": "user", "content": "Camera turns left" }; 3. Intent Classification: LLM categorizes user input into a predefined instruction category based on training / prompt words.

[0081] For example, "Turn the camera left" is categorized under "CameraControl"; "Switch to Mercator coordinate system" is categorized under "CoordinateSystemSwitch". 4. Parameter Extraction: While classifying, extract the parameter values ​​required for the operation: "Turn the camera left" → {"direction": "left", "angle": 15} (default angle 15°, rules can be agreed upon in the prompt). "Switch to Mercator coordinate system" → {"target_system": "Mercator"}.

[0082] 5. Output structured results: LLM ultimately returns a structured JSON, which the MCP Server can use directly. { "intent": "CameraControl", function": "rotateCamera", "parameters": { "direction": "left", "angle": 15 } }; or: { "intent": "CoordinateSystemSwitch", "function": "switchCoordinateSystem", "parameters": { "target_system": "Mercator" } }; Step 4: The LLM calls the corresponding tool registered with the MCP based on the parsing results to generate structured instructions. Specifically: (1) LLM and MCP tool registration mechanism: The Model Context Protocol (MCP) provides a tool registration mechanism (e.g., using the `@tool` decorator or configuration file declaration). Each tool needs to define the following upon registration: tool name (unique identifier); callable function name; input parameter type (schema); and return result type (schema). The LLM learns of the existence of these tools and their invocation rules through the system prompt.

[0083] In this way, after parsing the user's intent, LLM can automatically call the corresponding MCP tool and output structured calling instructions that conform to the MCP protocol.

[0084] (2) LLM generates structured instructions: After step 3, which involves intent classification and parameter extraction, LLM generates a structured invocation instruction (conforming to the MCP tool invocation specification). This instruction is characterized by: explicitly pointing to the tool registered in the MCP; including the function name; carrying all necessary parameters; and attaching the invocation context (such as request_id / callback_id, used for subsequent responses).

[0085] Example: The user says "Turn the camera 15 degrees to the left"; the structured instructions generated by LLM (sent to the MCP Server, not directly to UE5): json{ "tool": "CameraControl", function": "rotateCamera", "parameters": { "direction": "left", "angle": 15 }, "callback_id": "req_20250902_001" }; (3) Interaction between structured instructions and the MCP Server: After receiving the structured instructions from the LLM, the MCP Server (Modular Control Protocol Server) performs two steps: 1. Verify the validity of the instructions: Check whether `tool` and `function` are registered and whether the parameters conform to the schema. 2. Translate into a JSON instruction packet: Encapsulate it into standardized JSON according to the UE5 TCP protocol format, ready to send it to UE5.

[0086] In this process, the MCP Server acts as a "protocol converter" and "scheduler".

[0087] (4) JSON command packet generated by MCP Server (sent to UE5): Continuing with the example above, the JSON command package generated by the MCP Server is as follows: json{ function": "rotateCamera", "parameters": { "direction": "left", "angle": 15 }, "callback_id": "req_20250902_001", "timestamp": "2025-09-02T15:30:00Z" }; Among them, `function`: corresponds to the blueprint / C++ function name already encapsulated in the UE5 plugin; `parameters`: can be directly executed by UE5; `callback_id`: used to associate the request when the result is returned; `timestamp`: optional field, convenient for logging and debugging.

[0088] (5) Complete process summary (use cases): The user said: "Switch to Mercator coordinate system"; 1. Speech-to-text output: "Switch to Mercator coordinate system"; 2. Step 3 → LLM parsing results: json { "intent": "CoordinateSystemSwitch", "function": "switchCoordinateSystem", "parameters": { "target_system": "Mercator"} }; 3. Step 4 → LLM generates structured instructions based on the MCP tool registration information: json { "tool": "CoordinateSystem", "function": "switchCoordinateSystem", "parameters": { "target_system": "Mercator"}, "callback_id": "req_20250902_002" }; 4. The MCP Server translates the data into a TCP JSON command packet (sent to UE5): json { "function": "switchCoordinateSystem", "parameters": { "target_system": "Mercator"}, "callback_id": "req_20250902_002", "timestamp": "2025-09-02T15:31:00Z" }; It should be noted that the LLM in this application does not directly output natural language, but generates structured instructions based on the MCP tool call specification; the MCP Server acts as a protocol converter, converting the abstract tool calls of the LLM into UE5 executable JSON instruction packages; the introduction of the callback_id mechanism supports multi-round interaction and asynchronous response, enhancing the robustness of the system.

[0089] Step 5: The MCP Server sends a JSON command packet (containing function name, parameters and callback ID) to UE5 via the TCP protocol to establish a persistent connection.

[0090] It's important to note that the UE5 plugin acts as a "TCP server," waiting for instructions from the MCP Server. The "client logic" of the MCP protocol itself (tool registration, instruction generation) is handled on the MCP Server side. The UE5 side only needs to implement: a TCP server listener; a JSON parser (converting JSON into UE5 internal function calls); a function library (blueprints or C++ wrappers, such as rotateCamera, switchCoordinateSystem); and a result feedback mechanism (returning the execution results back to the MCP Server via TCP).

[0091] It's important to note that the UE5 client doesn't require a full MCP client; instead, it implements a lightweight TCP server and an MCP command interpreter. The actual MCP client role resides on the MCP Server side, responsible for interacting with the LLM, generating commands according to the MCP standard, and pushing them to UE5.

[0092] Step 6: After UE5 parses the JSON, it dynamically calls the Blueprint or C++ function to perform 3D scene operations.

[0093] Step 7: The execution result is returned to MCP via TCP, including status code (success / failure) and detailed data (such as new view coordinates).

[0094] Step 8 (optional): MCP returns the results to LLM, generating natural language feedback (such as "Switch to Mercator coordinate system").

[0095] To better understand, the following further explanation is provided: I. Participating Parties and Pre-Agreements: UE5 Plugin (Server): Enables TCP listening at startup (e.g., 0.0.0.0:27015), has built-in JSON parsing and function routing (Blueprint / C++), and supports callbacks.

[0096] MCP Server (Client): Maintains the Tool Registry and parameter schema, and is responsible for bidirectional protocol conversion and scheduling with LLM / UE5.

[0097] LLM service: Supports "function calling" style output; only produces structured calling instructions, not directly produces UE5 executable JSON.

[0098] Unified Identifier: The entire process uses callback_id (or request_id) for request association and idempotency.

[0099] II. End-to-end implementation process: Step A: UE5 retrieves the text and sends it to LLM (via MCP); (1) UE5 obtains text: Source: ASR or UI input.

[0100] Example text (in Chinese): "Turn the camera 15 degrees to the left and zoom in to the building entrance."

[0101] UE5 sends the text to the MCP Server (HTTP / WS is acceptable; HTTP POST is recommended).

[0102] (2) MCP encapsulates LLM requests: 2.1. System Prompt (Key Points): Role: 3D engine instruction parser; Objective: Output a structured call format consisting of "tool name + function name + parameters"; Constraints: Unspecified parameters are completed using the default strategy; units / coordinates are standardized; callback_id is generated; 2.2. Tools List (from MCP Registry) (Example Entries): CameraControl.rotateCamera(angle: number, direction: enum[left|right]); CameraControl.zoomTo(target: enum[poi|bbox|latlon], value: any); CoordinateSystem.switch(target_system: enum[Mercator|WGS84|ENU]); User Prompt: Raw text + optional context (current coordinate system, camera pose, etc.).

[0103] It should be noted that in this step, LLM can only "see and call the tool," but does not generate the final JSON for UE5.

[0104] Step B: LLM generates structured call instructions (for MCP, not for UE5); The output is a structured call (single or pipeline / batch) of the "tool layer" for MCP retranslation.

[0105] It should be noted that steps support orchestration (wait_for builds dependencies), demonstrating innovative chained tool calls. At this stage, LLM can perform parameter completion / unit standardization and explicitly record it in assumptions.

[0106] Step C: MCP sends the structured call to UE5 instruction JSON (TCP delivery); (1) Schema validation and normalization: Verify whether the tool / function is registered and whether the parameter type / enumeration / range matches.

[0107] Units and coordinate systems are normalized again (if necessary, a preliminary transformation step is triggered, such as switching the coordinate system first).

[0108] (2) Instruction mapping and packaging (for the "execution layer" of UE5): Mapping rules: CameraControl.rotateCamera → UE5 rotateCamera (Blueprint / C++ public function); Parameter mapping: angle_deg → angle (float), direction → direction (enum); Similarly, package the second step of zoomTo and send it in dependency order / single connection order (or perform queue scheduling on the MCP side).

[0109] (3) TCP sending and persistent connections: MCP (Client) → UE5 (Server) persistent connection, sequentially sending JSON command packets.

[0110] Connection disconnection retry, sequence number and idempotency policy (duplicate packets are directly discarded or overwritten).

[0111] Step D: UE5 execution and feedback; (1) UE5 parsing and routing: Receive JSON → Verify version / opcode / function / parameters → Route to Blueprint / C++.

[0112] Execute functions (such as camera rotation, scaling, coordinate system switching, etc.).

[0113] (2) Execution results are sent back to MCP: (3) MCP summary and (optional) feedback LLM: MCP summarizes the results of each step with pipeline_id; if step 8 is enabled, the structured results can be sent back to LLM to generate natural language feedback (Chinese / English multilingual).

[0114] III. Error Handling and Robustness Key Points: Double-layer verification: LLM-side schema constraints + MCP-side strong verification to prevent dirty parameters from being implemented.

[0115] Idempotent / Deduplication: Use callback_id / sequence number as the key; UE5 returns the most recent result directly when it receives a duplicate packet.

[0116] Parameter strategy: Default values ​​are explicitly recorded (assumptions.defaults); thresholds / clamps can be set in MCP.

[0117] Dependencies / Transactions: Supports steps.wait_for; provides atomic batches (commit only if all batches succeed).

[0118] Coordinate / Unit Standardization: LLM / MCP uses "internal standards" (such as angles °, meters, Mercator) for consistent input / output conversion.

[0119] Multilingual: ASR is labeled with lang, LLM parsing does not depend on language, only on semantics; feedback is output by lang.

[0120] Security: MCP→UE5 packets can be signed (e.g., sig field, HMAC-SHA256); UE5 executes after signature verification.

[0121] To better understand, the following specific embodiment is illustrated: Example 1: Single instruction chain (left turn + amplification): (1) Input text (UE5→MCP): "Turn the camera 15 degrees to the left and zoom in to the building entrance"; (2) LLM structured calls (to MCP): (3) MCP→UE5 instruction packet (sent sequentially): (4) UE5→MCP execution return: A01_s1 succeeded, returning the current camera pose; A01_s2 succeeded, returning the new FOV / distance of the camera; (5) MCP→LLM→Natural Language Feedback: "Turn 15° to the left and zoomed in to the entrance."

[0122] It should be noted that this application has the following innovative aspects: 1. LLM only produces structured instructions for the tool layer, and MCP performs secondary compilation of the protocol and semantics to ensure the stability and verifiability of the engine side.

[0123] 2. Supports programmable multi-step pipelines (steps + wait_for), allowing complex operations across tools, coordinates, and time-dependent processes to be completed in a single natural language session.

[0124] 3. The two-layer schema is consistent with the default strategy / dimension, adding an engineering safeguard (safety / idempotency / traceability) on top of the uncertainty of LLM.

[0125] 4. End-to-end traceable link (pipeline_id / callback_id / trace), facilitating playback, auditing, and problem localization.

[0126] 5. Automatic normalization of coordinates / units and injection of necessary pre-steps (such as switching systems before flying) significantly improves success rate and user experience.

[0127] Furthermore, the UE5 engine portion of this application is integrated as a plugin, featuring a robust modular structure and platform compatibility. TCP server listening is automatically started upon project startup. The plugin includes a built-in JSON parsing module, converting JSON into functions and parameters, supporting function calling. It also encapsulates function libraries for camera control, coordinate system transformation, and other functionalities.

[0128] The MCP service component designs a tool registration mechanism (such as the @tool decorator) according to the MCP standard specification. It implements API interfaces with LLM (such as OpenAI Function Calling). A TCP client module is developed to connect UE5 and the MCP server via TCP.

[0129] The LLM interaction design specification includes system prompts, defining AI roles, defining tool invocation specifications, establishing communication between LLM and MCP, and implementing error handling mechanisms.

[0130] Furthermore, in response to the problems of unintuitive interaction methods, the need to memorize complex operation commands, and complex and redundant UI interfaces in related technologies, this application introduces natural language interaction. Users only need to describe their intentions with voice, without having to memorize complex operations, thus improving the intuitiveness of the interaction.

[0131] To address the problem that related technologies cannot understand the complex spatial operation intentions expressed in natural language, this application combines LLM (Large Language Model) for semantic parsing, converting the user's natural language into precise operation instructions to achieve intelligent control.

[0132] To address the issue of poor system scalability and the need to modify core code for new functions in related technologies, this application adopts the MCP (Modular Control Protocol) architecture, which encapsulates functions as independent tools. Through LLM dynamic scheduling, adding new functions only requires registering the new tool in the MCP without modifying the main program, thus improving scalability.

[0133] Corresponding to the embodiments of the foregoing methods, this specification also provides embodiments of the apparatus and the terminal to which it is applied.

[0134] The embodiments of the control device for the 3D geographic information system described in this specification can be applied to computer equipment, such as servers or terminal devices. The device embodiments can be implemented through software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor controlling the 3D geographic information system loading the corresponding computer program instructions from non-volatile memory into memory for execution. From a hardware perspective, such as... Figure 4 The diagram shown is a hardware structure diagram of a computer device housing the control device of a three-dimensional geographic information system according to an embodiment of this specification. Except for... Figure 4 In addition to the processor 410, memory 430, network interface 420, and non-volatile memory 440 shown, the server or electronic device where the device 431 is located in the embodiment may also include other hardware depending on the actual function of the computer device, which will not be described in detail here.

[0135] like Figure 5 As shown, Figure 5 This specification is a block diagram illustrating a control device for a three-dimensional geographic information system according to an exemplary embodiment. The device is applied to Unreal Engine and includes: The acquisition module 510 is used to acquire target voice commands, wherein the target voice commands are used to instruct the three-dimensional geographic information system to perform target operations. The determination module 520 is used to convert the target speech into text instructions and determine the operation instructions corresponding to the text instructions through a large language model. The execution module 530 is used to perform the target operation on the three-dimensional geographic information system according to the operation instructions.

[0136] The control device of the aforementioned 3D geographic information system revolutionizes the traditional operation mode of 3D GIS by leveraging intelligent parsing of natural language and large language models. Users issue natural language commands, and the Unreal Engine then calls the LLM (Language Model) to determine the corresponding operation instructions and execute the appropriate actions on the 3D GIS. This process avoids lengthy manual operations and complex interface navigation, not only improving control efficiency but also making operation more intuitive and convenient, greatly enhancing the user experience. Especially in emergency or professional scenarios, it can quickly respond to user needs and achieve efficient 3D spatial control.

[0137] The specific implementation process of the functions and roles of each module in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0138] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of the solution in this specification according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0139] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, performs the steps in any of the above method embodiments.

[0140] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0141] Specific examples in this embodiment can be found in the examples described in the above method embodiments and exemplary implementations, and will not be repeated here.

[0142] Embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the steps in any of the above method embodiments.

[0143] Embodiments of this application also provide an electronic device, including a memory and a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps in any of the above method embodiments.

[0144] In one exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0145] Specific examples in this embodiment can be found in the examples described in the above method embodiments and exemplary implementations, and will not be repeated here.

[0146] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0147] Other embodiments of this specification will readily occur to those skilled in the art upon consideration of the specification and practice of the invention claimed herein. This specification is intended to cover any variations, uses, or adaptations that follow the general principles of this specification and include common knowledge or customary techniques in the art not claimed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this specification are indicated by the following claims.

[0148] It should be understood that this specification is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this specification is limited only by the appended claims.

[0149] The above description is merely a preferred embodiment of this specification and is not intended to limit this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of protection of this specification.

Claims

1. A control method for a three-dimensional geographic information system, characterized in that, Applied to Unreal Engine, including: Obtain text commands, wherein the text commands are used to instruct the execution of target operations on the three-dimensional geographic information system, and the text commands are commands obtained by converting acquired voice commands or commands obtained through an interactive interface; The operation instructions corresponding to the text instructions are determined by a large language model; The target operation is performed on the three-dimensional geographic information system according to the operation instructions.

2. The method according to claim 1, characterized in that, The operation instructions corresponding to the text instructions are determined through a large language model, including: The text instruction is sent to the large language model, wherein, after receiving the text instruction, the large language model performs semantic parsing on the text instruction to obtain a semantic parsing result; the large language model generates a structured calling instruction that conforms to the MCP tool calling specification based on the semantic parsing result, and sends the structured calling instruction to the MCP server; The operation instruction sent by the MCP server is obtained, wherein after obtaining the structured call instruction, the MCP server encapsulates the structured call instruction into a standardized operation instruction according to the transmission protocol format between the MCP server and the Unreal Engine; the operation instruction is a JSON instruction package executable by the Unreal Engine.

3. The method according to claim 2, characterized in that, After performing the target operation on the three-dimensional geographic information system according to the operation instructions, the method further includes: The execution result of the target operation is announced verbally; and / or The execution results are displayed through the interactive interface; and / or The execution result of the target operation is sent to the MCP server. After obtaining the execution result, the MCP server sends the execution result to the large language model, and the large language model performs result analysis on the execution result.

4. The method according to claim 2, characterized in that, The semantic parsing results include: intent classification results and parameter values ​​for performing the target operation; The large language model generates a structured invocation instruction conforming to the MCP tool invocation specification based on the semantic parsing result, including: determining the system prompt words and tool list provided by the MCP server, wherein the system prompt words are used to indicate the output format and output constraints of the large language model, and the tool list contains the registration information of each tool in the MCP server; generating the structured invocation instruction based on the system prompt words and tool list, wherein the structured invocation instruction includes target information, the target information including: the tool name pointing to the tool registered in the MCP server, the function name of the function that the tool is allowed to call, the parameter value for executing the target operation, and a first identifier; the first identifier is used to identify a function call.

5. The method according to claim 4, characterized in that, When the target operation includes multiple sub-operations, the structured invocation instruction includes: target information and a second identifier corresponding to each sub-operation; wherein, the second identifier corresponding to the sub-operation is used to point to the first identifier corresponding to the remaining sub-operations, and is used to control the execution order of the sub-operations.

6. The method according to claim 4, characterized in that, In the process of generating a structured invocation instruction that conforms to the MCP tool invocation specification based on the semantic parsing result, when the parameter value for executing the target operation cannot be quantitatively determined, the large language model performs parameter value completion processing according to preset rules, and records the completion processing process in the structured invocation instruction through the field value of the target field.

7. The method according to claim 1, characterized in that, After performing the target operation on the three-dimensional geographic information system according to the operation instructions, the method further includes: If the target operation includes multiple sub-operations, and only some of the sub-operations are successfully executed on the 3D geographic information system, the 3D geographic information system is rolled back to the state before the target operation was executed.

8. A control device for a three-dimensional geographic information system, characterized in that, Applied to Unreal Engine, including: The acquisition module is used to acquire text commands, wherein the text commands are used to instruct the execution of target operations on the three-dimensional geographic information system, and the text commands are commands obtained by converting acquired voice commands or commands obtained through an interactive interface. The determination module is used to determine the operation instruction corresponding to the text instruction through a large language model; An execution module is used to perform the target operation on the three-dimensional geographic information system according to the operation instructions.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of the method described in any one of claims 1 to 7.

10. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method according to any one of claims 1 to 7.

Images