Engineering intelligent design method and system based on LLM and MCP

Abstract

The invention discloses an intelligent engineering design method and system based on LLM and MCP. The method comprises the following steps: receiving a design demand described by a natural language of a user; using a large language model to query registration metadata of a pre-created MCP operator, and based on the registration metadata, performing analysis and intelligent arrangement on the design demand to generate an executable MCP operator calling sequence subjected to dependency sorting; and scheduling the corresponding MCP operator by utilizing an execution engine according to the MCP operator calling sequence to complete the design requirement. According to the method, the cognitive ability of a large language model is combined with reliable execution of a professional operator, through metadata driving and process tracing, uncertainty caused by direct dependence on the large model is effectively avoided, and efficient, safe and controllable intelligent man-machine collaborative design of complex engineering is achieved.

Description

Technical Field

[0001] This invention relates to the field of railway survey and design technology, specifically to an engineering intelligent design method and system based on LLM and MCP. Background Technology

[0002] Currently, engineering design software for various disciplines such as railway lines, stations, bridges and tunnels, electrification, and signaling has developed into relatively complete algorithm systems and digital design functions after years of development. However, existing software is mostly driven by fixed processes, with high coupling between modules and insufficient data connection across disciplines and design stages, making it difficult to adapt to complex and ever-changing engineering design needs.

[0003] To improve flexibility, some studies have attempted optimization through expert systems / rule bases or digital design platforms, but the following shortcomings still exist:

[0004] 1. Flexibility issues of existing expert systems and rule bases: To reduce development complexity, existing expert systems or rule bases typically employ fixed rule combinations and simplified logical judgment processes. Their intelligence level is essentially a mechanical matching of preset rules. This results in a lack of adaptability and inference capabilities when dealing with real-world engineering scenarios with complex design constraints and personalized solution requirements, thus limiting their applicability.

[0005] 2. Automation and Usability Issues of Digital Design Platforms: While existing digital design platforms integrate design tools, they still heavily rely on engineers to perform extensive and tedious manual data configuration and process operations before executing specific design tasks. The complex operation process and low level of automation not only result in low efficiency but also significantly limit the flexibility and responsiveness of design exploration.

[0006] 3. Issues with the flexibility and traceability of data management: While existing solutions have improved data flow across disciplines and design phases, they typically rely on fixed data formats and predefined interaction interfaces, lacking flexibility. Furthermore, the general lack of a complete data lineage tracing and version management mechanism leads to opaque design processes, difficulty in assessing the impact of data changes, and an inability to quickly locate and trace problems after they occur, thus hindering the efficiency and quality of collaborative design. Summary of the Invention

[0007] This invention provides an intelligent engineering design method and system based on LLM (Large Language Model) and MCP (Model Context Protocol) to solve the above-mentioned technical problems existing in traditional engineering design methods.

[0008] According to the first aspect, one embodiment provides an engineering intelligent design method based on LLM and MCP, the method comprising:

[0009] Receive design requirements from users described in natural language;

[0010] By using a large language model to query the registration metadata of pre-created MCP operators, and based on the registration metadata, the design requirements are parsed and intelligently orchestrated to generate an executable MCP operator call sequence with dependency sorting.

[0011] The execution engine utilizes the MCP operator call sequence to schedule the corresponding MCP operator to fulfill the design requirements.

[0012] Furthermore, the use of large language models to analyze and intelligently orchestrate design requirements specifically includes:

[0013] When parsing and intelligently orchestrating design requirements based on registration metadata, collaborative reasoning is also performed by retrieving a pre-built domain knowledge base. This method can adaptively handle both mature fixed processes and non-fixed, complex, and flexible design requirements.

[0014] Furthermore, collaborative reasoning is performed by retrieving a pre-built domain knowledge base, specifically including:

[0015] Based on the professional rules in the domain knowledge base, the logical compliance verification and optimization of the MCP operator call sequence are performed.

[0016] Furthermore, the method also includes:

[0017] This method identifies the type of design requirement and obtains a predefined sequence of standard design requirement decomposition steps by calling a preset prompt word template. It is applicable to design requirements with mature and fixed processes.

[0018] Furthermore, the registration metadata for the MCP operator is created, specifically including:

[0019] The functions, interfaces, constraints, and dependencies of the MCP operators are registered in the MCP server as metadata.

[0020] Furthermore, the execution engine utilizes the MCP operator call sequence to schedule the corresponding MCP operator to fulfill the design requirements, specifically including:

[0021] The implementation of the MCP operator includes microservices, plugins, or modular design functional units. The execution engine invokes the MCP operator by calling microservices deployed locally or remotely, or by loading and running local plugins or component libraries.

[0022] Furthermore, the method also includes:

[0023] Throughout the entire process of executing design requirements, operator call associations and data lineage information are recorded to support traceability and verification of the execution process.

[0024] Furthermore, the method also includes:

[0025] The design results and execution process status are fed back to the user, forming a closed-loop human-computer collaborative interaction.

[0026] According to the second aspect, one embodiment provides an engineering intelligent design system based on LLM and MCP, the system comprising:

[0027] The user interaction module is used to receive natural language design requirements input by users and provide feedback on the results.

[0028] The task parsing and orchestration module is used to use a large language model to query the registration metadata of pre-created MCP operators, and based on the registration metadata, parse and intelligently orchestrate the design requirements to generate an executable MCP operator call sequence after dependency sorting.

[0029] The execution engine module is used to schedule the corresponding MCP operators according to the MCP operator call sequence to complete the design requirements;

[0030] The data management module is used to store MCP operator metadata, prompt word templates, domain knowledge base, and record data lineage information during task execution.

[0031] According to a third aspect, one embodiment provides an electronic device, the device comprising: a processor and a memory;

[0032] The memory is used to store one or more program instructions;

[0033] The processor is configured to run one or more program instructions to perform the steps of an engineering intelligent design method based on LLM and MCP as described in any of the preceding claims.

[0034] Compared with existing railway engineering design software that relies on fixed processes, manual configuration, and expert rule bases, this invention provides an intelligent engineering design method and system based on LLM and MCP, which has the following advantages:

[0035] 1. Intelligent parsing mechanism for "demand-capability" matching and verification driven by "operator metadata".

[0036] Behavior: Before executing a task, the command big language model actively queries and uses the structured metadata (including functional description, input and output constraints, and execution dependencies) of the MCP operator to ensure that the granularity of the task decomposition matches the function of the operator.

[0037] Effect: By strictly constraining the parsing results of natural language requirements within the capabilities of the system's callable operators, it ensures that each step in the entire chain from requirements analysis to result generation corresponds to an executable professional operator, fundamentally ensuring the feasibility and technical feasibility of the design scheme.

[0038] 2. It achieves "dynamic workflow planning" instead of "fixed instruction translation".

[0039] Behavior: Based on the queried MCP operator metadata, the system infers and dynamically combines in real time to generate a non-preset, dependency-ordered executable operator call sequence;

[0040] Results: It overcomes the limitation of existing technologies that can only execute predefined fixed processes, enabling the system to autonomously cope with complex, variable, and non-standardized engineering design needs, and achieves true intelligent design automation.

[0041] 3. A collaborative decision-making framework based on "multi-level structured information sources" was constructed.

[0042] Behavior: In the decision-making process, three major information sources are used collaboratively: operator metadata (defining "what can be done" and dependencies), domain knowledge base (providing the logic and rules of "why do it this way"), and prompt word templates (providing guidance on "how to parse efficiently").

[0043] Results: A triple guarantee of "capability boundaries - professional logic - parsing efficiency" has been formed, which has significantly improved the accuracy, compliance and intelligence of task parsing and orchestration.

[0044] 4. A management and traceability method based on "end-to-end data lineage" has been introduced.

[0045] Behavior: During task execution, automatically record the call relationships between operators, data flow paths, and version information to build a complete data lineage graph;

[0046] Results: It achieves full transparency in the design process, supports rapid traceability, impact analysis, and technical verification of any design outcome, and greatly enhances the quality control and credibility of the design process.

[0047] 5. Possesses dual-mode processing capabilities that allow for "adaptive adaptation between fixed and flexible processes".

[0048] Behavior: The system can intelligently identify task types and adaptively select the optimal processing strategy: for mature and fixed requirements, it matches and calls preset template sequences; for complex and flexible requirements, it starts the dynamic metadata orchestration engine.

[0049] Results: It balances high efficiency in handling standardized tasks with high flexibility in dealing with innovative tasks, giving the system both industrial-grade reliability and cutting-edge exploration capabilities. Attached Figure Description

[0050] Figure 1 A flowchart illustrating an engineering intelligent design method based on LLM and MCP, provided as an embodiment of the present invention;

[0051] Figure 2 This is an overall implementation architecture diagram of an engineering intelligent design method based on LLM and MCP, provided as an embodiment of the present invention;

[0052] Figure 3 This is a schematic diagram illustrating the automatic execution of workflow in an engineering intelligent design method based on LLM and MCP, provided as an embodiment of the present invention.

[0053] Figure 4 This is a schematic diagram of the execution engine call + microservice in an engineering intelligent design method based on LLM and MCP provided in one embodiment of the present invention. Detailed Implementation

[0054] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0055] Furthermore, the features, operations, or characteristics described in the specification can be combined in any suitable manner to form various embodiments. At the same time, the steps or actions in the method description can be rearranged or adjusted in a manner obvious to those skilled in the art. Therefore, the various orders in the specification and drawings are only for the clear description of a particular embodiment and do not imply a necessary order, unless otherwise stated that a particular order must be followed.

[0056] This invention provides an intelligent engineering design method based on LLM and MCP, the main process of which is as follows: Figure 1 As shown, the overall implementation architecture supporting method implementation is as follows: Figure 2As shown. The overall architecture adopts a microservice, layered, and decoupled architecture. Each layer's modules utilize microservices and are deployed in containers such as Docker. It mainly includes: an interaction layer, an intelligent engine layer, a service layer, and a data layer. The components of each layer and their connections are as follows:

[0057] 1. Interaction Layer

[0058] (1) Components: natural language human-computer interaction interface, graphical user interface, and templated input assistant.

[0059] (2) Connection Relationship: The interaction layer serves as the system entry point, allowing users to submit design requirements, constraints, and modification instructions through natural language, graphical operations, or form-based input based on prompt word templates. The system will preprocess the input (such as calling the prompt word template library for standardization) before passing it to the intelligent engine layer. At the same time, the interaction layer is responsible for providing visual feedback on the design results and process status to the user.

[0060] 2. Intelligent Engine Layer (The System's "Brain")

[0061] (1) Components:

[0062] 1) Task Analysis and Orchestration Module

[0063] It has a built-in Large Language Model (LLM) and prompt word template library. Its core function is to perform intelligent planning based on user needs from a global perspective.

[0064] The prompt word template library, as an input specification constraint mechanism for low-code LLMs, is used to transform natural language or graphical user interface input into structured task descriptions, avoiding inaccurate task arrangement caused by the arbitrariness of natural language. The template mechanism's function is:

[0065] Consistency: Ensure that the requirements for common tasks (such as the design of the main wiring diagram and longitudinal profile design of the traction station) are stable and standardized to avoid misunderstandings in LLM;

[0066] Flexibility: It supports users to fill in variables (parameterization) on the basis of fixed templates, so as to meet complex and personalized design needs without losing control;

[0067] Scalability: Templates can be added / iterated incrementally, which is equivalent to "low-code task blocks" and supports cross-disciplinary design requirement combinations.

[0068] The template supports professional classification (such as lines, bridges and tunnels, electrification, signaling, etc.) and task-level classification (such as planning level, scheme level, construction drawing level). It can also define input data, output results and constraints through placeholders, thereby ensuring that the orchestration logic generated by LLM is accurate and controllable.

[0069] After being parsed by LLM, the prompt word template is translated into a standard MCP operator call sequence.

[0070] 2) Execution Engine: Embedded workflow engine and state manager. Its core function is to strictly follow the established rules for executing sequences and manage the state and exceptions during execution. LLM, combined with prompt word templates, transforms user input into an ordered sequence of MCP operator calls.

[0071] The integration point between prompt word templates and MCP:

[0072] The template defines the "intent-operator mapping" relationship of the design task (e.g., "Generate catenary suspension scheme" → call the suspension simulation operator);

[0073] LLM automatically generates the MCP workflow by parsing the template filling results (which can be seen as a transformation from semi-structured prompt words to structured operator chains).

[0074] (2) Connection relationship: This layer is the core of the system. The input of the interaction layer is passed to the task parsing and orchestration module; this module calls LLM, and with the help of the prompt word template library, decomposes the design requirements into an ordered sequence of MCP operator calls (i.e., workflow), and sends the sequence to the execution engine; the execution engine receives the sequence and strictly calls the MCP operators in the service layer in sequence according to the preset input and output specifications, dependencies and execution constraints of each MCP operator, and monitors its execution status.

[0075] 3. Service Layer (the system's "hands and feet," based on a microservices architecture and hosted in containers such as Docker)

[0076] (1) Components: Multiple MCP operators.

[0077] (2) Connection relationship: Each MCP operator is an independently encapsulated microservice that communicates with the execution engine through the standard MCP protocol.

[0078] Operators are categorized by specialty, such as: power electrification operators (e.g., traction substation layout operators, traction power supply simulation calculation operators, traction substation main wiring design operators, traction substation equipment layout operators, contact network suspension simulation operators, cable route planning operators, etc.), line operators (e.g., horizontal alignment operators, longitudinal profile design operators), and bridge and tunnel operators (e.g., bridge layout operators, tunnel entrance positioning operators), etc. They receive call instructions and input data from the execution engine, execute the core algorithm, and return the results.

[0079] The connection between the prompt template and the MCP is as follows:

[0080] The template can be predefined with the calling mode of the MCP operator, for example:

[0081] Template: Design of a 110kV traction substation pavilion → MCP operator sequence: [Site selection operator] → [Main wiring design operator] → [Equipment layout operator]

[0082] LLM generates the input data structure for the corresponding operator call based on the parameters filled in the template (such as voltage level, scale, geographical constraints).

[0083] The execution engine then sends the task to the MCP operator for execution.

[0084] 4. Data layer (stores registration, metadata, lineage and version, logs, etc., and non-service layer operator data)

[0085] (1) Components: The data layer of this system is mainly responsible for process data, metadata, and relationship management, rather than replacing the original professional database. Its main components include:

[0086] 1) Metadata and registry: Used to register and manage the metadata of all MCP operators, including the operator's unique identifier, functional description, the mode and source of its required input data (e.g., from which interface of which original system), the mode of output data, constraint relationships, and a list of other operators that its execution depends on.

[0087] 2) Process Status Database: Used to store and manage the execution status of the design workflow, the call logs of each MCP operator, and key intermediate calculation results and data identifiers (not the complete data itself). For example, it stores the storage ID and version number of the output file of the "Longitudinal Section Design Operator" in the original design platform.

[0088] 3) Lineage and Version Management Database: Used to record the entire lineage of data generation (DataLineage). This lineage records the data's origin path (i.e., which version and which operator generated a particular output data from, and based on which data sources), rather than storing the data entity itself. Simultaneously, this database manages snapshots and relationships between design process versions and key output versions.

[0089] 4) Version management of prompt word templates: The template itself is also stored as a "metadata object" in the metadata and registry, supporting updates, version rollback and cross-professional sharing.

[0090] (2) Connection relationship: This layer provides scheduling basis and management support for the intelligent engine layer, and together with the service layer (MCP operator) it forms a complete data flow.

[0091] 1) Relationship with the intelligent engine layer: When orchestrating and executing workflows, the execution engine queries the metadata and registry to obtain the operator dependencies and constraints; the status and logs during workflow execution are persisted to the process status database; the lineage of all data is automatically recorded by the execution engine in the lineage and version management database.

[0092] 2) Relationship with the Service Layer (MCP Operator): The MCP operator is the execution channel for the data flow. During execution, the operator retrieves the required raw input data from the original digital design system platform through its own encapsulated original system interface, based on the identifier in the input parameters. After completing the calculation, the operator returns the main version and identifier of the generated result data to the execution engine, which records the lineage and simultaneously stores the complete data back to the original digital design system platform or a designated file storage.

[0093] 3) Core principle: This system does not replace the original project database, but interacts with it through MCP operators, and focuses on intelligent scheduling and management of the design process, data relationships and data versions.

[0094] The following is combined with Figure 1 The engineering intelligent design method based on LLM and MCP proposed in the embodiments of the present invention will be described in detail.

[0095] like Figure 1 As shown, in step S100, the design requirements described by the user in natural language are received.

[0096] Specifically, users input a natural language description of the design task through the interaction layer, such as "Find the xxxx high-speed rail project in the project library, conduct the scheme design, design the traction power supply system for a passenger dedicated line with a speed of 250 km / h, adopt the autotransformer supply system of electric traction (AT) power supply method, and estimate the number of main electrical equipment and cable lengths."

[0097] like Figure 1 As shown, in step S200, the large language model is used to query the registration metadata of the pre-created MCP operator, and the design requirements are parsed and intelligently orchestrated based on the registration metadata to generate an executable MCP operator call sequence that has been sorted by dependencies.

[0098] Specifically, the task parsing and orchestration module calls LLM and combines it with prompt word templates (such as "Please identify the [specialty], [stage], [key parameters], and [output requirements] in the task") to parse the fuzzy requirements into structured task description objects, and identify the specialty (electrification), key parameters (speed 250km / h, AT power supply), and expected output (number of devices, cable length).

[0099] The task parsing and orchestration module, based on the generated structured task description, and guided by the MCP operator library, operator dependency constraints, industry standard clauses, and prompt word template extension rules, uses LLM for reasoning to automatically generate an executable MCP operator call sequence. For example, the generated sequence might be: ["Calling mutual data from various disciplines such as line operation" → "Traction substation layout operator" → "Traction power supply simulation calculation" → "Traction substation capacity configuration operator" → "Traction substation design operator" → "Contact wire suspension type selection and simulation operator" → "Main equipment material list generation operator"]. MCP configuration constraints ensure the execution order and correctness of the workflow; for example, capacity configuration can only be triggered after simulation calculation is completed, and contact wire selection must rely on line longitudinal profile data.

[0100] like Figure 1 As shown, in step S300, the execution engine uses the corresponding MCP operator to complete the design requirements by scheduling the MCP operator according to the MCP operator call sequence.

[0101] The execution engine receives the MCP operator call sequence generated by S200 and calls the corresponding MCP operators in the service layer in sequence. During execution, each MCP operator must strictly follow the predefined input / output pattern and execution constraints to ensure the correctness and integrity of the computation chain.

[0102] Depending on the requirements, some operators support manual interaction for reviewing and correcting key steps, ensuring the reliability of the results. The output of the previous operator is automatically converted into the input format required by the next operator, enabling seamless data transfer across disciplines. The entire process is monitored by the execution engine, and any operator failure will automatically trigger rollback, fault tolerance, or user intervention notification.

[0103] End-to-end traceability and version management: During operator execution, the lineage and version management database automatically records the source, transformation process, and output results of all data, forming a complete data lineage map. The calling process and generated version number of each MCP operator are automatically recorded, allowing users to trace any design outcome and query its generation path and dependent original data and intermediate parameters.

[0104] In addition, the system supports version snapshots of design results at key nodes, allowing users to perform historical comparisons and iterate solutions, thereby improving the transparency and reliability of engineering design.

[0105] Results Feedback and Interaction Optimization: Final design results, intermediate results, or problems encountered during execution (such as missing parameters or conflicting solutions) are fed back to the user through the interaction layer. Users can review the results and choose to modify input parameters or adjust the task template before re-triggering execution from any step.

[0106] The prompt template system plays a crucial role here: users can adjust their input requirements based on the templates without having to relearn complex command syntax. This creates a closed-loop human-machine collaboration process: "request submission → intelligent generation → manual review → iterative optimization."

[0107] This invention adopts an open, layered, and decoupled system architecture, possessing high technical compatibility and evolutionary flexibility. With the rapid development of low-code platforms, AI (Artificial Intelligence) native workflow systems, and intelligent agents, each core component of this system can be extended, enhanced, or replaced using modern technology stacks, ensuring its adaptability and continued leading-edge performance across different development stages, deployment environments, and engineering application scenarios. Specific scalable or replaceable solutions are as follows:

[0108] 1. Expansion of interaction methods

[0109] In addition to natural language input and a graphical interface, this invention can also be extended to support voice recognition, gesture interaction, or augmented reality (AR) / virtual reality (VR) operation. By combining with a 3D visualization system, intuitive input of design tasks and review of results can be achieved in an immersive environment, improving the efficiency of human-computer interaction.

[0110] 2. Replacement of prompt template mechanism

[0111] The prompt word templates can be implemented not only as regularized text but also as a graphical low-code configurator, allowing users to build task logic through drag-and-drop, form filling, and other methods. Furthermore, by combining expert knowledge bases, industry standard databases, or domain ontology, automatic template generation and semantic enhancement can be achieved, improving the system's intelligence level.

[0112] 3. Alternative implementation methods for the intelligent engine layer

[0113] In addition to using large language models for task parsing and process orchestration, the following technical approaches can also be introduced:

[0114] 1) Based on a low-code workflow platform that supports natural language driving, the structured instructions generated by LLM are mapped into executable processes to achieve dynamic orchestration and visual management of tasks;

[0115] 2) Construct a multi-agent collaborative system, in which each agent encapsulates domain-specific knowledge and operator invocation capabilities, and completes the decomposition and execution of complex design tasks through a negotiation mechanism;

[0116] 3) Use reinforcement learning models to train on historical task execution data, optimize operator call sequences, and improve process execution efficiency.

[0117] 4. Extension of the MCP operator protocol

[0118] Although MCP is used as the standard communication protocol, the system is also compatible with mainstream microservice interface protocols such as REST API, gRPC, and GraphQL. MCP operators can be encapsulated as independent service modules, deployed in containerized environments, and asynchronous communication and decoupled scheduling can be achieved through message queues. As long as the operators have clear input / output patterns and dependency constraints, modular combination and cross-platform calls can be realized.

[0119] 5. Enhanced data lineage and version management

[0120] In addition to a dedicated lineage database, the system can also integrate blockchain or distributed ledger technology to achieve trusted evidence storage and tamper-proof traceability of critical design processes. It can also employ a graph database to store data dependencies, supporting complex path queries and impact analysis to meet the auditing needs of highly compliant scenarios.

[0121] 6. Expansion of applicable fields

[0122] Although this invention is primarily applied to railway engineering design, its architecture is versatile. By replacing the corresponding MCP operator library and prompt word templates, it can be migrated to other complex engineering fields such as highways, municipal works, power, and energy, achieving cross-industry reuse of intelligent design capabilities.

[0123] In summary, this invention not only possesses practicality under current technological conditions, but also supports continuous evolution in interaction methods, intelligent engines, operator protocols, data management, and application areas through its open architecture design. Particularly at the intelligent engine layer, it can integrate cutting-edge technologies such as AI-native workflows and multi-agent systems, ensuring the system maintains its technological advancement and industrial applicability in the long term.

[0124] The following combination Figure 3 , Figure 4 Taking a typical engineering task as an example, the implementation process of the present invention is explained in detail as follows:

[0125] 1) Design task description: Locate the "XX High-speed Railway Project" in the project database, conduct the scheme design, design a traction power supply system for a passenger dedicated line with a speed of 250 km / h, adopt the AT power supply method, and estimate the number of main electrical equipment and cable length.

[0126] 2) The specific implementation steps are as follows:

[0127] Step 1: Requirements Input and Structured Analysis

[0128] The user inputs the above task instructions into the natural language input module of the interaction layer. The system automatically matches the "Electrification_Traction Power Supply Scheme Design_v1" template from the prompt word template library. This template predefines the input fields, output targets, and standard MCP operator call paths.

[0129] The task parsing and orchestration module calls the Large Language Model (LLM), combines this template to perform semantic parsing of user input, extracts key parameters, and generates a structured task description.

[0130] json{"project_name":"XX High-speed Rail Project",

[0131] "design_phase":"Solution design",

[0132] "speed_target":"250km / h",

[0133] "supply_mode":"AT",

[0134] "output_requirements":["Number of main transformers","Number of circuit breakers","Total length of feeder cables"]}

[0135] Based on this, LLM infers and generates a strictly ordered sequence of MCP operator calls according to the pre-defined professional logic and dependencies in the template (e.g., simulation calculations must be performed after the facility is deployed, and capacity configuration must depend on simulation results).

[0136] json[

[0137] {"operator_id":"interdisciplinary_data_exchange","desc":"Invoking interdisciplinary data exchange for routes, train operation, and other related fields"}

[0138] {"operator_id":"traction_substation_layout","depends_on":interdisciplinary_data_exchange"},

[0139] {"operator_id":"traction_power_flow_simulation","depends_on":"traction_substation_layout"},{"operator_id":"substation_capacity_configuration","depends_on":"traction_power_flow_simulation"},

[0140] {"operator_id":"substation_detailed_design","depends_on":"substation_capacity_configuration"},

[0141] {"operator_id":"catenary_type_selection_and_simulation","depends_on":"interdisciplinary_data_exchange"},

[0142] {"operator_id":"bom_generation","depends_on":["substation_detailed_design","catenary_type_selection_and_simulation"]}]

[0143] This sequence ensures that the design process complies with railway engineering professional standards: first, external professional input is obtained, then the layout and simulation of the station are carried out, the capacity is configured based on the simulation results, then the detailed design of the station and the catenary is completed, and finally the equipment and material list is generated.

[0144] Step 2: Workflow Execution

[0145] After receiving the MCP call sequence, the execution engine strictly schedules the operators in the service layer to be executed in order of dependency, as follows:

[0146] S201: Calling data from various disciplines such as track operation and train operation. The system calls this operator through the MCP protocol to obtain the horizontal and vertical profile data of the track discipline and the basic parameters of the train operation diagram from the BIM collaboration platform as input for subsequent design.

[0147] S202: Traction substation layout operator, which automatically generates a preliminary layout plan for traction substations based on the line route, load distribution and terrain conditions (such as setting them at K50+000, K120+500, etc.).

[0148] S203: Traction power supply simulation calculation operator, which performs power flow calculation, short-circuit analysis, and voltage level verification based on the layout scheme. If the voltage in a certain section is too low, the system pauses the process, prompting the user to adjust the spacing between substations or add substations. After confirmation, the process will automatically restart.

[0149] S204: Traction substation capacity configuration operator, which automatically calculates the capacity and specifications of key equipment such as main transformers and circuit breakers based on the maximum load current and short-circuit level in the simulation results.

[0150] S205: Traction substation pavilion design operator, calls the electrical main wiring design module, generates a main wiring diagram that conforms to the specifications, and completes the equipment layout.

[0151] S206: Selection of catenary suspension type and simulation operator. Based on the target line speed (250km / h) and terrain conditions, select the appropriate fully compensated simple catenary suspension scheme and conduct pantograph-catenary dynamic performance simulation.

[0152] S207: Main equipment bill of materials generation operator, which summarizes the equipment parameters and cable path data output by the preceding operators to generate a structured bill of materials, including:

[0153] Main transformers: 2×40MVA, 4 units in total;

[0154] Feeder circuit breakers: rated current 3150A, 16 units in total;

[0155] Total length of feeder cable: approximately 18.6 kilometers.

[0156] Step 3: Full-process traceability and result feedback

[0157] During execution, the call status and input / output parameters of each MCP operator are recorded in the process status database. The lineage and version management database automatically constructs data lineage chains, for example:

[0158] "Equipment list (ID:BOM-2025-001) ← sourced ← from the design results of the inn (ID:DES-340-08) and the catenary scheme (ID:SIM-350-05) ← sourced ← from the capacity configuration results (ID:CAP-330-06) ← sourced ← from the simulation report (ID:SIM-320-07)."

[0159] Users can view the complete lineage map through a graphical interface, supporting the tracing of the generation path of any result, the versions of parameters used, and the source of the original data. The final design results are output in PDF report and Excel list format for manual review and archiving.

[0160] As demonstrated by the above embodiments, this invention can guide LLM to generate a strict MCP operator call sequence that conforms to professional logic for complex engineering tasks described in natural language through prompt word templates. The execution engine then drives the automatic execution of various professional operators, achieving an end-to-end intelligent design closed loop from requirement input to output. By enabling cross-system capability calls through the MCP protocol and ensuring full traceability through a lineage database, it effectively solves problems such as low collaborative efficiency, numerous human errors, and chaotic version management in traditional design processes, significantly improving the intelligence, standardization, and reliability of railway engineering design.

[0161] Corresponding to the aforementioned disclosed intelligent engineering design method based on LLM and MCP, this invention also discloses an intelligent engineering design system based on LLM and MCP, which specifically includes:

[0162] The user interaction module is used to receive natural language design requirements input by users and provide feedback on the results.

[0163] The task parsing and orchestration module is used to use a large language model to query the registration metadata of pre-created MCP operators, and based on the registration metadata, parse and intelligently orchestrate the design requirements to generate an executable MCP operator call sequence after dependency sorting.

[0164] The execution engine module is used to schedule the corresponding MCP operators according to the MCP operator call sequence to complete the design requirements;

[0165] The data management module is used to store MCP operator metadata, prompt word templates, domain knowledge base, and record data lineage information during task execution.

[0166] It should be noted that for a detailed description of the engineering intelligent design system based on LLM and MCP provided in the embodiments of the present invention, please refer to the relevant description of the engineering intelligent design method based on LLM and MCP provided in the embodiments of the present invention, which will not be repeated here.

[0167] In addition, embodiments of the present invention also provide an electronic device, the device comprising: a processor and a memory; the memory being used to store one or more program instructions; the processor being used to execute one or more program instructions to perform the steps of an engineering intelligent design method based on LLM and MCP as described in any of the preceding embodiments.

[0168] It should be noted that for a detailed description of an electronic device provided in the embodiments of the present invention, please refer to the relevant description of an engineering intelligent design method based on LLM and MCP provided in the embodiments of the present invention, which will not be repeated here.

[0169] Those skilled in the art will understand that all or part of the functions of the various methods in the above embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, which may include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to achieve the above functions. For example, the program can be stored in the memory of a device, and when the program in the memory is executed by the processor, all or part of the above functions can be achieved. In addition, when all or part of the functions in the above embodiments are implemented by computer programs, the program can also be stored in a server, another computer, disk, optical disk, flash drive, or external hard drive, etc., and can be downloaded or copied to the memory of a local device, or the system of the local device can be updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be achieved.

[0170] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the principles of this invention.

Claims

1. An intelligent engineering design method based on LLM and MCP, characterized in that, The method includes: Receive design requirements from users described in natural language; By using a large language model to query the registration metadata of pre-created MCP operators, and based on the registration metadata, the design requirements are parsed and intelligently orchestrated to generate an executable MCP operator call sequence with dependency sorting. The execution engine utilizes the MCP operator call sequence to schedule the corresponding MCP operator to fulfill the design requirements.

2. The engineering intelligent design method based on LLM and MCP according to claim 1, characterized in that, The method further includes: When parsing and intelligently orchestrating design requirements based on registration metadata, collaborative reasoning is also performed by retrieving a pre-built domain knowledge base.

3. The engineering intelligent design method based on LLM and MCP according to claim 2, characterized in that, Collaborative reasoning is performed by retrieving a pre-built domain knowledge base, specifically including: Based on the professional rules in the domain knowledge base, the logical compliance verification and optimization of the MCP operator call sequence are performed.

4. The engineering intelligent design method based on LLM and MCP according to claim 1, characterized in that, Utilizing large language models to analyze and intelligently orchestrate design requirements, specifically including: Identify the design requirement type and obtain a predefined sequence of standard design requirement decomposition steps by calling a preset prompt word template.

5. The engineering intelligent design method based on LLM and MCP according to claim 1, characterized in that, The registration metadata for the MCP operator is created, specifically including: The functions, interfaces, constraints, and dependencies of the MCP operators are registered in the MCP server as metadata.

6. The engineering intelligent design method based on LLM and MCP according to claim 1, characterized in that, The execution engine utilizes the MCP operator invocation sequence to fulfill the design requirements, specifically including: The implementation of the MCP operator includes microservices, plugins, or modular design functional units. The execution engine invokes the MCP operator by calling microservices deployed locally or remotely, or by loading and running local plugins or component libraries.

7. The engineering intelligent design method based on LLM and MCP according to claim 1, characterized in that, The method further includes: Throughout the entire process of executing design requirements, operator call associations and data lineage information are recorded to support traceability and verification of the execution process.

8. The engineering intelligent design method based on LLM and MCP according to claim 1, characterized in that, The method further includes: The design results and execution process status are fed back to the user, forming a closed-loop human-computer collaborative interaction.

9. An engineering intelligent design system based on LLM and MCP, characterized in that, The system includes: The user interaction module is used to receive natural language design requirements input by users and provide feedback on the results. The task parsing and orchestration module is used to use a large language model to query the registration metadata of pre-created MCP operators, and based on the registration metadata, parse and intelligently orchestrate the design requirements to generate an executable MCP operator call sequence after dependency sorting. The execution engine module is used to schedule the corresponding MCP operators according to the MCP operator call sequence to complete the design requirements; The data management module is used to store MCP operator metadata, prompt word templates, domain knowledge base, and record data lineage information during task execution.

10. An electronic device, characterized in that, The device includes: a processor and a memory; The memory is used to store one or more program instructions; The processor is configured to run one or more program instructions to perform the steps of an engineering intelligent design method based on LLM and MCP as described in any one of claims 1 to 8.

Images