Personalized electronic sand table recommendation system in virtual construction system based on GraphRAG

Abstract

The invention discloses an electronic sand table personalized recommendation system in a virtual construction system based on GraphRAG. The electronic sand table personalized recommendation system comprises a data acquisition module used for acquiring user behavior data, scene perception data and industry knowledge data; the data processing and analysis module is used for analyzing user behavior data and scene recognition and comparison; the GraphRAG retrieval enhancement module is used for constructing a knowledge graph, retrieving a template and a component library and enhancing the retrieval capability; the personalized recommendation module is used for recommending sand table templates and sand table components; the personalized explanation generation module is used for generating personalized explanation for the recommended templates and components by utilizing a GraphRAG technology; the visual interaction module comprises a template, component interaction and dialogue interaction; and the dynamic adjustment updating module is used for scene change identification and user feedback analysis. According to the method, the behavior data of the user and the three-dimensional scene information are combined, and the industry knowledge graph is utilized, so that scene perception and personalized recommendation are realized, and the intelligent level of virtual construction is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of virtual construction and intelligent recommendation, and in particular to a GraphRAG-based electronic sandbox personalized recommendation system in a virtual construction system. Background Art

[0002] In virtual construction systems, users usually need to create various 3D scenes according to different design requirements, and select suitable sandbox templates and components for design and display. However, existing virtual construction systems usually rely on static template libraries and component libraries, lacking the ability to intelligently perceive and dynamically respond to user-specific needs and behaviors. When faced with a large number of templates and component choices, users often feel confused and find it difficult to quickly find the most suitable design solution, which in turn affects design efficiency and quality.

[0003] In addition, with the increasing complexity of virtual construction application scenarios, such as transportation infrastructure, urban planning, and environmental protection, users' design needs have become more diverse. Existing technologies are unable to fully utilize user behavior data and scenario information to perform effective scenario recognition and personalized recommendations. This results in the virtual construction system lacking pertinence and intelligence when applied to complex scenarios, and failing to meet users' efficient design needs.

[0004] Existing technologies face the following major problems in virtual construction systems, which limit the intelligence and efficiency of the systems in complex design scenarios:

[0005] 1. Insufficient scene recognition and perception capabilities: The existing system lacks the ability to automatically recognize three-dimensional scenes and cannot automatically determine the type and requirements of the current design task based on the scenes and behavior data created by the user, resulting in a mismatch between the recommended content and the actual needs of the user.

[0006] 2. Insufficient personalized recommendations: Existing systems usually rely on static templates and component libraries, which makes it difficult to provide personalized recommendations based on users' individual needs and preferences, affecting users' design efficiency.

[0007] 3. Insufficient dynamic response and adjustment capabilities: The existing system lacks the ability to dynamically adjust recommended content during the user design process and cannot respond to user design changes and feedback in real time, resulting in delayed or inaccurate recommendations.

[0008] 4. Lack of professionalism in recommendation explanations: The existing system lacks the ability to professionally explain the recommended content, making it difficult for users to understand the basis for the recommendations, which affects the confidence and accuracy of design decisions.

[0009] In view of the above problems, the present invention proposes a virtual construction system based on GraphRAG, which realizes dynamic adjustment and professional interpretation of recommended content through intelligent scene recognition and personalized recommendation, combined with industry knowledge graph, and significantly improves the intelligence level of the system and user experience.

[0010] The information disclosed in this background technology section is only intended to deepen the understanding of the overall background technology of the present invention, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art already known to those skilled in the art. Summary of the invention

[0011] The purpose of the present invention is to provide a personalized electronic sandbox recommendation system in a virtual construction system based on GraphRAG. The system realizes intelligent analysis of three-dimensional scenes and user behavior data through the collaborative work of multiple functional modules, provides personalized sandbox templates and component recommendations, and supports dynamic updating and interpretation generation.

[0012] In order to achieve the above object, the present invention adopts the following technical solutions:

[0013] The present invention provides an electronic sandbox personalized recommendation system in a virtual construction system based on GraphRAG, comprising:

[0014] Data collection module: used to collect user behavior data, scene perception data and industry knowledge data, and send the collected data to the data processing and analysis module;

[0015] Data processing and analysis module: used to analyze user behavior data and scene identification and comparison; and send the processing and analysis results to the GraphRAG retrieval enhancement module;

[0016] GraphRAG retrieval enhancement module: used for knowledge graph construction, template and component library retrieval, and enhanced retrieval capabilities; and sends the processed data to the personalized recommendation module;

[0017] Personalized recommendation module: used to recommend sandbox templates and sandbox components, and send the recommendation results to the personalized explanation generation module;

[0018] Personalized explanation generation module: Utilizes GraphRAG technology and combines industry knowledge graphs to generate personalized explanations for recommended templates and components, helping users understand the basis for recommendations and enhancing their confidence in decision-making; and sends the generated results to the visualization interaction module;

[0019] Visual interaction module: including template, component interaction and dialogue interaction; and feeding back the interaction content to the dynamic adjustment and update module;

[0020] Dynamic adjustment and update module: used for scene change recognition and user feedback analysis, and feeds back the results of scene change recognition and user feedback analysis to the GraphRAG retrieval enhancement module.

[0021] Furthermore, the data acquisition module specifically includes:

[0022] User behavior data collection submodule: This module is responsible for collecting user behavior data in the virtual construction system, including user operation records, design preferences, basic information and historical design data; by analyzing these data, it provides basic support for subsequent scene recognition and recommendation;

[0023] Scene perception data collection submodule: This module collects data on the three-dimensional scene created by the user in the system, including the terrain, building model, and environmental parameters in the scene; the system identifies the current design scene type by analyzing these data;

[0024] Industry knowledge data collection submodule: collect relevant data from the transportation infrastructure and urban planning industry knowledge base, build an industry knowledge graph, and support subsequent recommendation and explanation generation.

[0025] Furthermore, the data processing and analysis module specifically includes:

[0026] User behavior data analysis: Analyze user behavior data through machine learning algorithms to identify user design preferences and operating habits, and provide support for personalized recommendations;

[0027] Scene recognition and comparison: Use pattern recognition technology to analyze the 3D scene created by the user and compare it with the pre-established scene model library to identify the current design scene type and provide a basis for recommendation decisions.

[0028] Furthermore, the GraphRAG search enhancement module specifically includes:

[0029] Knowledge graph construction: First, build a comprehensive knowledge graph that includes scenarios, templates, components, and industry knowledge. By classifying virtual construction scenarios, define different types of scenarios and associate these scenarios with corresponding sandbox templates and components. At the same time, integrate the professional knowledge of the transportation infrastructure industry, establish semantic relationships between scenarios, templates, components, and industry knowledge, and ensure the structuring and semantic richness of the knowledge graph.

[0030] Template and component library retrieval: This module is based on GraphRAG technology and retrieves sandbox templates and components that match the current scenario from the template and component database. By combining the user's design requirements and scenario characteristics, GraphRAG can quickly locate relevant templates and components in the knowledge graph and provide accurate recommendations to ensure that the recommended content meets the user's design intent and scenario requirements.

[0031] Enhanced retrieval capabilities: Improve the accuracy and relevance of retrieval through semantic enhancement of industry knowledge graphs; GraphRAG uses semantic relationships in industry knowledge graphs to better understand user intentions and needs, ensuring that recommended content is not only highly matched to the user's design scenario, but also meets professional standards and industry best practices, further optimizing the user's design decision-making process.

[0032] Furthermore, the personalized recommendation module specifically includes:

[0033] Sandbox template recommendation: The system analyzes the user's design requirements, behavior data, and current scenario characteristics, and uses GraphRAG technology to retrieve sandbox templates that match the current scenario in the knowledge graph, providing users with a variety of alternatives.

[0034] Sandbox component recommendation: After the user selects a sandbox template, the system further analyzes the characteristics of the template and the user's needs, and recommends components that match it. The component recommendation process is not only based on the association information between components and templates in the knowledge graph, but also combines the user's historical selection and feedback data. The system uses intelligent optimization algorithms to provide the best combination of components to ensure the coordination and functionality of components in the design.

[0035] Furthermore, the visual interaction module specifically includes:

[0036] Template and component interaction: Provides a three-dimensional visual display of sandbox components. Users can select, place and adjust components through the interactive interface to achieve intuitive design operations and real-time feedback.

[0037] Conversational interaction: Supports natural language conversational interaction. Users communicate with the system through voice or text commands to obtain recommendations, modify design plans, or obtain professional explanations.

[0038] Furthermore, the dynamic adjustment and update module specifically includes:

[0039] Scenario change recognition: The system monitors the user's design process in real time, identifies scenario changes, and dynamically adjusts the recommended content to ensure the real-time and accuracy of the recommendations;

[0040] User feedback analysis: Based on user feedback, the recommendation algorithm is adjusted in real time and the recommended content is optimized, so that the system can better adapt to changes in user needs and improve the intelligence and efficiency of the design.

[0041] By adopting the above technical solution, the present invention has the following beneficial effects:

[0042] The present invention provides an intelligent, personalized and dynamic design recommendation method for a virtual construction system, which can significantly improve design efficiency and user experience, and has broad application prospects and commercial value in the fields of transportation infrastructure, urban planning, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 A structural diagram of an electronic sandbox personalized recommendation system in a virtual construction system based on GraphRAG provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0047] Combination Figure 1 As shown, this embodiment provides a virtual construction system based on GraphRAG. The system realizes intelligent analysis of three-dimensional scenes and user behavior data through the collaborative work of multiple functional modules, provides personalized sandbox templates and component recommendations, and supports dynamic update and interpretation generation.

[0048] The specific components of the system include:

[0049] 1. Data acquisition module:

[0050] User behavior data collection submodule: This module is responsible for collecting user behavior data in the virtual construction system, including user operation records, design preferences, basic information (such as industry background, project type, etc.) and historical design data. By analyzing these data, it can provide basic support for subsequent scene identification and recommendation.

[0051] Scene perception data collection submodule: This module collects data on the three-dimensional scene created by the user in the system, including the terrain, building models, environmental parameters, etc. The system identifies the current design scene type by analyzing these data.

[0052] Industry knowledge data collection submodule: collect relevant data from industry knowledge bases such as transportation infrastructure and urban planning, build industry knowledge graphs, and support subsequent recommendation and explanation generation.

[0053] 2. Data processing and analysis module:

[0054] User behavior data analysis: Analyze user behavior data through machine learning algorithms, identify user design preferences and operating habits, and provide support for personalized recommendations.

[0055] In this embodiment, users frequently select and adjust templates and components during the design process. The system needs to analyze the user's operation behavior, identify their preferences, and perform personalized optimization in subsequent recommendations. The specific implementation process is as follows:

[0056] 1) Deep learning algorithm selection: Preferably, this application adopts the CNN-LSTM combined model:

[0057] In order to analyze user behavior data, this application uses a combination model of convolutional neural network (CNN) and long short-term memory network (LSTM). This model can simultaneously capture the spatial characteristics and time series characteristics of user behavior, thereby more accurately identifying user design preferences and operating habits.

[0058] 2) The algorithm flow is as follows:

[0059] 1. Data collection and preprocessing

[0060] (1) Data input: The system collects various data types from users’ operation behaviors, including click behaviors, browsing history, selection frequency, operation sequence, etc. These behavior data will be encoded into time series data and input into the deep learning model.

[0061] (2) Data feature extraction: The system encodes the user's operation behavior and converts each operation (such as clicking, selecting, or adjusting templates and components) into a feature vector to form a user's behavior feature matrix. The specific example is as follows:

[0062] User click frequency: Users select a certain template or component multiple times;

[0063] Operation order: users add components in a fixed order (e.g., select a template first, then a component);

[0064] Time distribution: the user's high-frequency operations within a certain period of time (such as frequently selecting a certain template in the later stage of a project).

[0065] 2.CNN module: spatial feature extraction

[0066] (1) Convolutional layer: First, the user behavior data is subjected to spatial feature extraction through the convolutional layer. The convolution kernel is used to scan the user operation behavior feature matrix to capture the local features of user preferences, such as:

[0067] Users repeatedly choose a certain template in multiple design projects, which means that the template may be a common design element for users;

[0068] If users click on a certain type of component continuously in a short period of time, it indicates that they have a high preference for that type of component.

[0069] (2) Pooling layer: The Max Pooling layer is used to reduce the dimension of features, retain the most significant feature information, and enhance the robustness of features.

[0070] 3.LSTM module: time series feature extraction

[0071] (1) Time series processing: The LSTM layer inputs the spatial features from the CNN to further analyze the temporal patterns of user behavior. For example, when users tend to choose basic templates at the beginning of a project and prefer to add decorative components at the end of the project, LSTM can capture these changes in preference over time series.

[0072] (2) Hidden state update: LSTM continuously updates the hidden state through the memory and forget gate mechanism, recording the user’s behavior habits and their evolution process, so as to better capture the user’s design preferences at different stages.

[0073] 4. Fully connected layer and output layer

[0074] (1) The output of LSTM is processed by the fully connected layer to generate a probability distribution of user preferences. The output layer assigns a recommendation probability to each electronic sandbox template and component to form a personalized recommendation list.

[0075] (2) Cross-Entropy Loss: used to evaluate the prediction effect of the model and optimize the model through gradient descent during training:

[0076]

[0077] Among them, y i Indicates the actual user selection (1 for selection, 0 for not recommended); Represents the recommendation probability of the model.

[0078] 3) Specific examples of personalized recommendations

[0079] 1. Case Background

[0080] Assume that a user uses the system to design a sandbox model of an intelligent transportation hub in a construction project. The user exhibits the following behaviors during the design process:

[0081] Click several times and select the template of the "Traffic Station" category;

[0082] The "high-density flow" component is frequently selected to show the congestion of transportation hubs;

[0083] During the operation, always give priority to the basic template and then add functional components.

[0084] 2. Recommendation process

[0085] (2) Data preprocessing

[0086] The system converts user behavior data (number of clicks, browsing time, selection order, etc.) into behavior feature matrix and time series features.

[0087] Example of behavior feature matrix: Users select the “transportation station” template more frequently (eigenvalue = 0.9), while users select the “commercial district” template less frequently (eigenvalue = 0.3).

[0088] (2) Feature extraction and preference analysis

[0089] CNN extracts user spatial preferences: Through the convolutional layer, the system recognizes that users prefer to choose the "transportation station" template and click on the "high-density pedestrian flow" component multiple times, which shows that users pay more attention to the visualization of pedestrian flow when designing transportation hubs.

[0090] LSTM analyzes user time preference: The LSTM layer finds that users tend to choose basic templates at the beginning of the project, but pay more attention to adding functional components in the later stage of the project, which shows that users' design habit is to build the infrastructure first and then add details.

[0091] 3. Generate recommendation results

[0092] (1) Based on the output of the CNN-LSTM model, the system generates a personalized electronic sandbox template and component recommendation list, including:

[0093] Recommended template: Traffic station template, with a recommendation probability of 85%;

[0094] Recommended components: high-density flow components, with a recommended probability of 90%;

[0095] Recommended order: Select the basic template first, then add functional components.

[0096] (2) The recommended results will be presented in an intuitive manner on the user interface, along with the reasons for the recommendation (e.g., “You have selected the traffic station template multiple times in your last five projects”).

[0097] Scene recognition and comparison: Use pattern recognition technology to analyze the 3D scene created by the user and compare it with the pre-established scene model library to identify the current design scene type (such as bridge construction, road planning, etc.) and provide a basis for recommended decisions.

[0098] 3.GraphRAG search enhancement module:

[0099] Knowledge graph construction: First, build a comprehensive knowledge graph that includes scenarios, templates, components, and industry knowledge. By classifying virtual construction scenarios, define different types of scenarios (such as bridge construction, water transport planning, etc.), and associate these scenarios with corresponding sandbox templates and components. At the same time, integrate professional knowledge of industries such as transportation infrastructure, establish semantic relationships between scenarios, templates, components, and industry knowledge, and ensure the structuring and semantic richness of the knowledge graph.

[0100] In this embodiment, the knowledge graph is constructed in a modular and extensible manner, as follows:

[0101] 1. Modular design:

[0102] 1. Scene module:

[0103] This module is responsible for building the structure of the virtual construction scene and classifying different types of scenes, such as "transportation hub", "commercial area", "industrial park", etc.

[0104] Implementation method: Each scenario is a node in the knowledge graph and has relevant attribute information (such as scenario type, scale, application field, etc.).

[0105] Example: In a transportation hub scenario, the system will build a "transportation hub" node and associate it with the corresponding visualization templates (such as "subway station template", "bus hub template"). This modular structure allows the system to dynamically expand new nodes and relationships based on different scenario types.

[0106] 2. Sandbox template module:

[0107] This module is responsible for building node information related to the visualization large screen template, including template type, layout structure, applicable scenarios, etc. The template module and the scenario module are associated through semantic relationships.

[0108] Implementation method: The template node contains attributes such as template ID, template name, template type, template layout, etc., and establishes an "applicable" relationship with the corresponding scene node.

[0109] For example, when a user selects the "subway station" scenario, the system retrieves the "subway station template" node and associates it with the "transportation hub" scenario node. The extensibility of this association allows new templates to be added to the knowledge graph without affecting the structure of existing templates.

[0110] 3. Sandbox component module:

[0111] This module is responsible for building node information of visualization large screen components, including component type, functional attributes, applicable templates, etc. The component module and the template module are associated through semantic relationships.

[0112] Implementation method: The component node contains attributes such as component ID, component name, component type, function description, etc., and establishes an "include" relationship with the corresponding template node.

[0113] Example: In the "subway station template", the system will construct nodes such as "passenger flow component" and "train arrival information component", and associate them with the template nodes through the "include" relationship. This modular design allows the system to dynamically adjust the graph structure according to the changes of different components.

[0114] 4. Industry knowledge module:

[0115] This module integrates professional knowledge in the field of transportation infrastructure and establishes semantic relationships between scenarios, templates, components and industry knowledge. For example, the "transportation hub" scenario may be associated with knowledge such as industry standards, design guidelines and safety regulations.

[0116] Implementation method: Industry knowledge nodes include industry standards, design specifications, best practices, etc. The system connects these nodes with related scenarios, templates, and components through semantic relationships (such as "conforms to" and "based on").

[0117] For example, on the "Transportation Hub" scene node, the system can associate an "Industry Standard" node, which contains standards such as "Crowd Control" and "Evacuation Guidance", and establishes a "Comply" relationship with related templates and components. This modular structure ensures the dynamic update and expansion of industry knowledge.

[0118] 2. Scalability design form:

[0119] 1. Dynamic addition of new scenes and templates:

[0120] The system supports the dynamic addition of new scenarios and templates through the API interface. For example, when the system needs to add a new "smart parking lot" scenario, it only needs to add a new node and connect it to the corresponding template and component through the "applicable" relationship.

[0121] Expansion method: The addition of new scenarios and templates will not affect the existing graph structure, because the system adopts a modular node design that allows new nodes to be connected in parallel or cross-connected with existing nodes.

[0122] 2. Dynamic update of semantic relations:

[0123] When the system receives new industry standards or design specifications, the industry knowledge module can automatically update the corresponding nodes and relationships. For example, when the "separation of pedestrians and vehicles" industry standard is added, the node can automatically establish a "compliance" relationship with the "transportation hub" scenario and the corresponding templates and components.

[0124] Implementation: Through GrgphRAG technology, the system can automatically adjust and optimize the semantic structure of the graph according to the context of new knowledge.

[0125] 3. Cross-industry knowledge integration:

[0126] The system supports the dynamic integration of cross-industry knowledge, such as expanding the knowledge of smart transportation to the field of smart buildings. This expansion capability relies on the structure of the modular knowledge graph, allowing the system to add new industry knowledge nodes and their associations without changing the basic structure.

[0127] Implementation method: Through modular industry knowledge node design, the system can easily add new domain knowledge nodes and connect them across domains through semantic relationships such as "applicable" or "based on".

[0128] 3. Example: Modular and Scalable Knowledge Graph Construction

[0129] Assume that the system needs to build a knowledge graph for an intelligent transportation project. The specific implementation process is as follows:

[0130] 1. Scene node construction:

[0131] The system first adds the "Intelligent Transportation Hub" scenario node, defines the scenario type as "Transportation Infrastructure", the scale as "Large", and the application area as "Urban Transportation".

[0132] 2. Template node construction:

[0133] In the "Smart Transportation Hub" scenario, the system adds a "Public Transportation Hub Template" node and establishes an "Applicable" relationship with the scenario node.

[0134] The "public transport hub template" node has attributes such as template type "transportation station" and layout structure "multi-level streamlined layout".

[0135] 3. Component node construction:

[0136] The system adds the "passenger flow monitoring component" and "vehicle dispatching component" nodes under the "public transport hub template" and establishes an "inclusion" relationship with the template nodes.

[0137] The component node includes the component type "data visualization", the functional description "real-time crowd flow monitoring and scheduling", and the applicable template is "traffic station template".

[0138] 4.Industry knowledge node integration:

[0139] The system adds the "Industry Standard for Pedestrian and Vehicle Diversion" node to the "Intelligent Transportation Hub" scene node, and associates it with the scene, template, and component through the "Conform to" relationship.

[0140] The "Industry Standard for Separating Pedestrians and Vehicles" node has the standard type "Traffic Design Standard" and the applicable scope "High-density Pedestrian Areas". The system adjusts the design suggestions for scenes and templates based on the standard requirements.

[0141] Through modular and scalable design, the system's knowledge graph can dynamically adapt to different virtual construction scenarios, visualization large screen templates and components, and changes in industry knowledge in the field of transportation infrastructure. This design not only improves the flexibility and scalability of the system, but also ensures the structuring and semantic richness of the knowledge graph.

[0142] Template and component library retrieval: This module is based on GraphRAG technology and retrieves sandbox templates and components that match the current scenario from the template and component database. By combining the user's design requirements and scenario characteristics, GraphRAG can quickly locate relevant templates and components in the knowledge graph and provide accurate recommendations to ensure that the recommended content meets the user's design intent and scenario requirements.

[0143] Enhanced retrieval capabilities: Improve the accuracy and relevance of retrieval through semantic enhancement of industry knowledge graphs. GraphRAG uses semantic relationships in industry knowledge graphs to better understand user intentions and needs, ensuring that recommended content is not only highly matched to the user's design scenario, but also meets professional standards and industry best practices, further optimizing the user's design decision-making process.

[0144] 4. Personalized recommendation module:

[0145] Sandbox template recommendation: The system analyzes the user's design requirements, behavior data, and current scene characteristics, and uses GraphRAG technology to retrieve sandbox templates that match the current scene in the knowledge graph, providing users with a variety of alternatives. These solutions may cover different design styles and component types, and users can choose according to specific project requirements and personal preferences to ensure design diversity and flexibility.

[0146] Sandbox component recommendation: After the user selects a sandbox template, the system further analyzes the characteristics of the template and the user's needs, and recommends matching components. The component recommendation process is not only based on the association information between components and templates in the knowledge graph, but also combines the user's historical selection and feedback data. The system uses an intelligent optimization algorithm to provide the best combination of components to ensure the coordination and functionality of the components in the design. This optimized combination not only improves the overall efficiency of the design, but also ensures the accuracy and professionalism of the design results, allowing users to complete high-quality design solutions in a shorter time.

[0147] In this embodiment, in order to further illustrate how the present application provides the best combination of components through the intelligent optimization algorithm, the following is a detailed description with reference to specific examples:

[0148] Case background: Assume that the user is designing a large-screen visualization sandbox for a city traffic dispatch center. The system needs to recommend the most suitable component combination for the user based on the visualization large-screen template selected by the user. These components include: real-time traffic flow diagram, traffic accident alarm window, dispatch strategy chart, meteorological data display, etc.

[0149] User needs analysis and sandbox template recommendation:

[0150] (1) Requirements Analysis: The system first analyzes the user’s design requirements and current scenario characteristics. For example, in the scenario of a traffic dispatch center, the user’s main concerns are real-time traffic flow, accident alarm information, and dispatch strategy charts.

[0151] (2) Template recommendation: The system uses GraphRAG technology to retrieve a sandbox template (such as "Urban Real-time Traffic Template") that matches "Traffic Dispatch Center" in the knowledge graph and recommends the template to the user. The template has a layout and basic component structure suitable for traffic management scenarios.

[0152] Recommended and intelligent optimization algorithms for electronic sandbox components:

[0153] (1) After the user selects the "Urban Real-time Traffic Template", the system further analyzes the characteristics of the template (such as its emphasis on displaying real-time data and traffic status information) and the user's specific needs, and recommends a matching large-screen visualization component.

[0154] (2) Recommended algorithm selection: Multi-objective optimization algorithm:

[0155] The system uses a multi-objective optimization algorithm to combine the functionality, coordination, user preferences and layout characteristics of components to recommend the best component combination for users.

[0156] Multi-objective optimization algorithm description:

[0157] (1) Objective function: The objective function of the multi-objective optimization algorithm is as follows:

[0158] O total =max(α·P u +β·F c +γ·C t );

[0159] Among them, O total is the total optimization score of the component combination; P u is the user preference, that is, the frequency of components selected by the user in history; F c Functionality score, which evaluates the functional match between the component and the template; C t It is the coordination score between components, that is, the layout compatibility of different components under the same template; α, β, γ are the weights of each goal, which are adjusted according to actual needs.

[0160] Recommended process for electronic sandbox component assembly:

[0161] (1) Step 1: Enter user preferences and template characteristics

[0162] User preference data and selected template features are used as input data. For example, users prefer “real-time traffic flow map” and “accident alarm window”, while the template focuses on displaying traffic status information and dynamic scheduling strategies.

[0163] (2) Step 2: Calculate component functionality score

[0164] The system retrieves the association information between components and templates through the knowledge graph. For example, the functionality score of "real-time traffic flow map" and "city real-time traffic template" is 85%, and the functionality score of "accident alarm window" is 90%.

[0165] (3) Step 3: Evaluate the coordination between components

[0166] The system evaluates the layout compatibility of different components in the same template based on the historical combination data of components in the knowledge graph. For example, the coordination score of "real-time traffic flow map" and "dispatching strategy chart" is 80%, and the coordination score of "traffic accident alarm window" and "weather data display" is 70%.

[0167] Intelligent optimization and selection of the best component combination:

[0168] (1) Calculate the total optimization score: The system uses a multi-objective optimization algorithm to calculate the optimization scores of different component combinations. Assume there are four optional component combinations:

[0169] Combination A: includes “real-time traffic flow diagram” and “dispatching strategy diagram”, with a total score of 0.85;

[0170] Combination B: includes “traffic accident alarm window” and “weather data display”, with a total score of 0.80;

[0171] Combination C: includes “real-time traffic flow map”, “traffic accident alarm window” and “dispatching strategy chart”, with a total score of 0.90;

[0172] Combination D: includes “traffic accident alarm window”, “weather data display” and “dispatching strategy chart”, with a total score of 0.75.

[0173] (2) The system selects the combination with the highest score (such as combination C) and recommends it to the user.

[0174] Visualization of recommendation results and user feedback:

[0175] (1) Visual display: The system displays the recommended component combination on the user interface and provides reasons for the recommendation. For example, “Combination C has the highest score because it contains the main components preferred by the user and has the best functionality and coordination.”

[0176] (2) User feedback and dynamic adjustment:

[0177] If the user adjusts the recommended results (such as replacing components or rearranging the layout), the system will recalculate the optimization score based on the new operation data and update the recommended combination.

[0178] Algorithm optimization and continuous learning:

[0179] (1) Real-time feedback mechanism: The system adjusts the weight parameters of the multi-objective optimization algorithm based on the user’s choices and feedback to better meet the user’s needs. For example, if the user selects “real-time traffic flow map” multiple times, the system will increase its weight in future recommendations.

[0180] (2) Model update: The system uses GraphRAG technology to add users’ new selection records to the knowledge graph, thereby improving the intelligence and personalization of recommendation results.

[0181] Through the above specific embodiments, the system uses GraphRAG technology and multi-objective optimization algorithms to recommend the best component combination that matches the visualization large screen template for users. The optimization algorithm achieves the best component recommendation combination through multi-objective comprehensive calculations such as user preferences, component functionality and coordination. At the same time, the system has real-time feedback and dynamic adjustment functions to ensure the accuracy of the recommendation results and user satisfaction.

[0182] 5. Personalized explanation generation module:

[0183] Using GraphRAG technology and combining it with industry knowledge graphs, we generate personalized explanations for recommended templates and components to help users understand the basis for the recommendations and enhance their confidence in making decisions. By generating models, we transform professional knowledge into easy-to-understand explanation texts, improving users’ understanding and acceptance of recommended content.

[0184] 6. Visual interaction module:

[0185] Template and component interaction: Provides a three-dimensional visual display of sandbox components. Users can select, place and adjust components through the interactive interface to achieve intuitive design operation and real-time feedback.

[0186] Conversational interaction: Supports natural language conversational interaction. Users can communicate with the system through voice or text commands to obtain recommendations, modify design plans, or obtain professional explanations.

[0187] 7. Dynamically adjust and update modules:

[0188] Scenario change recognition: The system monitors the user's design process in real time, identifies scenario changes (such as bridge construction changing to highway construction), and dynamically adjusts the recommended content to ensure the real-time and accuracy of the recommendations.

[0189] In this embodiment, in order to further illustrate how the application dynamically adjusts the recommended content, the following is a detailed description with reference to specific examples:

[0190] Preferably, the present application adds a reinforcement learning algorithm to the dynamic adjustment module.

[0191] Example: 3D scene change recognition and dynamic recommendation adjustment:

[0192] Case background: Assume that the user is designing a series of 3D scenes. The initial design scene is a "transportation construction scene" that includes models of roads, bridges, and traffic lights. The user then switches to a "water conservancy construction scene" that includes models of dams, channels, and hydrological monitoring stations. The system needs to recognize changes in the 3D scene in real time and adjust the recommended sandbox components (visualization large screen components).

[0193] 1. Application of scene perception data acquisition submodule:

[0194] (1) Data collection process:

[0195] The system monitors the user's 3D design scene in real time through the scene perception data collection submodule. The collected information includes:

[0196] Terrain data: such as flat roads, rolling hills, rivers and dams;

[0197] Building models: such as bridges, tunnels, channels and hydrological monitoring stations;

[0198] Environmental parameters: such as traffic flow, rainfall, wind speed, etc.

[0199] (2) Data analysis and scene recognition:

[0200] The system uses this data to analyze the current 3D scene type. For example, when the system identifies the terrain data as rivers and valleys, and detects the building models as dams and hydrological monitoring stations, it marks the scene type as "water conservancy construction scene."

[0201] The identified scene changes will trigger corresponding recommendation strategy adjustments. For example, when the system changes from "transportation construction scene" to "water conservancy construction scene", it will give priority to recommending components related to water conservancy construction (such as channel monitoring charts and dam control models).

[0202] 2. Dynamic adjustment and optimization of reinforcement learning algorithms:

[0203] To ensure the real-time and accuracy of recommendations, the system uses reinforcement learning algorithms to dynamically adjust and optimize after identifying scene changes.

[0204] (1) Reinforcement Learning Model Structure

[0205] Reinforcement Learning Model Selection: Deep Q-Network (DQN):

[0206] DQN is a reinforcement learning algorithm suitable for dynamic scene changes. It estimates the Q value of state and action through a neural network to learn and adjust the optimal action strategy.

[0207] Definition of state, action and reward:

[0208] State (S): the current 3D scene type (such as traffic construction scene or water conservancy construction scene), selected components, and user design preferences.

[0209] Action (A): The component adjustment action recommended by the system in the current state, such as "recommend hydrological monitoring chart" or "replace road monitoring camera".

[0210] Reward (R): The effect of an action is evaluated based on user feedback. When a user accepts a recommended component or adjusts an action, the system gives a positive reward; when a user rejects a recommendation, a negative reward is given.

[0211] Reward function: R = α·U s +β·M c -γ·D t ;

[0212] Among them, U s Indicates the user's satisfaction with the recommended content (such as click rate or selection rate); M c Indicates the matching degree between the recommended component and the current scene; D t represents the recommended delay time; α, β, and γ are the weights of each parameter.

[0213] (2) Training and dynamic adjustment process of reinforcement learning

[0214] Q value update and strategy optimization:

[0215] After each scene change, the system updates the Q value based on the new state and the action taken to optimize the next action. The Q value update formula is:

[0216]

[0217] Among them, Q(s, a) represents the Q value of executing action a under s; η represents the learning rate, which indicates the speed of updating the Q value; R represents the current reward value; γ represents the discount factor, which measures the importance of future rewards; max a′ Q(s′, a′) represents the maximum possible Q value under the new state s′.

[0218] (3) Specific application of reinforcement learning in three-dimensional scene changes

[0219] Scene change recognition and action selection:

[0220] When the system recognizes that the scene switches from "transportation construction" to "water conservancy construction", the DQN model calculates the optimal recommended action based on the new state, such as "recommended hydrological monitoring chart".

[0221] Real-time recommendations and feedback:

[0222] The system displays the recommended components to the user. If the user selects the recommended "Hydrological Monitoring Chart", the system gives a positive reward and updates the Q value; if the user selects other components, the system gives a negative reward and adjusts the strategy.

[0223] Continuous optimization and dynamic adjustment:

[0224] The reinforcement learning algorithm is updated after each scene change and user feedback to ensure real-time optimization of the recommendation strategy. For example, if a user selects the dam control component in the "water conservancy construction scene" multiple times, the system will increase its priority in future recommendations.

[0225] 3. Advantages and effects:

[0226] (1) Real-time response: The reinforcement learning algorithm can adjust its strategy according to the real-time changes of the three-dimensional scene to ensure a high degree of match between the recommended content and the scene.

[0227] (2) Continuous learning: The system optimizes recommendation strategies by continuously learning from user feedback and scenario changes, thereby improving the accuracy of recommended content and user satisfaction.

[0228] This embodiment realizes the real-time recognition of 3D scene changes and dynamic optimization of recommendation strategies through the combination of scene perception data collection submodule and reinforcement learning algorithm. The reinforcement learning algorithm enables the system to quickly adjust the recommended content after the scene switches, improves the real-time and accuracy of the recommendation, and thus enhances the intelligence and practicality of the system in virtual construction.

[0229] User feedback analysis: Based on user feedback, the recommendation algorithm is adjusted in real time and the recommended content is optimized, so that the system can better adapt to changes in user needs and improve the intelligence and efficiency of the design.

[0230] Compared with the prior art, the present invention has the following significant technical advantages and beneficial effects:

[0231] 1. Intelligent scene perception:

[0232] Through comprehensive analysis of three-dimensional scenes and user behavior data, the system realizes intelligent perception and recognition of complex design scenes, and can provide highly matched recommended content based on the characteristics of different scenes, significantly improving the intelligence level of the virtual construction system.

[0233] 2. Personalized recommendation precision:

[0234] By using GraphRAG technology combined with industry knowledge graphs, the system can provide accurate personalized recommendations based on users' individual needs and design preferences, improve design efficiency, and help users quickly find the most suitable sandbox templates and components.

[0235] 3. Dynamic adjustment in real time:

[0236] The system supports real-time analysis and dynamic adjustment of scene changes and feedback during the user's design process, ensuring that the recommended content is always consistent with the user's design needs, improving the response speed and intelligence of the design process.

[0237] 4. Personalized professional interpretation:

[0238] By combining industry knowledge graphs, the system generates professional and personalized explanations for recommended content, helping users understand the basis for recommendations, enhancing their confidence in design, and improving the user experience of the system.

[0239] 5. Visualization and interactive friendliness:

[0240] The system provides an intuitive 3D visual interactive interface and natural language dialogue interaction functions, making user operations more convenient and feedback more timely during the design process, thereby improving the overall user experience and design efficiency.

[0241] The present invention provides an intelligent, personalized and dynamic design recommendation method for a virtual construction system, which can significantly improve design efficiency and user experience, and has broad application prospects and commercial value in the fields of transportation infrastructure, urban planning, etc.

[0242] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A personalized recommendation system for electronic sandbox in a virtual construction system based on GraphRAG, characterized in that: include: Data collection module: used to collect user behavior data, scene perception data and industry knowledge data, and send the collected data to the data processing and analysis module; The data processing and analysis module is used to analyze user behavior data and scene identification and comparison; and send the processing and analysis results to the GraphRAG retrieval enhancement module; The GraphRAG retrieval enhancement module is used for knowledge graph construction, template and component library retrieval, and enhanced retrieval capabilities; and the processed data is sent to the personalized recommendation module; The personalized recommendation module is used to recommend sandbox templates and sandbox components, and sends the recommendation results to the personalized explanation generation module; The personalized explanation generation module: uses GraphRAG technology and combines industry knowledge graphs to generate personalized explanations for recommended templates and components, helping users understand the basis of recommendations and enhancing user decision-making confidence; And send the generated results to the visual interaction module; The visual interaction module includes templates, component interactions and dialogue interactions; And feed back the interactive content to the dynamic adjustment and update module; The dynamic adjustment and update module is used for scene change recognition and user feedback analysis, and feeds back the scene change recognition and user feedback analysis results to the GraphRAG retrieval enhancement module.

2. The electronic sandbox personalized recommendation system in the virtual construction system based on GraphRAG according to claim 1 is characterized in that: The data acquisition module specifically includes: User behavior data collection submodule: This module is responsible for collecting user behavior data in the virtual construction system, including user operation records, design preferences, basic information and historical design data; by analyzing these data, it provides basic support for subsequent scene recognition and recommendation; Scene perception data collection submodule: This module collects data on the three-dimensional scene created by the user in the system, including the terrain, building model, and environmental parameters in the scene; the system identifies the current design scene type by analyzing these data; Industry knowledge data collection submodule: collect relevant data from the transportation infrastructure and urban planning industry knowledge base, build an industry knowledge graph, and support subsequent recommendation and explanation generation.

3. The electronic sandbox personalized recommendation system in the virtual construction system based on GraphRAG according to claim 1 is characterized in that: The data processing and analysis module specifically includes: User behavior data analysis: Analyze user behavior data through machine learning algorithms to identify user design preferences and operating habits, and provide support for personalized recommendations; Scene recognition and comparison: Use pattern recognition technology to analyze the 3D scene created by the user and compare it with the pre-established scene model library to identify the current design scene type and provide a basis for recommendation decisions.

4. The electronic sandbox personalized recommendation system in the virtual construction system based on GraphRAG according to claim 1 is characterized in that: The GraphRAG retrieval enhancement module specifically includes: Knowledge graph construction: First, build a comprehensive knowledge graph that includes scenarios, templates, components, and industry knowledge. By classifying virtual construction scenarios, define different types of scenarios and associate these scenarios with corresponding sandbox templates and components. At the same time, integrate the professional knowledge of the transportation infrastructure industry, establish semantic relationships between scenarios, templates, components, and industry knowledge, and ensure the structuring and semantic richness of the knowledge graph. Template and component library retrieval: This module is based on GraphRAG technology and retrieves sandbox templates and components that match the current scenario from the template and component database. By combining the user's design requirements and scenario characteristics, GraphRAG can quickly locate relevant templates and components in the knowledge graph and provide accurate recommendations to ensure that the recommended content meets the user's design intent and scenario requirements. Enhanced retrieval capabilities: Improve the accuracy and relevance of retrieval through semantic enhancement of industry knowledge graphs; GraphRAG uses semantic relationships in industry knowledge graphs to better understand user intentions and needs, ensuring that recommended content is not only highly matched to the user's design scenario, but also meets professional standards and industry best practices, further optimizing the user's design decision-making process.

5. The electronic sandbox personalized recommendation system in the virtual construction system based on GraphRAG according to claim 1 is characterized in that: The personalized recommendation module specifically includes: Sandbox template recommendation: The system analyzes the user's design requirements, behavior data, and current scenario characteristics, and uses GraphRAG technology to retrieve sandbox templates that match the current scenario in the knowledge graph, providing users with a variety of alternatives. Sandbox component recommendation: After the user selects a sandbox template, the system further analyzes the characteristics of the template and the user's needs, and recommends components that match it. The component recommendation process is not only based on the association information between components and templates in the knowledge graph, but also combines the user's historical selection and feedback data. The system uses intelligent optimization algorithms to provide the best combination of components to ensure the coordination and functionality of components in the design.

6. The electronic sandbox personalized recommendation system in the virtual construction system based on GraphRAG according to claim 1, characterized in that: The visual interaction module specifically includes: Template and component interaction: Provides a three-dimensional visual display of sandbox components. Users can select, place and adjust components through the interactive interface to achieve intuitive design operations and real-time feedback. Conversational interaction: Supports natural language conversational interaction. Users communicate with the system through voice or text commands to obtain recommendations, modify design plans, or obtain professional explanations.

7. The electronic sandbox personalized recommendation system in the virtual construction system based on GraphRAG according to claim 1 is characterized in that: The dynamic adjustment and update module specifically includes: Scenario change recognition: The system monitors the user's design process in real time, identifies scenario changes, and dynamically adjusts the recommended content to ensure the real-time and accuracy of the recommendations; User feedback analysis: Based on user feedback, the recommendation algorithm is adjusted in real time and the recommended content is optimized, so that the system can better adapt to changes in user needs and improve the intelligence and efficiency of the design.

Images