An intelligent project special consultation management system and method based on multi-source data fusion

Abstract

The application discloses an intelligent project special consultation management system and method based on multi-source data fusion, and belongs to the technical field of engineering project management. The method comprises the following steps: accessing multi-source heterogeneous data through a unified data platform, and performing deep fusion based on a semantic alignment algorithm to generate semantic enhanced structured data packets; dynamically scheduling special work flows and tasks based on reinforcement learning; realizing multi-special model coupling simulation through a model federation interface, and performing policy analysis and risk prediction by using AI; constructing a knowledge graph based on a graph neural network to realize intelligent causal reasoning and detection of cross-special conflicts; finally, generating a visual report in a multi-person collaborative three-dimensional immersive environment, and archiving the whole-process knowledge to a graph database to support system self-evolution. The application realizes the whole-process intelligentization from data fusion, intelligent analysis, collaborative management to knowledge sedimentation, and significantly improves the efficiency, quality and collaboration level of special consultation.

Description

Technical Field

[0001] This invention relates to the field of engineering project management technology, and in particular to an intelligent project-specific consulting management system and method based on multi-source data fusion. Background Technology

[0002] In the preliminary planning and approval stages of large-scale engineering projects (such as commercial complexes, industrial parks, and transportation hubs), the preparation and review of multiple specialized consulting reports, including environmental impact assessments, traffic impact assessments, safety assessments, and geological hazard risk assessments, must be completed simultaneously or sequentially. This multi-specialized parallel management process has long faced the following specific and complex technical bottlenecks in practice: First, there are serious problems of "multi-source heterogeneity" and "semantic barriers" at the data level. Various specialized analyses rely on GIS spatial data (topography, water systems, road networks), planning indicator data (plot ratio, building area), dynamic monitoring data, and unstructured policy and regulatory texts. These data come from diverse sources, have different formats, and use different coordinate systems. For example, CAD road network files used by transportation engineers cannot be directly overlaid and analyzed with GIS-formatted water system data required by environmental impact assessment engineers; the textual description of "no less than 100 meters from the water source protection area" in policy provisions cannot be automatically recognized by the system and used to calculate spatial distances to the vector boundaries of project sites. Data preprocessing heavily relies on manual conversion and subjective interpretation, resulting in low efficiency and a high risk of introducing errors, creating isolated "data silos."

[0003] Secondly, the business process suffers from the dilemma of "passive collaboration" and "postponed logical conflicts." Each specialized task is typically carried out separately by different teams using independent software (such as noise prediction software and traffic simulation software), with communication between teams mainly relying on loose methods such as meetings and emails. There are close data dependencies and logical connections between specialized reports; for example, the future traffic flow output from the traffic report is a key input for noise prediction. However, due to the lack of a systematic collaboration platform, this dependency often relies on manual document transfer, making it difficult for downstream data to be updated synchronously once upstream data changes. More importantly, potential contradictions between the conclusions of various specialized tasks (such as conflicts between traffic flow design values ​​and noise control targets, and inconsistencies between building layout and aviation height restrictions) cannot be detected in real time and automatically during the preparation process. They usually only surface during the final internal or external review, leading to significant rework, project delays, and a surge in coordination costs.

[0004] Furthermore, the analysis and decision-making process relies excessively on personal experience, lacking sufficient intelligence. The quality of report preparation heavily depends on the professional skills and experience of engineers. Faced with massive amounts of dynamically updated regulations, manual retrieval and interpretation easily overlook key constraints. The setting of model parameters and the identification of risk points also rely heavily on experience-based judgment, lacking quantitative prediction and early warning support based on historical project big data. This results in inconsistent report quality and makes it difficult to achieve standardized knowledge accumulation and efficient reuse.

[0005] Finally, the report output and knowledge management methods are outdated. Report preparation is a tedious "copy-paste-integrate" process, and charts need to be manually exported and formatted from various software programs, which is prone to errors and inconsistent in format. After the project is completed, valuable intermediate process data, model parameters, review comments, etc., are scattered on personal computers or different systems, failing to form a structured project knowledge asset. This makes it impossible to provide intelligent recommendations and decision support for subsequent similar projects, resulting in the loss of knowledge assets and duplication of work.

[0006] Currently, while independent project management software, GIS platforms, automated reporting tools, and various professional analysis models exist, they all focus on solving problems in a single stage or are merely simple aggregations of tools. For example, the traditional "workflow system + BIM" emphasizes progress and visualization management but lacks deep integration with multi-source data and professional models; some report generation systems can only fill in templates and cannot understand the logical relationships between report content. Existing technologies have not yet been able to build a closed-loop management system that integrates "intelligent fusion of multi-source data, cross-professional model collaborative calculation, real-time detection of logical conflicts, and self-evolution of knowledge," thus failing to systematically solve the core pain points in the entire process of project-specific consulting.

[0007] Therefore, there is an urgent need for an innovative intelligent method and system that can fundamentally break down data barriers, achieve proactive collaboration, enhance analytical intelligence, and accumulate project knowledge, thereby comprehensively improving the efficiency, quality, and reliability of specialized consulting management. Summary of the Invention

[0008] To address the technical problems existing in the prior art, this invention provides a specialized consulting management method based on multi-source data fusion and intelligent collaboration, comprising the following steps: S1. Project Initialization and Intelligent Data Access: Based on the project type entered when creating the project, the necessary list of specialized consulting reports is automatically generated through a pre-set project type-specialized mapping relationship library; through a unified data platform, a multi-source heterogeneous initial dataset consisting of planning indicators, geospatial data, and policy and regulatory texts is retrieved in parallel from the project basic database, geospatial database, and various specialized basic databases according to the project type and geographical location. S2. Multi-source data semantic fusion and structured preprocessing: Under a unified spatiotemporal benchmark, coordinate system unification and spatial overlay analysis are performed to generate a fused spatial dataset. Simultaneously, a semantic alignment algorithm based on named entity recognition and geocoding is used to extract spatial entities and their attributes from policy and regulatory texts. These entities are then matched to GIS vector layers or temporary semantic boundaries are generated through geocoding. The spatial topological relationships between project plots and various semantic boundaries are dynamically calculated to generate a compliance heatmap. The fused spatial dataset, processed structured data, and text index are integrated into a semantically enhanced structured data package. S3. Adaptive Workflow Configuration and Intelligent Task Scheduling: Configure standardized workflows for each item in the special consultation report list; construct a directed acyclic graph containing all special task nodes and dependencies, and introduce an adaptive scheduling engine based on reinforcement learning. The scheduling engine achieves dynamic task allocation, adaptive adjustment of approval levels, and real-time workflow reconstruction in new projects by modeling task nodes as states, defining multi-dimensional reward functions, and training strategy models based on historical project data. S4. Multi-model Coupled Simulation and AI-assisted Analysis: The semantically enhanced structured data package is pushed to the corresponding special project interface; the integrated special project analysis model is called for calculation, and through standardized model interfaces and real-time data pipelines, federated coupled simulation between cross-special project models is realized, so that the output of the upstream model is used as the streaming input of the downstream model, and closed-loop parameter adjustment based on the results is supported; in parallel, through the AI-assisted analysis unit, the pre-trained model is used to intelligently parse the policy provisions to extract key control indicators, and the machine learning model trained based on historical data is used for project risk prediction and early warning. S5. Intelligent Conflict Detection and Causal Reasoning Based on Knowledge Graph: When the preliminary analysis results are received, the graph neural network reasoning engine is started based on the multi-dimensional knowledge graph of "project-environment-regulation-model-personnel" constructed by the graph database. It traverses the related subgraphs to perform causal reasoning, identifies cross-specific logical conflicts and risk transmission paths, and generates a causal analysis report containing the root cause of the conflict, the impact link and mitigation suggestions. S6. Immersive Report Generation and Closed-Loop Knowledge Management: Based on templates, data, charts, and conclusions are automatically extracted from the system to generate draft reports; visual review and real-time annotation are performed in a multi-person collaborative 3D immersive environment built on WebGL technology; after the report is finalized, the final report, full-process data, model parameters, causal analysis graphs, and collaborative trajectories are packaged and archived in the form of associated data to a special report archive. The archive is built based on a knowledge graph to support intelligent tracing, case recommendation, and to provide training data for the AI-assisted analysis unit and graph neural network inference engine, thereby realizing the continuous evolution of the system's intelligence level.

[0009] Furthermore, the semantic alignment algorithm in step S2 specifically includes: using a pre-trained BERT-CRF named entity recognition model to extract "water source protection area", "ecological red line area" and their protection level and boundary description from the policy text; calling a geocoding service to convert the entity name into coordinates or administrative division code, matching it with the existing GIS layer, and automatically generating a temporary vector layer labeled "legal semantic boundary" if no match is found; using the project plot boundary as a benchmark, calculating the intersection area, nearest distance and buffer overlap rate in real time, and visualizing the intensity of spatial compliance conflict in the form of a heat map.

[0010] Furthermore, in step S3, the adaptive scheduling engine based on reinforcement learning models the state of tasks, including task type, risk level, responsible person's skill label, current load, and historical completion quality. The reward function comprehensively considers task completion time, quality score, number of conflict triggers, and resource utilization. It uses proximal policy optimization or deep Q-network algorithm for policy learning. When a task execution deviation or high-risk warning is detected, workflow reconstruction is automatically triggered, including upgrading the task to expert review or downgrading it to fast-track approval.

[0011] Furthermore, the federated coupling simulation in step S4 specifically involves: each specialized analysis model providing a standardized calling interface following the RESTful or gRPC protocol, with inputs and outputs defined using JSON Schema; using a Kafka real-time message queue, the future hourly traffic matrix output by the traffic flow prediction model is pushed as streaming data to the noise diffusion model in real time, triggering its secondary calculation; if the noise calculation result exceeds the regulatory limit, the system automatically triggers the traffic model in reverse, suggesting adjustments to the entrance / exit layout or traffic flow parameters, forming a closed-loop optimization iteration between models.

[0012] Furthermore, the graph neural network causal reasoning in step S5 is specifically as follows: when the rule engine detects the condition "traffic flow is greater than threshold X", the graph neural network reasoning engine takes this as the starting node in the knowledge graph, traverses the nodes connected to it such as "noise prediction value", "greening noise reduction plan", "building layout" and "historical treatment case", calculates the association weights through the graph attention network, infers the key impact path and root cause, and outputs a structured causal chain report.

[0013] Furthermore, the multi-user collaborative 3D immersive environment in step S6 specifically includes: loading the project's 3D terrain, BIM model, noise isosurface, and traffic flow analysis results based on the Cesium.js and Three.js fusion engine; supporting multiple users to perform real-time spatial annotation, distance measurement, and 3D section analysis in the same scene, and synchronously recording all operations to generate a decision trajectory map; providing a virtual meeting room function, where participants enter the scene through virtual avatars to conduct immersive discussions and voting.

[0014] This invention also provides an intelligent project-specific consulting management system based on multi-source data fusion, used to implement the methods described above. The system adopts a microservice architecture and includes the following core modules that interact via API: A unified data platform is used for the access, governance, integration, and unified service publishing of multi-source heterogeneous data; The specialized workflow engine module includes a process model designer, a process execution engine, a progress monitoring and early warning system, and a reinforcement learning-based intelligent scheduler. The multi-source data fusion and intelligent analysis module includes a data fusion unit for semantic alignment and spatial fusion, a model computing unit that supports a model federation interface, and an AI-assisted analysis unit that integrates a regulatory analysis model, a risk prediction model, and a graph neural network inference engine. The collaboration and conflict detection module includes a conflict rule knowledge base, a rule engine, a graph neural network causal reasoning engine, and a causal chain analyzer; The visualization report generation module includes a template management center, a data extraction and rendering engine, and a 3D visualization component that supports multi-person collaborative annotation and virtual review. The Special Report Archive module, built on a knowledge graph database, is used for structured storage of project assets throughout their entire lifecycle and supports intelligent retrieval and reasoning.

[0015] Furthermore, the graph neural network inference engine in the AI-assisted analysis unit operates based on the knowledge graph built from the Neo4j graph database; the model computing unit encapsulates each specialized analysis model in a containerized manner and provides a unified streaming data exchange channel to achieve real-time coupling between models.

[0016] The present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method described above.

[0017] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0018] Compared with existing technologies, the intelligent project-specific consulting management system and method based on multi-source data fusion provided by this invention has achieved the following significant technical progress and positive effects by introducing a series of synergistic technical means: 1. It effectively solves the technical challenge of automatically associating and integrating non-spatial text data with spatial data, improving data preprocessing efficiency and the accuracy of compliance review.

[0019] In existing technologies, spatial constraints described in policy and regulatory texts (such as "100 meters from a water source protection area") require manual interpretation and annotation in a geographic information system, which is inefficient and prone to errors. This invention employs a semantic alignment algorithm based on named entity recognition and geocoding to automatically extract spatial entities and their attributes from policy texts and map them to vector boundaries or semantic layers in GIS. The direct effect of this technology is the automated and structured association between text and spatial data, and the automatic calculation of spatial topological relationships (such as nearest distance and overlap area) between project boundaries and various regulatory control areas. The beneficial technical effects are: firstly, it significantly reduces the workload and subjective errors of manual data preprocessing; secondly, the generated "compliance heatmap" provides intuitive and quantifiable compliance risk warnings in the early stages of a project, providing an accurate, reliable, and clearly regulatory-related data foundation for subsequent specialized analyses, thus preventing systemic design errors caused by data misunderstandings.

[0020] 2. It overcomes the shortcomings of rigid workflow scheduling and inability to dynamically respond to changes in task status in traditional project management, and realizes intelligent dynamic allocation of task resources.

[0021] Traditional project management software relies on static task allocation based on the critical path method, which cannot dynamically adjust according to real-time personnel workload and changes in task risk. This invention introduces a reinforcement learning-based adaptive scheduler into the workflow engine, modeling multi-dimensional factors such as task nodes, executor states, and historical performance into a state space, and training strategies based on historical data. The direct effect of this technique is that the system can dynamically recommend and execute optimal task allocation and approval path planning for new project instances, and automatically trigger process refactoring (such as upgrading review levels) when task delays or increased risks are detected. The beneficial technical effects are: significantly improved overall efficiency and resilience in scheduling complex, multi-disciplinary parallel tasks; avoidance of idle or bottlenecked human resources; and ensuring that high-risk tasks receive more rigorous review, thereby systematically reducing project delays and quality risks caused by unreasonable task allocation or process mismatches.

[0022] 3. It breaks down the barriers of independent operation of various special analysis models and lagging data exchange, and realizes real-time data-driven and collaborative simulation between models.

[0023] In existing technologies, analytical models for different specialties (such as traffic simulation and noise prediction) are usually independent software. Data transfer between models relies on manual export and file import, a cumbersome process that cannot achieve real-time iteration. This invention achieves a key breakthrough in technical performance by designing a standardized interface protocol for the model calculation unit and establishing a real-time data pipeline (such as based on message queues): the calculation output of the upstream model (such as the traffic model) can be used as streaming data to automatically drive the downstream model (such as the noise model) to start a new round of calculations in real time. This not only eliminates manual intermediaries and improves the automation and speed of the analysis process, but more importantly, it enables the analysis of multiple specialties to be based on data snapshots at the same time and unified boundary conditions, ensuring the internal logical consistency and timeliness of the analysis results. This provides a technical possibility for discovering and solving cross-disciplinary dynamic coupling problems (such as the immediate impact of traffic adjustments on noise).

[0024] 4. It has improved the depth and intelligence of cross-disciplinary conflict detection, upgrading it from simple rule matching to causal reasoning based on correlation.

[0025] Existing conflict detection methods largely rely on predefined hard rules for numerical or conditional matching, failing to uncover implicit or indirect logical contradictions and lacking interpretability. This invention constructs a multi-dimensional knowledge graph of "project-environment-regulation-model" and utilizes graph neural networks for reasoning, achieving a more advanced technical effect: when a specific data anomaly is detected (such as excessive traffic flow), the system can automatically traverse the associated entities and relationships in the graph (such as associated noise prediction nodes, existing noise reduction measure nodes, and historical similar case nodes) to deduce potential conflict chains and root causes. The beneficial effect is that the system outputs no longer a simple conflict alert, but a causal report containing impact paths and root cause analysis, greatly assisting professionals in locating the essence of the problem, conducting targeted coordination and modification, and improving the efficiency and depth of cross-disciplinary problem-solving.

[0026] 5. Improved the interactive experience and knowledge management methods for report review, realizing the transformation from static document management to dynamic, structured knowledge asset accumulation.

[0027] Traditional report review relies on two-dimensional documents, which lacks intuitive understanding of spatial relationships, and project knowledge is often stored in unstructured, scattered documents. This invention integrates a WebGL-based multi-user collaborative 3D visualization environment and constructs a specialized report archive using a knowledge graph database, achieving the following technical effects: First, reviewers can interactively review and annotate spatial analysis results in a unified 3D scene, improving the efficiency and accuracy of communication regarding complex spatial information. Second, the system structurally stores process data, model parameters, and decision records from the entire project lifecycle in a knowledge graph as related data. This not only supports efficient source tracing and intelligent recommendation of similar cases, but more importantly, it provides a continuous, high-quality source of training data for AI models (such as risk prediction models and reinforcement learning schedulers) that rely on historical data for learning, forming a closed loop of "application-accumulation-learning-optimization," enabling the system's intelligence level to continuously and autonomously evolve with usage. Attached Figure Description

[0028] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart of an intelligent project-specific consulting management method based on multi-source data fusion provided by an embodiment of the present invention.

[0031] Figure 2 This is a schematic diagram of the structure of an intelligent project-specific consulting management system based on multi-source data fusion provided in an embodiment of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0033] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0034] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" and "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention. Example 1

[0035] like Figure 1 As shown, this invention provides an intelligent project-specific consulting management method based on multi-source data fusion. Specifically, it includes the following steps: S1. Project Initialization and Intelligent Data Access Users create new projects through the system's interactive interface, entering core project attributes via a form. These attributes include project name, project type (building construction, municipal engineering, water system, etc.), project location (specific latitude and longitude coordinates or a site map with clearly defined boundaries), project scale (e.g., land area, building area, investment amount), and key boundary conditions (e.g., planning and design guidelines, list of surrounding sensitive targets). Based on the project type, the system automatically matches and generates a list of required specialized consulting reports for the project using a pre-built project type-specialized mapping database. This list includes at least several reports from environmental impact assessment, traffic impact assessment, safety assessment, soil and water conservation plan, and geological hazard risk assessment. This information is then packaged to generate a structured project file.

[0036] The structured project file initiates data requests to the unified data platform through the system's built-in standard data interface. Based on the project type and geographical location, the unified data platform retrieves planning indicators such as boundary coordinates, plot ratio, and greening rate of the corresponding plot from the project's basic database; it retrieves topographic elevation data, water system distribution vector data, road network vector data, and multi-period satellite remote sensing imagery data within a preset buffer radius centered on the project coordinates from the geospatial database; and it retrieves distribution maps of environmentally sensitive areas (such as water source protection areas and nature reserves), seismic zone parameter zoning maps, acoustic environment functional zoning maps, and currently effective legal and policy texts from various specialized basic databases in parallel. All retrieved data is returned through the standard data interface, accompanied by data source, version, and coordinate system-unified metadata, and is associated with the structured project file using a project identifier, together forming the project's multi-source initial dataset.

[0037] S2, Multi-source data fusion and structured preprocessing The system inputs multiple initial datasets into a unified spatiotemporal reference (GIS platform) for fusion. Within the GIS platform, a coordinate system unification operation is first performed using a coordinate transformation algorithm to convert all data with spatial attributes (such as land parcel boundaries, terrain, roads, distribution of sensitive areas, etc.) to a preset unified plane coordinate system and elevation reference.

[0038] Next, the system employs a semantic alignment algorithm based on geocoding and named entity recognition to map non-spatial text data (such as "primary water source protection area" in policy provisions) to spatial vector data. The specific steps are as follows: First, text semantic parsing is performed. The system calls a finely tuned BERT-CRF named entity recognition model deployed in the natural language processing service to parse the policy text and extract spatial entities (such as "water source protection area" and "ecological red line area") and their attributes (such as protection level and boundary description).

[0039] Next, geocoding and boundary reconstruction are performed. The extracted entity names are converted to standard geographic coordinates or administrative division codes using a geocoding service, and spatial and attribute queries are performed against the existing vector layer library on the GIS platform. If a match is found, the vector layer is directly associated. If no corresponding layer is found, the system executes a descriptive boundary reconstruction sub-process. First, the attribute field "boundary description" of the spatial entity is parsed (e.g., the text description "extending 100 meters outwards from the river centerline on both sides"). Then, buffer analysis and polygon generation operators in the GIS spatial analysis library (such as GEOS) are called to generate the corresponding polygon vector based on key distances in the descriptive text and reference features (whose geometry is obtained from the basic geographic database). Finally, this polygon vector is stored as a temporary layer for "legal semantic boundaries".

[0040] Finally, dynamic calculation of topological relationships and compliance strength assessment are carried out. Based on the boundaries of the project plot, the system calls the spatial relationship calculation function of GIS to calculate the spatial topological relationships with various semantic boundaries in real time, including intersection area, closest distance, and buffer overlap rate. Then, the system accesses the pre-set rule library for judging regulations-spatial conflicts, and sets the conflict level and threshold calculation logic for each type of sensitive area. For example, for the "first-class water source protection area", the rule is set as: "If the intersection area is greater than zero, the conflict strength is directly set to the maximum value of 1.0"; for the "noise-sensitive area", the conflict strength is calculated through a pre-set piecewise function based on the distance and predicted noise value. The system calculates a conflict strength value for the project plot relative to each type of sensitive area according to the calculated topological relationship values and the corresponding rules. The system synthesizes the conflict strength values of all sensitive areas, generates a continuous conflict strength raster surface in the project plot and its surrounding areas through the inverse distance weighted interpolation method, and renders it as a "compliance heat map" to visually display the spatial conflict strength distribution.

[0041] Meanwhile, the system generates a digital elevation model surface for the terrain elevation data through a spatial interpolation algorithm; overlays vector data such as road networks and water systems with the DEM surface to calculate terrain relationship attributes such as road longitudinal slope and river channel slope; conducts spatial intersection analysis on thematic layers such as environmental sensitive areas and seismic zoning with the project plot to identify and mark various sensitive areas and zoning categories involved within the project scope, and finally generates a multi-layer fusion spatial dataset that is completely aligned on the spatio-temporal benchmark and semantically associated.

[0042] Meanwhile, the system performs data cleaning and standardization preprocessing on the non-spatial structured data (such as planning indicators) and unstructured text data (such as policy articles) in the multi-source initial dataset: checks and completes missing values for the structured data, and unifies the units into international standard units; for the text data, performs word segmentation, removes stop words, and standardizes professional terms (such as unifying "concrete" as "concrete"). Finally, the processed fusion spatial dataset, structured data table, and text data index are integrated and output as a unified and standardized semantic-enhanced structured data package for all subsequent special analyses.

[0043] S3. Dynamic Adaptive Workflow Configuration and Intelligent Task Scheduling The special workflow engine reads the overall project progress plan timeline and key nodes plan for filing and construction application in the structured project archive, and configures a standardized four-stage workflow for each report in the special report list in combination with the legal requirements and internal operation standards of each special report: report compilation stage, internal review stage, modification and improvement stage, external submission / review stage.

[0044] The system achieves adaptive scheduling through an intelligent scheduling agent microservice. The intelligent scheduling agent microservice collaborates with the workflow engine via a RESTful API. The specific workflow is as follows: When the workflow engine creates a new task instance, it automatically sends an HTTP POST request to the scheduling request interface of the intelligent scheduling agent service. The request body contains the task's unique identifier, type code, planned duration, and other core metadata.

[0045] Upon receiving a request, the intelligent scheduling agent service first constructs a state vector. Based on the received task type code, the service queries the historical task analysis database to obtain the historical average delay rate and rework rate for similar tasks, and quantifies these into an initial risk level value using pre-defined rules (e.g., 0.2 represents low risk, 0.5 represents medium risk, and 0.8 represents high risk). Next, the service queries the organizational database to obtain a list of engineers whose skills match the task type and whose current status is "idle" or "working." For each engineer in the list, the service calculates their current workload rate in real time (total planned hours of all unfinished tasks under their name / 40 standard hours per week) and queries their historical average quality score for this type of task from the personnel performance database. Finally, the service combines the task attributes (type code, risk level) with each engineer's attributes (engineer ID, ability vector, current workload rate, historical average quality score) to form a set of candidate task-engineer matching state feature vectors.

[0046] Subsequently, the intelligent scheduling agent service performs policy model inference. Internally, the intelligent scheduling agent service loads a pre-trained deep reinforcement learning policy model. After standardizing all the constructed state feature vectors, the intelligent scheduling agent service inputs them in batches into the deep reinforcement learning policy model. The deep reinforcement learning policy model outputs an expected value score for each state vector.

[0047] Next, the intelligent scheduling agent service performs allocation decisions and system integration. The scheduling service selects the engineer ID corresponding to the state vector with the highest expected value score. Then, the intelligent scheduling agent service sends a command to the workflow engine's REST API, specifying in the request body that the current task be assigned to that engineer. Upon receiving this command, the workflow engine updates the executor field of the task instance in the database to the specified engineer, thus completing the formal assignment of the task.

[0048] Finally, the system implements reward feedback and continuous model learning. Upon task completion, the system calculates the comprehensive reward value for this task execution according to a predefined weighted formula, based on the actual completion time, final quality score, and the number of collaborative conflicts triggered during task execution. This reward value, along with the state feature vector recorded at the task's allocation time and the final executor's ID, is stored as a complete training sample in the experience replay database. The system sets up offline, timed training tasks to sample data from the experience replay database for incremental training of the policy model, updating its network parameters and achieving continuous optimization of the scheduling strategy.

[0049] Based on the above-mentioned specialized workflow configuration, the specialized workflow engine constructs a directed acyclic graph containing all task nodes of all specialized projects. The dependencies between nodes include: 1) intra-project stage sequence dependencies; 2) cross-project data dependencies. For example, the preparation of the traffic impact assessment report requires the construction scale from the project's basic data, while the noise prediction in the environmental impact assessment report requires the traffic flow results from the traffic report as an input source. The specialized workflow engine uses the critical path method to extrapolate based on the overall project plan, calculating the earliest start time and latest end time for each task node.

[0050] Next, based on personnel skill tags in the organizational database, current workload, and the requirements of task nodes for responsible roles, the system dynamically allocates each task to the most suitable professional engineer through the aforementioned intelligent scheduling agent microservice. The allocation results, along with the time plan, are visualized on the project collaboration interface in the form of a Gantt chart. The system automatically monitors the actual start and completion times of tasks and compares them with the planned times. When the deviation exceeds a preset threshold, a progress anomaly alert message is triggered to the task responsible person and the project manager.

[0051] S4, Multi-model Coupled Simulation and Intelligent Auxiliary Analysis Each specialist engineer enters their respective work interface, and the system automatically pushes the semantically enhanced structured data package generated in step S2, which is relevant to the current project, to that work interface for the specialist engineer to use. On this work interface, the specialist engineer can access various classic analysis models integrated into the system's model calculation unit for each project.

[0052] The system supports real-time coupled simulation in a federated model architecture, specifically implemented as follows: All models are deployed as containerized microservices and provide a unified gRPC interface. The input and output data structures of each model are strictly defined using ProtocolBuffers files. The system manages data flow between models through a "model orchestration and message bus service" (e.g., based on Apache Kafka). For example, after completing its calculations, the traffic flow prediction model serializes its output "future hourly traffic matrix" into a message of a specific format and publishes it to a Kafka topic. The noise diffusion model service subscribes to this topic; upon receiving a new message, it immediately deserializes the data and uses it as input parameters to initiate a new round of noise calculations, thus achieving real-time cascading triggering of models. Furthermore, the system can be configured with feedback rules: if the noise model's calculation result exceeds a limit, it can automatically send a new calculation request containing adjusted parameters (e.g., reducing peak travel rate assumptions) to the traffic model service, forming a closed-loop optimization iteration.

[0053] For example, in the environmental impact assessment project, a professional engineer clicks to run the noise diffusion model on the interface. The system backend then calls the model's gRPC service, inputting the sound source parameters extracted from structured data packets and real-time traffic flow data obtained from the Kafka bus. After the model service completes the calculation in the computing cluster, it returns and displays the generated dynamic noise distribution map data. For the traffic impact assessment project, the process of calling the traffic flow prediction model is similar; its output is simultaneously stored in the database and published to the message bus. For the geological hazard assessment project, the system calls the GIS-based information method model service, inputting parameters such as topography and lithology, and returns hazard zoning raster data.

[0054] Professional engineers use these modeling tools to perform calculations and analyses, generate preliminary analysis results, and submit them to the system. The preliminary analysis results include analysis charts, data results, and conclusion descriptions.

[0055] In parallel, the system's AI-assisted analysis unit continues to operate. Its natural language processing subunit uses a pre-trained BERT model to semantically understand the latest policy texts obtained from the data platform, automatically identifying, extracting, and structuring key control indicators (such as emission concentration limits, safety distance requirements, and energy-saving design standards), and associating them with corresponding special projects and project parameters. Its machine learning prediction subunit, based on massive amounts of data from a historical project archive, trains special risk prediction models for different project types. After the structured data package of the current project is input into the special risk prediction model, it outputs the potential risk level and early warning prompts for the project across various special project dimensions. These early warning prompts are then linked in real-time to the corresponding special tasks or project overviews, forming proactive risk warnings.

[0056] S5. Intelligent Conflict Detection and Causal Reasoning Based on Graph Neural Networks When the system detects the preliminary analysis results submitted by each project, the collaboration and conflict detection module within the system is activated.

[0057] The collaboration and conflict detection module not only relies on a hard rule base but also introduces a causal reasoning engine based on graph neural networks. First, the system constructs a multi-dimensional relational knowledge graph based on the Neo4j graph database, consisting of "project entities - environmental elements - legal clauses - model parameters - personnel roles". Nodes in the graph have types and attributes, and edges have relation types.

[0058] When a preliminary analysis result (such as "Service level of intersection A is F") is submitted, the data is first transformed into one or more knowledge graph nodes or attribute updates. Subsequently, the system's "causal reasoning service" is triggered. This service performs the following steps: First, it traverses the knowledge graph with a fixed number of hops (e.g., 3 hops) centered on the triggering node (e.g., "Service level of intersection A = F"), extracting a local subgraph. Then, it transforms the node and edge information of this subgraph into a graph data structure and inputs it into a pre-trained graph neural network model (e.g., the GAT attention network). The GAT model, through multiple rounds of message passing and attention aggregation, calculates a correlation score for other nodes in the subgraph that have potential causal relationships with the triggering node (e.g., "Traffic flow at project entrance / exit B", "Sound level at noise prediction point C", "Solution of historical similar case D"). The service sorts the nodes and links based on the scores, selecting the most relevant nodes and connection paths.

[0059] Finally, the system generates a causal analysis report that clearly identifies the root cause of the conflict, the complete transmission path of its impact, and targeted mitigation recommendations, rather than simply issuing a conflict alert.

[0060] The collaboration and conflict detection module includes a built-in extensible logical rule library. These rules define the logical contradictions that may exist between different specialized conclusions. For example: Rule 1: A conflict exists if the predicted peak hour traffic volume in the traffic impact assessment report is greater than X pcu / h and the predicted noise level at sensitive points in the environmental impact assessment report is less than Y dB(A).

[0061] Rule 2: If the building height in the planning and design scheme is greater than H meters, and the conclusion of the aviation obstacle assessment report is that there is no need to install obstruction lights, then there is a conflict.

[0062] Rule 3: If the safety protection distance of the hazard source in the safety assessment report is D1 meters, and the distance to the nearest sensitive building in the site layout plan is D2 meters, then there is a conflict.

[0063] The collaboration and conflict detection module matches the preliminary analysis results with the rule base and uses natural language processing technology to extract key information and compare semantic similarity in the text conclusions. Simultaneously, it initiates graph neural network inference services to perform multi-dimensional causal association analysis. Once a conflict or risk transmission path is detected, the system automatically generates a causal conflict analysis report, clearly identifying the conflicting parties, points of contention, impact links, and mitigation suggestions. The report is pushed in real-time to the responsible persons, project managers, and chief engineers of relevant projects via system messages and emails, prompting the initiation of cross-disciplinary review and coordination. The coordination process may generate new analysis results or modification suggestions, requiring relevant engineers to update their project analysis results in the system. The updated analysis results will be resubmitted, triggering a new round of conflict detection and graph updates until all detected conflicts have been resolved.

[0064] S6, Immersive Visual Report Generation and Intelligent Archiving Once the system detects that all specific tasks have been completed and the collaboration and conflict detection module has not found any unresolved conflicts, it enters the report generation stage. The user (usually the project leader or the report editor) selects a standardized report template that meets the requirements of the target approval department from the system's report template library, or makes custom adjustments based on the standard template.

[0065] The visualization report generation module automatically extracts all relevant data from the system based on the structure of the selected template: basic project information from structured project files; basic data and geographic information maps from structured data packages; analytical charts and data tables from the calculation results of various specialized models; conclusive text from the conclusions submitted by engineers from various disciplines; and relevant key regulatory requirements from AI-assisted analysis units. Through the template engine, the visualization report generation module accurately fills the extracted content into the corresponding chapters and chart positions of the template, automatically generating a complete and formatted draft of a specialized consulting report.

[0066] The system supports a collaborative, immersive 3D review environment for multiple users. Based on a fusion engine of Cesium.js and Three.js, it constructs a WebGL 3D scene, supporting the simultaneous overlay and display of project area terrain, building information models, and various analysis results (such as noise isosurfaces, traffic flow lines, and geological zoning) within the same 3D scene. It supports real-time annotation, distance and area measurement, and arbitrary sectioning analysis by multiple users within the same 3D scene; all operations are synchronously recorded and generate replayable decision trajectory maps. The system can also create virtual meeting rooms, allowing participants to enter the same 3D scene using digital identities (avatars) for immersive discussions and voting on solutions.

[0067] The generated draft report is submitted to the review process. Reviewers can annotate and provide suggestions online within the system, and can also perform spatial annotation in the 3D scene. After the responsible person makes revisions based on the suggestions, the system records the version change history. The final draft report can be exported as a formal document in Word or PDF format with one click for external submission. After the report is submitted, the system packages the final report file, all process data, model parameters, intermediate analysis results, review records, conflict coordination records, causal analysis graphs, 3D collaborative trajectories, etc., and archives them in a structured form in a special report archive, forming project knowledge assets. The system also provides an interface to track the report's circulation status in external approval departments and return approval opinions for archiving, forming a complete closed-loop management from drafting to approval. Finally, the complete knowledge assets formed by this project archive will be used as new training samples, incorporated into the historical project archive on which the AI-assisted analysis unit described in S4 depends, and the graph neural network model and reinforcement learning strategy will be updated, thereby achieving continuous evolution and improvement of the system's intelligence level. Example 2

[0068] This invention also provides an intelligent project-specific consulting management system based on multi-source data fusion. This system implements the method described in Embodiment 1. The system adopts a microservice architecture and specifically includes a core module for data interaction and message passing via a RESTful API. Unified Data Platform The unified data platform serves as the system's data hub, responsible for the aggregation, governance, storage, and service of multi-source heterogeneous data. Internally, it includes a data access gateway, a data storage center, and a data governance and service engine.

[0069] The data access gateway provides a variety of standard data interfaces, including HTTP API, WebSocket, and database connectors, for connecting to external geographic information public service platforms, government department public databases, enterprise internal project management databases, IoT sensor networks, etc., to achieve automatic or semi-automatic collection and access of basic project data, geospatial data, and dynamic policy data.

[0070] The data storage center employs a hybrid storage architecture, using relational databases (such as PostgreSQL / MySQL) to store highly structured project attribute data, task data, and user data; time-series databases (such as InfluxDB) to store dynamically changing market material prices and other time-series data; and distributed file systems (such as HDFS) or object storage (such as MinIO) to store large-scale unstructured or semi-structured data such as remote sensing images, 3D models, and document reports. All data entities are associated through globally unique identifiers (UUIDs).

[0071] The data governance and service engine performs cleaning, transformation, and fusion operations on the incoming data to create high-quality fused data products. Cleaning includes deduplication and error correction; transformation includes coordinate transformation and format standardization; and fusion includes spatial overlay and attribute association. The engine provides on-demand, secure data access services to upper-layer modules through a unified data service API, using standard JSON or GeoServices formats.

[0072] Specialized Workflow Engine Module The specialized workflow engine module is the core driver of business processes, built upon workflow engines such as Activiti and Camunda. Its core components include a process model designer, a process instantiation and execution engine, a progress monitoring and early warning system, a reinforcement learning scheduler, and an adaptive rule engine.

[0073] The process model designer provides a visual interface that allows administrators to drag and drop to configure standardized business process models (BPMNs) for different types of projects and different special projects (environmental impact assessment, traffic impact assessment, safety impact assessment, etc.), defining task nodes, gateways, events, participant roles, form variables, and deliverable templates.

[0074] The process instantiation and execution engine is responsible for instantiating the corresponding main process and various specialized sub-processes according to the project type when a new project is created, driving the process to flow according to the definition, automatically creating and allocating task instances, managing task status (pending, in progress, completed), and enforcing the dependencies and sequence constraints between tasks.

[0075] The progress monitoring and early warning system calculates the planned and actual progress of each task in the process instance in real time, and displays it visually through Gantt charts and dashboards. Built-in monitoring rules automatically trigger early warning messages (internal messages, emails, SMS) to notify relevant stakeholders when a task is delayed, the critical path is blocked, or a specific event occurs.

[0076] The reinforcement learning scheduler trains a deep reinforcement learning model based on historical task data to achieve dynamic task allocation and process optimization.

[0077] The adaptive rule engine supports automatically adjusting the approval path and node permissions based on the project's risk level. For example, it can automatically upgrade the approval level when the risk level is high.

[0078] Multi-source data fusion and intelligent analysis module The multi-source data fusion and intelligent analysis module is the intelligent analysis brain of the system, consisting of three collaborative sub-units: a data fusion unit, a model calculation unit, and an AI-assisted analysis unit. The data fusion unit, centered on an open-source GIS server (such as GeoServer) or commercial GIS components, receives multi-source spatial and attribute data from the data platform. It then calls spatial analysis operator libraries (such as GDAL / OGR library functions) to perform coordinate system transformation, spatial overlay analysis (Intersect, Union), buffer analysis, network analysis, and other operations. This achieves deep fusion and correlation of data on a unified spatiotemporal benchmark, outputting a dataset that integrates spatial and non-spatial attributes, directly usable by various specialized models. The data fusion unit also includes a semantic alignment algorithm submodule to achieve the mapping and fusion of text and spatial data.

[0079] The model computation unit is a containerized environment integrating multiple specialized analysis models. It encapsulates the computational kernel of commercial software (such as noise prediction software and traffic simulation software) and also integrates open-source numerical computation models or self-developed algorithm models. This unit provides a standard model calling interface, receives input parameters, performs calculations in the background, and returns structured result data. Model management supports version control, parameter configuration, and dynamic scheduling of computational resources. Furthermore, the model computation unit supports a model federation interface, providing a streaming data exchange channel and supporting real-time coupled simulation between multiple models.

[0080] The AI-assisted analysis unit is built upon machine learning frameworks (such as TensorFlow and PyTorch) and comprises two main functional models: a regulatory intelligent parsing model and a project risk prediction model. The regulatory intelligent parsing model employs pre-trained large language models (such as BERT and ERNIE) for fine-tuning, specifically designed to identify and extract entities (such as limits, distances, and standard names) and relationships from policy and regulatory texts, and stores them in a structured manner in a knowledge graph. The project risk prediction model uses ensemble learning algorithms (such as random forests and XGBoost), trained using historical project data, and can predict the probability of significant approval issues or technical risks in various specific areas for new projects based on their feature vectors, providing key risk factors. The AI-assisted analysis unit also includes a graph neural network inference engine, supporting causal analysis based on the knowledge graph.

[0081] Collaboration and Conflict Detection Module The collaboration and conflict detection module is key to achieving cross-disciplinary collaboration, and its core is a rule-based reasoning system. It consists of a conflict rule knowledge base, a rule engine, an early warning and collaboration interface, and a graph neural network causal reasoning engine.

[0082] The conflict rule knowledge base stores predefined conflict detection rules in a structured manner. Each rule includes a rule ID, triggering conditions (composed of specific data fields from different specializations using logical operators), conflict description, severity level, and suggested coordinating party. The rule base supports CRUD operations by administrators through a user interface and is scalable.

[0083] The rule engine employs open-source rule engines such as Drools. When each sub-project submits its analysis conclusions, the engine matches the conclusion data against the conditions in the rule base. The matching process not only supports precise numerical comparisons but also performs fuzzy matching of text conclusions by integrating a semantic similarity calculation component.

[0084] When the rules engine detects a conflict, the alert and collaboration interface automatically generates an alert event containing detailed conflict information. This event is published to the system's message bus, and the collaboration interface component is responsible for distributing the alert information to relevant users, and can trigger collaborative tasks such as creating collaborative review meetings and online annotation discussions.

[0085] The graph neural network causal reasoning engine builds a project knowledge graph based on Neo4j, uses the R-GCN model to achieve multi-dimensional correlation reasoning, generates causal chain analysis reports, and provides decision support for conflict resolution.

[0086] Visual report generation module The visualization report generation module is responsible for transforming the analysis process and results into intuitive and standardized deliverables. Its core components include a template management center, a data extraction and rendering engine, and 3D visualization components.

[0087] The template management center manages standard templates for various specialized reports. Templates are defined using markup languages ​​(such as XML and Markdown extensions), and include static text, style definitions, and dynamic data embedding points (placeholders). Placeholders are associated with specific fields or query statements in the system's data model.

[0088] Based on the user-selected template, the data extraction and rendering engine parses placeholders and automatically extracts corresponding text, values, and charts (such as image URLs and vector graphics data) from the database, file system, and model calculation results through predefined query paths. Then, it uses document rendering libraries (such as Apache POI for Word and ReportLab for PDF) or web rendering technologies to populate the data into the template and generate a formatted report document.

[0089] The 3D visualization component is developed based on WebGL technologies (such as Cesium.js and Three.js), providing a 3D scene that can run in a browser without plugins. This component can load 3D terrain and building BIM models (if applicable) of the project area, and display analysis results such as noise isosurfaces, traffic flow heatmaps, and geological hazard zoning as overlaid layers, particle effects, and dynamic streamlines within the 3D scene. It supports interactive rotation, scaling, and sectioning analysis. The 3D visualization component also supports multi-user collaborative annotation and virtual review, providing an immersive decision support environment. Furthermore, report generation supports causal chain graph embedding, enhancing the logical interpretability of the report.

[0090] Special Report Archive Module The Special Report Archive module serves as the system's knowledge repository, built upon a non-relational database (such as MongoDB) or a knowledge graph database (such as Neo4j). It not only stores the final report documents but also stores process assets throughout the entire project lifecycle in the form of linked data. Through graph neural networks, it continuously learns and updates these relationships, supporting intelligent reasoning and case recommendation. These lifecycle process assets include: raw input data, model runtime parameters and logs, intermediate analysis results, draft reports from version iterations, internal and external review comments, approval process status, and approval documents. By establishing multi-dimensional relationships between projects, data, models, reports, and personnel, this archive supports complex source tracing queries, intelligent recommendation of similar projects, reuse of historical case experience, and provides a data foundation for continuous training of AI-assisted analysis units.

[0091] The specific implementation methods of the above-mentioned modules correspond to the specific implementation methods of the aforementioned method embodiments, and together they constitute a complete intelligent project-specific consulting management system that integrates data, performs intelligent analysis, collaborative management and knowledge accumulation. Example 3

[0092] The present invention also provides an electronic device, including: a processor, a transmitting device, an input device, an output device, and a memory. The processor may be implemented using a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit, or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory may be implemented using a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM), and is used to store computer program code. The computer program code includes computer instructions. When the processor executes the computer instructions, the electronic device executes a method as described in any of the above possible implementation methods. Example 4

[0093] The present invention also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor of an electronic device, cause the processor to perform a method as described in any of the above possible implementations.

[0094] In the description of this specification, the references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0095] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A specialized consulting management method based on multi-source data fusion and intelligent collaboration, characterized in that, Includes the following steps: S1. Project Initialization and Intelligent Data Access: Based on the project type entered when creating the project, the necessary list of specialized consulting reports is automatically generated through a pre-set project type-specialized mapping relationship library; through a unified data platform, a multi-source heterogeneous initial dataset consisting of planning indicators, geospatial data, and policy and regulatory texts is retrieved in parallel from the project basic database, geospatial database, and various specialized basic databases according to the project type and geographical location. S2. Multi-source data semantic fusion and structured preprocessing: Under a unified spatiotemporal benchmark, coordinate system unification and spatial overlay analysis are performed to generate a fused spatial dataset. Simultaneously, a semantic alignment algorithm based on named entity recognition and geocoding is used to extract spatial entities and their attributes from policy and regulatory texts. These entities are then matched to GIS vector layers or temporary semantic boundaries are generated through geocoding. The spatial topological relationships between project plots and various semantic boundaries are dynamically calculated to generate a compliance heatmap. The fused spatial dataset, processed structured data, and text index are integrated into a semantically enhanced structured data package. S3. Adaptive Workflow Configuration and Intelligent Task Scheduling: Configure standardized workflows for each item in the special consultation report list; construct a directed acyclic graph containing all special task nodes and dependencies, and introduce an adaptive scheduling engine based on reinforcement learning. The scheduling engine achieves dynamic task allocation, adaptive adjustment of approval levels, and real-time workflow reconstruction in new projects by modeling task nodes as states, defining multi-dimensional reward functions, and training strategy models based on historical project data. S4. Multi-model Coupled Simulation and AI-Assisted Analysis: The semantically enhanced structured data package is pushed to the corresponding special project interface; the integrated special project analysis model is called for calculation, and through standardized model interfaces and real-time data pipelines, federated coupled simulation between cross-special project models is realized, so that the output of the upstream model is used as the streaming input of the downstream model, and closed-loop parameter adjustment based on the results is supported; in parallel, through the AI-assisted analysis unit, the pre-trained model is used to intelligently parse the policy provisions to extract key control indicators, and the machine learning model trained based on historical data is used to predict and warn of project risks; S5. Intelligent Conflict Detection and Causal Reasoning Based on Knowledge Graph: When the preliminary analysis results are received, the graph neural network reasoning engine is started based on the multi-dimensional knowledge graph of "project-environment-regulation-model-personnel" constructed by the graph database. It traverses the related subgraphs to perform causal reasoning, identifies cross-specific logical conflicts and risk transmission paths, and generates a causal analysis report containing the root cause of the conflict, the impact link and mitigation suggestions. S6. Immersive Report Generation and Closed-Loop Knowledge Management: Based on templates, data, charts, and conclusions are automatically extracted from the system to generate draft reports; visual review and real-time annotation are performed in a multi-person collaborative 3D immersive environment built on WebGL technology; after the report is finalized, the final report, full-process data, model parameters, causal analysis graphs, and collaborative trajectories are packaged and archived in the form of associated data to a special report archive. The archive is built based on a knowledge graph to support intelligent tracing, case recommendation, and to provide training data for the AI-assisted analysis unit and graph neural network inference engine, thereby realizing the continuous evolution of the system's intelligence level.

2. The method according to claim 1, characterized in that, The semantic alignment algorithm in step S2 specifically includes: using a pre-trained BERT-CRF named entity recognition model to extract "water source protection area", "ecological red line area" and their protection level and boundary description from policy text; calling geocoding services to convert entity names into coordinates or administrative division codes, matching them with existing GIS layers, and automatically generating a temporary vector layer labeled "legal semantic boundary" if no match is found; using the project plot boundary as a benchmark, calculating the intersection area, nearest distance and buffer overlap rate in real time, and visualizing the intensity of spatial compliance conflict in the form of a heat map.

3. The method according to claim 1, characterized in that, In step S3, the adaptive scheduling engine based on reinforcement learning models the state of tasks, including task type, risk level, responsible person's skill label, current load, and historical completion quality. The reward function comprehensively considers task completion time, quality score, number of conflict triggers, and resource utilization. It uses near-end policy optimization or deep Q-network algorithm for policy learning. When a task execution deviation or high-risk warning is detected, the workflow is automatically restructured, including upgrading the task to expert review or downgrading it to fast-track approval.

4. The method according to claim 1, characterized in that, The federated coupling simulation in step S4 is as follows: each specialized analysis model provides a standardized calling interface that follows the RESTful or gRPC protocol, and the input and output are defined using JSONSchema; through the Kafka real-time message queue, the future hourly traffic matrix output by the traffic flow prediction model is pushed as streaming data to the noise diffusion model in real time, triggering its secondary calculation; If the noise calculation results exceed the regulatory limits, the system will automatically trigger the traffic model in reverse, suggesting adjustments to the entrance / exit layout or traffic flow parameters to form a closed-loop optimization iteration between models.

5. The method according to claim 1, characterized in that, The graph neural network causal reasoning in step S5 is as follows: When the rule engine detects the condition "traffic flow is greater than threshold X", the graph neural network reasoning engine takes this as the starting node in the knowledge graph, traverses the nodes connected to it such as "noise prediction value", "greening and noise reduction plan", "building layout" and "historical treatment cases", calculates the association weights through the graph attention network, infers the key impact path and root cause, and outputs a structured causal chain report.

6. The method according to claim 1, characterized in that, The multi-user collaborative 3D immersive environment in step S6 specifically includes: loading the project's 3D terrain, BIM model, noise isosurface, and traffic flow analysis results based on the Cesium.js and Three.js fusion engine; supporting multiple users to perform real-time spatial annotation, distance measurement, and 3D section analysis in the same scene, and synchronously recording all operations to generate a decision trajectory map; providing a virtual meeting room function, where participants enter the scene through virtual avatars to conduct immersive discussions and voting.

7. An intelligent project-specific consulting management system based on multi-source data fusion, used to implement the method of any one of claims 1-6, characterized in that, The system adopts a microservice architecture and includes the following core modules that interact via API: A unified data platform is used for the access, governance, integration, and unified service publishing of multi-source heterogeneous data; The specialized workflow engine module includes a process model designer, a process execution engine, a progress monitoring and early warning system, and a reinforcement learning-based intelligent scheduler. The multi-source data fusion and intelligent analysis module includes a data fusion unit for semantic alignment and spatial fusion, a model computing unit that supports a model federation interface, and an AI-assisted analysis unit that integrates a regulatory analysis model, a risk prediction model, and a graph neural network inference engine. The collaboration and conflict detection module includes a conflict rule knowledge base, a rule engine, a graph neural network causal reasoning engine, and a causal chain analyzer; The visualization report generation module includes a template management center, a data extraction and rendering engine, and a 3D visualization component that supports multi-person collaborative annotation and virtual review. The Special Report Archive module, built on a knowledge graph database, is used for structured storage of project assets throughout their entire lifecycle and supports intelligent retrieval and reasoning.

8. The system according to claim 7, characterized in that, The graph neural network inference engine in the AI-assisted analysis unit operates based on the knowledge graph built from the Neo4j graph database; the model computing unit encapsulates various specialized analysis models in a containerized manner and provides a unified streaming data exchange channel to achieve real-time coupling between models.

9. An electronic device comprising a processor, a memory, and a computer program stored in the memory, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1-6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-6.

Images