Method and system for intelligent generation and dynamic optimization of whole-process consulting feasibility study report based on large model

Abstract

This invention discloses an intelligent generation and dynamic optimization method and system for full-process consulting feasibility study reports based on a large model. It constructs an integrated intelligent system whose core architecture consists of a multi-source data real-time fusion module, a professional computing module, an intelligent report generation module, and a dynamic optimization and sensitivity analysis module. These four modules work collaboratively to form a closed-loop workflow from data input to report generation and then to dynamic optimization. This not only solves the problems of low data integration efficiency, difficulty in cross-disciplinary collaboration, and subjective sensitivity analysis in existing technologies, but also upgrades traditional static feasibility study reports into interactive and simulated dynamic decision-making tools, comprehensively improving the generation efficiency, data accuracy, and decision-making reference value of full-process consulting feasibility study reports.

Description

Technical Field

[0001] This invention belongs to the field of engineering management information technology, specifically referring to a method and system for intelligent generation and dynamic optimization of feasibility study reports for the entire consulting process based on a large model. Background Technology

[0002] In the initial stages of project investment decision-making, the feasibility study report (hereinafter referred to as "feasibility report") is a crucial basis for assessing project feasibility and making scientific decisions. However, the traditional preparation process has the following three major difficulties: (I) Low efficiency and lack of timeliness in data integration Preparing a high-quality feasibility study report requires the manual collection, organization, and analysis of data from multiple dimensions, including policies and regulations, macroeconomic markets, building material prices, land planning, and geological surveys. This process is not only time-consuming and labor-intensive, but also highly susceptible to deviations from the actual situation due to outdated data (e.g., the release of new national environmental standards, adjustments to local land transfer policies, or drastic fluctuations in key building material price indices). This can lead to misleading investment decisions.

[0003] (II) Barriers to cross-disciplinary collaboration and information silos Feasibility study reports involve multiple professional fields such as economics, technology, finance, and environment, and the data and analytical logic of each field are often disconnected. For example, there is a lack of effective dynamic linkage between the technical parameters provided by engineering technicians, the environmental constraints of environmental assessment experts, and the financial models of economic analysts. This information silo effect leads to a large amount of cross-departmental communication and manual proofreading, which not only prolongs the preparation cycle but may also cause errors due to information distortion. (III) Subjectivity and Limitations of Sensitivity Analysis Sensitivity analysis is a crucial step in assessing a project's resilience. It aims to measure the impact of changes in key variables (such as construction investment, operating costs, and construction period) on the project's financial indicators (such as internal rate of return (IRR) and payback period). Traditional methods heavily rely on expert experience to define the range of changes and scenarios. The analysis process is relatively static and subjective, making it difficult to comprehensively and dynamically simulate complex risk scenarios with multiple overlapping factors in the real world. Furthermore, it lacks support from risk response strategies based on historical data.

[0004] Among the existing technologies, the closest to the technical solution of this invention are patents CN202510153221.0 (A method for intelligent analysis, research and automatic generation of feasibility study reports for projects), CN120765387A (A method and system for generating enterprise due diligence reports driven by large models) and CN120258875A (A feasibility analysis system for scientific and technological achievements). However, the above-mentioned existing technologies still have the following significant shortcomings and are fundamentally different from this invention: 1. CN202510153221.0 lacks a multimodal data deep analysis mechanism designed for engineering scenarios, only processing conventional text and tabular data. It does not integrate semantic parsing and structured extraction technologies for unstructured engineering data such as geological survey drawings. Furthermore, data synchronization relies on periodic database updates rather than real-time connection to official data sources, resulting in discrepancies between the reported data and the actual geological conditions and latest policy requirements of the project, affecting the accuracy of investment estimates and economic evaluations. It also lacks a professional error correction mechanism, resulting in insufficient accuracy in the calculation of core financial indicators, and lacks multi-factor coupled dynamic simulation functions.

[0005] 2. CN120765387A focuses on enterprise due diligence scenarios, with its core function being the processing of enterprise-related data such as business registration, finance, legal matters, and public opinion. It emphasizes enterprise risk identification and correlation analysis, but does not address core needs specific to the engineering field, such as geological survey data analysis, dynamic synchronization of engineering building material prices, and engineering risk response. Its multimodal processing method is designed for enterprise document design and cannot be adapted to special data such as engineering drawings. Its dynamic update mechanism only addresses changes in enterprise business registration and legal matters, which is not compatible with the dynamic optimization needs of engineering scenarios.

[0006] 3. CN120258875A focuses on the feasibility analysis of scientific and technological achievements. Its core is to evaluate the technical, economic, market, and operational feasibility of scientific and technological achievements. Its data entry module only realizes simple text and image sample entry and consistency analysis. It does not design a dedicated analysis process for engineering geological survey drawings, nor does it involve functions such as intelligent generation of feasibility study reports, multi-factor coupled dynamic simulation, and matching of engineering-specific risk response strategies. It is significantly different from the core requirements of the present invention for the generation and optimization of feasibility study reports for the whole process of engineering consultation.

[0007] 4. In addition, existing technologies that apply large models to the generation of feasibility study reports generally suffer from problems such as insufficient adaptability to engineering fields, low accuracy of core indicator calculations, single dynamic simulation, and lack of engineering-specific risk response strategies. Furthermore, they have not formed a closed-loop system of coordinated linkage among various modules, which cannot meet the professional and precise needs of consulting throughout the entire engineering process.

[0008] Therefore, there is an urgent need for an intelligent generation and dynamic optimization method and system that can adapt to engineering scenarios to address the shortcomings of the existing technologies. Summary of the Invention

[0009] The main objective of this invention is to provide a method and system for intelligent generation and dynamic optimization of feasibility study reports based on a large model throughout the entire consulting process. This system achieves a closed-loop workflow from data input to report generation and dynamic optimization. Through a series of innovative designs specific to the engineering field, it addresses the shortcomings of existing technologies and enhances the professionalism, accuracy, and practicality of feasibility study reports. It primarily solves the following problems: (1) Solve the problems of incomplete, non-real-time, and poor engineering adaptability of existing multi-source data fusion, realize real-time synchronization of multi-dimensional engineering-related data such as policy, market, and geology, and achieve in-depth analysis and structured extraction of unstructured engineering data (geological survey drawings, etc.); (2) Solve the problems of lack of standardized algorithm support, insufficient accuracy and poor adaptability to engineering scenarios in the calculation of existing technical professional indicators, and ensure the accurate calculation of core indicators such as internal rate of return and dynamic investment payback period to adapt to the needs of different types of engineering scenarios; (3) Solve the problem of static and singular sensitivity analysis of existing technologies, realize real-time simulation and risk assessment of multi-factor coupling in engineering scenarios, and improve the comprehensiveness and objectivity of risk analysis; (4) To solve the problems of data fragmentation and poor module collaboration in existing technologies, establish a dynamic linkage mechanism for data in fields such as technology, finance, and environment, and improve the efficiency of cross-disciplinary collaboration. (5) To address the problem that existing technology risk response strategies lack engineering specificity and have low matching accuracy, customized and implementable engineering risk response solutions are provided through engineering domain knowledge graphs and improved matching algorithms.

[0010] To achieve the above objectives, one solution of the present invention is: A method for intelligent generation and dynamic optimization of feasibility study reports for the entire consulting process based on a large model, comprising: Step 1: The multi-source data real-time fusion module connects to the data source and performs real-time synchronization. It uses a multimodal large language model based on the Transformer architecture, which is dedicated to the engineering field, to perform semantic parsing and structured extraction on the data source to obtain cross-modal data. The multimodal large language model is pre-trained and optimized using corpora from the fields of engineering geology and engineering economics, and a dedicated parsing process is designed for geological exploration drawings. Step 2: Calculate the internal rate of return using a specialized algorithm module. and dynamic investment payback period The calculation results are then formatted for use by subsequent modules. Step 3: The intelligent report generation module integrates and refines the cross-modal data from Step 1 with the calculation results from Step 2 to form a visualization result; the visualization result is an engineering-specific chart and supports cross-disciplinary data linkage updates; Step 4: The dynamic optimization and sensitivity analysis module, combined with the multi-source data real-time fusion module, professional calculation module, and intelligent report generation module, performs real-time scenario simulation and visualization analysis of multi-factor coupling. Based on the engineering domain knowledge graph and historical project database, it matches risk response strategies through an improved cosine similarity algorithm; the improved cosine similarity algorithm introduces engineering weight factors.

[0011] In step one, the data sources accessed include a policy and regulation database, a market database, and a project basic database. The policy and regulation database is equipped with customized web crawlers and API interfaces specifically for the engineering field, and connects in real time to official information release systems for engineering construction. The market database is designed with a data synchronization mechanism for engineering building materials and labor costs. The project basic database supports the uploading and parsing of special format files for engineering projects.

[0012] Preferably, the policy and regulation database contains relevant standards and specifications. Through pre-set frequency scheduled tasks and keyword subscriptions, it automatically captures text information related to the project and establishes an AI knowledge base that is synchronized in real time with the information release system, with a built-in engineering policy classification tag system. The market database connects to building material price information platforms, bulk commodity trading data centers, and financial data service providers through API interfaces. It uses a sliding window caching mechanism combined with the timeliness requirements of engineering data for data synchronization, and sets an early warning threshold for abnormal fluctuations in engineering and building material prices and establishes cross-platform consistency verification rules. The project basic database supports users to upload core basic information about the project and automatically verifies it to avoid duplicate uploads, while storing the information in batches and categories.

[0013] Step one employs a multimodal large language model based on the Transformer architecture. It is pre-trained and optimized using corpora from the fields of engineering geology and engineering economics, encoding and mapping data from different modalities into a unified engineering semantic representation space. The specific processing flow for geological survey drawings is as follows: (1) Image recognition and segmentation First, the improved U-Net semantic segmentation model based on convolutional neural networks is used to preprocess the drawings. The model parameters are optimized according to the features of the geological exploration drawings, and different regions are segmented. (2) Extraction of key information 1) Text information: For legend areas, table areas and annotation text, a deep learning OCR model specifically for the engineering field is used for text recognition. The recognition algorithm is optimized for engineering terms and geological parameters, and the text information is extracted. 2) Graphical information: For borehole columnar sections and geological profiles, an end-to-end computer vision model specifically designed for the engineering field is used, combined with target detection and image classification, to convert the graphics into structured geological parameter data; 3) Construct a knowledge graph in the field of engineering geology, which includes core entities and their relationships in terms of stratigraphic types, soil and rock properties, and engineering specifications. During the information extraction process, the identified entities are linked to nodes in the knowledge graph, and the entity relationships defined in the graph are used for verification and semantic completion. At the same time, engineering specifications constraints are introduced to ensure that the extracted data complies with industry standards.

[0014] In step two It is the discount rate when the cumulative net present value is 0. This refers to the time it takes for the cumulative net present value to turn from negative to positive. The relevant calculation formulas for both are as follows: ; ; in, Represents the cumulative net present value; Indicates the discount rate; Indicates the total number of years; Indicates the year of the calculation period, i.e. ; Indicates the first Annual cash inflow; Indicates the first Annual cash outflow.

[0015] In step four, when performing real-time scenario simulation and visualization analysis involving multiple factors, the dynamic optimization and sensitivity analysis module uses a human-computer interaction interface to allow users to input "hypothetical scenarios" via natural language and then performs the following workflow: 1) A multimodal large language model for engineering fields receives and understands the natural language input by users, identifies the sensitive factors that need to be adjusted and their magnitude of change, and performs factor correlation analysis in combination with the characteristics of engineering scenarios to obtain the change parameters. 2) The changing parameters are transmitted to the professional computing module in real time, and at the same time, the multi-source data real-time fusion module is triggered to update the relevant related data; 3) The professional calculation module recalculates and updates the key indicators related to the project based on the changed parameters and the updated related data, and outputs an indicator change analysis report; 4) Conduct several simulations involving multiple factors to simulate the project's performance under different combinations of factors, and present the simulation results in visualization charts specific to the engineering field.

[0016] In step four, when performing risk response strategies based on similar projects, the dynamic optimization and sensitivity analysis module constructs and maintains a dedicated historical project database for the engineering field in real time. This database contains basic information about historical projects, engineering parameters, various engineering risk events encountered during construction and operation, response strategies adopted at the time, and final implementation results. When a high-risk factor is identified in a project, the large model will retrieve the most similar historical project from the historical project database using an improved cosine similarity algorithm, extract the successful response strategies used to handle the high-risk factor, and adapt and adjust them in conjunction with the specific engineering parameters of the current project to form a customized risk response plan. Subsequently, the large model will visualize the corresponding historical project, the adapted response strategy, and implementation suggestions.

[0017] Preferably, the improved cosine similarity algorithm introduces an engineering weight factor, which includes a geological condition weight of 0.3, a construction scale weight of 0.25, an industry type weight of 0.25, and a sensitive factor type weight of 0.2.

[0018] Preferably, a cosine similarity algorithm is used for project matching. The matching dimensions include industry type, construction scale, geological conditions, and sensitive factor type. The similarity threshold is set to 0.8. When a project with a similarity of ≥0.8 is retrieved, its risk response strategy is automatically extracted and adapted.

[0019] The second solution of the present invention is: An intelligent system includes a multi-source data real-time fusion module, a professional computing module, a report intelligent generation module, and a dynamic optimization and sensitivity analysis module, used to execute the intelligent generation and dynamic optimization method for the whole-process consulting feasibility study report based on a large model. The multi-source data real-time fusion module, the professional computing module, and the report intelligent generation module respectively execute steps one to three, and the four modules work together through an engineering data linkage bus to perform step four.

[0020] By adopting the above technical solution, this invention constructs an integrated intelligent system. Its core architecture consists of a multi-source data real-time fusion module, a professional computing module, a report intelligent generation module, and a dynamic optimization and sensitivity analysis module. These four modules work collaboratively through an engineering data linkage bus, forming a closed-loop workflow from data input to report generation and then to dynamic optimization. Compared with existing technologies, this invention has the following significant advantages: 1. Strong Adaptability to Engineering Fields: It innovatively adopts a multimodal large language model specifically for engineering fields, designs a dedicated parsing process for geological survey drawings, and combines engineering geology knowledge graphs to improve the professionalism and accuracy of data extraction. This solves the problems of existing multimodal processing methods being unable to adapt to engineering scenarios and low parsing accuracy of unstructured engineering data. The data source access and processing are designed for engineering scenarios, comprehensively covering all types of data required for the preparation of engineering feasibility study reports, which is different from the technical solutions for other scenarios such as enterprise due diligence and scientific and technological achievement analysis.

[0021] 2. Accurate and reliable calculation of core indicators: The professional algorithm module has a built-in engineering-specific error correction mechanism and multi-scenario adaptation algorithm, which automatically adjusts the calculation parameters for different engineering types to ensure the calculation accuracy of core financial indicators such as IRR and PBP. This solves the problems of insufficient accuracy and poor engineering adaptability of existing general calculation methods, and provides reliable data support for investment decisions.

[0022] 3. More comprehensive dynamic simulation and risk analysis: It innovatively realizes real-time scenario simulation of multi-factor coupling in engineering scenarios, breaking through the limitations of traditional single-factor simulation. Combined with engineering-specific visualization charts to present simulation results, it improves the comprehensiveness and intuitiveness of risk analysis. The improved cosine similarity algorithm introduces engineering weight factors and combines engineering domain knowledge graphs and historical project databases to provide customized and implementable engineering risk response strategies, solving the problems of lack of specificity and low matching accuracy of existing technology risk response strategies.

[0023] 4. High efficiency of cross-disciplinary collaboration: Real-time synchronization and interaction of data from various modules are achieved through the engineering data linkage bus, establishing a dynamic linkage mechanism for data in the fields of technology, finance, environment, etc., solving the problems of fragmented cross-disciplinary data and poor collaboration in existing technologies, significantly shortening the feasibility study report preparation cycle, reducing the workload of manual proofreading, and avoiding information transmission distortion.

[0024] 5. Highly practical and easy to implement: It upgrades the traditional static feasibility study report into an interactive and simulated dynamic decision-making tool, which is adapted to the professional needs of the whole process of engineering consulting. The generated feasibility study report has the characteristics of standardization, professionalism and accuracy. The risk response strategy is combined with engineering practice experience and can be directly implemented. It comprehensively improves the generation efficiency, data accuracy and decision reference value of the whole process consulting feasibility study report. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating a specific embodiment of the present invention. Detailed Implementation

[0026] To further explain the technical solution of the present invention, the present invention will be described in detail below through specific embodiments.

[0027] To address the problems of existing technologies, this invention constructs an integrated intelligent system. Its core architecture consists of four main modules: a multi-source data real-time fusion module, a professional computing module, a report intelligent generation module, and a dynamic optimization and sensitivity analysis module. These four modules work collaboratively to form a closed-loop workflow from data input to report generation and then to dynamic optimization. Specifically, refer to... Figure 1 As shown, this invention discloses an intelligent generation and dynamic optimization method for full-process consulting feasibility study reports based on a large model, including: Step 1: The multi-source data real-time fusion module connects to the data source and performs real-time synchronization. It employs a Transformer-based engineering-specific multimodal large language model (MLLM) to perform semantic parsing and structured extraction of the data source, obtaining cross-modal data to build the foundation for the project's "digital twin." This engineering-specific MLLM is pre-trained and optimized using corpora from engineering geology and engineering economics, and features a dedicated parsing process designed for geological survey drawings. Unlike general multimodal processing methods, it accurately adapts to the unstructured data processing needs of engineering scenarios. This module is the "data hub" of the entire system, its core task being to accurately and in real-time aggregate and process various heterogeneous data related to the project, providing high-quality "raw materials" for subsequent intelligent generation and analysis.

[0028] Furthermore, in step one above, the data sources accessed include policy and regulatory databases, market databases, and project foundation databases: The Policy and Regulation Database is a customized web crawler (developed using the Scrapy framework and with targeted crawling rules for official engineering and construction websites) and API interface specifically designed for the engineering field, connecting in real-time to official information release systems for engineering and construction. Specifically, the database contains relevant standards and specifications and automatically retrieves the latest infrastructure policies, environmental regulations, land use plans, and other project-related text information through pre-set scheduled tasks (e.g., three times daily: 8 AM, 12 PM, and 8 PM) and keyword subscriptions in the engineering field (e.g., "municipal engineering environmental protection standards," "land policy for building construction projects," etc.). Simultaneously, an AI knowledge base for the engineering field is established, synchronized in real-time with the information release system, incorporating a built-in engineering policy classification tagging system to achieve precise matching of policies and projects, ensuring the timeliness and authority of information and avoiding decision-making errors due to information delays.

[0029] The market database features a data synchronization mechanism designed for engineering construction materials and labor costs. Specifically, it connects to mainstream building material price information platforms (such as Mysteel and 100-Year Construction Network) and commodity trading data centers via API interfaces to obtain real-time building material price indices and regional labor costs. Simultaneously, it connects to financial data service providers to obtain the latest Loan Prime Rate (LPR) and related industry financing rates. A sliding window caching mechanism (setting a 1-hour sliding window) is used to synchronize data in accordance with the timeliness requirements of engineering data. An early warning threshold for abnormal fluctuations in engineering construction material prices is set (e.g., a daily fluctuation exceeding ±10% triggers manual review). Cross-platform consistency verification rules are also established (e.g., if the price of the same product category is averaged from more than 3 platforms, a deviation exceeding 5% is flagged as abnormal), ensuring time alignment and data consistency, providing accurate input for financial evaluation.

[0030] The project foundation database supports the uploading and parsing of engineering-specific format files (CAD drawings, geological survey reports in PDF format, etc.), differing from general data storage and access methods. Specifically, the project foundation database allows users to upload core basic data of the project, including but not limited to geological survey reports (PDF / CAD format), topographic maps, planning red line maps (DWG / DXF format), preliminary design schemes, and other engineering-specific format files; uploaded files are automatically checked for MD5 to avoid duplicate uploads; folder directories are created according to four main categories: "Project Overview, Geological Survey, Planning and Design, and Preliminary Approvals," and users can customize subdirectories; geological survey files are automatically associated with project latitude and longitude information and geological condition tags, supporting the retrieval of geological data for similar projects by region and project type, facilitating data reuse and comparison of similar projects.

[0031] Meanwhile, step one above employs a multimodal large language model based on the Transformer architecture. Pre-training and optimization are performed using corpora from the fields of engineering geology and engineering economics to encode and map data from different modalities (such as text, images, and engineering drawings) into a unified engineering semantic representation space. This achieves deep fusion and understanding of cross-modal information. The specific processing flow for geological exploration drawings is as follows: (1) Image recognition and segmentation First, an improved U-Net semantic segmentation model based on convolutional neural networks (CNN) is used to preprocess the drawings. Model parameters are optimized for features such as stratigraphic boundaries and borehole locations in geological exploration drawings, identifying and segmenting different areas such as legends, borehole columnar sections, cross-sections, and tables. This step effectively handles multi-scale features and complex backgrounds in the drawings, solving the problems of inaccurate detail recognition and blurred boundaries in traditional segmentation models, thus improving segmentation accuracy.

[0032] (2) Extraction of key information 1) Text information: For legend areas, table areas and annotation text, use engineering-specific deep learning OCR models (such as DeepSeek-OCR) for high-precision text recognition. Optimize the recognition algorithm for engineering terms and geological parameters, and extract text information such as stratum names and rock and soil physical and mechanical properties to reduce recognition errors. 2) Graphical Information: For borehole columnar sections and geological profiles, an end-to-end computer vision model specifically designed for the engineering field is used. Combining target detection (identifying stratigraphic boundaries) and image classification (identifying soil and rock types), the graphical geological stratification information, groundwater level, etc., are converted into structured geological parameter data, including core engineering parameters such as stratigraphic number, layer thickness, and soil and rock mechanical properties, such as "{'Stratigraphic Number':'1','Name':'Plain Fill','Bottom Depth':'2.5m'}".

[0033] 3) Knowledge Graph Enhancement: To further improve the accuracy of information extraction, this invention constructs a knowledge graph in the field of engineering geology, which includes core entities and their relationships such as stratigraphic types, soil and rock properties, and engineering specifications. During the information extraction process, the identified entities (such as "silty clay") are linked to nodes in the knowledge graph. The entity relationships defined in the graph (such as the bearing capacity characteristic value range of "silty clay") are used for verification and semantic completion. At the same time, engineering specifications constraints are introduced to ensure that the extracted data conforms to industry standards. This approach differs from the application of general knowledge graphs and effectively solves problems such as inaccurate terminology recognition and missing key parameters.

[0034] Through the above process, this invention can accurately transform a complex, unstructured geological survey map into structured parameters that can be directly accessed by a computer, breaking down the barriers between engineering data and economic analysis, and solving the problem that traditional data processing methods are unable to handle unstructured maps and reports.

[0035] Step Two: For highly specialized calculations, such as "financial evaluation," the large model does not directly "create" numbers. Instead, it calls upon built-in, rigorously validated professional algorithm modules to calculate the internal rate of return. and dynamic investment payback period The calculation results are formatted for subsequent modules to use; the specialized algorithm module has a built-in error correction mechanism and multi-scenario adaptation algorithm, which ensures calculation accuracy through double verification. The error correction mechanism and multi-scenario adaptation algorithm are designed specifically for the engineering field, solving the problems of insufficient accuracy and poor adaptability of general calculation methods in engineering scenarios.

[0036] Furthermore, in step two above... It is the discount rate when the cumulative net present value is 0. This refers to the time it takes for the cumulative net present value to turn from negative to positive. The relevant calculation formulas for both are as follows: ; ; in, Represents the cumulative net present value; Indicates the discount rate; Indicates the total number of years; Indicates the year of the calculation period, i.e. ; Indicates the first Annual cash inflow; Indicates the first Annual cash outflow.

[0037] The error correction mechanism of the aforementioned professional algorithm module reduces calculation errors by introducing benchmark parameters from the engineering industry (such as a benchmark discount rate of 8% for municipal engineering, 8% for building construction, and 7.5% for highway engineering) to calibrate deviations. The multi-scenario adaptation algorithm can automatically adjust calculation parameters according to the type of project. For example, the construction period for municipal engineering is usually 3-5 years and the operation period is 20-30 years, while the construction period for building construction is usually 1-3 years and the operation period is 15-20 years. The algorithm can automatically match the corresponding parameters to ensure the accuracy of calculation results under different engineering scenarios.

[0038] Meanwhile, the calculation results of the aforementioned specialized algorithm modules are in JSON format as follows: {"Calculation Parameters":{"Project Type":"Municipal Engineering","Discount Rate":8.0%,"Calculation Period":25 years (3 years construction period, 22 years operation period),"Benchmark Rate of Return":8.0%},"Intermediate Data":{"Annual NPV":[-12 million, -18 million, -9 million, 3 million, 5.2 million,...],"Critical Year T":3 (3rd year NPV = -9 million, 4th year NPV = 3 million)},"Final Indicator":{"IRR":10.25%,"PBP":3.75 years},"Calculation Accuracy":{"IRR Iteration Error":3.2×10^-7,"PBP Calculation Error":0.01 years},"Error Correction Record":{"Benchmark Parameter Adaptation":"Municipal Engineering Benchmark Discount Rate 8%","Deviation Calibration Value":0.002}}.

[0039] Step 3: The intelligent report generation module integrates and refines the cross-modal data from Step 1 with the calculation results from Step 2, organizing them into a complete text report using coherent, professional, and fluent language, and generating visualization results such as charts. The visualization results include engineering-specific charts such as geological stratification diagrams and financial evaluation sensitivity curves, and support cross-professional data linkage updates. It can automatically adapt to the standardized structure of engineering feasibility study reports, improving the professionalism and efficiency of report generation.

[0040] Furthermore, the specific settings for the text report generated in step three are as follows: 1. Standardized design of report structure The feasibility study report adopts a standardized structure of "chapter + appendix," covering core chapters such as project overview, market analysis, construction plan, investment estimation, financial evaluation, and risk analysis. Each chapter has fixed templates and variable fields, with variable fields automatically linked to multi-source data and calculation results. For example, the "financial evaluation" chapter automatically inserts calculation results such as IRR and PBP, as well as cash flow statements and break-even analysis charts; the "construction plan" chapter automatically inserts parameters extracted from geological exploration and geological stratification diagrams.

[0041] 2. Text Integration and Polishing Rules The text polishing focuses on optimizing logical coherence, data consistency, and format standardization to adapt to the professional expression habits of engineering feasibility study reports; it supports user-defined report styles and provides three templates: "Concise Version," "Detailed Version," and "Professional Version," allowing adjustment of the level of detail in chapters and the density of charts to meet the needs of different scenarios.

[0042] Step Four: The Dynamic Optimization and Sensitivity Analysis module, combined with the multi-source data real-time fusion module, professional calculation module, and intelligent report generation module, performs real-time scenario simulation and visualization analysis of multi-factor coupling. Based on the engineering domain knowledge graph and historical project database, it uses an improved cosine similarity algorithm to match risk response strategies, thereby providing interactive intelligent decision support. The improved cosine similarity algorithm introduces engineering weight factors to improve the accuracy of similar project matching, solving the problems of insufficient targeting and accuracy of traditional matching methods. Thus, this module transforms the feasibility study report from a static document into an interactive and simulated dynamic decision-making tool, greatly enhancing its application value.

[0043] Furthermore, in step four above, when conducting real-time scenario simulation and visualization analysis involving multi-factor coupling, the dynamic optimization and sensitivity analysis module uses a human-computer interaction interface to allow users to input hypothetical scenarios using natural language, such as: "What would happen if steel prices increased by 15%?" or "Assuming the project duration is extended by 6 months, what impact would that have on the investment payback period?", and then performs the following workflow: 1) Intent understanding: The engineering-specific multimodal large language model receives and understands the natural language input by the user, accurately identifies the sensitive factors that need to be adjusted (such as building material prices, construction period, geological conditions, etc.) and their change range (+15%), and performs factor correlation analysis in combination with the characteristics of the engineering scenario to obtain the change parameters; 2) Data and model linkage: Changes in parameters are transmitted to the professional computing module in real time, while triggering the multi-source data real-time fusion module to update relevant related data; 3) Real-time recalculation and result feedback: The professional calculation module recalculates based on the changed parameters and updated related data, updates key project-related indicators such as total investment estimate, total project cost, internal rate of return, and investment payback period, and outputs an indicator change analysis report. 4) Visual Presentation: Several multi-factor coupled simulations are conducted to simulate the project's performance under different combinations of factors. The simulation results (e.g., the change in IRR when building material prices fluctuate within a range of -20% to +20%) are presented using engineering-specific visualization charts (such as tornado diagrams, break-even analysis charts, etc.), which differs from the traditional approach of single-factor simulation. This graphical presentation allows users to intuitively identify the most critical and sensitive factors affecting the project's profitability.

[0044] Specifically, when recognizing natural language, the model outputs four elements: "sensitivity factors + magnitude of change + impact dimensions + correlation analysis". For example, if a user inputs "If labor costs increase by 8%, how will the project's profitability be?", the recognition result is: Sensitivity factors = labor costs, magnitude of change = +8%, impact dimensions = IRR, PBP, total cost, correlation analysis = Increased labor costs will lead to increased annual cash outflows during the operating period, thereby reducing the cumulative net present value, causing IRR to decrease and PBP to lengthen.

[0045] The entire process is real-time, allowing users to continuously adjust assumptions and explore various possibilities, much like having a conversation.

[0046] Meanwhile, in step four above, when matching risk response strategies based on similar projects, the dynamic optimization and sensitivity analysis module constructs and maintains a large-scale historical project database specifically for the engineering field in real time. This database contains basic information about historical projects, engineering parameters, various engineering risk events encountered during construction and operation, the response strategies adopted at the time, and the final implementation results. When a high-risk factor is identified in a project (such as "extremely high risk of project delay"), the large model will retrieve the most similar historical projects from the historical project database using an improved cosine similarity algorithm and extract the successful response strategies adopted when dealing with the high-risk factor. For example, "signing long-term supply contracts with core suppliers to lock in prices to cope with the risk of material price fluctuations" or "adopting a segmented construction and parallel operation construction organization method to avoid the impact of the rainy season on earthwork engineering." These strategies are then adapted and adjusted in conjunction with the specific engineering parameters of the current project to form a customized risk response plan. Subsequently, the large model will visualize the corresponding historical projects, the adapted response strategies, and implementation suggestions, providing decision-makers with data-supported and practice-tested risk avoidance solutions. Among them, the improved cosine similarity algorithm introduces engineering weight factors to improve matching accuracy. The engineering weight factors include geological conditions weight of 0.3, construction scale weight of 0.25, industry type weight of 0.25, and sensitive factor type weight of 0.2.

[0047] Specifically, this invention employs a cosine similarity algorithm for project matching. Matching dimensions include industry type, construction scale, geological conditions, and types of sensitive factors. The similarity threshold is set to 0.8. When a project with a similarity ≥ 0.8 is retrieved, its risk response strategy is automatically extracted and adjusted accordingly. When a project with a similarity between 0.6 and 0.8 is retrieved, it is marked as "moderately relevant," and the response strategy is supplemented and improved using a knowledge graph specific to the engineering field. When a project with a similarity < 0.6 is retrieved, a completely new risk response strategy is generated based on engineering standards and industry expert experience. The output format is presented in a structured table of "risk factors - response strategy - implementation steps - expected results," accompanied by case studies (showing the implementation results of historical projects using this strategy). Users can download the strategy list and case reports for easy implementation.

[0048] Furthermore, this invention also discloses an intelligent system, including a multi-source data real-time fusion module, a professional computing module, a report intelligent generation module, and a dynamic optimization and sensitivity analysis module. This system is used to execute a method for intelligent generation and dynamic optimization of feasibility study reports based on a large model throughout the entire consulting process. The multi-source data real-time fusion module, professional computing module, and report intelligent generation module respectively execute steps one through three. The four modules collaborate through an engineering data linkage bus to perform step four. Utilizing the engineering data linkage bus enables real-time synchronization and interaction of data between modules, solving the problem of cross-disciplinary data fragmentation, improving system collaboration efficiency, and differing from existing simple data transfer methods between modules.

[0049] Through the above-described scheme, this invention constructs an integrated intelligent system. Its core architecture comprises a multi-source data real-time fusion module, a professional computing module, a report intelligent generation module, and a dynamic optimization and sensitivity analysis module. These four modules work collaboratively to form a closed-loop workflow from data input to report generation and dynamic optimization. The process begins with the multi-source data real-time fusion module synchronizing and deeply analyzing multi-dimensional data in real time, providing high-quality "raw materials" for subsequent work. Next, the professional computing module performs precise calculations of core indicators based on standardized algorithms, ensuring data reliability. Then, the report intelligent generation module integrates and refines the preceding data and calculation results, generating professional and fluent text reports and visual charts. Finally, the dynamic optimization and sensitivity analysis module achieves dynamic iteration and decision support for the report through real-time scenario simulation, visual analysis, and risk strategy matching with similar projects. This closed-loop workflow not only solves the problems of low data integration efficiency, difficulty in cross-disciplinary collaboration, and subjective sensitivity analysis in existing technologies, but also upgrades the traditional static feasibility study report into an interactive and simulated dynamic decision-making tool, comprehensively improving the generation efficiency, data accuracy, and decision-making reference value of the entire process of consulting feasibility study reports. It should be noted that the technical solution of the present invention is not limited to the specific embodiments described above. Any simple replacement or adjustment of the module architecture, algorithm parameters, data processing flow, etc., based on the core concept of the present invention, or the adaptive application in different engineering scenarios (municipal, building construction, highway, etc.), shall fall within the protection scope of the present invention.

[0050] The above embodiments and figures are not intended to limit the product form and style of the present invention. Any appropriate changes or modifications made by those skilled in the art should be considered as not departing from the patent scope of the present invention.

Claims

1. A method for intelligent generation and dynamic optimization of feasibility study reports for the entire consulting process based on a large model, characterized in that: include: Step 1: The multi-source data real-time fusion module connects to the data source and performs real-time synchronization. It uses a multimodal large language model based on the Transformer architecture, which is dedicated to the engineering field, to perform semantic parsing and structured extraction on the data source to obtain cross-modal data. The multimodal large language model is pre-trained and optimized using corpora from the fields of engineering geology and engineering economics, and a dedicated parsing process is designed for geological exploration drawings. Step 2: Calculate the internal rate of return using a specialized algorithm module. and dynamic investment payback period The calculation results are then formatted for use by subsequent modules. Step 3: The intelligent report generation module integrates and refines the cross-modal data from Step 1 with the calculation results from Step 2 to form a visualization result; the visualization result is an engineering-specific chart and supports cross-disciplinary data linkage updates; Step 4: The dynamic optimization and sensitivity analysis module, combined with the multi-source data real-time fusion module, professional calculation module, and intelligent report generation module, performs real-time scenario simulation and visualization analysis of multi-factor coupling. Based on the engineering domain knowledge graph and historical project database, it matches risk response strategies through an improved cosine similarity algorithm; the improved cosine similarity algorithm introduces engineering weight factors.

2. The intelligent generation and dynamic optimization method for full-process consulting feasibility study reports based on a large model as described in claim 1, characterized in that: In step one, the data sources accessed include a policy and regulation database, a market database, and a project basic database. The policy and regulation database is equipped with customized web crawlers and API interfaces specifically for the engineering field, and connects in real time to official information release systems for engineering construction. The market database is designed with a data synchronization mechanism for engineering building materials and labor costs. The project basic database supports the uploading and parsing of special format files for engineering projects.

3. The intelligent generation and dynamic optimization method for full-process consulting feasibility study reports based on a large model as described in claim 2, characterized in that: The policy and regulation database contains standards and specifications related to the entire consulting process. Through pre-set scheduled tasks and keyword subscriptions, it automatically captures text information related to the project and establishes an AI knowledge base that is synchronized in real time with the information release system, with a built-in engineering policy classification tag system. The market database connects to building material price information platforms, bulk commodity trading data centers, and financial data service providers through API interfaces. It uses a sliding window caching mechanism combined with the timeliness requirements of engineering data for data synchronization, and sets early warning thresholds for abnormal fluctuations in engineering and building material prices and establishes cross-platform consistency verification rules. The project basic database supports users to upload core basic information about the project and automatically verifies it to avoid duplicate uploads, while storing the information in batches and categories.

4. The intelligent generation and dynamic optimization method for full-process consulting feasibility study reports based on a large model as described in claim 1, characterized in that: Step one employs a multimodal large language model based on the Transformer architecture. It is pre-trained and optimized using corpora from the fields of engineering geology and engineering economics, encoding and mapping data from different modalities into a unified engineering semantic representation space. The specific processing flow for geological survey drawings is as follows: (1) Image recognition and segmentation First, the improved U-Net semantic segmentation model based on convolutional neural networks is used to preprocess the drawings. The model parameters are optimized according to the features of the geological exploration drawings, and different regions are segmented. (2) Extraction of key information 1) Text information: For legend areas, table areas and annotation text, a deep learning OCR model specifically for the engineering field is used for text recognition. The recognition algorithm is optimized for engineering terms and geological parameters, and the text information is extracted. 2) Graphical information: For borehole columnar sections and geological profiles, an end-to-end computer vision model specifically designed for the engineering field is used, combined with target detection and image classification, to convert the graphics into structured geological parameter data; 3) Construct a knowledge graph in the field of engineering geology, which includes core entities and their relationships in terms of stratigraphic types, soil and rock properties, and engineering specifications. During the information extraction process, the identified entities are linked to nodes in the knowledge graph, and the entity relationships defined in the graph are used for verification and semantic completion. At the same time, engineering specifications constraints are introduced to ensure that the extracted data complies with industry standards.

5. The intelligent generation and dynamic optimization method for full-process consulting feasibility study reports based on a large model as described in claim 1, characterized in that: In step two It is the discount rate when the cumulative net present value is 0. This refers to the time it takes for the cumulative net present value to turn from negative to positive. The relevant calculation formulas for both are as follows: ； ； in, Represents the cumulative net present value; Indicates the discount rate; Indicates the total number of years; Indicates the year of the calculation period, i.e. ; Indicates the first Annual cash inflow; Indicates the first Annual cash outflow.

6. The intelligent generation and dynamic optimization method for full-process consulting feasibility study reports based on a large model as described in claim 1, characterized in that: In step four, when performing real-time scenario simulation and visualization analysis involving multiple factors, the dynamic optimization and sensitivity analysis module uses a human-computer interaction interface to allow users to input "hypothetical scenarios" via natural language and then performs the following workflow: 1) A multimodal large language model for engineering fields receives and understands the natural language input by users, identifies the sensitive factors that need to be adjusted and their magnitude of change, and performs factor correlation analysis in combination with the characteristics of engineering scenarios to obtain the change parameters. 2) The changing parameters are transmitted to the professional computing module in real time, and at the same time, the multi-source data real-time fusion module is triggered to update the relevant related data; 3) The professional calculation module recalculates and updates the key indicators related to the project based on the changed parameters and the updated related data, and outputs an indicator change analysis report; 4) Conduct several simulations involving multiple factors to simulate the project's performance under different combinations of factors, and present the simulation results in visualization charts specific to the engineering field.

7. The intelligent generation and dynamic optimization method for full-process consulting feasibility study reports based on a large model as described in claim 1, characterized in that: In step four, when performing risk response strategies based on matching similar projects, the dynamic optimization and sensitivity analysis module constructs and maintains a dedicated historical project database for the engineering field in real time. This database contains basic information about historical projects, engineering parameters, various engineering risk events encountered during construction and operation, response strategies adopted at the time, and final implementation results. When a high-risk factor is identified in a project, the large model will retrieve the most similar historical project from the historical project database using an improved cosine similarity algorithm, extract the successful response strategies used to handle the high-risk factor, and adapt and adjust them in conjunction with the specific engineering parameters of the current project to form a customized risk response plan. Subsequently, the large model will visualize the corresponding historical projects, the adapted response strategies, and implementation suggestions.

8. The intelligent generation and dynamic optimization method for full-process consulting feasibility study reports based on a large model as described in claim 7, characterized in that: The improved cosine similarity algorithm introduces an engineering weight factor, which includes a geological condition weight of 0.3, a construction scale weight of 0.25, an industry type weight of 0.25, and a sensitive factor type weight of 0.

2.

9. The intelligent generation and dynamic optimization method for full-process consulting feasibility study reports based on a large model as described in claim 7, characterized in that: The cosine similarity algorithm is used for project matching. The matching dimensions include industry type, construction scale, geological conditions, and sensitive factor type. The similarity threshold is set to 0.

8. When a project with a similarity of ≥0.8 is retrieved, its risk response strategy is automatically extracted and adapted.

10. An intelligent system, characterized in that, It includes a multi-source data real-time fusion module, a professional calculation module, a report intelligent generation module, and a dynamic optimization and sensitivity analysis module, used to execute the intelligent generation and dynamic optimization method for full-process consulting feasibility study reports based on a large model as described in any one of claims 1 to 9. The multi-source data real-time fusion module, the professional calculation module, and the report intelligent generation module respectively execute steps one to three, and the four modules work together through an engineering data linkage bus to perform step four.

Images