Bill of quantities automatic preparation and auditing method and system based on industrial big data

Abstract

This invention relates to the fields of data processing and engineering cost technology, and discloses a method for automatic compilation and review of engineering quantity lists based on industrial big data, including the following steps: S1: Constructing an industrial big data pool, which aggregates heterogeneous data sources from IoT sensors for engineering materials, project BIM design model databases, historical engineering databases, and market price information platforms in real time or with a delay not exceeding a preset time threshold. By constructing an industrial big data pool and integrating natural language processing models and BIM technology, this invention can automatically parse design documents, intelligently extract project features, and perform mandatory verification using three-dimensional spatial quantity calculation. This significantly reduces routine manual intervention and avoids misunderstandings and calculation errors caused by human negligence. Simultaneously, by using an intelligent matching engine to call historical data for automated correction and completion of the quantity list, it ensures the standardization and completeness of the descriptions of item features in the quantity list.

Description

Technical Field

[0001] This invention relates to the fields of data processing and engineering cost technology, specifically to a method and system for automatic compilation and review of engineering quantity lists based on industrial big data. Background Technology

[0002] The bill of quantities is the core document for construction project cost management. Its accuracy and efficiency in preparation and review directly affect project investment control, the fairness of bidding, and the smooth progress of subsequent construction and settlement. Currently, the preparation and review of the bill of quantities mainly rely on manual labor or rudimentary information technology tools, and there are two long-standing technical challenges that have not been effectively resolved: First, the compilation process relies excessively on manual experience, resulting in low efficiency and poor consistency. Current methods typically involve cost engineers manually reviewing massive amounts of design drawings and tender documents, extracting project features, calculating quantities, and applying quotas. This process is not only time-consuming and labor-intensive but also limited by the engineer's individual experience and skill level, easily leading to misunderstandings of drawings, omissions or duplications in project feature descriptions, and errors in quantity calculations. While BIM-based quantity calculation software (such as Revit and Glodon BIM Quantity Calculation) and bill of quantities compilation assistance systems exist in the market, these tools still have significant limitations: on the one hand, they primarily focus on… While automated calculations focused on geometric quantities are sufficient, they cannot intelligently extract semantic information from unstructured texts (such as design specifications and technical specifications), and the completeness and compliance of project features still require manual judgment. On the other hand, even if the system integrates historical databases, its matching methods are mostly keyword searches or simple classifications, lacking a deep understanding of engineering semantics and failing to achieve intelligent correction and recommendation based on multi-dimensional similarity. Furthermore, the lack of systematic and intelligent technical means to effectively utilize the massive historical project data accumulated within the enterprise to provide intelligent reference and correction for new projects makes it difficult to transform valuable experience data into productivity.

[0003] Secondly, the review process lacks intelligent and forward-looking risk warning capabilities. Currently, the review of the checklist mainly relies on auditors to conduct manual checks after the fact, based on standards and experience. This is a static and passive quality control method. Existing review systems are mostly based on fixed rule bases (such as pricing specification items) or simple numerical range checks. They cannot systematically discover hidden logical contradictions within the checklist (such as duplicate or omitted work content), unbalanced pricing risks that are out of touch with market prices, and high-risk items that can be predicted based on historical change data. The quality of the review heavily depends on the individual ability and energy of the review experts, and it is impossible to form an audit knowledge system that can be accumulated, iterated, and automatically executed. In particular, it is impossible to conduct closed-loop correlation analysis between the project's subsequent change order data, actual consumption data, final settlement data and the original checklist, so as to dynamically optimize the review rules and realize the transformation from "post-event correction" to "pre-event prevention". Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for automatic compilation and review of bill of quantities based on industrial big data, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for automatic compilation and review of bill of quantities based on industrial big data, comprising the following steps: S1: Construct an industrial big data pool, which aggregates heterogeneous data sources from engineering material IoT sensors, project BIM design model databases, historical engineering databases, and market price information platforms in real time or with a delay not exceeding a preset time threshold. S2: Based on the natural language processing model, the design documents and tender documents of the project to be compiled are parsed, the project feature structure tree and key process parameters are extracted, and the BIM model data in the industrial big data pool are combined to perform three-dimensional spatial quantity calculation benchmark verification and generate the initial bill of quantities items. S3: Call the intelligent matching engine to perform multi-dimensional similarity matching between the initial bill of quantities items and the historical engineering data in the industrial big data pool. The multi-dimensional similarity includes structural feature dimension, process and method dimension, and material specification dimension. Based on the matching results, adaptively modify the engineering quantity and project feature description of the initial bill of quantities items, and associate and recommend the comprehensive unit price and resource consumption index of historical projects. S3a: Based on real-time market price information in the industrial big data pool, dynamically adjust the comprehensive unit price of the historical projects recommended in step S3, generate a benchmark unit price that conforms to the current market conditions, and use the benchmark unit price as the comparison benchmark for the unbalanced pricing risk warning in step S4; S4: Construct a dynamic audit rule base based on a reinforcement learning model. The training data for the reinforcement learning model includes historical audit records, change visa data, and final settlement data. The dynamic audit rule base performs real-time compliance review, logical contradiction detection, and unbalanced pricing risk warning on the list items generated in step S3. S5: Output the final bill of quantities after review and simultaneously generate a report on the bill of quantities preparation and review process. The report visually displays the matching source of key items, the basis for review, and risk warnings.

[0006] As a further optimization of the technical solution of this invention, its innovation lies in the following collaborative working mechanism: First, by constructing an industrial big data pool that integrates BIM geometric data, IoT working condition data, historical engineering semantic data, and market price time series data, a unified data foundation is provided to solve the problems of data isolation and poor timeliness in traditional methods. Second, an innovative three-layer progressive list generation and verification mechanism of "natural language parsing → BIM spatial verification → historical semantic matching" is adopted. Among them, the semantic matching link introduces the Sentence-BERT model, which has been fine-tuned with engineering corpus, to solve the problem of the general model's accuracy in matching professional terms. To address the issue of insufficient accuracy, a reinforcement learning model was introduced and trained using historical review, change order, and settlement data. This resulted in the construction of a dynamic review rule base with self-evolution capabilities, enabling a leap from static rule review to dynamic risk warning. Finally, through closed-loop optimization design, construction process and settlement result data were fed back to drive continuous iterative optimization of the historical database and review model, giving the system the intelligent characteristic of "becoming more accurate with use." Experimental results show that this method can significantly improve the efficiency and quality of bill of quantities compilation, with an estimated increase of over 60% in efficiency, a project feature completeness rate of over 98%, and an unbalanced pricing risk warning accuracy rate of over 90%.

[0007] As a preferred technical solution of the present invention, the processing of heterogeneous data sources in step S1 includes: labeling IoT sensor data with working condition context; performing standardized cleaning and reconstruction of historical engineering data based on a unified project decomposition structure; and using a time series analysis model to perform trend filtering and outlier removal on market price information.

[0008] As a preferred technical solution of the present invention, the three-dimensional spatial quantity calculation benchmark verification in step S2 specifically involves: spatially mapping and comparing the list items obtained from natural language parsing with the component quantities automatically extracted from the BIM model. When the deviation exceeds a preset threshold, a manual review reminder is triggered and the reason for the deviation is recorded in the industrial big data pool for optimization of subsequent parsing models. This step constitutes the first layer of mandatory verification of "semantic-geometric" cross-validation, effectively intercepting hard errors in engineering quantities caused by misunderstanding of drawings or omissions in description.

[0009] As a preferred embodiment of the present invention, the process of multi-dimensional similarity matching performed by the intelligent matching engine in step S3 includes: S31: Based on the feature structure tree of the current project, filter out the candidate project set from the historical project database; S32: Using a pre-trained engineering semantic model, semantic vectorization is performed on the corresponding descriptions of the current list item and the candidate item set, and a matching index reflecting the semantic correlation between the two is generated. The engineering semantic model is a Sentence-BERT model fine-tuned with engineering domain corpus. The fine-tuning process includes: using paired synonymous or near-synonymous engineering description texts extracted from historical engineering databases as positive samples, and randomly paired different engineering description texts as negative samples, and using a contrastive learning loss function to fine-tune the model to optimize its vector representation, maximizing the cosine similarity between positive samples and minimizing the cosine similarity between negative samples. This fine-tuning process enables the model to accurately understand the semantic equivalence of engineering professional expressions such as "C30 concrete rectangular column" and "rectangular column, concrete strength grade C30", solving the problem of inaccurate matching of general semantic models in professional domains.

[0010] S33: Based on the comprehensive semantic relevance matching index, the proportion of engineering quantities, and the matching degree of process and method, output a comprehensive matching confidence score; S34: Only when the overall matching confidence is higher than a set threshold, the historical data of the associated recommendations are used to correct and complete the current list items. This threshold mechanism ensures that automatic correction is only performed when the matching is highly reliable, avoiding interference from low-quality historical data on the compilation results.

[0011] As a preferred embodiment of the present invention, the process of constructing the dynamic audit rule base in step S4 includes: S41: Based on decisions and modification suggestions in historical audit records, generate basic compliance rules through supervised learning; S42: Based on the correlation analysis between change visa data and corresponding list items in historical projects, potential risk correlation patterns are mined through unsupervised learning to form logical contradiction detection rules; S43: Based on the retrospective comparison between the final settlement data and the winning bid list data, the triggering conditions for the risk warning of unbalanced pricing are dynamically optimized through a reinforcement learning model. This construction process integrates supervised, unsupervised and reinforcement learning, enabling the rule base to not only encode explicit audit experience, but also to mine implicit risk patterns from massive project post-effect data, and to achieve adaptive optimization of the warning strategy through reinforcement learning.

[0012] As a preferred technical solution of the present invention, after step S5, a closed-loop optimization step S6 is also included: the actual resource consumption data collected by IoT sensors during the actual construction process and the final list adjustment data determined in the project settlement stage are fed back to the industrial big data pool as feedback data streams for incrementally updating the historical project database and training the reinforcement learning model. This step constitutes the core evolution mechanism of the system, enabling the system's matching accuracy and auditing capabilities to continuously improve with the implementation of more projects, realizing the transformation from a "tool" to an "intelligent agent".

[0013] A system for implementing the above-described method for automatic compilation and review of bill of quantities based on industrial big data includes: The industrial big data aggregation and management module is used to build and maintain the industrial big data pool, and to realize the access, cleaning, fusion and storage of multi-source heterogeneous data; The intelligent parsing and initial compilation module integrates a natural language processing model and a BIM interface. It is used to parse the design documents and tender documents of the project to be compiled, extract the project feature structure tree and key process parameters, and perform three-dimensional spatial quantity calculation benchmark verification in combination with BIM model data to generate initial bill of quantities items. The data intelligent matching and list optimization module integrates the intelligent matching engine, which is used to perform multi-dimensional similarity matching between the initial bill of quantities items and historical engineering data in the industrial big data pool, and adaptively correct the engineering quantity, project feature description, and associated recommendation of the comprehensive unit price and resource consumption index of historical projects based on the matching results. The dynamic intelligent audit and risk warning module integrates the dynamic audit rule base based on the reinforcement learning model, which is used to conduct real-time compliance review, logical contradiction detection and unbalanced pricing risk warning on the optimized list items; The Bill of Quantities Output and Visual Reporting module is used to output the final, approved Bill of Quantities and simultaneously generate a report on the Bill of Quantities preparation and review process.

[0014] As a preferred technical solution of the present invention, the industrial big data aggregation and management module includes a working condition context labeling unit, a data standardization and cleaning unit, and a market information processing unit; the data intelligent matching and list optimization module includes a candidate project screening unit, an engineering semantic model unit, and a confidence calculation and decision-making unit.

[0015] As a preferred technical solution of the present invention, the dynamic intelligent audit and risk warning module includes a rule generation unit, a pattern mining unit, and a rule dynamic update unit, which are respectively used to generate basic compliance rules based on historical audit records, form logical contradiction detection rules based on risk association patterns mined from change visa data, and dynamically optimize the triggering conditions for unbalanced pricing risk warning based on final settlement data.

[0016] As a preferred embodiment of the present invention, the system further includes a system coordination and data bus module; The industrial big data aggregation and management module, intelligent analysis and initial compilation module, data intelligent matching and list optimization module, dynamic intelligent audit and risk warning module, and list output and visualization report module are all connected and exchange data through the system collaboration and data bus module. The system coordination and data bus module is configured as follows: Receive initial bill of quantities items from the intelligent parsing and initial compilation module, and based on preset process rules, distribute the initial bill of quantities items and their associated BIM verification data to the data intelligent matching and bill of quantities optimization module. Receive optimized list items and associated historical data recommendation reliability from the data intelligent matching and list optimization module, and trigger the dynamic intelligent review and risk warning module to load the corresponding review rule subset for review; The system receives audit results and risk markers from the dynamic intelligent audit and risk warning module. Based on the level of risk markers, it controls the list output and visualization report module to generate visualization reports with different levels of detail, or feeds back the list items marked with high risk and the audit basis to the intelligent parsing and initial compilation module to initiate the manual review process. The system's collaboration module adopts a message queue mechanism based on Apache Kafka, which realizes decoupled, asynchronous, and traceable data flow and process scheduling between various intelligent modules, ensuring the stable and efficient execution of complex processing processes.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention, by constructing an industrial big data pool and integrating natural language processing models and BIM technology, can automatically parse design documents, intelligently extract project features, and use three-dimensional spatial quantity calculation for mandatory verification. This significantly reduces routine manual intervention and avoids misunderstandings and calculation errors caused by human negligence. At the same time, by using an intelligent matching engine to call historical data to automatically correct and complete the bill of quantities, it ensures the standardization and completeness of the description of the bill of quantities' features, significantly improving the quality of the initial compilation results. In particular, through a three-layer progressive mechanism of "semantic parsing - BIM verification - historical matching," it forms a three-dimensional, high-precision compilation and verification capability for the bill of quantities.

[0018] 2. By adding step S3a, this invention dynamically adjusts the historical recommended unit price using real-time market price information from the industrial big data pool, generates a benchmark unit price that conforms to the current market conditions, and directly applies it to the comparison benchmark for unbalanced pricing risk warning. This effectively solves the problem of warning deviation caused by poor timeliness of historical data, and further improves the accuracy and timeliness of risk warning. This mechanism transforms the static historical reference unit price into a dynamic market benchmark unit price, so that the risk warning benchmark is updated in real time with market fluctuations.

[0019] 3. In this invention, the Sentence-BERT model, finely tuned with engineering corpus, is used in the intelligent matching engine. The model is specially optimized by contrastive learning loss function, which enables it to accurately understand the semantic associations of engineering professional expressions, significantly improving the accuracy and reliability of multi-dimensional matching and enhancing the intelligence level of the system. This optimization enables the model to effectively distinguish subtle professional differences such as "marble paving" and "granite paving", while general models perform poorly in such tasks.

[0020] 4. This invention creatively constructs a dynamic audit rule base based on a machine learning model. This rule base can not only integrate historical audit experience for real-time compliance review, but also proactively identify potential logical contradictions and unbalanced pricing risks by mining risk correlation patterns in historical engineering data. In particular, through a reinforcement learning mechanism, the system can continuously optimize early warning rules based on actual project feedback (such as settlement data), enabling the audit capability to have self-evolutionary characteristics. It upgrades from traditional static experience-based judgment to dynamic, data-driven intelligent early warning, significantly improving the coverage and accuracy of risk identification. Compared with existing fixed rule base audit systems, this system's rule base has the dual capabilities of "learning from data" and "optimizing in practice".

[0021] 5. The invention’s unique closed-loop optimization design feeds back the actual resource consumption data and final settlement results during the construction process to the industrial big data pool for updating the historical database and training models. This allows the system’s intelligent matching and review rules to evolve and become more accurate as more projects are implemented. The system is no longer a fixed tool, but a “growing” intelligent agent that can transform scattered project experience into reusable structured knowledge and automation capabilities for enterprises, achieving a leap from single-point efficiency improvement to overall management system optimization.

[0022] 6. The system corresponding to this invention adopts a modular design with clear responsibilities for each functional module. It achieves efficient and orderly communication and process scheduling through a unified system collaboration and data bus module. This architecture not only ensures the stable and automated execution of complex processes from data aggregation, intelligent parsing, matching optimization to review and output, but also makes the system easy to integrate with existing BIM platforms, enterprise ERP systems, IoT platforms, etc., and has good engineering practicality and promotion value. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall structure of the automatic compilation and review method for bill of quantities based on industrial big data of the present invention. Figure 2 A schematic diagram of the data aggregation and processing architecture of the industrial big data pool constructed in step S1 of the present invention; Figure 3This is a schematic diagram illustrating the principle of the intelligent parsing and three-dimensional spatial quantity calculation benchmark verification process in step S2 of the present invention. Detailed Implementation

[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0025] Example 1: General Implementation Process of Automatic Bill of Quantities Compilation and Review Method Based on Industrial Big Data Step S1: Construct an industrial big data pool Build an industrial big data pool to aggregate data from the following heterogeneous data sources in real time or with a delay not exceeding a preset time threshold: IoT sensors for engineering materials, BIM design model databases for projects, historical engineering databases, and market price information platforms.

[0026] The following processing is performed on heterogeneous data sources: Add contextual annotations to IoT sensor data; Standardized cleaning and reconstruction of historical engineering data based on a unified project breakdown structure; A time series analysis model is used to perform trend filtering and outlier removal on market price information.

[0027] Step S2: Intelligent parsing and initial compilation The design documents and tender documents of the project to be prepared are analyzed using a natural language processing model. The project feature structure tree and key process parameters are extracted, and the BIM model data in the industrial big data pool are combined to perform three-dimensional spatial quantity calculation benchmark verification to generate the initial bill of quantities items.

[0028] Specifically, the three-dimensional spatial quantity calculation benchmark verification involves spatial mapping and comparison between the list items obtained from natural language parsing and the component quantities automatically extracted from the BIM model. When the deviation exceeds a preset threshold, a manual review reminder is triggered, and the reason for the deviation is recorded in the industrial big data pool for optimization of subsequent parsing models.

[0029] Step S3: Intelligent Data Matching and List Optimization The intelligent matching engine is invoked to perform multi-dimensional similarity matching between the initial bill of quantities items and historical engineering data in the industrial big data pool. The multi-dimensional similarity includes structural feature dimension, process and method dimension, and material specification dimension. Based on the matching results, the engineering quantity and project feature description of the initial bill of quantities items are adaptively modified, and the comprehensive unit price and resource consumption index of historical projects are associated and recommended.

[0030] The intelligent matching engine performs multi-dimensional similarity matching, including the following processes: 1. Based on the feature structure tree of the current project, filter out a set of candidate projects from the historical project database; 2. Using a pre-trained engineering semantic model, semantic vectorization is performed on the corresponding descriptions of the current list item and the candidate item set, and a matching index reflecting the semantic correlation between the two is generated. 3. Based on the comprehensive semantic relevance matching index, the ratio of engineering quantities, and the matching degree of process and method, output the comprehensive matching confidence score; 4. Only when the overall matching confidence level is higher than the set threshold will the historical data of the associated recommendations be used to correct and complete the current list items.

[0031] The engineering semantic model is a Sentence-BERT model fine-tuned with engineering domain corpus. During the fine-tuning process, paired synonymous or near-synonymous engineering description texts extracted from historical engineering databases are used as positive samples, and randomly paired different engineering description texts are used as negative samples. A contrastive learning loss function is used to fine-tune the model to optimize its vector representation, maximizing the cosine similarity between positive samples and minimizing the cosine similarity between negative samples. Step S3a: Dynamic adjustment of unit price and establishment of benchmark Based on real-time market price information in the industrial big data pool, the comprehensive unit price of the historical projects recommended in step S3 is dynamically adjusted to generate a benchmark unit price that conforms to the current market conditions. This benchmark unit price is then used as the comparison benchmark for the unbalanced pricing risk warning in step S4.

[0032] Step S4: Dynamic Intelligent Review and Risk Warning A dynamic audit rule base is constructed based on a reinforcement learning model. The training data for the reinforcement learning model includes historical audit records, change visa data, and final settlement data. The dynamic audit rule base performs real-time compliance review, logical contradiction detection, and unbalanced pricing risk warning on the list items generated in step S3.

[0033] The process of building a dynamic audit rule base includes: 1. Based on decisions and modification suggestions in historical audit records, basic compliance rules are generated through supervised learning; 2. Based on the correlation analysis between change visa data and corresponding list items in historical projects, potential risk correlation patterns are discovered through unsupervised learning to form logical contradiction detection rules; 3. Based on the retrospective comparison between the final settlement data and the winning bid list data, the triggering conditions for unbalanced pricing risk warning are dynamically optimized through a reinforcement learning model.

[0034] Step S5: Output and Report Generation Output the final, approved bill of quantities and simultaneously generate a report on the bill of quantities preparation and review process. The report visually displays the matching sources of key items, the review basis, and risk warnings. Step S6: Closed-loop optimization The actual resource consumption data collected by IoT sensors during the actual construction process and verified, as well as the final list adjustment data determined during the project settlement stage, will be fed back to the industrial big data pool as feedback data for incremental updates to the historical project database and training of reinforcement learning models.

[0035] Example 2: Application of the method in a specific engineering project This embodiment uses a large commercial complex project in East China as an example to demonstrate the specific implementation process of the above method in a real-world scenario.

[0036] Project Information: Project code HD-CS-2023-001, is a frame-core tube structure, with a building area of ​​approximately 120,000 square meters.

[0037] Step S1 is implemented in detail as follows: Real-time data collection of material consumption data, such as steel bars and concrete, is achieved through RFID and weight sensors deployed on-site. Integrate a BIM model with LOD 350 precision created using Autodesk Revit; Extract data from 50 similar historical projects from the enterprise engineering data center; Regularly capture material prices and labor indices published on engineering cost information websites and Mysteel.com.

[0038] In terms of data processing, the system automatically adds working condition labels to sensor data, such as "foundation slab pouring stage - C35P8 commercial concrete - pumping". Historical data is standardized and cleaned, and the names and units of measurement of list items are unified. Market prices are filtered using the Holt-Winters exponential smoothing method.

[0039] Step S2 is implemented in detail as follows: The system loads the project's PDF drawings and tender documents, and extracts project features such as "Structure Type: Frame-Core Tube" and "Seismic Resistance Level: Level II" through a pre-trained engineering semantic parsing model, forming a feature structure tree. At the same time, the total volume of "rectangular columns" extracted from the BIM model is 8673 cubic meters, while the NLP parsing value is 8500 cubic meters, a deviation of about 2.0%, exceeding the preset 1.5% threshold, triggering manual review. The engineer confirms that the BIM data is correct, and the system records this deviation case for subsequent model optimization.

[0040] Step S3 is implemented in detail as follows: The system selected eight similar historical projects as a candidate set based on project characteristics. For the item "rectangular column, C50, cross-sectional perimeter within 1.8m", the semantic matching degree was calculated using a fine-tuned Sentence-BERT model, reaching a maximum of 0.92. Combining the proportion of engineering quantities and the matching degree of processes, the overall matching confidence level was 0.85, exceeding the threshold of 0.80. The system automatically supplemented the project characteristics from the best-matching historical items to "rectangular column, concrete strength grade C50, cross-sectional perimeter within 1.8m, pumping, curing", and recommended a historical comprehensive unit price of 852.36 yuan / cubic meter.

[0041] Step S3a is implemented in detail as follows: Based on real-time market prices, the cost of concrete and labor has increased by about 5%. The system will dynamically adjust the historical unit price to a benchmark unit price of 895 yuan / cubic meter as a benchmark for subsequent risk warnings.

[0042] Step S4 is implemented in detail as follows: The dynamic audit rule base scans the list. The system finds that the pipe diameter specification is not specified in the "Electrical Piping" item, triggering a "Project Features Incomplete" warning. At the same time, the system detects that a certain bid price of 1200 yuan / cubic meter deviates from the benchmark unit price of 895 yuan / cubic meter by more than 20%, triggering an "Unbalanced Bidding Risk Warning".

[0043] Step S5 is implemented in detail as follows: The system outputs the final list and generates a visual report showing the matching source, benchmark unit price, quotation deviation, and risk level of the "Rectangular Column C50" item.

[0044] Step S6 is implemented in detail as follows: During project construction, the actual concrete consumption data transmitted by the Internet of Things was 3% lower than the bill of quantities. During the settlement phase, some bill of quantities items were reduced. This data was fed back to the industrial big data pool to update the historical database and optimize the audit model.

[0045] Example 3: System Architecture and Module Collaborative Implementation This embodiment corresponds to the system architecture and demonstrates the specific deployment and module collaboration workflow of the system.

[0046] The system is deployed using a cloud-based microservice architecture, with the following modules: Industrial Big Data Aggregation and Management Module: Based on Apache NiFi, it enables multi-source data access and cleaning, and stores the data on Alibaba Cloud MaxCompute.

[0047] Intelligent parsing and initial compilation module: It is equipped with an engineering semantic parsing model deployed with TensorFlow Serving, receives documents through a RESTful API, and integrates BIM interface to realize 3D verification.

[0048] Data intelligent matching and list optimization module: It includes a candidate project screening unit based on Elasticsearch, a Sentence-BERT semantic model unit, a confidence calculation and decision-making unit, and a unit price dynamic adjustment sub-unit.

[0049] Dynamic intelligent review and risk warning module: integrates a dynamic rule base based on reinforcement learning, including a rule generation unit, a pattern mining unit, and a rule dynamic update unit.

[0050] List output and visualization report module: Developed based on Vue.js and Node.js, it supports list export and report visualization.

[0051] System collaboration and data bus module: Apache Kafka message queues are used to achieve decoupled communication between modules. The specific process is as follows: 1. The intelligent parsing module publishes the initial list to the specified topic; 2. The data bus distributes it to the matching module; 3. After the matching module completes its processing, release the optimization list; 4. Subscribe to and perform reviews in the review module, and publish the review results; 5. The data bus generates or triggers manual review based on the risk level control report.

[0052] Example 4: Experimental Verification and Data Analysis To verify the effectiveness of the method of the present invention, the following control experiment was conducted: Experimental setup: Experimental group: using the method and system of this invention; Control group A: Traditional manual compilation combined with quota software; Control group B: A basic system with only basic BIM quantity calculation and simple historical query functions.

[0053] The test projects consisted of three newly built residential projects, P1, P2, and P3, that were not included in the training.

[0054] Experimental Data Recording Table

[0055] Experimental data comprehensively verify the significant advantages of this invention in the preparation and review of bills of quantities, as detailed below: In terms of efficiency improvement, the experimental group took an average of only 8 person-days to compile, which is far lower than the 25 person-days of traditional manual compilation and the 19 person-days of the primary system. This directly proves that the synergistic effect formed by the present invention through automated analysis, intelligent matching and real-time review can replace a large amount of repetitive and experience-dependent manual labor, and improve the overall compilation efficiency by more than 60%.

[0056] In terms of compilation quality, the experimental group showed a comprehensive improvement, with a project feature description completeness rate of 98.5%. The fundamental reason for this is that the intelligent matching engine can proactively extract and complete project feature descriptions from standardized historical engineering data, effectively overcoming the subjectivity and arbitrariness of manual descriptions. The detection rate of obvious errors in engineering quantities reached 100%, which is entirely due to the three-dimensional spatial quantity calculation benchmark verification mechanism. This mechanism uses natural language parsing results and geometric engineering quantities extracted from the BIM model for mandatory cross-verification, eliminating hard errors in engineering quantities caused by misreading drawings or calculation mistakes from the source. The first-time review pass rate of the compilation results was as high as 95%, which fully demonstrates the proactive risk interception capability of the dynamic review rule base. This rule base not only encodes explicit review experience but also mines hidden risk patterns from historical change and settlement data, thereby discovering and warning of compliance and logical issues in advance during the compilation stage.

[0057] In terms of intelligent risk prevention and control, the experimental group achieved a 92% accuracy rate in risk warning of unbalanced pricing. This result is significant, as it demonstrates the effectiveness of the real-time market price dynamic adjustment mechanism and reinforcement learning optimization strategy introduced in this invention. By combining historical recommended unit prices with real-time market conditions, the system generates dynamically updated benchmark unit prices, providing a more accurate and timely comparison benchmark for risk warning. At the same time, the reinforcement learning model can continuously optimize the warning triggering conditions based on actual project settlement feedback, enabling the risk identification strategy to have adaptive evolution capabilities.

[0058] In terms of intelligent system evolution, the closed-loop optimization mechanism demonstrates long-term value. Experimental data shows that after multiple rounds of project data feedback, the matching confidence of the core intelligent matching engine of the system has increased by an average of 15%. This indicates that the "data feedback-model iteration" closed loop designed in this invention is effective. The system is no longer a static software tool, but an intelligent agent that can continuously learn from practice, accumulate knowledge, and optimize itself. This "the more you use it, the more accurate it becomes" characteristic makes the application value of the system continuously increase over time.

[0059] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A method for automatic compilation and review of bill of quantities based on industrial big data, characterized in that: Includes the following steps: S1: Construct an industrial big data pool, which aggregates heterogeneous data sources from engineering material IoT sensors, project BIM design model databases, historical engineering databases, and market price information platforms in real time or with a delay not exceeding a preset time threshold. S2: Based on the natural language processing model, the design documents and tender documents of the project to be compiled are parsed, the project feature structure tree and key process parameters are extracted, and the BIM model data in the industrial big data pool are combined to perform three-dimensional spatial quantity calculation benchmark verification and generate the initial bill of quantities items. S3: Call the intelligent matching engine to perform multi-dimensional similarity matching between the initial bill of quantities items and the historical engineering data in the industrial big data pool. The multi-dimensional similarity includes structural feature dimension, process and method dimension, and material specification dimension. Based on the matching results, adaptively modify the engineering quantity and project feature description of the initial bill of quantities items, and associate and recommend the comprehensive unit price and resource consumption index of historical projects. S3a: Based on real-time market price information in the industrial big data pool, dynamically adjust the comprehensive unit price of the historical projects recommended in step S3, generate a benchmark unit price that conforms to the current market conditions, and use the benchmark unit price as the comparison benchmark for the unbalanced pricing risk warning in step S4; S4: Construct a dynamic audit rule base based on a reinforcement learning model. The training data for the reinforcement learning model includes historical audit records, change visa data, and final settlement data. The dynamic audit rule base performs real-time compliance review, logical contradiction detection, and unbalanced pricing risk warning on the list items generated in step S3. S5: Output the final bill of quantities after review and simultaneously generate a report on the bill of quantities preparation and review process. The report visually displays the matching source of key items, the basis for review, and risk warnings.

2. The method according to claim 1, characterized in that, The processing of heterogeneous data sources in step S1 includes: labeling IoT sensor data with operating condition context; performing standardized cleaning and reconstruction of historical engineering data based on a unified project decomposition structure; and using a time series analysis model to perform trend filtering and outlier removal on market price information.

3. The method according to claim 1, characterized in that, The specific steps of the three-dimensional spatial quantity calculation benchmark verification in step S2 are as follows: spatial mapping and comparison are performed between the list items obtained from natural language parsing and the component engineering quantities automatically extracted from the BIM model. When the deviation exceeds the preset threshold, a manual review reminder is triggered and the reason for the deviation is recorded in the industrial big data pool for optimization of subsequent parsing models.

4. The method according to claim 1, characterized in that, The process of multi-dimensional similarity matching performed by the intelligent matching engine in step S3 includes: S31: Based on the feature structure tree of the current project, filter out the candidate project set from the historical project database; S32: Using a pre-trained engineering semantic model, semantic vectorization is performed on the corresponding descriptions of the current list item and the candidate item set, and a matching index reflecting the semantic correlation between the two is generated. The engineering semantic model is a Sentence-BERT model fine-tuned with engineering domain corpus. The fine-tuning process includes: using paired synonymous or near-synonymous engineering description texts extracted from historical engineering databases as positive samples, and randomly paired different engineering description texts as negative samples, and using a contrastive learning loss function to fine-tune the model to optimize its vector representation, so as to maximize the cosine similarity between positive samples and minimize the cosine similarity between negative samples. S33: Based on the comprehensive semantic relevance matching index, the proportion of engineering quantities, and the matching degree of process and method, output a comprehensive matching confidence score; S34: Only when the overall matching confidence level is higher than the set threshold, use historical data of related recommendations to correct and complete the current list items.

5. The method according to claim 1, characterized in that, The process of constructing the dynamic audit rule base in step S4 includes: S41: Based on decisions and modification suggestions in historical audit records, generate basic compliance rules through supervised learning; S42: Based on the correlation analysis between change visa data and corresponding list items in historical projects, potential risk correlation patterns are mined through unsupervised learning to form logical contradiction detection rules; S43: Based on the retrospective comparison between the final settlement data and the winning bid list data, the triggering conditions for unbalanced pricing risk warning are dynamically optimized through a reinforcement learning model.

6. The method according to claim 1, characterized in that, The step S5 is followed by a closed-loop optimization step S6: the actual resource consumption data collected by IoT sensors during the actual construction process and the final list adjustment data determined in the project settlement stage are fed back to the industrial big data pool as feedback data streams for incrementally updating the historical project database and training the reinforcement learning model.

7. A system for implementing the automatic compilation and review method of bill of quantities based on industrial big data as described in any one of claims 1-6, characterized in that, include: The industrial big data aggregation and management module is used to build and maintain the industrial big data pool, and to realize the access, cleaning, fusion and storage of multi-source heterogeneous data; The intelligent parsing and initial compilation module integrates a natural language processing model and a BIM interface. It is used to parse the design documents and tender documents of the project to be compiled, extract the project feature structure tree and key process parameters, and perform three-dimensional spatial quantity calculation benchmark verification in combination with BIM model data to generate initial bill of quantities items. The data intelligent matching and list optimization module integrates the intelligent matching engine, which is used to perform multi-dimensional similarity matching between the initial bill of quantities items and historical engineering data in the industrial big data pool, and adaptively correct the engineering quantity, project feature description, and associated recommendation of the comprehensive unit price and resource consumption index of historical projects based on the matching results. The dynamic intelligent audit and risk warning module integrates the dynamic audit rule base based on the reinforcement learning model, which is used to conduct real-time compliance review, logical contradiction detection and unbalanced pricing risk warning on the optimized list items; The Bill of Quantities Output and Visual Reporting module is used to output the final, approved Bill of Quantities and simultaneously generate a report on the Bill of Quantities preparation and review process.

8. The system according to claim 7, characterized in that, The industrial big data aggregation and management module includes a working condition context labeling unit, a data standardization and cleaning unit, and a market information processing unit; the data intelligent matching and list optimization module includes a candidate project screening unit, an engineering semantic model unit, and a confidence calculation and decision-making unit.

9. The system according to claim 7, characterized in that, The dynamic intelligent audit and risk warning module includes a rule generation unit, a pattern mining unit, and a rule dynamic update unit, which are respectively used to generate basic compliance rules based on historical audit records, form logical contradiction detection rules based on risk association patterns mined from change visa data, and dynamically optimize the trigger conditions for unbalanced pricing risk warnings based on final settlement data.

10. The system according to claim 7, characterized in that, The system also includes a system coordination and data bus module; The industrial big data aggregation and management module, intelligent analysis and initial compilation module, data intelligent matching and list optimization module, dynamic intelligent audit and risk warning module, and list output and visualization report module are all connected and exchange data through the system collaboration and data bus module. The system coordination and data bus module is configured as follows: Receive initial bill of quantities items from the intelligent parsing and initial compilation module, and based on preset process rules, distribute the initial bill of quantities items and their associated BIM verification data to the data intelligent matching and bill of quantities optimization module. Receive optimized list items and associated historical data recommendation reliability from the data intelligent matching and list optimization module, and trigger the dynamic intelligent review and risk warning module to load the corresponding review rule subset for review; Receive audit results and risk markers from the dynamic intelligent audit and risk warning module, and based on the level of risk markers, control the list output and visualization report module to generate visualization reports with different levels of detail, or feed back the list items marked with high risk and the audit basis to the intelligent parsing and initial compilation module to initiate the manual review process.

Images