BIM-based dynamic cost allocation calculation method for the entire life cycle of construction projects

Abstract

This invention discloses a BIM-based method for dynamic cost allocation and calculation throughout the entire lifecycle of construction projects. Belonging to the field of cost control technology for the entire lifecycle of construction projects, this method utilizes BIM technology to construct a standardized set of components using a lightweight analytical model. It combines machine learning to mine resource time-series characteristics and deduce cost consumption sequences. By comparing schedule deviations and incorporating engineering changes to correct dynamic costs, it uses a multi-dimensional model to trace the costs of all elements, break down the responsibilities of each party, correct allocation distortions, and iterates the model to form a closed-loop calculation throughout the entire process. This effectively solves the problem of static cost calculation deviations, accurately adapts to dynamic changes in construction, and achieves traceable costs, quantifiable responsibilities, and calibrable allocation. This significantly improves the accuracy of cost calculation and allocation, ensuring the implementation of a closed-loop project cost control system.

Description

Technical Field

[0001] This invention relates to the field of cost control technology for the entire life cycle of construction projects, and in particular discloses a BIM-based method for dynamic allocation and calculation of the entire life cycle cost of construction projects. Background Technology

[0002] Cost control throughout the entire lifecycle of construction projects is a core element in controlling project investment, clarifying cost responsibilities among participating parties, and reducing settlement disputes. Utilizing BIM models for dynamic cost calculation and multi-party cost allocation is a crucial direction for the industry's digital transformation. A search reveals that existing BIM cost accounting-related patent applications (CN114969888A and CN112785257A) only rely on BIM for static quantity surveying and periodic cost summarization. They lack lightweight BIM model analysis and standardized component cleaning and classification, and are deficient in an integrated full-cycle calculation mechanism that includes machine learning-based temporal feature mining, dynamic cost adjustment for schedule deviations, precise matching of changed quantities, reverse tracing of all cost elements, dynamic cost allocation based on performance ratio, and closed-loop iterative optimization of the model.

[0003] The shortcomings and corresponding consequences of existing technologies are as follows: First, directly loading a complete BIM model for quantity calculation without lightweight analysis, deduplication of duplicate components, and cleaning of invalid data leads to loading lag and low calculation efficiency in large projects, and lacks a unified mapping standard for the physical and process attributes of components. Second, using only a fixed-quota static pricing method fails to uncover the phased resource input patterns of components, cannot identify the impact of price adjustments due to project delays, and makes it difficult to accurately match design changes and on-site visa quantities with components, resulting in delayed and distorted cost calculation results. Third, cost data lacks multi-dimensional reverse traceability. First, the source link, material costs, labor costs, and machinery costs cannot be linked to corresponding components, construction logs, and participating entities, making it difficult to define the boundaries of responsibility for each party when cost overruns occur. Second, the use of a fixed ratio for simple cost allocation without considering the contribution to the project schedule, penalties for breach of contract, and dynamic accounting of the responsibilities of each entity for quality assurance is prone to distortion in cost allocation and prominent contradictions in settlement and reconciliation. Third, the lack of a closed-loop mechanism for iteratively updating the benchmark budget and schedule model with the calculated data makes it impossible to continuously optimize the accuracy of subsequent project cost predictions. It can only achieve static accounting after completion and lacks the ability to dynamically allocate and control costs throughout the entire construction process.

[0004] Therefore, how to extrapolate the full-cycle resource consumption based on lightweight BIM component data, dynamically adjust the cost based on schedule deviations and engineering changes, achieve accurate cost allocation for multiple participating parties, and continuously optimize the calculation benchmark based on historical calculation samples to build a full-process closed-loop dynamic cost allocation calculation system is a key technical problem that urgently needs to be solved in the field of digital cost control of engineering projects. Summary of the Invention

[0005] This invention provides a BIM-based method for dynamic cost allocation and calculation throughout the entire lifecycle of construction projects. It aims to address the technical shortcomings of existing construction project cost calculation methods, which are mostly static and difficult to adapt to dynamic scenarios such as construction progress deviations and engineering changes. These methods suffer from inaccurate cost allocation, unclear cost traceability, unclear division of responsibilities, poor calculation closure, and inability to accurately achieve dynamic cost control throughout the entire lifecycle.

[0006] This invention relates to a BIM-based method for dynamically allocating and calculating the total lifecycle cost of a construction project, comprising the following steps:

[0007] S100. Perform lightweight analysis on the BIM engineering information model of the construction project, extract basic data of engineering components, establish the mapping relationship between components and multi-dimensional attributes based on the physical attributes and construction process attributes of each component, and construct a standardized first component set after deduplication of duplicate components and cleaning of invalid data.

[0008] S200. Import the project's preset baseline schedule data. Based on the first component set, use a pre-trained machine learning feature extraction model to mine the resource input time sequence characteristics of each component at different construction stages. Combine the engineering pricing rules to conduct a full life cycle value consumption deduction and generate the first value consumption sequence.

[0009] S300. Compare the time nodes of each resource consumption in the first value consumption sequence with the preset benchmark progress data to identify the progress deviation status of component resource consumption. When there is a progress deviation, extract the multi-dimensional structured first time-series difference features.

[0010] S400: Obtain incremental data of engineering quantity changes corresponding to design changes and on-site visas; classify and match changed engineering quantities based on BIM component codes; adjust the price for schedule lag based on the first time-series difference characteristics; and obtain the first dynamic cost after the budget rate is adapted to dynamic pricing.

[0011] S500 uses BIM engineering quantities, construction logs, and subcontracting contracts as traceability data sources. Through a preset multi-dimensional cost traceability model, it reverse-traces the cost of all elements including material costs, labor costs, machinery usage fees, and management fees in the first dynamic cost, and generates a multi-level first traceability path that binds the performance information of each participating party.

[0012] S600. Based on the scope of duties and contribution ratio of the responsible entity corresponding to the first traceability path, the first dynamic cost is broken down into multiple cost components. After deducting the penalty for breach of contract and the amount reserved for quality assurance, the first responsibility cost corresponding to each responsible entity is calculated.

[0013] S700: Compare the primary responsibility cost of each responsible entity with the corresponding initial budget amount for each phase to identify cost allocation distortions; when allocation distortion is determined to exist, construct differential allocation weights based on the actual project performance cycle and the degree of overall progress deviation, and redistribute the current period's responsibility cost to obtain the corrected first settlement bill; use the entire process cost, progress, and allocation data as samples to iteratively update the benchmark progress and budget calculation model, and realize dynamic closed-loop allocation calculation of the construction project's full life cycle cost.

[0014] Furthermore, step S100 specifically includes:

[0015] S110. Perform lightweight analysis and 3D bounding box collision compliance verification on the BIM engineering information model, and extract basic parameters such as component quantity, material code, construction section number, planned start time, and planned completion time in batches to construct the basic feature vector corresponding to each component.

[0016] S120. Calculate the cosine similarity of the basic feature vectors of any two components using the cosine similarity algorithm, and identify duplicate components by combining the preset duplicate component similarity threshold; retain the legal reuse scenarios of the same component assigned to different construction sections, and only remove duplicate redundant components and invalid attribute data.

[0017] S130. Using construction sections and professional engineering types as dual search indexes, the cleaned effective component feature vectors are aggregated to construct a standardized first component set with unique BIM codes and attribute search indexes.

[0018] Furthermore, step S200 specifically includes:

[0019] S210. Set a fixed sliding time sequence sampling window based on the planned time nodes of each sub-project within the preset baseline schedule. When the project undergoes major design changes or the overall project duration is extended, the window time span is adaptively adjusted to construct a time sequence sampling time set.

[0020] S220. Use a one-dimensional effective convolution feature extraction operator to perform temporal feature mining on the first component set, and extract the temporal feature vectors of the input of three types of resources: manpower, materials and machinery at each sampling time.

[0021] S230. Based on the engineering quota pricing rules, the resource input at each time point is matched with the current information unit price to calculate the current resource consumption cost. All current resource consumption costs are collected along the sliding time sequence sampling window to generate the first value consumption sequence that represents the capital consumption pattern of the entire project life cycle.

[0022] Furthermore, step S300 specifically includes:

[0023] S310. For each resource consumption node in the first value consumption sequence, the single-node schedule deviation is obtained by subtracting the baseline planned time from the actual occurrence time.

[0024] S320. Compare the absolute value of the single-node schedule deviation with the preset allowable deviation threshold, and extract features only for deviation nodes that exceed the deviation threshold.

[0025] S330, absolute duration of splicing progress lag, relative amount of current period resource over-investment, and process interleaving time offset coefficient are used to construct a multi-dimensional structured first time-series difference feature vector.

[0026] Furthermore, step S400 specifically includes:

[0027] S410. Classify and match the changed quantities of work according to the unique BIM component codes. Based on the first time-series difference feature vector, calculate the first norm of the eigenvector and combine it with the lag price adjustment sensitivity coefficient preset based on the historical full life cycle cost sample to solve the schedule lag price adjustment correction coefficient.

[0028] S420. Retrieve the official unit prices of labor, materials and machinery for the current period and calculate the total cost of the current period's change based on the resource consumption adjusted by the progress lag price adjustment factor.

[0029] S430, the project's phased base contract cost and the current period's total change cost are combined and then summarized after non-negative constraint verification to obtain the project's current period's first dynamic cost.

[0030] Furthermore, step S500 specifically includes:

[0031] S510. Using three types of data—BIM component quantities, time-series construction logs, and subcontracting contracts—as the basic data sources for traceability, a multi-dimensional rule-matching traceability mapping function is used to locate the unique code of the responsible party for each expense in the first dynamic cost.

[0032] S520. Construct a multi-level reverse traceability link according to the time sequence of cost occurrence and the level of responsibility of participating construction units. Encapsulate the unique code of the responsible entity for cost, the code of each level of participating construction unit, the time of cost occurrence, and the amount of individual cost into a structured first traceability path.

[0033] Furthermore, step S600 specifically includes:

[0034] S610. Based on the first traceability path, the actual completed work volume and the contracted planned work volume of each responsible entity are statistically analyzed, and a minimum zero constant is introduced to calculate the performance rate of the responsible entity.

[0035] S620. Normalize the project schedule performance rate of all responsible parties, and use the normalized project schedule performance rate as the cost allocation weight to decompose the first dynamic cost of the project level by level to obtain the initial allocated cost of each responsible party.

[0036] S630. Based on the initial allocated cost, deduct the current period's default deductions and warranty reserve amount, and ensure that the settlement cost is non-negative through maximum value constraints, to calculate the first liability cost corresponding to each responsible party.

[0037] Furthermore, step S700 specifically includes:

[0038] S710. Calculate the cost deviation rate of each responsible entity's current period responsibility cost relative to the phased budget amount, and compare the absolute value of the cost deviation rate with the preset allocation distortion deviation rate threshold to determine whether there is cost allocation distortion in the current period responsibility cost.

[0039] S720. Set the performance cycle weight coefficient and schedule deviation weight coefficient, and calculate the first difference apportionment weight of each responsible party by combining the actual performance period of the responsible party and the overall schedule deviation.

[0040] S730. First, calculate the total over-budget amount of the project in the current period, constrain the total over-budget amount to be non-negative, and redistribute the total over-budget amount among the responsible entities based on the first difference allocation weight. The over-budget amount is then added to the current budget amount to obtain the final settlement cost of each responsible entity after distortion correction. The final settlement costs of all responsible entities together constitute the first settlement bill of the project.

[0041] S740. Using the project schedule deviation, change costs, multi-party allocation weights, and settlement bill data as training samples, the gradient descent algorithm is used to iteratively update the weight parameters of the baseline schedule and budget calculation model, thereby realizing iterative optimization of the enterprise cost database and closed-loop management of project cost throughout its entire life cycle.

[0042] The beneficial effects achieved by this invention are as follows:

[0043] This invention provides a BIM-based dynamic cost allocation calculation method for the entire lifecycle of construction projects. Addressing the technical problems of existing construction project cost calculations, which are mostly static and difficult to adapt to dynamic scenarios such as construction schedule deviations and engineering changes, this method suffers from distorted cost allocation, unclear cost traceability, ambiguous responsibility division, poor calculation closure, and inability to accurately achieve dynamic cost control throughout the entire lifecycle, this invention utilizes BIM technology to construct a standardized set of components using a lightweight analytical model. It combines machine learning to mine resource time-series characteristics and deduce cost consumption sequences. By comparing schedule deviations and correcting dynamic costs based on engineering changes, it uses a multi-dimensional model to trace the costs of all elements, break down the costs of each party's responsibilities, correct the allocation distortion problem, and iterates the model to form a closed-loop calculation throughout the entire process. This effectively solves the problem of static cost calculation deviations, accurately adapts to dynamic changes in construction, achieves traceable costs, quantifiable responsibilities, and calibrable allocation, significantly improves the accuracy of cost calculation and allocation, and ensures the implementation of closed-loop project cost control. The specific beneficial effects achieved are as follows:

[0044] 1. This invention performs lightweight analysis on BIM engineering models. By mapping multi-dimensional attributes of components, removing duplicate components, and cleaning invalid data, a standardized set of components is constructed. This effectively solves the problems of redundant model data, messy component information, and low data standardization in traditional BIM cost estimation. It achieves the normalization of basic engineering data and provides an accurate and unified data foundation for subsequent time-series cost extrapolation and dynamic allocation calculation.

[0045] 2. This invention combines project baseline schedule data and relies on machine learning feature extraction models to mine the time-series characteristics of resource input across construction stages of components. It combines engineering pricing rules to complete the full life cycle value consumption extrapolation, breaking through the limitations of traditional static cost accounting that only focuses on a single construction node and ignores the differences in time-series resource input. It realizes the full-process, dynamic prediction and extrapolation of cost consumption.

[0046] 3. This invention accurately identifies deviations in construction progress and extracts structured time-series differences by comparing resource consumption time nodes with baseline progress. It can effectively capture hidden progress disturbances such as project delays and work process misalignments, solving the shortcomings of traditional cost estimation that cannot correlate with progress deviations and has delayed cost corrections, and providing accurate deviation basis for subsequent dynamic price adjustments.

[0047] 4. This invention relies on BIM component coding to achieve accurate classification and matching of incremental engineering quantities such as design changes and on-site visas. It combines the characteristics of schedule time difference to complete the delayed price adjustment correction and dynamic pricing processing, realizing the coupled cost correction of changed engineering quantities and schedule deviations. It gets rid of the problem of the traditional change pricing model being rigid and unable to adapt to the fluctuation of the construction period, and improves the fit and rationality of dynamic cost accounting.

[0048] 5. This invention uses BIM engineering quantities, construction logs, and subcontracting contracts as multi-dimensional traceability data sources. Through a cost traceability model, it achieves reverse traceability of costs for all elements, including labor, materials, machinery, and management. It constructs a multi-level traceability path that binds the performance information of participating parties, solving the pain points of vague traceability links and lack of basis for responsibility division in traditional cost accounting. This enables the source of cost to be traceable and the responsibility to be determined.

[0049] 6. This invention breaks down the costs of multiple parties based on their scope of duties and contribution ratios, and accurately calculates the costs of each party's responsibilities by combining default penalties and warranty reservation rules. This achieves a deep link between cost allocation and performance, changes the irrationality of traditional average allocation and fixed-ratio allocation models, and ensures the fairness and relevance of cost allocation among multiple parties.

[0050] 7. This invention identifies distorted cost allocation by comparing actual cost with phased budgets, constructs differentiated allocation weights based on project performance cycle and schedule deviation, completes secondary cost correction and settlement optimization, and simultaneously updates the baseline schedule and budget model through full-process data iteration, constructing a dynamic closed-loop cost calculation system for the entire lifecycle. This invention connects the entire business chain of BIM model data, construction progress, change pricing, responsibility allocation, and model iteration, solving the industry problems of static and fixed cost calculation, poor change adaptation, ambiguous responsibility allocation, and distorted settlement in traditional construction project cost calculation. It significantly improves the intelligence, dynamism, and accuracy of full-cycle cost control in construction projects, and has strong engineering adaptability and practicality. Attached Figure Description

[0051] Figure 1 This is a flowchart illustrating an embodiment of the BIM-based dynamic cost allocation calculation method for the entire lifecycle of construction projects according to the present invention.

[0052] Figure 2 This is a detailed flowchart illustrating an embodiment of step S100 in the BIM-based dynamic cost allocation calculation method for the entire lifecycle of a construction project according to the present invention. Detailed Implementation

[0053] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0054] like Figure 1As shown, the first embodiment of this invention proposes a BIM-based dynamic cost allocation calculation method for the entire lifecycle of construction projects. This method is applicable to dynamic cost management throughout the entire lifecycle of various EPC and general contracting construction projects, including building construction, municipal engineering, rail transit, and water conservancy. It addresses industry pain points such as traditional static cost accounting, delayed change order adjustments, lack of basis for cost allocation among multiple parties, and the inability to achieve closed-loop self-optimization of cost data. Through standardized processing of lightweight BIM components, machine learning-based time-series resource consumption projection, dynamic cost adjustment for schedule deviations, precise matching of changed work quantities, multi-dimensional cost reverse tracing, automatic decomposition of multi-party responsibility costs, reconstructed settlement of distorted allocation weights, and a full-process sample iterative model, it achieves dynamic, traceable, closed-loop, and iterative intelligent cost allocation calculation for construction projects throughout the entire lifecycle from design, construction, completion to quality assurance. The method includes the following steps:

[0055] Step S100: Perform lightweight analysis on the BIM engineering information model of the construction project, extract basic data of engineering components, establish the mapping relationship between components and multi-dimensional attributes based on the physical attributes and construction process attributes of each component, and construct a standardized first component set after deduplication of duplicate components and cleaning of invalid data.

[0056] The project's complete IFC standard BIM engineering information model is read, and lightweight mesh simplification and redundant attribute stripping algorithms are used to complete the lightweight analysis of the model, eliminating redundant geometric patches and non-cost-related auxiliary attributes. Basic geometric and measurement data of all engineering components, including walls, beams, slabs, columns, pipelines, and equipment, are extracted in batches. Physical attributes (dimensions, materials, volume, self-weight) and construction process attributes (construction section, process number, planned start / completion nodes, work team) are bound to each type of component, establishing a structured mapping relationship between unique component codes and multi-dimensional attributes. All component datasets are traversed, and duplicate components with identical geometric parameters and attributes are merged and deduplicated. Invalid dirty data lacking measurement parameters or construction process bindings are removed, and standardized field formats are unified to construct a standardized first component set that can be directly used for time-series cost estimation.

[0057] Lightweight BIM engineering information model parsing involves stripping redundant geometric and auxiliary attributes from the BIM model, such as rendering and decoration, that are irrelevant to cost estimation, retaining only core data related to quantity measurement and construction procedures. Basic component data consists of fundamental parameters for pricing and measurement of a single type of component, including core measurement fields such as component code, geometric dimensions, quantity, and component type. Multidimensional attribute mapping relationships for components use a structured data table mapping rule with the component's unique code as the primary key, linking physical attributes and construction procedure attributes. Duplicate component deduplication involves identifying components of the same type with identical parameters and procedures, merging duplicate samples, and compressing data volumes. Invalid data cleaning removes abnormal component data with missing parameters, unbound procedures, or those unsuitable for pricing. The standardized first component set is a standardized dataset of all project components after lightweighting, deduplication, cleaning, and attribute mapping, providing a unified input data source for cost time-series projection.

[0058] Step S200: Import the project's preset baseline schedule data. Based on the first component set, use a pre-trained machine learning feature extraction model to mine the resource input time sequence characteristics of each component at different construction stages. Combine the engineering pricing rules to conduct a full life cycle value consumption deduction and generate the first value consumption sequence.

[0059] The project imports the overall control plan and segmented construction plan prepared in the early stage to form the preset benchmark progress data. The benchmark progress includes the start and end times of each construction stage and the planned resource input quota. The standardized first component set is input into the machine learning feature extraction model that has been trained offline through a large number of historical project progress-cost samples. The model automatically mines the time sequence changes of material, labor and mechanical resource input in each stage of foundation, main structure, decoration and installation for each type of component. Based on national standards, local engineering pricing quotas and fee rules, combined with the resource input of components in each stage, the project's total funds and resource consumption amount from the start to the warranty period is deduced node by node. The cumulative and current consumption costs of each node are stored in a time sequence structure, generating a continuous time sequence of first value consumption.

[0060] The preset baseline schedule data consists of the overall project construction control plan and segmented work plan, including the planned construction periods and standard resource input quotas for each component; the machine learning feature extraction model is a time-series feature mining model trained on massive historical project component-schedule-cost samples, used to automatically identify the phased resource input patterns of components; the resource input time-series features are the time-series patterns of material, labor, and machinery input for the same component at different construction stages; the full life-cycle value consumption projection is a numerical projection calculation that combines construction time sequence and pricing rules to predict the cost consumption at each time node of the entire project life cycle; the first value consumption sequence is a standardized time-series dataset that is sorted chronologically and records the current and cumulative cost consumption for each period.

[0061] Step S300: Compare the time nodes of each resource consumption in the first value consumption sequence with the preset baseline progress data to identify the progress deviation status of component resource consumption. When there is a progress deviation, extract the multi-dimensional structured first time-series difference features.

[0062] The process iterates through each resource consumption time node within the first value consumption sequence, performing frame-by-frame temporal matching and comparison between the actual construction period and resource input time of the component corresponding to each resource consumption time node and the planned operation time within the preset baseline schedule. It then compares and determines three types of schedule deviation states: early, late, and stopped. If a late or early deviation is determined, it extracts multi-dimensional quantitative parameters such as deviation duration, deviation operation segment, current over / under input of resources, corresponding cost difference, and affected construction scope. These parameters are integrated into a multi-dimensional structured first-series difference feature, serving as the input basis for subsequent change-based price adjustments. Schedule deviation states refer to the early, late, or stopped deviation states caused by the actual resource input time of component construction relative to the baseline planned operation time. The first-series difference feature is a multi-dimensional structured feature vector that quantitatively records the schedule deviation duration, resource input deviation, cost deviation, and affected operation scope.

[0063] Step S400: Obtain incremental data of engineering quantity changes corresponding to design changes and on-site visas; classify and match changed engineering quantities based on BIM component codes; adjust the pricing for delayed progress based on the first time-series difference characteristics; and obtain the first dynamic cost after adapting the budget rate to dynamic pricing.

[0064] The system receives design change orders and site visa orders online, analyzes the incremental data of added, deleted, and adjusted engineering quantities in the documents, extracts the component types and size parameters corresponding to the changes, and classifies and matches them one by one with the standard components in the first component set through the unified BIM component code to achieve accurate binding of the changed engineering quantities to the corresponding pricing components; retrieves the first time-series difference features generated in step S300, and makes a delayed price adjustment to the basic unit price of the changes based on the progress lag time and resource deviation (compensating for labor idleness, machinery idleness, and material price increase lag losses); combines the current market material price, management fee rate, profit margin, and regulatory fees and taxes to complete the dynamic adaptation pricing, superimposes the benchmark extrapolation cost and the change correction cost, and outputs the overall first dynamic cost of the current project.

[0065] The incremental data for changes in quantities includes measurement data for increases, decreases, and dimensional adjustments in quantities resulting from design changes and on-site approvals; BIM component coding classification and matching uses unified component codes as the primary key to match changed quantities to standard components for accurate pricing; schedule lag price adjustment compensates for idle labor, equipment downtime, and material price differences caused by schedule lags by adjusting the comprehensive unit price of changes; rate-adapted dynamic pricing dynamically calculates the comprehensive unit price of changes by matching the current market prices of labor, materials, and machinery with the current fee standards; and the first dynamic cost is the real-time dynamic total cost of the project in the current period after integrating the benchmark time-series cost, change order price adjustment, and schedule lag compensation.

[0066] Step S500: Using BIM engineering quantities, construction logs, and subcontracting contracts as traceability data sources, the cost of all elements including material costs, labor costs, machinery usage fees, and management fees in the first dynamic cost is traced back through a preset multi-dimensional cost traceability model, generating a multi-level first traceability path that binds the performance information of each participating party.

[0067] Establish a three-source collaborative traceability data source: BIM standardized component quantities, daily on-site construction logs (work teams, equipment shifts, material requisition records), and general contractor / subcontractor / material supplier subcontracting contracts (unit price, scope of work, performance terms). Utilize a pre-built multi-dimensional cost traceability model to break down the first dynamic cost into four cost elements: material costs, labor costs, machinery usage fees, and enterprise management fees. Tracee back layer by layer the components, construction periods, work teams, and suppliers corresponding to each cost consumption. Bind each traceability node to the corresponding general contractor, professional subcontractors, material suppliers, equipment leasing companies, and other participating entities' performance records, completed work quantities, and contract duration information, constructing a hierarchical multi-level first traceability path from total cost → itemized cost → element cost → participating units.

[0068] The multi-dimensional cost traceability model is a mathematical model that uses BIM engineering quantities, construction logs, and subcontracting contracts as three-dimensional traceability basis to achieve reverse decomposition and traceability of all cost elements. The total cost element consists of four core components of engineering cost: material cost, labor cost, machinery usage cost, and management cost. The multi-level primary traceability path is a cost traceability link that is hierarchical from top to bottom and binds the performance information of all participating parties, so that the responsible party can be located for each cost.

[0069] Step S600: Based on the scope of responsibilities and contribution ratio of the responsible entity corresponding to the first traceability path, the first dynamic cost is broken down into multiple cost components. After deducting the penalty for breach of contract and the amount reserved for quality assurance, the first responsibility cost corresponding to each responsible entity is calculated.

[0070] The system reads the contractual responsibilities, actual completed work volume, and contribution percentage of the entire project lifecycle for each participating entity within the multi-level primary traceability path. Based on the contribution weight, it automatically splits the overall primary dynamic cost into multiple components. It retrieves the project contract's preset default penalty rules (penalties for delayed construction, penalties for quality defects, and penalties for safety breaches) and the quality assurance deposit reservation ratio. It then simultaneously deducts the default amount and the current period's quality assurance deposit reservation amount from the corresponding entity's cost breakdown. After the deduction is completed, it separately calculates the current period's primary responsibility cost borne and settled by the general contractor, each professional subcontractor, material supplier, and equipment leasing company.

[0071] The contribution ratio of the responsible entity to the project's schedule performance is the weighted ratio of the actual completed work volume and effective construction period of each participating unit to the total project volume / total project period; the multi-party cost breakdown is the allocation of the overall dynamic project cost to each participating responsible entity according to the weight of the performance contribution; the penalty for breach of contract is the settlement amount deducted for breaches of schedule, quality, and safety as stipulated in the contract; the quality assurance reserve amount is the quality assurance deposit that is temporarily withheld in the current settlement according to a certain proportion and returned without interest after the quality assurance period expires; the primary responsibility cost is the amount of responsibility-shared cost that should be settled by each participating entity in the current period after deducting the penalty for breach of contract and the quality assurance deposit.

[0072] Step S700: Compare the primary responsibility cost of each responsible entity with the corresponding initial budget amount for each phase to identify cost allocation distortion conditions; when allocation distortion is determined to exist, construct differential allocation weights based on the actual project performance cycle and the degree of overall progress deviation, and redistribute the current period's responsibility cost to obtain the corrected first settlement bill; use the entire process cost, progress, and allocation data as samples to iteratively update the benchmark progress and budget calculation model to achieve dynamic closed-loop allocation calculation of the construction project's full life cycle cost.

[0073] The initial phased budgets prepared in the early stages of the project, divided by main body and construction phase, are retrieved. The difference between the current period's primary responsibility cost of each responsible entity and the corresponding phased budget value is compared. If the actual responsibility cost deviates significantly from the budget threshold, the current period's cost allocation is determined to be distorted. Under distorted conditions, a standardized differential allocation weight is constructed by comprehensively considering the overall actual performance period of the project, the cumulative progress deviation time of the entire project, and the lagging contribution of each entity. Based on the weight, the current period's allocated cost of each participating entity is readjusted to eliminate the allocation distortion problem caused by the large budget deviation, and a corrected and compliant first settlement bill is generated. All BIM component data, benchmark progress, value consumption sequence, change and price adjustment records, traceability paths, responsibility allocation, and settlement bills of this entire process are collected as training samples and input into the benchmark progress model and budget calculation model to complete the reverse iterative optimization of parameters, continuously improving the accuracy of subsequent project cost extrapolation and allocation calculation, forming a closed loop of "BIM analysis - time sequence extrapolation - dynamic price adjustment - traceability allocation - settlement correction - model iteration", and completing the dynamic closed loop allocation calculation of the construction project's full life cycle cost.

[0074] The initial phased budget is a phased control budget amount prepared in the early stage of the project according to the participating entities and construction stages; the cost allocation distortion condition is a condition in which the actual responsibility cost of the main entity in the current period deviates from the phased budget by more than a preset threshold, and the original allocation result is distorted and cannot be directly settled; the difference allocation weight is a weight coefficient constructed by combining the project performance cycle and the overall progress deviation, which is used to redistribute the responsibility cost of each entity under the distortion condition; the first settlement bill is the standardized settlement document of each participating entity in the current period after the distortion weight has been redistributed and corrected; the model iteration update uses the current real cost, progress, and allocation sample to back-optimize the benchmark progress and budget calculation model parameters to improve the accuracy of subsequent calculations; the full life cycle cost dynamic closed-loop allocation calculation is a full-link, iterative, and dynamically adjustable cost allocation calculation closed-loop system from BIM data input to model self-optimization.

[0075] The BIM-based dynamic cost allocation calculation method for the entire lifecycle of construction projects provided in this embodiment has the following beneficial effects compared with existing technologies:

[0076] 1. This invention employs lightweight BIM analysis, duplicate component deduplication, and invalid data cleaning to construct a standardized component dataset. It unifies the mapping of component physical and process multi-dimensional attributes, solving the problems of data redundancy, messy component pricing data, and insufficient standardization of pre-cost estimation data in traditional BIM models, and significantly improving the efficiency of time-series cost extrapolation calculation.

[0077] 2. By leveraging machine learning models to mine the phased resource input time-series characteristics of components and combining them with pricing rules to deduce the full-cycle value consumption sequence, the cost can be transformed from traditional static cost accounting upon completion to phased time-series dynamic prediction, which can identify the cost fluctuation risk caused by schedule deviation in advance.

[0078] 3. By accurately matching change order quantities through BIM component coding, and combining the characteristics of schedule differences, the pricing of delayed work and idle losses is adjusted and corrected. With the support of real-time dynamic pricing, the industry pain points of static fixed change order pricing, inability to quantify compensation for delayed work losses, and numerous settlement disputes are solved. The dynamic cost calculation closely matches the actual working conditions on site.

[0079] 4. Construct a multi-dimensional cost traceability model that integrates BIM project quantities, construction logs, and subcontracting contracts to achieve reverse traceability of all elements including material costs, labor costs, machinery costs, and management costs. Generate multi-level traceability paths that bind the performance information of each participating party, ensuring that the cost allocation among multiple parties has complete data traceability basis and that the division of responsibilities is clear and dispute-free.

[0080] 5. Based on the scope of performance and contribution ratio of each entity, the dynamic cost is automatically broken down, and the default deduction and quality assurance reserve are standardized to obtain the cost of each entity's responsibility. For the distorted allocation of work due to budget deviation, the cost is redistributed by constructing a two-dimensional difference allocation weight based on the performance cycle and schedule deviation, and outputting a compliant and corrected settlement bill, which greatly reduces the conflict of settlement among multiple parties.

[0081] 6. Real data on cost, schedule, and allocation throughout the entire process are automatically accumulated as training samples for the model. The baseline schedule and budget calculation model are iteratively optimized in reverse, enabling cost calculation to continuously adapt to changes in project conditions and forming a complete self-optimizing closed loop. This truly realizes dynamic, accurate, traceable, and iterative intelligent cost allocation calculation for the entire lifecycle of a construction project, from commencement to quality assurance.

[0082] Furthermore, the BIM-based dynamic cost allocation calculation method for the entire lifecycle of a construction project provided in this embodiment specifically includes step S100 as follows:

[0083] Step S110: Perform lightweight parsing and 3D bounding box collision compliance verification on the BIM engineering information model, extract basic parameters such as component quantity, material code, construction section number, planned start time, and planned completion time in batches, and construct the basic feature vector corresponding to each component.

[0084] Read the IFC standard BIM engineering information model, perform lightweight simplification of mesh patches, and lightweight analysis calculation of stripping redundant attributes not related to cost; simultaneously carry out 3D bounding box collision compliance verification of components, and eliminate invalid components with geometric collisions and abnormal dimensions; batch extract five basic measurement process parameters for each type of component: engineering quantity, unique material code, construction section number, and planned start / completion time, uniformly normalize the dimensions of each parameter, and arrange them in a fixed dimension order to generate a standardized basic feature vector for a single component.

[0085] The 3D bounding box collision compliance verification is to screen out unqualified component data with geometric conflicts and dimensional anomalies by detecting the spatial intersection of component bounding boxes; the component basic feature vector is a one-dimensional standardized feature vector composed of engineering quantity, material code, contract section, and planned start and end dates, which is used to determine the similarity of duplicate components.

[0086] The formula for constructing the basic feature vector is:

[0087] (1)

[0088] In formula (1), For the first The standardized basic feature vector of each engineering component is dimensionless and a fixed 5-dimensional column vector; it serves as a structured data carrier obtained after lightweight BIM parsing, realizing the regular encapsulation of non-heterogeneous BIM information, and is a unified input data source for subsequent time-series feature mining, resource consumption prediction, and change quantity matching. For the first The quantity of each component, in units or value range It is used for the initial measurement of component costs and the accounting of resource consumption. The component material is coded using discrete integer numerical codes, which are dimensionless; this is used to distinguish different types of building materials and is suitable for quota pricing and material cost classification and traceability accounting. It serves as the construction section number, a dimensionless discrete code, used for batch collection of components under multiple sections of a project, statistical analysis of progress deviations by zone, and division of responsibilities among participating parties. For the first The planned start time for each component is in days, used for aligning the baseline construction schedule and extrapolating the timing of resource input. For the first The planned completion time for each component is in days, used for determining project performance, quantifying schedule deviations, and adjusting costs in case of delays. As a vector transpose operator, the one-dimensional row vector is normalized into a standard column vector, unifying the input dimension format of the machine learning model and avoiding data reading anomalies caused by dimension disorder. The control logic of formula (1) is to perform lightweight analysis and three-dimensional bounding box collision verification on the BIM model, extract five types of original engineering parameters for each component: engineering quantity, material, section, and planned start and completion time; encapsulate the five types of parameters into a one-dimensional array in a fixed order, and normalize them into a 5-dimensional standard column feature vector through vector transpose; based on this structured feature vector, complete the deduplication of duplicate components and the cleaning of invalid and abnormal engineering quantity data, and construct a standardized component dataset. Formula (1) encapsulates and standardizes five types of original BIM attribute parameters—the quantity of work, material code, construction section, planned start time, and planned completion time of a single component—into a standard column vector according to a fixed dimension, thereby realizing the structured conversion of heterogeneous BIM engineering data. It solves the problem that traditional BIM data is difficult to model in batches and time series, and provides a standardized structured data source for subsequent construction resource time series extrapolation, progress deviation identification, dynamic cost pricing, and multi-party cost responsibility tracing and allocation, thus establishing the data foundation for the BIM model to achieve closed-loop calculation of dynamic cost throughout the entire life cycle.

[0089] Step S120: Calculate the cosine similarity of the basic feature vectors of any two components using the cosine similarity algorithm, and identify duplicate components by combining the preset duplicate component similarity threshold; retain the legal reuse scenarios of the same component assigned to different construction sections, and only remove duplicate redundant components and invalid attribute data.

[0090] Traverse all component basic feature vectors and use the cosine similarity algorithm to calculate feature matching similarity for each pair of vectors; preset a similarity threshold for duplicate components; if the similarity between two pairs of components is greater than the preset similarity threshold, they are judged as duplicate redundant components; distinguish between two types of duplicate scenarios: reuse of the same component across different sections is a legal working condition and is retained; duplicate components with completely identical parameters and no reuse value in the same section are merged and removed; simultaneously remove invalid dirty data with missing attributes and blank process parameters to complete the component dataset cleaning.

[0091] Component feature cosine similarity is a dimensionless index that quantifies the degree of matching between the basic features of two sets of components, with a value range of... Legitimate reuse scenarios are compliant working conditions where the same standard component is assigned to multiple different construction sections for repeated use without the need for deletion; redundant components are duplicate component samples from the same construction section with completely identical parameters and no reuse value.

[0092] The formula for calculating the cosine similarity of component features is:

[0093] (2)

[0094] In formula (2), For the first The component and the first The feature cosine similarity of each component, dimensionless, with a range of values. The closer the value is to 1, the higher the consistency of the two components in terms of multi-dimensional attributes such as project quantity, material, contract section, and planned construction period, and the greater the probability of them being identified as duplicate components. , Components ,member The 5-dimensional standardized basic feature vector obtained through BIM analysis. It is the inner product of two feature vectors, used to quantify the overall degree of matching in the same direction across multiple attribute dimensions. , These are the L2 norms (Euclidean norms) of the two eigenvectors, used to normalize the vectors, eliminate the interference of eigenvector magnitudes on the similarity calculation results, and constrain the similarity to within a certain range. Within the range. L2 norm is the L2 vector norm, a mathematical operator used for cosine similarity normalization. It eliminates the interference of eigenvector magnitude differences on the similarity results, ensuring that the similarity value stably falls within a certain range. The interval allows for a unified and fair comparison benchmark for threshold determination. The control logic of formula (2) is to traverse the standardized basic feature vectors of all BIM components, calculate the inner product of each pair of vectors and their respective L2 norms; obtain the cosine similarity by dividing the inner product by the product of the L2 norms of the two sets of vectors, and quantify the similarity of the multidimensional engineering attributes of the two components; compare the calculated similarity with the preset similarity threshold of 0.95 for duplicate components, and if it is higher than the preset similarity threshold for duplicate components, it is initially determined to be a duplicate component; additional business rule verification: if two similar components belong to different construction sections, they are determined to be legally reused components and are not removed; other scenarios are determined to be redundant duplicate components, and data cleaning and deduplication are performed. Formula (2) obtains cosine similarity by performing the inner product of the basic feature vectors of each pair of components and L2 norm normalization, which quantifies the similarity of the multidimensional engineering attributes of different components; combined with the 0.95 similarity threshold, it realizes the automatic identification and cleaning of redundant and duplicate components, and retains the legal reuse data of components through cross-section business rules, which solves the pain points of low efficiency and easy deletion of effective data in traditional BIM component manual deduplication, and provides a reliable data preprocessing mechanism for the construction of standardized component datasets.

[0095] The formula for determining duplicate components is:

[0096] (3)

[0097] In formula (3), The binary indicator (indicator function) for duplicate components takes only 0 or 1 values. A value of 1 indicates that the two components have highly overlapping multidimensional attributes, and they are judged as redundant duplicate components, and data removal and cleaning operations are performed. A value of 0 indicates that the two components have significantly different engineering attributes, and they are judged as valid independent components, which are retained and included in the standardized component dataset. For components With components The feature cosine similarity is calculated by formula (2), and its value range is [range missing]. . To preset the similarity threshold for duplicate components, this embodiment is fixed at 0.95, serving as the critical boundary for determining whether components are duplicated. The control logic of formula (3) reads the similarity between components calculated by the cosine similarity formula. The similarity score is compared with a preset threshold of 0.95 for duplicate components; if the similarity is greater than or equal to 0.95, the component is identified as a duplicate. Assigning a value of 1 marks the group of components as duplicate and redundant data, triggering the removal process. Simultaneously, business rule validation is applied: if the two components belong to different construction sections, they are considered legally reused components and are retained even if marked as 1; if the similarity is less than 0.95, they are marked as duplicates. The value is set to 0, which is determined to be a valid component and directly retained in the first component set. This formula (3) uses 0.95 as the preset similarity threshold for duplicate components, and performs standardized binary classification judgment on the cosine similarity results of components through a binary indicator function, and outputs the component duplicate identifier; realizes the automated redundancy removal of massive BIM components, and avoids the accidental deletion of valid data by combining cross-section legal reuse business rules, solves the pain points of low efficiency, missed detection and misjudgment of traditional manual component deduplication, and provides a standardized and auditable data cleaning mechanism for the construction of standardized component datasets.

[0098] Step S130: Using construction section and professional engineering type as dual search indexes, the cleaned effective component feature vectors are collected to construct a standardized first component set with unique BIM code and attribute search index.

[0099] Using the construction section number and the engineering type (civil engineering / mechanical / electrical / decoration) as the two-level retrieval primary key, the feature vectors of the cleaned and effective components are classified and collected; a globally unique BIM component code is bound to each component data, a multi-level retrieval index table of section-profession-component code is established, the data storage field specifications are unified, and a standardized first component set that can be directly input into the machine learning time series inference model is generated.

[0100] The dual-search index is a two-level search structure built by construction section and professional engineering type to enable rapid classification and retrieval of components; the standardized first component set is a structured dataset of all project components after lightweighting, collision checking, deduplication and cleaning, and index encapsulation.

[0101] The formula for constructing the first set of components is:

[0102] (4)

[0103] In formula (4), The first component set is a BIM component feature vector dataset after cleaning up redundant data. Each element in the set is a 5-dimensional structural feature vector of the component's basic structure. It binds a unique BIM component code and supports fast retrieval via a two-dimensional index. For the first The standardized basic feature vector of each component is generated by formula (1), which carries multiple engineering attributes such as engineering quantity, material, section, and planned start and completion time. The component selection constraints are set to retain only valid components with a duplicate identifier of 0, and components that are judged to be redundant or duplicated are removed, thus completing the deduplication and cleaning of the original BIM data. This refers to the overall set of all construction sections of the project. Indicates the first The construction section to which each component belongs falls within the scope of a legitimate project section. It is a global set consisting of all professional engineering types in the project. Indicates the first The professional type to which each component belongs is a preset valid engineering category of the project. For the first The component is coded with discrete, dimensionless professional engineering types for the classification and indexing of sub-items of civil engineering, mechanical and electrical engineering, decoration, etc. The set condition separator indicates that only feature vectors that satisfy all subsequent constraints can be included in the first component set. The control logic of formula (4) is to traverse all basic feature vectors of all components and select them through constraints. Select non-duplicate valid components and remove redundant duplicate data; perform a second verification of the construction section to which the component belongs. Professional engineering type coding Only components belonging to the legal sections and valid professional types of the project are included in the dataset; a two-dimensional retrieval index is constructed based on the construction section + professional engineering type, and a globally unique BIM code is assigned to each component in the set, forming a standardized structured dataset that can be quickly filtered and batch collected. Formula (4) filters valid components based on the duplicate component judgment identifier, and completes data filtering by combining the dual legal scope constraints of construction section and professional engineering type, and constructs a standardized BIM component dataset with a two-dimensional retrieval index; it realizes the standardized collection and rapid batch filtering of massive BIM components, and provides a structured and traceable basic engineering data source for the time-series dynamic cost calculation of the entire life cycle of construction projects.

[0104] Preferably, the BIM-based dynamic cost allocation calculation method for the entire life cycle of a construction project provided in this embodiment includes step S200 as follows:

[0105] Step S210: Set a fixed sliding time sequence sampling window based on the planned time nodes of each sub-project within the preset baseline schedule. When a major design change occurs in the project or the overall project duration is extended, the time span of the window is adaptively adjusted to construct a time sequence sampling time set.

[0106] Based on the project's preset baseline overall control progress and the planned start and end nodes of sub-projects, a fixed-duration sliding time-series sampling window is initialized; project change instructions are monitored in real time, and if major design changes or overall project duration extensions are identified, the window time span is adaptively scaled according to the project duration extension ratio; the sampling points of the window throughout the entire construction cycle are traversed, and all valid sampling time points are collected to construct an ordered time-series sampling time set, which serves as the time-series benchmark for resource feature mining.

[0107] The sliding time-series sampling window is a fixed / adaptive duration statistical interval that slides along the construction time sequence and is used for phased resource consumption statistics; the time-series sampling time set is a standardized time point sequence for extracting the features of all resource inputs throughout the entire construction cycle.

[0108] The formula for constructing the time series sampling time set is:

[0109] (5)

[0110] In formula (5), This is the set of time points for the time series sampling window, starting from the reference start time. Initially, continuous sampling time nodes are generated sequentially according to a fixed time step to uniformly segment the construction process throughout the entire project lifecycle and unify the sampling time granularity of engineering resource input data at each stage. The total duration of the sliding timing sampling window is expressed in days. In this embodiment, it is fixed at 30 days, representing the time span of a single sliding sampling. When there are major design changes or extensions in the overall project schedule, the duration of this window can be adaptively adjusted. The starting baseline schedule time for the sliding sampling window, in days, is taken from the start baseline time of the overall project schedule and serves as the initial time origin for global timing alignment. From the start time The sampling time nodes are sequentially increased to obtain a total of W time-series sampling points, covering all sampling times within a complete 30-day sliding cycle. The control logic of formula (5) uses the start time of the overall project schedule as the starting reference time of the window. Set a fixed total sampling duration of 30 days for sliding time series sampling. ;from Generate sequentially according to a uniform time step. to All time-series nodes are encapsulated to form a standardized set of time-series sampling times. It iterates through the standardized BIM component set, aligning the engineering quantity and resource input data of each component to the corresponding time-series sampling nodes, achieving alignment of the time dimension characteristics of components from different construction sections and different professional types; if major design changes or extensions to the overall construction period occur, the total window duration is adaptively adjusted. The time-series sampling set is regenerated to adapt to the changed project schedule. Formula (5) takes the project's benchmark start time as the starting point and constructs a sliding time-series sampling time set with a fixed duration of 30 days to achieve uniform time-series segmentation throughout the project's life cycle; it unifies the time sampling granularity of resource input data for multiple sections and multiple professional components, eliminates the problem of inconsistent feature dimensions caused by differences in construction cycles, and supports adaptive window adjustment under changes in the construction period, providing a standardized time-series data source for component resource input time-series feature mining and dynamic extrapolation of the full life cycle cost.

[0111] Step S220: Use a one-dimensional effective convolution feature extraction operator to perform temporal feature mining on the first component set, and extract the temporal feature vectors of the input of three types of resources, namely manpower, materials and machinery, at each sampling time.

[0112] The pre-trained machine learning feature extraction model is called to call the built-in one-dimensional effective convolution operator. Using the set of time-series sampling moments as the sliding reference, segmented time-series convolution operations are performed on the standardized first component set. The time-series fluctuation features of three types of resources—labor hours input, material consumption, and machine shift consumption—are extracted at each sampling moment and integrated to generate a three-dimensional resource input time-series feature vector at a single sampling moment.

[0113] To enable technical personnel in the fields of BIM engineering cost estimation and time-series intelligent calculation to fully reproduce this pre-trained machine learning feature extraction model, this embodiment fully discloses all training parameters, network architecture, sample rules, dataset partitioning, and hyperparticipation convergence criteria of the model, as follows:

[0114] 1. Training sample size and automatic annotation rules

[0115] We collected full-cycle time-series data from 1,472 completed housing construction, municipal engineering, and rail transit projects over the past eight years. Each project had at least 21,600 time-series samples, resulting in a total of approximately 31.8 million training time-series samples. Each sample uniformly included six structured fields: BIM component characteristics, monthly labor, material, and equipment input, quota cost, change order amount, progress lag days, and final audit settlement. We adopted a fully automated dual-label annotation system. The primary truth label was taken from the actual cost in the final audit ledger, while the secondary progress label was automatically divided into three levels: normal, slightly delayed, and severely delayed, based on a 7-day allowable deviation threshold. No manual annotation was required throughout the entire process.

[0116] 2. Complete model network architecture

[0117] The optimal implementation model of this scheme is a one-dimensional CNN temporal feature extraction network, with the following layer sequence: standardized BIM feature input layer → one-dimensional effective convolutional layer (9 kernels, 3 stride) → ReLU activation layer → one-dimensional max pooling layer → two-layer one-dimensional convolution (5 kernels, 2 stride) → global average pooling → three-dimensional human, material and machine feature output layer; LSTM, GRU, Transformer and other temporal networks are all equivalent and general solutions in this field. The core innovation of this invention focuses on the one-dimensional effective convolution operator, and the other temporal models are only equivalent replacement methods.

[0118] 3. Dataset hierarchical partitioning rules

[0119] The dataset is randomly segmented into layers based on project type and duration to avoid leakage of time-series data from the same project. The segmentation ratio is 75% for training set, 20% for validation set, and 5% for test set. The layering dimension is divided into three categories: short-term (≤360 days), medium-to-long-term (360~720 days), and long-term (>720 days) projects. This ensures a balanced distribution of working conditions across the three datasets and suppresses model overfitting.

[0120] 4. Hyperparameter mesh search and optimal fixed parameters

[0121] The hyperparameter traversal search range is: batch size {16, 32, 64}, initial learning rate {1e-3, 5e-4, 1e-3}, convolution kernel size {3, 5, 9}, L2 regularization {1e-4, 1e-5}, and a maximum of 300 iterations. After filtering by minimizing the loss on the validation set, the optimal hyperparameters are fixed as follows: batch size 32, learning rate 5e-4, convolution kernel size 9, and L2 regularization coefficient 1e-5.

[0122] 5. Convergence Criteria for Dual Indicators

[0123] Training can be stopped only if two conditions are met simultaneously: ① The decrease in MA on the validation set for 20 consecutive rounds is <1×10⁻ 5 ② The difference in MA error between the validation set and the test set is <0.008; if the maximum iteration of 300 rounds is still not satisfied, the weight of the point with the lowest validation loss is selected and fixed as an offline pre-trained model.

[0124] One-dimensional effective convolution feature extraction operator is a convolution operation unit adapted to single-dimensional construction time series data, used to mine the fluctuation pattern of phased resource input; the resource input time series feature vector is a multi-dimensional time series feature vector composed of the real-time input of three types of resources: labor, materials, and machinery.

[0125] The formula for extracting time-series feature vectors is:

[0126] (6)

[0127] In formula (6), For the first The time-series feature vector of the input of three types of resources (human, material, and machinery) at each time-series sampling point is dimensionless and is a 3-dimensional standardized column vector, which carries the dynamic input time-series information of the three core cost resources of human, material, and machinery during the current construction period. for The cumulative amount of human resources invested at any given time, expressed in man-days, represents the scale of labor consumption during the current construction phase. for The cumulative amount of main material resources invested at any given time, in tons, represents the total amount of building materials purchased and consumed in the current period. for The amount of construction machinery used in shifts at any given time, expressed in shifts, represents the current consumption level of large construction equipment. The first set of standardized BIM components serves as the original structured input data source for convolutional feature extraction. The one-dimensional temporal convolution kernel is pre-trained offline and is dimensionless. It is trained using historical project duration-resource samples and is used to capture the temporal trend of resource input within a continuous sliding window. It is a one-dimensional effective mode convolution operator that only outputs the convolution calculation results of the effective region, does not perform boundary zero padding, avoids feature distortion of edge temporal sampling points, and ensures that the feature vector dimension is strictly uniform under each sliding window. As a vector transpose operator, the three types of resource scalars are normalized into standard column vectors, unifying the input dimension specification of the machine learning model. The control logic of formula (6) is based on the standardized preprocessed BIM component dataset. As input, load a one-dimensional temporal convolutional kernel that has been trained offline. The one-dimensional effective convolution operator is invoked to traverse all components at each sampling time within a fixed 30-day sliding time window, and local weighted feature extraction is performed on the resource consumption sequence of labor, materials, and machinery within the continuous construction period. The time-series consumption values ​​of labor, materials, and machinery output by the convolution are encapsulated into a fixed 3-dimensional structured feature vector. ; Traverse all time-series sampling moments and generate resource time-series feature sequences in batches throughout the entire construction cycle, providing time-series input for subsequent full life-cycle value consumption extrapolation. Formula (6) uses a one-dimensional effective convolution operator, relying on offline trained and solidified time-series convolution kernels to perform sliding window time-series feature mining on standardized BIM component datasets, extracting the dynamic input of three types of resources—labor, materials, and machinery—at each sampling moment and encapsulating them into standardized time-series feature vectors; breaking through the limitations of traditional static quota single-point measurement, capturing the time-series evolution law of resource input throughout the construction process, and providing a time-series structured feature data source for dynamic cost extrapolation of the entire life-cycle of construction projects.

[0128] Step S230: Based on the engineering quota pricing rules, match the resource input at each time point with the current information unit price to calculate the current resource consumption cost, collect all current resource consumption costs along the sliding time sequence sampling window, and generate the first value consumption sequence that represents the capital consumption pattern of the entire project life cycle.

[0129] Based on the national standard engineering quota pricing rules, for each sampling moment within the sliding time-series sampling window, the total input of labor, main materials, and construction machinery shifts for the current period is taken. The basic resource cost is calculated by matching the corresponding official information unit price for the current period, and the enterprise comprehensive management fee rate, safety and civilized construction special fee rate, and construction industry value-added tax comprehensive tax rate are superimposed to obtain the resource consumption cost for each moment. Traversing all time-series nodes within the window, the current cost for each moment is arranged in order according to the construction sequence, and then vector transposed and normalized into a standard column vector to generate the first value consumption sequence representing the fluctuation law of the project's capital input throughout the entire cycle. This step relies on BIM time-series resource extraction data combined with dynamic market cost information to automatically and meticulously calculate the time-series cost in different time periods and unify the standardized time-series vector format. This solves the shortcomings of traditional static calculation of total cost, inability to reflect phased capital fluctuations, and low efficiency of manual segmented calculation. It outputs a standardized time-series dataset, providing a unified and batch-calculated quantitative time-series data source for subsequent progress deviation identification, change delay price adjustment, multi-party cost responsibility allocation, cash flow simulation, and time-series risk warning.

[0130] The current resource consumption cost is the cost of labor, materials, and machinery generated during the construction of internal components at a single time-series sampling node; the first value consumption sequence is a standardized time-series dataset that records the current and cumulative cost consumption at each point in time throughout the entire cycle.

[0131] The formula for calculating the current period's resource consumption cost is:

[0132] (7)

[0133] In formula (7), For the first The direct comprehensive cost at the time of sampling is expressed in ten thousand yuan and is the core output cost indicator of this formula. It represents the complete comprehensive cost of labor, materials, machinery and taxes for the construction of all components within a single time window and is the basic time-series data unit for constructing the full life-cycle value consumption sequence. For time sequence The total amount of manpower input at any given time is measured in man-days; it is output from the preceding temporal sliding window convolutional resource extraction step and represents the total man-days consumed by all construction procedures within this sampling period. For time sequence The total amount of main materials input at any given time is measured in tons; convolution extraction yields the total consumption of core main materials such as steel bars, concrete, and profiles in the current period, matching the measurement caliber of national / local standards. For time sequence The total number of construction machinery shifts at any given time, measured in shifts; including the current shift consumption of all construction machinery such as tower cranes, pump trucks, and mixing equipment. The unit price is the official labor cost information for the current period, in RMB 10,000 per man-day; it is taken from the current cost information platform of the housing and construction department and is the statutory benchmark unit price for quota pricing. This is the official unit price of the main building materials for the current period, in RMB 10,000 per ton; it is also matched with the market guidance price for building materials and the benchmark price of main building materials published by the cost station for the current period. The unit price is the official information price for the current period's machine shifts, in RMB 10,000 per shift; it includes the comprehensive fixed price for machine depreciation, fuel, and on-machine labor. The enterprise's comprehensive management fee rate is a dimensionless decimal; a fixed constant is preset based on the construction enterprise's qualifications and project category according to the national standard quota fee standard. The special fee rate for safe and civilized construction is a dimensionless decimal; the local housing and construction quotas have mandatory fee coefficients that cannot be adjusted arbitrarily. The comprehensive VAT rate for the construction industry is a dimensionless decimal; the current statutory VAT calculation coefficient for the construction industry is uniformly adapted to the tax-inclusive pricing rules. Formula (7) follows the national standard bill of quantities pricing specifications, first calculating the current basic resource costs by labor, main materials, and machinery shifts, and then superimposing enterprise management fees, safety and civilized construction fees, and comprehensive VAT rates to obtain the comprehensive tax-inclusive cost of a single time window; it automatically completes the refined cost calculation of the sliding sampling window, and generates a time series dataset of project life cycle capital consumption according to the time sequence, solving the defects of traditional static total cost that cannot represent the time-segmented capital fluctuations and the low efficiency of manual segmented calculation, and providing standardized quantitative time series data support for project cash flow simulation and time series cost risk warning.

[0134] The formula for generating the first value consumption sequence is:

[0135] (8)

[0136] In formula (8), The first value consumption sequence is the output of step S200, and the time series vector is a fixed dimension. The current resource consumption cost of each time node in the entire sliding sampling window is stored in chronological order to depict the evolution of capital investment time series consumption throughout the entire process from project commencement to completion. For the 0th, 1st...th The current resource consumption cost corresponding to each time-series sampling time is calculated by formula (7) time-by-time, with the unit being 10,000 yuan; The total duration of the sliding time window is fixed at 30 days. As a vector transpose operator, it converts a one-dimensional horizontal time series array into a standard column vector, unifying the data format of the time series and facilitating subsequent time series comparison, progress deviation identification, and machine learning time series modeling operations. The control logic of formula (8) is to traverse all time nodes within the sliding time series sampling window and call formula (7) to calculate the current total cost of labor, materials, and machinery for each node time by time. According to the chronological order of project construction, they will be carried out in sequence. to The current cost at all times is arranged in an orderly manner; the vector transpose is normalized into a standardized time series vector to generate a first value consumption sequence that can be batch-calculated and time-aligned; the time series cost sequence is compared with the project's preset benchmark progress cost sequence point by point to identify the cost deviation status corresponding to the progress ahead or behind in each construction stage, providing a time series cost basis for subsequent progress lag price adjustment and dynamic pricing of design changes. The consumption of manpower, materials and machinery resources at each time is multiplied by the current quota unit price to obtain the stage cost, and arranged in time to form the project's full life cycle value consumption sequence. Formula (8) normalizes the current resource consumption cost of each time series sampling node into a standardized time series vector according to the construction sequence, forming the project's full life cycle capital consumption time series sequence; relying on the BIM time series resource characteristics and combined with dynamic market information price, the stage dynamic cost extrapolation is realized, overcoming the defect that the traditional static one-time total cost calculation cannot reflect the capital fluctuations in the construction process, and providing a standardized time series cost data source for subsequent progress deviation identification, change lag price adjustment, and multi-party cost responsibility allocation.

[0137] Furthermore, the BIM-based dynamic cost allocation calculation method for the entire lifecycle of a construction project provided in this embodiment specifically includes step S300 as follows:

[0138] Step S310: For each resource consumption node in the first value consumption sequence, the single-node schedule deviation is obtained by subtracting the baseline planned time from the actual occurrence time.

[0139] Traverse all resource consumption time sequence nodes of the first value consumption sequence, read the actual time of resource input of the corresponding component and the planned input time within the preset baseline schedule; perform difference calculation by subtracting the planned time from the actual time, positive number represents schedule lag and negative number represents schedule advance, to obtain the quantitative value of single node schedule deviation.

[0140] Single-node schedule deviation is the time difference between the actual time of resource input for a single component and the baseline planned time, representing the duration of schedule deviation at a single point.

[0141] The formula for calculating the time sequence deviation of a single node is:

[0142] (9)

[0143] In formula (9), For the first The project duration deviation corresponding to each resource consumption time sequence node is in days; If the actual consumption of resources in the current period is later than the planned time, it is judged as a delay in construction progress. If the actual consumption of resources in the current period is earlier than the planned time, it is judged that the construction progress is ahead of schedule; The actual progress is completely consistent with the baseline planned progress, with no schedule deviation. for The actual time points of the arrival and consumption of human, material and equipment resources on site are recorded in days and are taken from construction logs, on-site progress ledgers, and actual construction records of BIM components. In the project's pre-set baseline schedule, the planned time node for the resource of this component should be invested, in days, is derived from the overall construction schedule and the start and completion attribute parameters of the BIM component plan. The control logic of formula (9) is to traverse all time-series sampling nodes in the first value consumption sequence, match the actual resource consumption time and the baseline plan investment time corresponding to each cost occurrence node; use the calculation method of subtracting the planned time from the actual time to calculate the duration of the schedule deviation for each node; mark the progress of each construction stage as ahead of schedule, behind schedule, or normal according to the positive and negative values ​​of the deviation, filter out the time-series nodes with schedule deviation, and extract the corresponding multi-dimensional time-series difference features. Formula (9) quantifies the duration of the schedule deviation for each cost occurrence node and distinguishes between the advanced and behind schedule states by performing a difference calculation between the actual resource consumption time and the baseline plan time; it uses the resource consumption time corresponding to the cost occurrence as the basis for schedule judgment, breaks through the limitations of the traditional rough judgment of progress based on the overall construction period of the component, accurately captures the actual progress deviation of each construction stage, and provides a quantitative basic indicator for subsequent extraction of time-series difference features and dynamic price adjustment for schedule lag.

[0144] Step S320: Compare the absolute value of the single-node schedule deviation with the preset allowable deviation threshold, and extract features only for deviation nodes that exceed the deviation threshold.

[0145] A threshold for minor progress fluctuations is preset. The absolute value of the time deviation of each single node is taken and compared logically with the preset allowable progress deviation threshold. If the absolute value of the time deviation of a single node is less than or equal to the preset allowable progress deviation threshold, it is determined to be a normal minor fluctuation, and the feature extraction process is skipped. If the absolute value of the time deviation of a single node is greater than the preset allowable progress deviation threshold, it is determined to be a valid progress deviation node, and the multidimensional difference feature extraction operation is started.

[0146] The preset allowable deviation threshold is a critical duration value used to filter out minor, normal schedule fluctuations that have no cost impact.

[0147] The formula for determining schedule deviation is:

[0148] (10)

[0149] In formula (10), This is a binary indicator for the progress deviation of the timing node, and can only take the value 0 or 1; If the absolute value of the schedule deviation at this time node exceeds the allowable fault tolerance threshold, it is determined that there is a substantial schedule deviation. The time difference characteristics of this node need to be extracted and included in the subsequent schedule lag adjustment calculation. If the schedule deviation is within a reasonable tolerance range of 7 days, it is considered to be a normal on-site process with minor short-term fluctuations in schedule. The progress is deemed compliant, and no difference feature extraction is performed, thus skipping the dynamic price adjustment process. For the first The absolute value of the schedule deviation at each time node eliminates the influence of positive and negative signs of ahead or behind schedule, and only measures the absolute duration of the schedule deviation in days. To preset the allowable deviation threshold for the schedule, this embodiment sets it to 7 days; this is used to distinguish between minor fluctuations in the schedule of normal processes and substantial schedule delays requiring cost adjustments. The control logic of formula (10) is to read the single-node schedule deviation calculated by formula (9). The absolute value of the deviation is calculated to eliminate interference from the leading and lagging directions on the threshold comparison. The absolute deviation is compared with the 7-day preset allowable deviation threshold one by one: if the absolute deviation is greater than 7 days, the binary indicator of the time node's progress deviation is set to 1, marking the node as a substantial progress deviation node and triggering the time difference feature extraction process; if the absolute deviation is less than or equal to 7 days, the binary indicator of the time node's progress deviation is set to 0, and it is determined to be a normal fluctuation in the on-site construction process. No deviation feature extraction is performed, and the process directly proceeds to the next time node verification. By filtering out a large number of nodes with small schedule fluctuations through thresholds, subsequent cost adjustment calculations are only performed on nodes with substantial progress deviations. This reduces the amount of invalid feature calculations and improves the efficiency of the algorithm while ensuring the accuracy of progress deviation identification. Formula (10) compares the absolute value of the time deviation of the construction period with the 7-day preset allowable deviation threshold, and uses a binary indicator function to mark whether the node has a substantial schedule deviation; it filters out small fluctuations in the construction period during normal construction, and only extracts time features for nodes that deviate beyond the threshold. This avoids cost disputes caused by misjudgment of the schedule, reduces invalid calculations, and improves the robustness of schedule deviation identification and the efficiency of time-series big data processing.

[0150] Step S330, the absolute duration of the splicing progress lag, the relative amount of excess investment in current resources, and the process interleaving time sequence offset coefficient are used to construct a multi-dimensional structured first time sequence difference feature vector.

[0151] For nodes that deviate from the schedule beyond the threshold, three types of quantitative indicators are solved simultaneously: absolute duration of schedule lag, relative proportion of excess current resources, and time offset coefficient of overlapping previous and subsequent processes. The three types of indicators are spliced ​​and integrated in a fixed dimension order to generate a multi-dimensional structured first time-series difference feature vector that can quantify the degree of cost disturbance caused by schedule deviation.

[0152] The first time-series difference feature vector is composed of lag duration, the proportion of excess resource input, and the process offset coefficient, and is a standardized feature vector used for adjusting price adjustments based on lag.

[0153] The formula for constructing the first time-series difference feature vector is:

[0154] (11)

[0155] In formula (11), For the first The first time-series difference feature vector corresponding to each schedule deviation node is a three-dimensional standardized column vector with no dimension. The engineering loss characteristics caused by schedule deviation are quantified from three dimensions: construction period, resource input, and construction procedures. This serves as the machine learning input feature for subsequent design change-based price adjustments and rate corrections. The absolute duration of the current schedule delay is in days. It is taken from the absolute value of the schedule deviation in formula (9), which represents the severity of the schedule delay and is a basic quantitative indicator that directly reflects the work stoppage and the lag in schedule management. It represents the relative amount of excess input of human, material, and machinery resources in the current period, and is dimensionless; it characterizes the proportion of extra input of human, material, and machinery exceeding the planned quota due to ahead-of-time or behind-time progress, and can quantify the resource losses caused by idle labor, material backlog, and idle equipment. This is the process timing offset coefficient, dimensionless, with a range of values. Negative values ​​represent the preceding and overlapping of processes, while positive values ​​represent the following and delayed processes. This reflects the additional management and construction losses caused by the disorder of quantitative flow construction and cross-operations. As a vector transpose operator, it converts a one-dimensional horizontal array into a standard column vector, unifying the data format of multi-dimensional features and adapting it to the input specifications of subsequent regression pricing models. The control logic of formula (11) is to filter out... The substantial schedule deviation from the time sequence node is extracted in sequence into three core deviation indicators: absolute schedule deviation duration, relative amount of excess resource input, and process time sequence deviation coefficient. The three indicators are spliced ​​together in a fixed dimension order and normalized into a three-dimensional structured time sequence difference feature vector through vector transpose. The multi-dimensional feature vector is used as input to train the price adjustment regression model and fit the mapping relationship between schedule deviation and cost increase / decrease rate to achieve refined dynamic price adjustment. Formula (11) splices the three indicators of absolute schedule lag duration, relative amount of excess resource input, and process time sequence deviation coefficient into a three-dimensional structured time sequence difference feature vector, which quantifies the engineering loss caused by schedule deviation from multiple dimensions of schedule, resources, and processes. It provides a refined and quantifiable feature basis for dynamic cost adjustment in the case of design changes and schedule lag, and overcomes the defects of traditional single schedule indicator price adjustment which is one-sided and prone to cost disputes.

[0156] Preferably, the BIM-based dynamic cost allocation calculation method for the entire life cycle of a construction project provided in this embodiment includes step S400 as follows:

[0157] Step S410: Classify and match the changed quantities of work according to the unique BIM component codes. Based on the first time-series difference feature vector, calculate the first norm of the variable vector and combine it with the pre-set lag adjustment sensitivity coefficient based on the historical full life cycle cost sample to solve the schedule lag adjustment correction coefficient.

[0158] The changed quantities are categorized and matched according to the unique BIM component codes. The normalized first-series difference feature vector output from the previous steps is taken, and the L1 norm of this vector is calculated to quantify the total deviation of the project's overall schedule. A lag adjustment sensitivity coefficient, adaptively calibrated according to the project type and based on historical full-lifecycle cost samples, is introduced and combined with a lag-free benchmark constant of 1 to construct the basic adjustment calculation value. Then, the calculation result is subjected to an interval constraint of 1 (lower limit 1) and 1.35 (upper limit 1.35) through a value range trimming operator to obtain the schedule lag adjustment correction coefficient. This schedule lag adjustment correction coefficient is: The dimensionless scaling factor is used to compensate for cost increases caused by idle labor, idle machinery depreciation, and delayed price increases of building materials, and uniformly adapts to the cost correction calculation of change orders. This step relies on BIM component time sequence data to quantify the global schedule deviation, adaptively matches the price adjustment sensitivity according to project type and sets hard constraints on the price adjustment range, solving the shortcomings of traditional delayed adjustment that rely on subjective manual estimation, do not distinguish between project types, have uncontrolled price increase range, and cannot achieve fine allocation based on components. It provides a standardized and reproducible quantitative compensation basis for dynamic cost calculation of change orders.

[0159] The delayed price adjustment sensitivity coefficient is a proportional coefficient based on historical cost samples, representing the degree of impact of schedule deviation on the comprehensive unit price; the schedule delayed price adjustment correction coefficient is a unit price correction scaling coefficient used to compensate for idle labor, idle machinery losses, and delayed material price increases.

[0160] The formula for calculating the adjustment factor for delayed pricing is as follows:

[0161] (12)

[0162] In formula (12), The adjustment factor for delayed schedule and the dimensionless scaling factor are the core outputs of this formula. They are used to compensate for and adjust the unit price of the current comprehensive cost, covering the cost increase caused by idle labor, idle machinery depreciation, and delayed price increases of building materials. Ultimately, they are used to dynamically allocate the changed cost. For range clipping operators, standard amplitude limiting mathematical operators; input calculated values Lower threshold Upper limit threshold Output rules: If Output ;like Output the original ;like Output It is used to constrain the fluctuation range of the price adjustment coefficient and avoid cost distortion caused by extreme project schedule deviations. This is the lower limit constant of the baseline without schedule lag; it means that when there is no schedule deviation, the correction factor is equal to 1, and no upward compensation is made for the cost. The upper limit of the price adjustment coefficient is a constant; the maximum price adjustment is forcibly limited to no more than 35%, avoiding the unlimited amplification of costs in extreme lagging conditions, and conforming to the housing construction price adjustment control regulations. The adaptive lag price adjustment sensitivity coefficient is a dimensionless piecewise function; input parameters Based on the project type (building construction / municipal construction / highway / water conservancy, etc.), it is obtained through offline calibration using a massive amount of historical full life cycle cost samples, and it represents the sensitivity of unit price increase corresponding to each unit of deviation in the progress of similar projects. The first time-series difference feature vector is normalized and comes from the output of the previous BIM progress time-series comparison step. It is obtained by normalizing and encapsulating the difference between the planned and actual construction periods of each component. The multi-dimensional vector represents the progress deviation distribution of each process in the whole project. The L1 norm of the normalized difference feature vector is a dimensionless scalar; the summation of the absolute values ​​of all dimensions of the vector quantifies the total amplitude of the overall project schedule deviation disturbance. The larger the value, the more serious the overall schedule lag. Formula (12) quantifies the global schedule deviation amplitude using the normalized time series difference feature vector L1 norm, and constructs a basic price adjustment model by combining the price adjustment sensitivity coefficient adaptively calibrated according to the project type with the lag-free benchmark value. Then, the price adjustment coefficient range [1, 1.35] is limited by the value range trimming operator, and the unit price scaling correction coefficient is output to adapt to idle work, idle machinery, and material price lag. It realizes the standardized and refined dynamic lag cost compensation based on BIM component change engineering quantity, solves the defects of traditional manual subjective adjustment, no distinction between engineering types, uncontrolled price adjustment, and inability to allocate by component, and provides an objective and reproducible quantitative basis for dynamic cost calculation of project change visa.

[0163] Step S420: Retrieve the official unit prices of labor, materials, and machinery for the current period, and calculate the total cost of the current period's change based on the resource consumption adjusted by the progress lag price adjustment factor.

[0164] The official unit prices for labor, main materials, and construction machinery shifts in the current period are retrieved. The resource consumption changes are adjusted using a schedule lag adjustment factor, and the total cost of compliant change orders for the current period is calculated hierarchically. The calculation first determines the net change in labor, main materials, and machinery shifts, i.e., the difference between the increase and decrease of each resource. Then, according to the national standard bill of quantities pricing rules, the net change in each resource is multiplied by the corresponding official unit price for the current period, and the sum is calculated. Enterprise management fees and the comprehensive VAT rate for the construction industry are then added to obtain the comprehensive cost of the change order excluding compensation for schedule lag losses. Finally, the schedule lag adjustment factor is introduced. The process involves adjusting the raw comprehensive cost by compensating for lag losses, and finally applying a value range trimming operator to the adjusted cost with a dual constraint of a lower limit of 0 and a pre-set upper limit for contract-preset change orders, outputting the total cost of compliant current change orders. This step integrates BIM change quantities, current official cost information, claims for delayed schedules, and contract change order limit control rules to achieve fully automated and standardized calculation of change order costs. It addresses the shortcomings of traditional change order pricing, such as lag losses from manual estimation, fragmented pricing logic, and the inability to automatically control over-limit change orders, providing a fully auditable quantitative calculation basis for dynamic cost allocation and change order settlement.

[0165] The formula for calculating the total cost of current change visas is as follows:

[0166] (13)

[0167] In formula (13), This represents the net change in labor, expressed in man-days. To add man-days for the change, The difference represents the final change in man-days for the visa, indicating a reduction in man-days. Net change in main material quantity, in tons; Changes to the quantity of new main materials Change the deduction of main material quantity. This represents the net change in the number of construction machinery shifts, expressed in units of shifts. For the addition of shifts, To reduce the number of shifts. The unit is RMB 10,000, representing the bare comprehensive cost of the change without the addition of compensation for delayed progress. Following the national standard list pricing rules, the net increase in labor, materials, and machinery is multiplied by the current official information price and summed, plus enterprise management fees and the comprehensive VAT rate. The official unit prices for labor, main materials, and machinery shifts for the current period; For enterprise management fee rates; The comprehensive VAT rate for the construction industry (using the fixed-price definition mentioned above). The output of the schedule lag adjustment correction coefficient for formula (12) is used to compensate for the costs of idle labor, idle machinery, and delayed material price increases. For the range trimming operator, constrain the lower limit of the cost range to 0 and the upper limit to 0. ;like Output 0 (no negative cost deduction); if the calculated value exceeds the cost cap threshold. Forced capping is To avoid excessively large visa applications. The upper limit threshold for cost control of a single visa is set as a constant based on the project contract and local cost control documents. The core output of this formula is the final total cost of compliant change orders in the current period, which is used to update the project's full life cycle capital consumption sequence. This formula (13) completes the change cost calculation in three progressive layers: the first layer calculates the net change resource quantity of labor, main materials, and machinery; the second layer calculates the basic comprehensive cost without delayed compensation based on the national standard list pricing rules; the third layer superimposes the schedule lag price adjustment correction coefficient and constrains the cost range through the value range trimming operator. It outputs the total cost of compliant change orders for the current period; integrates BIM change quantities, current official information prices, compensation for delays in construction period and contract limit control, and realizes standardized and automated cost calculation for various engineering changes. It solves the defects of traditional manual estimation, such as delayed claims, fragmented change pricing, and loss of control over large change orders, and provides objective, auditable and quantitative basis for dynamic cost allocation and change order settlement of construction projects.

[0168] Step S430: Overlay the project phased base contract cost and the current period change total cost, and after non-negative constraint verification, summarize to obtain the project's current period first dynamic cost.

[0169] The project's current phased benchmark contract cost is multiplied by the building materials comprehensive price adjustment index and then superimposed with the compliant change order total cost output in step S420. A dual-layer numerical constraint of a lower limit of 0 and a phased cost control upper limit is then implemented through a value range trimming operator to calculate the project's current phase's first dynamic cost. The value range trimming operator avoids negative cost spillover caused by large change deductions and limits the single-phase total cost to not exceeding the budget control threshold. This step integrates four cost disturbance factors: BIM benchmark static cost, main material cyclical price increases and adjustments, engineering design changes, and construction delays leading to idle labor losses. It outputs a single-phase real-time dynamic total cost that can be used for cash flow simulation and risk prediction. This addresses the shortcomings of traditional methods that rely solely on static contract prices, cannot comprehensively consider multi-dimensional cost fluctuations, and lack compliant numerical boundary constraints. It provides standardized time-series quantitative indicators for dynamic cash flow projection throughout the project's entire lifecycle, phased budget control, and cost risk classification and prediction.

[0170] The first dynamic cost is the real-time total cost of the project in the current period, which integrates the benchmark time-series cost, change order cost, and compensation for schedule delay losses.

[0171] The first dynamic cost calculation formula is:

[0172] (14)

[0173] In formula (14), The first dynamic cost of the project in the current period, in ten thousand yuan, is the core output of this formula; it integrates the phased benchmark contract cost, the comprehensive adjustment of building material prices, and the delayed compensation cost of change orders, representing the real-time dynamic total cost of the construction phase, which is used to update the time-series dataset of fund consumption throughout the project's entire life cycle. For the range clipping (limiting) operator, perform two-level constraint logic: lower threshold. To enforce non-negative dynamic costs and prevent negative cost overflows due to extreme changes and deductions; an upper limit threshold is set. To control the upper limit of single-phase construction cost, constants are preset based on construction contracts and annual investment plans to prevent single-phase costs from exceeding the budget and getting out of control. The estimated cost is the current phased contract cost, in ten thousand yuan. The benchmark static cost under the condition of no changes, no increase in building material prices, and no delay in progress is taken from the previous BIM phased project quantity quota calculation results. It is a dimensionless comprehensive price adjustment index for building materials in the current period; it is a weighted composite of the current price indices of main materials such as steel, concrete, and sand and gravel released by the housing and construction department, and is used to uniformly compensate for the cost of cyclical price increases of all categories of building materials. The total cost of compliant change orders for the current period is calculated using the output of formula (13) above. This includes net change costs for personnel, materials, and machinery, compensation for delayed labor / machinery idleness / material price increases, and the final change cost after the limit on a single change order. Formula (14) multiplies the phased base contract cost by the current period's comprehensive building materials price adjustment index, adds the total cost of compliant change orders after schedule lag correction, and then constrains the cost range through the value range trimming operator. The system integrates four major cost-influencing factors: benchmark engineering quantity, cyclical price increases of main materials, design changes, and construction period idle labor losses. It automatically generates the real-time dynamic total cost of a single construction phase, solving the problems of traditional static contract prices failing to represent multi-dimensional cost disturbances and lacking compliance constraints on numerical values. It provides standardized and quantitative time-series indicators for dynamic cash flow simulation, phased budget control, and cost risk prediction throughout the entire life cycle of construction projects.

[0174] Furthermore, the BIM-based dynamic cost allocation calculation method for the entire lifecycle of a construction project provided in this embodiment specifically includes step S500 as follows:

[0175] Step S510: Using three types of data—BIM component quantities, time-series construction logs, and subcontracting contracts—as the basic data sources for traceability, a unique code for the responsible party attributing each expense in the first dynamic cost is located through a multi-dimensional rule-matching traceability mapping function.

[0176] Construct a three-dimensional collaborative traceability data source: standardized BIM component measurement engineering quantities, daily time-series construction logs (team, shift, material requisition records), and general contracting / professional subcontracting / supplier subcontracting contracts; call the preset multi-dimensional rule matching traceability mapping function to perform layered matching on each material cost, labor cost, machinery cost, and management fee after the first dynamic cost breakdown, and accurately locate the unique identity code of the participating responsible entity corresponding to the cost.

[0177] The multi-dimensional rule-matching traceability mapping function is a mathematical mapping model that uses BIM, construction logs, and contract data as matching rules to automatically bind cost and responsible parties. The unique code of the responsible party is a globally unique identification code for the general contractor, subcontractor, material supplier, and equipment leasing company.

[0178] The formula for mapping the unique code of the responsible party for the cost is as follows:

[0179] (15)

[0180] In formula (15), It serves as a unique code for the responsible party attributing costs. It is a dimensionless code corresponding to the responsible party for each dynamic cost item. It is a globally unique identifier preset for the owner, general contractor, various professional subcontractors, and labor teams. It is used to achieve accurate attribution of each project cost, quantitative definition of responsibility, and archiving of settlement data ledgers. It is a multi-dimensional rule-matching traceability mapping function that uses explicit multi-condition logic to achieve data mapping and does not have implicit black-box operations. The function simultaneously accesses three types of constraint dimensions: the zoning and professional affiliation of BIM components, the time period of the construction log, and the scope of the contracted project stipulated in the subcontract. By matching these three conditions simultaneously, the unique responsible party is identified. For the first The data set of engineering quantities, professional types, construction zones, and sub-item features corresponding to each BIM component is used to locate the physical construction scope corresponding to the cost. for The structured feature vector of the construction site time-series construction log includes on-site performance records such as construction teams, personnel entering the site, machine shifts, and start and end times of operations. The feature vector of the subcontracting contract scope stores the contracting specialty, construction section, contract start and end dates, and contracted work content constraints of each subcontractor, serving as the legal basis for liability attribution. The control logic of formula (15) is to extract the BIM component quantity data corresponding to the current traceable costs. This involves identifying the construction specialty, spatial section, and sub-item scope corresponding to the cost; and matching the timing of the cost occurrence. Corresponding construction log vector Obtain information on the work teams and participating units that are actually carrying out the on-site operations; traverse the feature vectors of all subcontracting contracts. The system identifies a unique subcontractor whose contract scope simultaneously covers the BIM construction area, construction period, and professional type; and outputs a globally unique code for this responsible entity through a tracing mapping function that matches triple-condition rules. The single dynamic cost is bound to the responsibility code and archived to form a full-link audit traceability ledger of "cost-progress-BIM component-responsible unit". The formula (15) uses a multi-dimensional rule-based traceability mapping function to perform triple condition matching on three types of data sources: BIM component engineering quantity, time-series construction log, and subcontracting contract, and matches the corresponding unique code of the responsible entity for each dynamic cost of the first dynamic cost; relying on BIM, the engineering cost is accurately traced down from the general contract to the subcontractor and team level, solving the technical defects of traditional cost that can only be roughly collected and the responsibility is difficult to accurately define, and realizing full-link auditability and traceability of cost data.

[0181] This multi-dimensional rule-matching source tracing mapping function does not contain black-box logic. It fully discloses the three-layer matching priority rule base, conflict handling mechanism, and rule base construction method. Engineering information technology personnel can independently reproduce the entire process of matching cost responsibility entities.

[0182] 1. Three-level matching rules (priority from high to low, cannot be changed)

[0183] Priority 1 BIM zoning + professional matching: Only subcontractors whose contracted sections and professional types completely cover the current components are selected; multiple candidates automatically proceed to the next level of time sequence verification; Priority 2 Construction time sequence window matching: Filter out participating entities whose contract start and end dates cannot cover the time when costs occur; Priority 3 Sub-item work content matching: Compare the subcontracted items with the component construction type to lock in preliminary candidate entities.

[0184] 2. Multiple Candidate Conflict Handling Rule

[0185] When there are two or more candidate units after three-level matching, the construction team ledger for the day the cost occurred is retrieved, and the unit to which the team belongs is output as the sole directly responsible entity.

[0186] 3. Rule base construction method

[0187] Based on national standard EPC general contracting, professional subcontracting standard contract templates, and BIM zoning and item coding specifications, 126 general judgment rules are pre-built; during the project implementation phase, the project's bid section and subcontracting contract information are imported to complete instantiation and adaptation, and a project-specific matching rule set is automatically generated.

[0188] 4. Output constraint logic

[0189] Upon successful matching, a unique code for the responsible party is output; if no matching object is found, an abnormal warning indicator is output, prompting the user to supplement subcontract information and re-execute the matching operation.

[0190] Step S520: Construct a multi-level reverse traceability link according to the time sequence of cost occurrence and the level of responsibility of participating construction units, and encapsulate the unique code of the cost attribution responsibility entity, the code of each level of participating construction unit, the time of cost occurrence, and the amount of individual cost into a structured first traceability path.

[0191] Following a four-level reverse hierarchy of total cost → cost of individual items → cost of personnel, materials and machinery → participating units, a multi-level traceability link is constructed by combining the time sequence of cost occurrence; the codes of participating units at each level, the unique codes of the end responsible entities, the time of cost occurrence, and the amount of individual costs are uniformly encapsulated to generate a structured and complete first traceability path that can trace the source of costs and the attribution of responsibility.

[0192] The multi-level primary traceability path is a standardized cost traceability link that is hierarchical from top to bottom and binds the performance of all participating parties and cost details.

[0193] The formula for constructing the first tracing path is:

[0194] (16)

[0195] In formula (16), As the primary traceability path, it is dimensionless and a standardized structured column vector; it fully encapsulates multi-dimensional traceability information on the responsibility attribution, occurrence sequence, and cost amount of a single cost item throughout its entire lifecycle, serving as a standardized traceability carrier for engineering auditing, cost responsibility breakdown, and breach of contract accountability. The output of formula (15) is a unique code for the direct responsible party of the cost, which is generally the code of the construction subcontractor or labor team, and is the identifier of the direct performance responsibility level of the cost. It serves as a globally unique code for general contractors / specialized construction subcontractors and identifies the level of responsibility for on-site construction management. It serves as a unique code for the supervision unit, representing the third-party responsibility level for project quality and schedule control. It serves as a unique code for the construction unit (owner) and the top-level responsible entity for project investment management. This is the actual time of occurrence corresponding to the dynamic cost, accurate to the construction sequence sampling node, used to lock in the time of occurrence of schedule deviations and change events, and to achieve time-series traceability. The amount is the single dynamic cost amount after the progress lag correction, in ten thousand yuan, which is the settlement cost base corresponding to this traceability. The control logic of formula (16) is to extract the direct responsibility code, subcontractor code, supervision unit code, and construction unit code of a single cost in sequence to complete the collection of information on the four levels of participating responsibility; bind the two types of core business data, namely the cost occurrence sequence and the single settlement amount, corresponding to the cost; splice multiple heterogeneous data in a fixed dimension order, and generate a standardized structured first traceability path through vector transposition; bind an independent traceability path vector for each change and progress compensation cost and encrypt and archive it, so that the total project cost can be summarized in the forward direction according to the responsible subject, time period, and cost type, and the cost source can be located in the reverse direction through any dimension field, and the responsibility of each level of participating party can be traced layer by layer to form a closed-loop audit ledger. Formula (16) encapsulates the direct responsible entity code, subcontractor, supervisor, construction unit multi-level participation code, cost occurrence time, and single cost amount into a standardized and structured first traceability path, and builds a multi-level forward aggregation and reverse accountability cost traceability link; solves the defects of traditional engineering cost traceability data fragmentation, unclear responsibility level, and difficulty in auditing and evidence collection, meets the supervision and audit compliance requirements of the whole process cost settlement of construction projects, and provides standardized data support for subcontract cost splitting and breach of contract cost accountability.

[0196] Preferably, the BIM-based dynamic cost allocation calculation method for the entire life cycle of a construction project provided in this embodiment includes step S600, which specifically includes:

[0197] Step S610: Based on the first traceability path, the actual completed work volume and the contracted planned work volume of each responsible entity are statistically analyzed, and the minimum zero constant is introduced to calculate the performance rate of the responsible entity.

[0198] Based on the first traceability path, the actual completed work volume and contracted planned work volume of each responsible entity in the current period are statistically analyzed. A minimum positive constant to prevent zero is introduced and combined with the value range trimming operator to calculate the corresponding responsible entity's performance rate. During the calculation process, for the i-th responsible entity, its actual completed work volume in the current period is taken. Current contract planned work volume Simultaneously superimpose the minimum positive constants for preventing zero in both the numerator and denominator. Constructing fractions to avoid computational anomalies caused by denominators of 0, and then through... The operator performs range constraint clipping on the fractional calculation results, and outputs the project schedule fulfillment rate. Forced to be limited to Range; of which the project completion rate This is a dimensionless proportional indicator; the closer the value is to 1, the higher the completion rate of the current construction task for the responsible party. It is subsequently used for cost responsibility allocation, subcontractor performance assessment, and progress payment deduction calculation. It is a very small positive constant to prevent zeroing. To fix the minimum positive number, it is used to eliminate the division-by-zero error defect when the planned project quantity is 0; the value range clipping operator executes the segmented constraint logic: output 0 when the fraction calculation result is less than 0, output 1 when it is greater than 1, and output the original value when it is between 0 and 1. It unifies the measurement rules for project performance assessment, and realizes the automatic quantification of the construction completion of each subcontractor by relying on BIM traceability of project quantity. It solves the problems of division-by-zero collapse in traditional ratio calculation, no boundary of index value, and inconsistent measurement assessment caliber, and outputs standardized and reproducible performance quantification indicators.

[0199] The formula for calculating the project completion rate is:

[0200] (17)

[0201] In formula (17), For the first The current period performance rate of each responsible entity, a dimensionless proportional indicator, is the core output of this formula; its value, after being clipped, is fixed within a certain range. The closer the value is to 1, the higher the completion rate of the current construction task of the subject, which is used for the subsequent responsibility allocation, performance assessment and deduction calculation. For the first The actual completed work volume of the main structure in the current period is measured using a unified BIM component measurement standard; the total work volume of completed components of each subcontractor and construction unit in the current period is captured through a multi-level traceability path. For the contractual agreement The planned quantities of the main project in the current period are consistent with the actual quantities in terms of dimensions; they are taken from the phased quantities of the project construction schedule and the subcontract agreements. It is a zero-prevention constant for extremely small positive numbers, with a fixed value. It is a very small positive number; used to avoid planned project quantities. When the denominator is 0, it results in computational overflow and program errors, which is an engineering defect. For the range pruning operator, a double-layer boundary constraint rule applies: if the result of the fraction calculation is less than 0, the lower limit of the output is 0; if the result of the fraction calculation falls within the range of 0, the lower limit of the output is 0. The range is set to output the original value; if the result of the fractional calculation is greater than 1, the upper limit of 1 is output; the forced performance rate does not have a negative value and does not exceed 100%, which conforms to the industry rules for project performance assessment. This formula (17) adopts a symmetrical fractional structure with a minimum zero constant that is simultaneously superimposed on the numerator and denominator to avoid the calculation anomaly of the planned project quantity being 0, and then locks the result within the range by using the value range trimming operator. The interval outputs a standardized, dimensionless project completion rate; relying on BIM to trace the project quantity, the completion degree of each subcontractor's current construction task is automatically quantified, solving the defects of traditional ratio formulas such as division by zero collapse, no numerical boundary, and inconsistent assessment criteria, and providing objective and reproducible quantitative indicators for project subcontract performance assessment and cost responsibility allocation.

[0202] Step S620: Normalize the project schedule performance rate of all responsible parties, and use the normalized project schedule performance rate as the cost allocation weight to decompose the first dynamic cost of the project level by level to obtain the initial allocated cost of each responsible party.

[0203] Based on the project schedule fulfillment rates of each responsible entity output in step S610, a weighted normalization calculation is performed. The normalized comprehensive allocation weight is used to split the current dynamic total cost of the project, obtaining the initial allocation cost corresponding to each responsible entity. The calculation is first performed using the formula... Solve the first The overall original apportionment weight of each responsible entity The original weighting is integrated with the static price weighting of the contract. with standardized interval Construction period fulfillment rate Double factor; then all The overall original weights of all responsible entities are summed globally, and the proportion of each entity's original weight to the total global weight is used as the normalized allocation coefficient, combined with the project's current dynamic total cost. According to the formula Calculation yields the first The initial cost allocated to each responsible party The initial allocated cost The initial cost allocation for all entities, before deducting quality assurance deposits and penalties for breach of contract, is summed to equal the current dynamic total cost. This step also takes into account the contract size of each subcontractor and the actual level of performance in the current period. By unifying the global weights, the dynamic cost is fairly divided, overcoming the shortcomings of traditional methods that rely solely on the proportion of contract price to equally allocate costs and cannot distinguish between good and bad performance. This provides a standardized and quantitative calculation basis for phased fund settlement and cost deviation responsibility allocation that is traceable and auditable throughout the entire process.

[0204] The cost allocation weight is the normalized period performance rate, representing the proportion of each responsible party's contribution to the total cost in the current period.

[0205] The initial cost allocation calculation formula is as follows:

[0206] (18)

[0207] In formula (18), For the first The original apportionment weight of the responsible entity is dimensionless; it integrates the proportion of contract price and the current performance completion rate as the basic weight before normalization. Output the first value for formula (17) Main construction period fulfillment rate, value range This indicates the level of completion of the current construction tasks. For the first The main contract price has a static weight, which is dimensionless; it represents a fixed proportion of the subcontract amount to the total amount of all subcontracts in this phase. . For the first The initial apportioned cost of the responsible party before deducting warranty and breach of contract penalties, in ten thousand yuan, is the core output of this formula; it is the basic settlement amount attributable to the responsible party after the dynamic cost of the current period is split according to the performance + contract weight. The first dynamic total cost of the project in the current period is output for formula (14), which represents the total dynamic cost of all elements in this period. For all of the current period The total original weights of all participating entities are summed globally for global normalization, ensuring that the sum of the costs allocated to all entities equals the total cost of construction. . This refers to the total number of responsible entities, such as subcontractors and construction units, that participate in the cost sharing during the current period. The formula (18) first integrates the static contract price weight and the standardized construction period performance rate to generate the comprehensive original apportionment weight of each subject; secondly, it globally normalizes the weight of all subjects, splits the dynamic total cost of the project in the current period with a normalized ratio, and outputs the initial apportionment cost of each subcontractor without deducting quality assurance and breach of contract deductions; it realizes a fair dynamic cost split that takes into account both the contract scale and the actual performance of the current construction, solves the defects of the traditional single contract with equal proportion and no distinction between good and bad performance, and provides an auditable standardized quantitative basis for the phased settlement of funds and cost responsibility apportionment of multiple subcontractors in the construction project.

[0208] Step S630: Based on the initial allocated cost, deduct the current period's default deductions and warranty reserve amount, and ensure that the settlement cost is non-negative through maximum value constraints, to calculate the first liability cost corresponding to each responsible party.

[0209] Based on the initial allocated costs of each responsible party output in step S620, the current period's default deductions and warranty reserve amounts are deducted sequentially, and the deduction cap constraint and settlement cost non-negativity constraint are superimposed to calculate the first liability cost corresponding to each responsible party; in the calculation process, the first... The total amount to be deducted in the current period for each responsible party, including the default penalty and the warranty reserve, is used as the total amount to be deducted in the current period. A minimum value operator is used to limit the actual deduction amount to no more than 30% of the initial allocated cost of the responsible party, thus avoiding excessive deductions. Then, the actual deduction amount after the upper limit constraint is deducted from the initial allocated cost. A value range constraint operator is used to lock the lower limit of the deduction value to 0, and a non-negative first responsibility cost is forced to be output. The first responsibility cost is the amount that can be settled in the current period after deducting all compliant amounts, which is directly used for registration in the subcontract payment ledger and the final settlement ledger. This step integrates the simultaneous deduction of default penalties and warranty deposits, and adds a double layer of financial compliance constraints. It solves the defects of traditional project payment settlement, such as no cap on deductions, easy generation of negative accounts payable, and cumbersome calculation process for itemized deductions. It provides standardized and auditable quantitative accounting indicators for project subcontracting phased project payment and monthly settlement.

[0210] The primary liability cost is the preliminary apportionment settlement cost calculated by each responsible party after deducting default penalties and warranty reserve funds in the current period.

[0211] The formula for calculating the primary responsibility cost is:

[0212] (19)

[0213] In formula (19), For the first The primary liability cost of each responsible entity in the current period, expressed in ten thousand yuan, is the core output of this formula. It is the cost that the entity can settle and allocate in the current period after deducting default penalties and warranty reserves, and is verified by non-negative constraints. It is used for entry into the subcontract payment ledger and the final settlement ledger. The formula (18) outputs the initial apportionment cost of the main body, without deducting the base apportionment amount for breach of contract and warranty deductions. For the first The total amount of penalties for breach of contract by the main entity in the current period, in ten thousand yuan; including various contractual penalties such as fines for delayed construction, deductions for quality defects, and fines for violations of safety and civilized construction regulations. The amount reserved for quality assurance in the current period, in ten thousand yuan; the quality assurance deposit is accrued from the current period's allocated cost according to the contract quality assurance reservation ratio. Sub-item for deduction limit constraint: This represents the total amount deductible for the current period. The maximum deduction threshold for a single period is set, stipulating that the total deduction amount for the current period shall not exceed 30% of the initial apportioned cost of the entity; the smaller of the two values ​​shall be taken as the actual deduction amount to prevent excessive deductions from resulting in subcontractors having no settlement funds. As a non-negative value domain constraint operator, the lower limit of settlement cost is locked at 0; if the value after deducting all payments is negative, it is forced to output 0, thus preventing negative accounts payable and complying with the engineering financial accounting standards. Formula (19) combines the current period's default deduction and quality assurance reserve amount on the basis of the initial allocated cost of the main body, and limits the total deduction to no more than 30% of the initial cost through the minimum value operator; then, the lower limit of the settlement cost after deduction is locked at 0 through the maximum value operator, and the first responsibility cost of compliance is output; the default and quality assurance double payment deduction and double compliance constraints are completed in an integrated manner, solving the defects of traditional settlement without deduction cap, easy to have negative financial accounts, and cumbersome deduction item calculation, and providing standardized and auditable quantitative indicators for monthly settlement of subcontracting and payment of project funds.

[0214] Furthermore, the BIM-based dynamic cost allocation calculation method for the entire lifecycle of a construction project provided in this embodiment specifically includes step S700 as follows:

[0215] Step S710: Calculate the cost deviation rate of each responsible entity's current period responsibility cost relative to the phased budget amount. After taking the absolute value of the cost deviation rate, compare it with the preset allocation distortion deviation rate threshold to determine whether there is cost allocation distortion in the current period responsibility cost.

[0216] The standardized cost deviation rate of each responsible entity's current period cost relative to the corresponding phased budget amount is calculated separately. Then, based on the budget volume of each entity, an adaptive allocation distortion deviation threshold is calculated. The absolute value of the cost deviation rate is compared with this threshold to automatically determine whether there is distortion in the current period cost allocation. When calculating the standardized cost deviation rate, the difference between the first responsible cost output in step S630 and the current phased control budget is taken as the numerator. An adaptive denominator is constructed by taking the maximum value of the absolute value of the phased budget and the minimum positive zero-prevention constant to avoid calculation anomalies caused by a denominator of 0. Finally, the fractional result is constrained to the specified range using a value range pruning operator. The interval is used to obtain the dimensionless standardized cost deviation rate. In the distortion threshold calculation stage, a differentiated adaptive threshold is generated based on the proportion of each entity's current budget to the total subcontracted budget, combined with an adjustable benchmark adjustment constant and a square root attenuation term. This achieves a hierarchical control logic where the larger the budget, the wider the allowable deviation range. In the distortion judgment stage, the absolute value of the cost deviation rate is compared with the adaptive distortion threshold, and a binary distortion label is output. If the label is 1, the allocation is judged to be distorted and the cost redistribution iteration process is initiated. If the label is 0, the current allocation result is compliant and no recalculation is required. This step uses an adaptive zero-fractional and value range constraint operator to output a cost deviation index with unified boundaries, and is equipped with a dynamic distortion threshold linked to the budget volume to complete the automated judgment. This solves the defects of traditional deviation calculation, such as easy collapse, boundless index, single distortion judgment standard, and reliance on human subjective judgment. It provides an objective and auditable quantitative judgment basis for dynamic cost allocation fault-tolerant iteration and project budget risk classification and early warning.

[0217] Cost deviation rate is the proportion of deviation between the actual cost of the current period and the phased control budget; the preset allocation distortion deviation rate threshold is the critical deviation proportion for determining that the original cost allocation result is distorted and needs to be reallocated.

[0218] The formula for calculating the cost deviation rate is:

[0219] (20)

[0220] In formula (20), For the first The deviation rate of the responsible entity's budgeted cost, a dimensionless proportional indicator, is the core output of this formula; it is used to assess the deviation between the actual settlement cost of the current period and the phased budget, serving as the basis for determining allocation distortion. The value is taken within a fixed range after clipping. : The actual cost of responsibility exceeded the budget; The actual cost of liability was lower than the budget. The larger the value, the greater the deviation from the budget. Output the first value for formula (19) The entity's primary liability cost for the current period is the actual settlement cost for the current period after deducting default penalties and warranty deposits. For the first The current phased construction control budget amount is in ten thousand yuan; it is taken from the phased budget ledger calculated by BIM in the early stage of the project and is the upper limit of the cost control benchmark for the current period of this subcontractor. This is an adaptive zero-protection term for the denominator; This represents the absolute value of the installment budget. For extremely small positive numbers, prevent zero constants; when budgeted in installments The denominator will not be 0, avoiding computational overflow and program errors; when the budget is not 0, the minimal constant has no significant disturbance to the result. For the range clipping operator, the forced deviation rate falls within the range of... If the result of the fraction calculation Output upper limit 1; if the fraction calculation result Output lower limit -1; constrain the deviation rate to not amplify infinitely under extreme over-budget / significant savings conditions, and unify the horizontal comparison caliber. The formula (20) uses the difference between the actual responsibility cost and the phased budget amount as the numerator, and takes the maximum value of the absolute value of the budget and the minimum constant as the adaptive denominator to avoid division by zero anomalies. Then, the deviation rate is locked at the range by the value range trimming operator. The range outputs a standardized budget cost deviation rate; used to quantify the deviation of each subcontractor's current settlement cost from the control budget, serving as a quantitative indicator for the automated judgment of cost allocation distortion. This solves the defects of traditional deviation formulas, such as calculation collapse, lack of numerical boundaries, and subjective judgment of distortion, and provides an objective and auditable quantitative basis for dynamic cost allocation tolerance and budget risk classification and early warning of construction projects.

[0221] The formula for determining cost allocation distortion is:

[0222] (twenty one)

[0223] In formula (21), For the first The responsible entity sets a pre-defined threshold for the distortion deviation rate, which is dimensionless. The threshold changes dynamically with the proportion of the entity's budget to the total budget. The larger the budget, the higher the allowable deviation threshold, and the smaller the proportion of the budget, the stricter the control. The budget benchmark adjustment constant is preferably set to 0.15 in this embodiment; it can be manually adjusted according to the project type (building construction / municipal / highway) and project risk level to uniformly scale the entire subcontracting threshold range. For the first The current period's phased budget amount is defined using formula (20). The total budget amount for all participating entities sharing the responsibility in the current period. . The budget percentage is reduced by the square root of the decrease; adaptive control of budget volume is achieved: larger subcontracting has more room for error, while the control of deviations in small subcontracting is tightened, which is in line with the logic of project cost risk control. Output the standardized budget cost deviation rate for formula (20), in the range . The absolute value of the cost deviation rate represents the extent to which the actual cost of responsibility deviates from the budget. The binary distortion identifier variable is the core output of this formula: If the absolute value of the deviation exceeds the preset threshold for the allocated distortion deviation rate, it is determined that the allocation is distorted and the cost reallocation iteration process is initiated. If the deviation is within the preset allocation distortion deviation rate threshold, the current allocation result is compliant and there is no need to re-divide the cost. The first line of the formula (21) relies on the square root of the proportion of each subcontractor's current budget to the total budget, combined with an adjustable benchmark constant to solve the differentiated adaptive distortion threshold, so as to achieve tolerance for errors in large subcontracts and strict control over small subcontracts; the second line compares the absolute value of the cost deviation rate with the preset allocation distortion deviation rate threshold in segments, and outputs a 0 / 1 binary distortion label to automatically determine whether the current cost allocation is distorted and whether to start reallocation; it solves the defects of traditional fixed thresholds that do not distinguish budget volume and the subjective nature of distortion judgment, and provides a standardized and quantitative judgment basis for the automated error tolerance of dynamic cost allocation and hierarchical budget risk control of construction projects.

[0224] Step S720: Set the performance cycle weight coefficient and schedule deviation weight coefficient, and calculate the first difference allocation weight of each responsible party by combining the actual performance period of the responsible party and the overall schedule deviation.

[0225] The set value range is all Contract fulfillment period weighting coefficient Schedule Deviation Weighting Coefficient And the two coefficients are required to satisfy To achieve a unified total contribution, the original fusion weights of each entity are obtained by combining the proportion of each entity's actual performance period to the total overall performance period and the proportion of the cumulative absolute duration of progress deviation to the total overall deviation duration, through a double-coefficient weighted fusion solution. Then for all The original fusion weights of each responsible entity are summed globally, and the ratio of the original weight of each entity to the total global weight is normalized to obtain the standardized first difference allocation weight used for the secondary redistribution of the over-budget amount. The first difference allocation weight takes into account both the actual performance time of subcontractors and the responsibility for project delays. The sum of the first difference allocation weights of all entities is always equal to 1. It is specifically used when the cost reallocation process is initiated after the allocation distortion is determined in step S710. It solves the shortcomings of traditional secondary reallocation that only relies on the budget volume and cannot quantify the size of the responsibility of each entity related to the project period. It provides an auditable, standardized, and quantitative basis for the fair division of the total amount exceeding the budget and the cost error correction under distorted conditions.

[0226] The first difference allocation weight is a standardized weight constructed by combining the performance duration and the overall progress deviation, used for the secondary redistribution of the total amount exceeding the budget.

[0227] The formula for calculating the first difference allocation weight is:

[0228] (twenty two)

[0229] In formula (22), The performance period weighting coefficient is dimensionless and has a range of values. Control the contribution ratio of the actual contract performance period in the weight calculation. The schedule deviation weighting coefficient is dimensionless and has a range of values. Control the contribution percentage of schedule deviation time in the weighting calculation. Constraints To achieve dual-coefficient normalized coupling, the sum of the weight contributions of the two dimensions is always 1, avoiding an imbalance in the total weight. This can be adjusted according to the project management focus; for example, if the focus is on meeting the project schedule, the weight can be increased. Focusing on delayed claims increases . For the first The responsible party's actual and complete performance period of the contract; The total contractual period for all responsible parties during the current period. The weighting of each item in the performance cycle. For the first The cumulative progress of the main project deviates from the absolute duration (absolute value of the project delay). To ensure that the progress of all subcontracting projects deviates from the total duration, The weight of the percentage of progress deviation items. The original weights, which were not globally unified, were integrated with two core dimensions: contract fulfillment time and project delay, as the base values ​​for weight redistribution. For all of the current period The original fusion weights of each responsible entity are summed globally. This is the core output of the formula, the normalized first difference allocation weight, which is dimensionless; it satisfies... It is specifically designed for redistributing the secondary cost exceeding the budget under distorted working conditions. This represents the total number of responsible entities participating in the cost redistribution during the current period. The formula (22) first constrains the sum of the weighted coefficients for performance and schedule to be 1; then it integrates the proportion of actual performance period of each subcontractor and the proportion of cumulative schedule deviation time, and obtains the original integrated weight through double coefficient weighting; finally, it outputs the standardized first difference allocation weight with a sum of 1 by summing and normalizing the global weights; it is used for the secondary redistribution of the total amount exceeding the budget under the condition of distorted cost allocation, and simultaneously takes into account the performance completion degree and the responsibility for the schedule delay, solving the defects of the traditional redistribution based only on the budget and the inability to quantify the schedule responsibility, and providing a fair and auditable standardized splitting basis for the dynamic cost tolerance correction of construction projects.

[0230] Step S730: First, calculate the total over-budget amount for the current period of the project, constrain the total over-budget amount to be non-negative, and redistribute the total over-budget amount among the responsible entities based on the first difference allocation weight. The over-budget amount is then added to the current budget amount to obtain the final settlement cost of each responsible entity after distortion correction. The final settlement costs of all responsible entities together constitute the first settlement bill of the project.

[0231] First, the difference between the current primary responsibility cost and the phased budget amount for all responsible entities is summarized. The total global over-budget amount is calculated using a non-negative constraint operator. A secondary redistribution of over-budget costs is performed only when the project as a whole has an overspending situation; otherwise, the total over-budget amount is set to zero without reducing the costs of individual entities. Based on the normalized first difference allocation weight output in step S720, the total global over-budget amount is allocated to each responsible entity. This is then combined with the corresponding phased budget amount to obtain the pre-settlement amount. A value range trimming operator is used to impose a double constraint on the pre-settlement amount, with a lower limit of 0 and a maximum contract cost cap, outputting the compliant final settlement cost after distortion correction. Finally, an AES symmetric encryption operator is used to verify the final settlement cost for each entity. The settlement cost is bound to an audit index set containing the original BIM data of the entire process and encrypted to generate tamper-proof audit archive records. All encrypted archive records of the main entities are summarized to form the first settlement bill of the project. This step integrates the global excess base calculation, secondary cost redistribution based on performance and schedule responsibility, dual-layer compliance boundary constraints of settlement amount, and encrypted traceability archiving of settlement data. It solves the defects of traditional redistribution that do not distinguish the project balance status, do not quantify the period-related responsibilities of allocation, are easy to break through financial and contractual constraints in settlement amount, and are easy to tamper with settlement ledgers and cannot be fully audited and reproduced. It provides reproducible and tamper-proof quantitative settlement basis for standardized and compliant settlement after the correction of construction project distortion and financial audit.

[0232] The first settlement statement is a standardized settlement document from multiple parties that has been corrected twice for the distorted weighting and adapted to the actual project performance and progress.

[0233] The final settlement cost calculation formula is as follows:

[0234] (twenty three)

[0235] In formula (23), This is the sum of the current period's responsibility cost and the corresponding periodic budget difference for all subcontracted items; a positive sum indicates an overall budget overrun, while a negative sum indicates an overall budget surplus for the project. It is a non-negativity constraint operator; if the overall budget is in surplus, it forces the total excess to be... Costs are not reduced and allocated to subcontractors; reallocation is only initiated when there is an overall overspending. Output the first value for formula (19) The primary responsibility cost for the main entity in the current period; For the corresponding installment budget amount; This represents the total number of responsible entities in the current period. The output of this layer is the total amount of over-budget for the current period of the project, which serves as the base for the over-budget to be allocated in the distortion redistribution. The normalized first difference allocation weight is output for formula (22), which integrates the two dimensions of performance period and schedule lag, and is used as the excess cost allocation ratio. To be allocated to the first The amount of excess contribution by the main entity; and its own installment budget. The sums are used to obtain the pre-settlement amount. For the two-level value range pruning operator: Lower limit 0: to prevent negative settlement costs and comply with financial accounting standards; Upper limit The subcontract should stipulate a cap on the cost of each item to prevent the settlement amount from exceeding the contractual constraints after redistribution. This is the core cost output of this formula, and the compliant final settlement cost after distortion secondary allocation correction is used for subcontract payments and final settlement ledgers. The AES symmetric encryption operator is used to encrypt and store the final settlement cost to prevent settlement data from being tampered with and to meet the integrity requirements of financial and audit data. It is a full-chain audit index collection, containing a complete set of original data indexes for BIM component codes, time schedules, budget ledgers, change orders, and performance assessments, and is bound with encrypted records for complete traceability and reproduction afterward. To encrypt and archive audit records, all encrypted records of the main entities are aggregated to generate a standardized first settlement bill. The formula (23) firstly aggregates the total cost budget difference of all subcontracting and sets non-negative constraints to obtain the total over-budget amount of the project in the current period; the secondly allocates the over-cost with the difference allocation weight of performance-schedule dual dimensions, superimposes the phased budget, and outputs the compliant final settlement cost through dual-level value domain trimming; the thirdly uses AES symmetric encryption to bind the full-process BIM audit index to generate tamper-proof archived records; the integrated completion of distorted working condition over-allocation, settlement compliance constraints and audit traceability encryption solves the defects of traditional redistribution that does not distinguish project surplus, allocation without construction period responsibility, uncontrolled settlement values, and easy tampering and untraceability of ledgers, providing reproducible and tamper-proof quantitative settlement basis for standardized and compliant settlement of construction projects and financial audit.

[0236] Step S740: Using the project schedule deviation, change cost, multi-party allocation weight, and settlement bill data as training samples, the gradient descent algorithm is used to iteratively update the weight parameters of the baseline schedule and budget calculation model, thereby realizing iterative optimization of the enterprise cost database and closed-loop management of project cost throughout its entire life cycle.

[0237] The model training samples are collected from current project schedule deviations, changed costs, weights allocated to each responsible party, and encrypted archived settlement invoices. A gradient descent algorithm with L2 regularization and exponentially decaying learning rate is used to iteratively update the baseline schedule and budget calculation model weights, enabling autonomous iterative optimization of the enterprise cost database and forming a closed-loop cost management system throughout the project lifecycle. During iterative calculations, a joint loss function is first constructed, integrating the mean squared error of cost prediction and the L2 weight penalty term, to quantify the deviation between predicted costs and actual settlement costs and suppress model overfitting. Then, the real-time decaying learning rate is calculated based on the initial learning rate, decay coefficient, and the current iteration round. In the later stages of iteration, the weight update step size is gradually reduced to achieve smooth convergence. Finally, the model is updated along the negative gradient direction of the loss function. The entire weight matrix is ​​configured, and dual termination conditions are set: "the difference in loss between adjacent rounds is less than the convergence threshold" and "the number of iterations reaches the preset maximum number of rounds" to avoid oscillations, divergence, and infinite loops. Simultaneously, two operating modes are distinguished: in offline batch training mode, complete weight updates are performed and the enterprise cost database is updated synchronously; in online real-time cost allocation calculation mode, all model weights are frozen to ensure the stability of current settlement calculations. This step relies on real final settlement data to achieve fully automatic self-learning optimization of the BIM progress and budget calculation models, overcoming the shortcomings of traditional cost models that rely on manual parameter tuning, cannot automatically accumulate historical project cost patterns, and are easily affected by model training interference in online dynamic settlements. This establishes a fully automatic closed-loop dynamic cost control system covering the entire project lifecycle.

[0238] Gradient descent is a machine learning optimization algorithm that uses actual settlement deviations as loss to reverse-correct model weights and continuously improve the accuracy of cost estimation.

[0239] The parameter iterative update formula is:

[0240] (twenty four)

[0241] In formula (24), This is the overall loss function of the model and the core indicator of this layer. It is used to measure the deviation between the predicted cost and the actual settlement cost, and to guide the iterative optimization of the weights. This is the mean squared error loss term; The projected cost output by the baseline schedule / budget model; The true final settlement cost after distortion correction is output for formula (23); the global deviation between the predicted cost and the actual settlement is quantified. It is the L2 regularization coefficient, a non-negative adjustable constant; used to constrain the weight magnitude, suppress model overfitting, and avoid excessive interference from single project samples with the general cost estimation logic. Weight matrix The squared L2 norm is used to apply an amplitude penalty to all parameters of the model. Set the initial base learning rate for the model and preset hyperparameters for offline training; control the step size of the first round of weight updates. This is the learning rate decay coefficient, with a value of [value missing]. The attenuation effect increases with each iteration. This represents the current iteration round that has been completed; the higher the round, the smaller the learning rate, and the more stable the subsequent fine-tuning of the weights. The learning rate is decayed in real time during the epoch, and the update step size is gradually reduced in the later stages of the iteration to prevent loss oscillations from causing convergence. This is the original weight matrix of the model before iteration, which includes all branch parameters for progress projection and budget calculation. For loss function For the weight matrix The gradient represents the direction and magnitude of the influence of each parameter on the prediction bias. This is the new weight matrix updated in this iteration, used for loss calculation in the next round. (Or, condition one) The absolute value of the difference between the losses of two adjacent rounds is less than the model loss convergence threshold. If the model converges and the prediction accuracy meets the target, stop iterating; or if condition two is met. To reach the preset maximum number of iterations in the current iteration round To prevent infinite iteration and dead loops, training is forcibly terminated. Offline mode collects batches of historical project samples, performs a complete gradient descent weight update, and then iterates the updated enterprise cost model into the database. The online calculation mode freezes the weights completely when allocating the project's real-time dynamic cost, without modifying the parameters, ensuring the stability of the current settlement calculation. The formula (24) first layer constructs a joint loss function that integrates the mean square error of cost prediction and L2 regularization penalty; the second layer adopts an exponential decay strategy to dynamically adjust the iterative learning rate; the third layer updates the model weight matrix in reverse through the gradient descent algorithm; the fourth layer sets a double stopping condition of loss convergence and maximum iteration rounds; the fifth layer distinguishes between offline training with updateable weights and online calculation with frozen weights; relying on the real settlement data of the project, the BIM progress and budget calculation model is self-iteratively optimized, solving the defects of traditional cost models such as manual parameter adjustment, inability to learn historical completion rules independently, and online settlement being easily interfered with by training, and constructing a fully automatic closed-loop control system for dynamic cost of the entire life cycle of construction projects.

[0242] Optimal Network Structure for Baseline Schedule and Budget Calculation Model

[0243] A two-layer fully connected feedforward neural network is adopted: temporal feature input layer (output features of dimension matching step S230) → 128-neuron fully connected layer (ReLU activation) → 64-neuron fully connected layer (ReLU activation) → single-dimensional cost prediction output layer; linear regression and GBDT gradient boosting tree are equivalent replacement models in this field and do not change the core logic of iterative optimization of this invention.

[0244] The following detailed description of the BIM-based dynamic cost allocation calculation method for the entire lifecycle of construction projects provided by this invention uses specific embodiments:

[0245] 1. Basic test conditions for the project

[0246] 1.1 Project Overview

[0247] This is an EPC general contracting project for a 26-story high-rise residential building, with a total construction area of ​​58,000 square meters and a total construction period of 720 calendar days. The construction phase is selected as the calculation period for the main structure construction stage (day 121–day 420). Actual on-site conditions: the main structure construction is 12 days behind schedule; design changes have resulted in an additional 1200 tons of steel reinforcement; there are 7 participating entities: the construction unit, general contractor, civil engineering subcontractor, mechanical and electrical subcontractor, decoration subcontractor, steel reinforcement supplier, and tower crane rental company. The total budget for the main structure phase is controlled at 128 million yuan, of which the phased budget for civil engineering subcontracting is 76 million yuan, mechanical and electrical subcontracting is 32 million yuan, and decoration is 20 million yuan. Basic unit prices: labor 0.028 million yuan / man-day, steel reinforcement 0.42 million yuan / ton, tower crane 0.18 million yuan / shift; management fee rate 8%, construction industry value-added tax 9%.

[0248] 1.2 Experimental Grouping

[0249] Experimental group: This embodiment uses the complete algorithm for the entire process from S100 to S740 (with all thresholds in Table 1).

[0250] Comparative example: Existing traditional BIM static cost estimation schemes (corresponding patent backgrounds CN114969888A, CN112785257A, lacking lightweighting, time-series mining, schedule-based cost adjustment, source tracing, distortion correction, and model iteration).

[0251] Table 1. Complete set of control threshold parameters for this embodiment.

[0252] 2. The experimental group in this embodiment is calculated step by step (see Table 1 for threshold determination).

[0253] Step S100: Lightweight analysis of the BIM model to construct a standardized first component set.

[0254] Total number of original IFC complete components: 12,680; 42 components with abnormal geometric dimensions were removed by 3D bounding box collision check;

[0255] A 5-dimensional component feature vector was constructed according to the formula, and cosine similarity was calculated using a T1 threshold of 0.95 to remove duplicates: 1,860 duplicate and redundant components in the same contract section were removed, and all legally reused standard floor slabs, beams and columns across contract sections were retained; 79 invalid components with missing material or schedule parameters were removed.

[0256] Data was collected using a dual index of "construction section + specialized engineering" and assigned a globally unique BIM code, resulting in 10,701 standardized valid components. Quantitative data: Loading the original full model took 12 minutes; loading the lightweight dataset took 47 seconds, with a component data compression rate of 31.4%, and no valid components were mistakenly deleted.

[0257] Step S200: One-dimensional CNN temporal feature mining to generate the first value consumption sequence.

[0258] The basic sliding time series window is 30 days. There are no major design changes in this project, so the window span will not be adjusted. A full-cycle time series sampling time set will be generated.

[0259] Pre-trained one-dimensional CNN convolution operator extracts three-dimensional resource features of manpower, materials, and machinery for each sampling window; resource vector of sampling window on day 180: [1260 man-days, 3680 tons, 420 machine shifts];

[0260] Calculate the static cost of the window benchmark based on the pricing formula: The cost is calculated in ten thousand yuan; a continuous monthly value consumption sequence is generated by splicing the data in time sequence, and the full-cycle time-sharing capital data is output. Quantitative data: Traditional solutions only output one value, the total amount in installments. This embodiment outputs 10 sets of monthly time-series costs, which can be directly used for cash flow calculation.

[0261] Step S300: Compare the progress time series and extract the first time series difference features (T2=7 days).

[0262] The planned construction time for the reinforced steel components was 180 days, but the actual construction time was 192 days. The time deviation of a single node was ΔT = 192 - 180 = 12 days.

[0263] |ΔT|=12>7 (T2 fault tolerance threshold), indicating substantial progress delay, triggering price adjustment feature extraction;

[0264] Constructing a differential feature vector by splicing three-dimensional indicators Quantitative data: The proportional data does not have a schedule deviation identification logic, completely ignoring the idle labor and machinery costs caused by the 12-day delay.

[0265] Step S400: Change matching + delayed price adjustment, calculate the first dynamic cost (T3=[1, 1.35]).

[0266] calculate The L1 norm is 12.402, the sensitivity coefficient for lagged price adjustments in housing construction is 0.026, and the basic adjustment coefficient is 1 + 0.026 × 12.402 = 1.322, falling within the range of 1 to 1.35. ;

[0267] The net increase in steel reinforcement is 1200t, and the bare cost of the change is 1200 × 0.42 × 1.08 × 1.09 = 5,941,000 yuan; the total cost of the change after adding the delayed compensation is 5,941,000 × 1.322 = 7,854,000 yuan.

[0268] The current building materials price adjustment index is 1.02, and the first dynamic cost index is [missing data]. The estimated cost is 10,000 yuan. Quantitative data: The comparative static change cost is only 5,941,000 yuan, excluding construction period losses, with a relative deviation of 32.2%.

[0269] Step S500: Multi-dimensional reverse tracing to generate multi-level first tracing paths.

[0270] By matching BIM components, construction logs, and subcontracting contracts using three layers of rules, the system identifies the directly responsible civil engineering subcontractor (Sub01). A four-level structured traceability chain is generated (owner → supervisor → general contractor → civil engineering subcontractor), with each cost detail linked to a BIM code, construction period, and work team record. Quantitative data: Traditional manual cost tracing averages 8 hours per item; this system retrieves data in ≤10 seconds with a single click.

[0271] Step S600: Break down the performance rate and calculate the primary responsibility cost (T4 = 30% deduction cap).

[0272] The planned workload for the civil engineering subcontract was 76 million, with an actual completion of 69.2 million during the current period. The contract fulfillment rate was calculated using the T6 zero-prevention constant. ;

[0273] The static weight of the civil engineering contract = 7600 / 12800 = 0.5938, the original fusion weight = 0.5938 × 0.9105 = 0.5407, and the initial allocated cost after normalization = 2685.39 × 0.5407 = 1452.08 million yuan;

[0274] The current period's default penalty is 1.12 million, and the warranty reserve is 726,000, totaling 1.846 million in penalties; the penalty limit = 14.5208 million × 30% = 4.3562 million, which is not exceeded and will be deducted in full;

[0275] Civil engineering first responsibility cost The cost is 13.426 million yuan. Quantitative data: Based on the proportional allocation of fixed contract costs, the civil engineering cost allocation is 13.426 million yuan, which deviates from the actual cost of responsibility by 9.6%.

[0276] Step S700, distortion determination, weight correction, model closed-loop iteration (T5 / T7 / T8 / T9 / T10)

[0277] The current phased budget for civil engineering is 14.2 million, with a cost deviation rate of... ;

[0278] Adaptive distortion threshold , Without any distortion in cost allocation, 12.6748 million was directly used as the final settlement cost for civil engineering.

[0279] The entire process progress, changes, allocation, and settlement data are used as training samples, and gradient descent with L2 regularization is used for iteration; the loss difference for 20 consecutive rounds is <1×10⁻ 5 (T9) Iteration stops at 300 rounds (T10), and the enterprise cost benchmark model is updated offline. Quantitative data: Traditional solutions have no distortion correction, with a distortion rate of 21.3% for small-scale subcontracting; in this embodiment, the distortion rate is ≤1.2% after distortion correction; the cost prediction error for subsequent similar projects is reduced from 11.8% to 3.1%.

[0280] 3. Complete calculation process for comparative scale (traditional BIM static cost scheme)

[0281] 1. Directly loads all 12,680 BIM components without lightweighting or deduplication, resulting in model loading lag and extraction of only static quantities; 2. Uses fixed quotas for one-time pricing, without dividing construction timelines or resource consumption timelines; 3. Does not identify project delays, changes are calculated solely based on drawing quantities, without compensation for idle labor, idle machinery, or material delays and price increases; 4. Only allocates costs according to a fixed percentage of the subcontract amount, without dynamic adjustments based on performance rates or penalty deductions for breach of contract; 5. Lacks multi-dimensional traceability links, distortion identification, and secondary redistribution; calculated data is not fed back to optimize the baseline model.

[0282] Comparative core measurement results

[0283] The current static total cost = 1862.74 + 594.10 = 2456.84 million yuan; the fixed proportion of the civil engineering cost = 1342.60 million yuan; the settlement cost after deducting the fixed quality guarantee is 1270.00 million yuan; the cost loss caused by the delay due to the missed construction period is 191.30 million yuan; the error of the new project cost forecast remains at 11.8%.

[0284] Table 2. Quantitative Comparison of the Effects of This Example vs. Traditional Comparative Techniques

[0285] 4. Sufficient data to support feasibility

[0286] 4.1 Feasibility of Data Source Implementation

[0287] All input data in this solution (IFC-BIM, overall construction schedule, change orders, construction logs, subcontracts, and cost information) are mandatory archiving standard data for domestic housing construction, municipal, and rail projects. Statistics from 12 pilot EPC projects show that the additional workload for on-site record entry only increases by an average of 4%, and no new data collection equipment is required.

[0288] 4.2 Data on the Implementation of Hardware and Software Computing Power

[0289] 1. Lightweight BIM processing: A standard 8-core 16GB office PC can clean components for a 58,000㎡ project in ≤1 minute; traditional full model calculation with equivalent configuration takes ≥10 minutes. 2. One-dimensional CNN offline training: A standard 4-core 8GB cloud server can train a single project in ≤30 minutes; online real-time calculation of frozen model weights can calculate single phase cost in ≤3 seconds. 3. No reliance on high-end computing power; existing office equipment and basic cloud servers in construction companies can support the entire algorithm.

[0290] 4.3 Evidence of Industry Standard Adaptation

[0291] The complete set of thresholds and calculation methods fully match GB50500 "Construction Engineering Quantity List Pricing Specification", housing construction cost adjustment documents, and EPC general contracting model contracts; it can exchange data with mainstream BIM quantity calculation software such as Glodon and Luban, and existing cost systems only need to add algorithm modules to be modified without replacing the entire platform.

[0292] 4.4 General Verification Data for Multiple Engineering Types

[0293] Adjusting the T5 distortion baseline constant, sliding window duration, and lag price adjustment sensitivity coefficient makes it suitable for all types of projects: 1. Municipal roads: T5 is adjusted to 0.18, with an average calculation error of 3.6%; 2. Rail transit: the window is adjusted to 45 days, with an allocated distortion rate of <1.4%; 3. Industrial plants: the upper limit of the price adjustment coefficient is maintained at 1.35, with a prediction error of 2.9%. Multi-sector testing proves that the solution has no industry barriers to use.

[0294] 4.5 Feasibility of Industrialized Software Development

[0295] In this embodiment, each step from S100 to S740 provides complete mathematical formulas, judgment thresholds, and input / output boundaries, with no black-box logic. It can be divided into six independent modules: BIM analysis, time series characteristics, dynamic price adjustment, cost traceability, allocation correction, and model iteration. Two cost estimation software companies have already completed prototype development and piloted it in residential projects, making it ready for commercial product development.

[0296] 4.6 Economic benefits of project implementation (average of 5 pilot projects)

[0297] 1. The average completion settlement period is shortened by 28 days; 2. The cost audit period is shortened by 72%; 3. The risk of investment overruns is reduced by more than 70%; 4. Cost consulting service fees for similar long-term projects can be reduced by 12%.

[0298] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A BIM-based method for dynamic allocation and calculation of the entire lifecycle cost of a construction project, characterized in that, Includes the following steps: S100. Perform lightweight analysis on the BIM engineering information model of the construction project, extract basic data of engineering components, establish the mapping relationship between components and multi-dimensional attributes based on the physical attributes and construction process attributes of each component, and construct a standardized first component set after deduplication of duplicate components and cleaning of invalid data. S200. Import the project's preset baseline schedule data. Based on the first set of components, use a pre-trained machine learning feature extraction model to mine the resource input time sequence characteristics of each component at different construction stages. Combine the engineering pricing rules to conduct a full life cycle value consumption deduction and generate the first value consumption sequence. S300. Compare the time nodes of each resource consumption in the first value consumption sequence with the preset benchmark progress data in a time sequence to identify the progress deviation status of component resource consumption. When there is a progress deviation, extract the multi-dimensional structured first time sequence difference feature. S400: Obtain incremental data of engineering quantity changes corresponding to design changes and on-site visas; classify and match changed engineering quantities based on BIM component codes; adjust the price for schedule lag based on the first time sequence difference characteristics; and obtain the first dynamic cost after the budget rate is adapted to dynamic pricing. S500 uses BIM engineering quantities, construction logs, and subcontracting contracts as traceability data sources. Through a preset multi-dimensional cost traceability model, it reverse-traces the cost of all elements including material costs, labor costs, machinery usage fees, and management fees in the first dynamic cost, and generates a multi-level first traceability path that binds the performance information of each participating party. S600. Based on the scope of duties and contribution ratio of the responsible entity corresponding to the first traceability path, the first dynamic cost is broken down into multiple cost components. After deducting the default penalty and the quality assurance reserve amount, the first responsibility cost corresponding to each responsible entity is calculated. S700: Compare the primary responsibility cost of each responsible entity with the corresponding initial budget amount for each phase to identify cost allocation distortions; when allocation distortion is determined to exist, construct differential allocation weights based on the actual project performance cycle and the degree of overall progress deviation, and redistribute the current period's responsibility cost to obtain the corrected first settlement bill; use the entire process cost, progress, and allocation data as samples to iteratively update the benchmark progress and budget calculation model, and realize dynamic closed-loop allocation calculation of the construction project's full life cycle cost.

2. The method for dynamic allocation and calculation of the full life cycle cost of construction projects based on BIM as described in claim 1, characterized in that, Step S100 specifically includes: S110. Perform lightweight analysis and 3D bounding box collision compliance verification on the BIM engineering information model, and extract basic parameters such as component quantity, material code, construction section number, planned start time, and planned completion time in batches to construct the basic feature vector corresponding to each component. S120. Calculate the cosine similarity of the basic feature vectors of any two components using the cosine similarity algorithm, and identify duplicate components by combining the preset duplicate component similarity threshold; retain the legal reuse scenarios of the same component assigned to different construction sections, and only remove duplicate redundant components and invalid attribute data. S130. Using construction sections and professional engineering types as dual search indexes, the cleaned effective component feature vectors are aggregated to construct a standardized first component set with unique BIM codes and attribute search indexes.

3. The method for dynamic allocation and calculation of the full life cycle cost of a construction project based on BIM as described in claim 2, characterized in that, Step S200 specifically includes: S210. Set a fixed sliding time sequence sampling window based on the planned time nodes of each sub-project within the preset baseline schedule. When the project undergoes major design changes or the overall construction period is extended, the time span of the window is adaptively adjusted to construct a set of time sequence sampling moments. S220. Use a one-dimensional effective convolution feature extraction operator to perform temporal feature mining on the first component set, and extract the temporal feature vectors of the input of three types of resources, namely manpower, materials and machinery, at each sampling time. S230. Based on the engineering quota pricing rules, the resource input at each time point is matched with the current information unit price to calculate the current resource consumption cost. All current resource consumption costs are collected along the sliding time sequence sampling window to generate the first value consumption sequence that represents the capital consumption pattern of the entire project life cycle.

4. The method for dynamic allocation and calculation of the full life cycle cost of a construction project based on BIM as described in claim 1, characterized in that, Step S300 specifically includes: S310. For each resource consumption node in the first value consumption sequence, the single-node schedule deviation is obtained by subtracting the baseline planned time from the actual occurrence time. S320. Compare the absolute value of the single-node schedule deviation with the preset allowable deviation threshold, and extract features only for deviation nodes that exceed the deviation threshold. S330, absolute duration of splicing progress lag, relative amount of current period resource over-investment, and process interleaving time offset coefficient are used to construct a multi-dimensional structured first time-series difference feature vector.

5. The method for dynamic allocation and calculation of the full life cycle cost of a construction project based on BIM as described in claim 4, characterized in that, Step S400 specifically includes: S410. Classify and match the changed quantities of work according to the unique BIM component codes. Based on the first time-series difference feature vector, calculate the first norm of the eigenvector and combine it with the lag price adjustment sensitivity coefficient preset based on the historical full life cycle cost sample to solve the schedule lag price adjustment correction coefficient. S420. Retrieve the official unit prices of labor, materials and machinery for the current period and calculate the total cost of the current period's change based on the resource consumption after adjustment by the aforementioned progress lag price adjustment coefficient. S430. The project's phased base contract cost and the total change cost for the current period are combined and then summed after non-negative constraint verification to obtain the project's first dynamic cost for the current period.

6. The method for dynamic allocation and calculation of the full life cycle cost of a construction project based on BIM as described in claim 1, characterized in that, Step S500 specifically includes: S510. Using three types of data—BIM component quantities, time-series construction logs, and subcontracting contracts—as the basic data sources for traceability, a multi-dimensional rule-matching traceability mapping function is used to locate the unique code of the responsible party for each expense in the first dynamic cost. S520. Construct a multi-level reverse tracing link according to the time sequence of cost occurrence and the level of responsibility of participating construction units. Encapsulate the unique code of the responsible entity for the cost, the code of each level of participating construction unit, the time of cost occurrence, and the amount of each cost item into a structured first tracing path.

7. The method for dynamic allocation and calculation of the full life cycle cost of construction projects based on BIM according to claim 1, characterized in that, Step S600 specifically includes: S610. Based on the first traceability path, the actual completed work volume and the contracted planned work volume of each responsible entity are statistically analyzed, and a minimum zero constant is introduced to calculate the performance rate of the responsible entity. S620. Normalize the project schedule performance rate of all responsible parties, and use the normalized project schedule performance rate as the cost allocation weight to decompose the first dynamic cost of the project level by level to obtain the initial allocated cost of each responsible party. S630. Based on the initial allocated cost, deduct the current period's default deductions and warranty reserve amount, and ensure that the settlement cost is non-negative through maximum value constraints, to calculate the first liability cost corresponding to each responsible party.

8. The method for dynamic allocation and calculation of the full life cycle cost of a construction project based on BIM as described in claim 1, characterized in that, Step S700 specifically includes: S710. Calculate the cost deviation rate of each responsible entity's current period responsibility cost relative to the phased budget amount, and compare the absolute value of the cost deviation rate with the preset allocation distortion deviation rate threshold to determine whether there is cost allocation distortion in the current period responsibility cost. The formula for calculating the cost deviation rate is: ； in, For the first Budget cost deviation rate of the responsible entity For the first The primary responsibility cost of the main entity in the current period For the first The budget for phased construction of the main structure should be controlled. This is an adaptive zero-protection term for the denominator. This represents the absolute value of the installment budget. For extremely small positive numbers, zero-prevention constant, Range trimming operator; The formula for determining cost allocation distortion is: ； in, For the first The responsible party pre-sets a threshold for allocating the distortion deviation rate. This is the budget benchmark adjustment constant. For the first The main body's current period phased budget amount, The total budget amount for all participating entities sharing the responsibility in the current period. This represents the absolute value of the cost deviation rate. For binary distortion identification variables; S720. Set the performance cycle weight coefficient and schedule deviation weight coefficient, and calculate the first difference apportionment weight of each responsible party by combining the actual performance period of the responsible party and the overall schedule deviation. The formula for calculating the first difference allocation weight is: ； in, The performance period weighting coefficient, This is the schedule deviation weighting coefficient. For the first The responsible party's actual and complete performance period of the contract; The total contractual period for all responsible parties during the current period. For the first The cumulative progress of the main body deviates from the absolute duration. To ensure that the progress of all subcontracting projects deviates from the total duration, For the original weights that are not globally normalized, For all of the current period The original fusion weights of each responsible entity are summed globally. To normalize the first difference allocation weight; S730. First, calculate the total over-budget amount of the project in the current period, constrain the total over-budget amount to be non-negative, and redistribute the total over-budget amount among the responsible entities based on the first difference allocation weight. The over-budget amount is then added to the current budget amount to obtain the final settlement cost of each responsible entity after distortion correction. The final settlement costs of all responsible entities together constitute the first settlement bill of the project. S740. Using the project schedule deviation, change costs, multi-party allocation weights, and settlement bill data as training samples, the gradient descent algorithm is used to iteratively update the weight parameters of the baseline schedule and budget calculation model, thereby realizing iterative optimization of the enterprise cost database and closed-loop management of project cost throughout its entire life cycle.

9. The method for dynamic allocation and calculation of the full life cycle cost of a construction project based on BIM as described in claim 8, characterized in that, In step S730, the final settlement cost calculation formula is as follows: ； in, To sum the current period's responsibility cost and the corresponding installment budget difference for all subcontracting, It is a non-negative constraint operator. For the first The primary responsibility cost of the main entity in the current period To correspond to the installment budget amount, This represents the total amount of the project's current over-budget period. To be allocated to the first The amount of excess contribution by the main entity. For a two-level domain clipping operator, The subcontract stipulates a cap on the cost of each item. It is an AES symmetric encryption operator. This is a set of end-to-end audit indexes. Encrypt audit logs for archiving.

10. The method for dynamic allocation and calculation of the full life cycle cost of a construction project based on BIM as described in claim 9, characterized in that, In step S740, the parameter iterative update formula is: ； in, The overall loss function of the model. This is the mean square error loss term. To predict the cost, The true final settlement cost after distortion correction. The L2 regularization coefficient is... Weight matrix The square of the L2 norm, This is the initial base learning rate for the model. This is the learning rate decay coefficient. This represents the current completed iteration round. The learning rate decays in real time during the epoch. This is the original weight matrix of the model before iteration. For loss function For the weight matrix gradient, This is the new weight matrix after this iteration update. The absolute value of the difference between the losses of two adjacent rounds is less than the convergence threshold. , To reach the preset maximum number of iterations in the current iteration round , Offline mode is used to collect samples of historical projects in batches. In the online calculation mode, the weights are completely frozen during the real-time dynamic cost allocation of the project. For mathematical logic OR operators.

Images