Engineering cost intelligent calculation system and method based on multi-source heterogeneous data fusion

Abstract

The invention relates to the technical field of construction engineering cost management, and discloses an intelligent engineering cost calculation system based on multi-source heterogeneous data fusion, and the system comprises a multi-source data collection module which is used for collecting structured data and unstructured data from a design file, a market database, a construction monitoring system, a contract document, and a historical project library; and the heterogeneous data fusion module is connected with the multi-source data acquisition module and analyzes the risk terms in the contract text by adopting a natural language processing technology. According to the invention, the multi-source data acquisition module is used for widely collecting data in multiple aspects of design, market, construction, contract and the like, the problems of data splitting and information isolated island in traditional cost management are solved, integration of multi-source heterogeneous data is realized, and the heterogeneous data fusion module utilizes advanced technologies of natural language processing, image recognition and the like, so that the cost management efficiency is improved. Contract texts and design drawings can be efficiently analyzed, the processing capacity of unstructured data is improved, and the error rate and omission rate of manual interpretation are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of construction project cost management, and more specifically to an intelligent calculation system and method for construction cost based on multi-source heterogeneous data fusion. Background Art

[0002] In the construction industry, project cost management runs through the entire project lifecycle, encompassing multiple stages including feasibility studies, design, bidding, construction, completion, and operation and maintenance. Traditional project cost management relies heavily on manual operations and empirical judgment. Its core processes include manually compiling bills of quantities, itemizing material prices, labor costs, and equipment costs, and summarizing and reviewing data using paper or spreadsheets. However, as projects become more complex, larger, and more market-oriented, the shortcomings of traditional methods in terms of efficiency, accuracy, and collaboration are becoming increasingly prominent, manifesting in the following pain points:

[0003] 1. Data fragmentation and information silos

[0004] Difficulty integrating heterogeneous data from multiple sources: Construction costs involve multimodal information such as design drawings (CAD / BIM), contract documents, construction logs, market quotes (such as steel and cement prices), and equipment sensor data. Existing technologies typically store structured data (such as database tables) and unstructured data (such as drawings and text) in separate systems (such as ERP, BIM software, and supply chain management platforms), lacking unified data interfaces and standardized processing procedures. For example, design change information must be manually extracted from the BIM model and re-entered into the costing software, resulting in inefficiency and prone to errors.

[0005] Data gaps across project phases: Data between different project phases (e.g., design and construction) lacks coherence. On-site changes during the construction phase (e.g., adjustments to project quantities) are difficult to promptly communicate to the design department, leading to cumulative deviations between budget and actual results. Statistics show that approximately 30% of project cost overruns are due to data disconnections between the design and construction phases.

[0006] 2. Static models are out of touch with dynamic needs

[0007] Prediction model lag: Existing cost forecasts are often based on static models (such as linear regression and random forests) trained based on historical project data. These models are unable to respond in real time to changes in construction schedules, market price fluctuations (such as sudden material price increases), or policy adjustments (such as cost increases due to environmental regulations). For example, in 2021, a bridge project experienced a monthly steel price increase of over 15%. The static model failed to adjust the budget in a timely manner, resulting in a 12% cost overrun.

[0008] Inadequate risk quantification: Traditional methods for quantifying uncertainties (such as construction delays and geological disasters) rely on manual experience and lack probabilistic analysis based on big data. Research shows that manual risk assessments have an error rate as high as 20%-30% and are unable to capture long-tail risks.

[0009] 3. Lack of unstructured data processing capabilities

[0010] Low utilization of drawing and text information: Quantity information in design drawings (such as concrete usage) and risk clauses in contract documents (such as liquidated damages) often rely on manual interpretation, which is time-consuming and prone to missing key details. While existing OCR (optical character recognition) technology can extract text from drawings, it struggles to analyze geometric relationships (such as the quantity of work at beam-column joints), resulting in insufficient automation.

[0011] Limitations of semantic understanding: Ambiguous expressions in contract texts (such as "reasonable construction period") cannot be effectively converted into quantitative parameters by traditional NLP technology, which restricts the construction of risk models.

[0012] 4. Inefficient collaboration and difficult tracing

[0013] Barriers to cross-departmental collaboration: Data sharing between cost estimators, designers, and construction contractors relies on email or meetings, resulting in information delays of up to 3-5 days. On one subway project, delayed synchronization of change notifications resulted in additional rework costs of 800,000 yuan.

[0014] Data tampering risk: Paper documents and local spreadsheets lack tamper-proof mechanisms, making audit traceability difficult. Industry surveys show that 15% of project disputes stem from inconsistent cost data.

[0015] Summary of Industry Pain Points

[0016] Traditional construction cost management suffers from significant shortcomings in data integration, dynamic forecasting, unstructured data processing, and collaborative mechanisms, leading to cost overruns, project delays, and frequent disputes. While existing technologies have attempted to improve these aspects, a comprehensive solution encompassing multi-source data fusion, real-time model updates, and trusted collaboration has yet to emerge. Therefore, an intelligent construction cost calculation system and method based on multi-source heterogeneous data fusion is urgently needed to overcome existing technical bottlenecks and achieve precise, dynamic, and transparent management of construction costs. Summary of the Invention

[0017] In order to overcome the above-mentioned defects of the prior art, the present invention provides an intelligent calculation system and method for engineering cost based on multi-source heterogeneous data fusion to solve the problems existing in the above-mentioned background technology.

[0018] The present invention provides the following technical solution: an intelligent calculation system for engineering cost based on multi-source heterogeneous data fusion, comprising:

[0019] Multi-source data acquisition module: used to collect structured and unstructured data from design documents, market databases, construction monitoring systems, contract documents and historical project libraries;

[0020] Heterogeneous data fusion module: connects to the multi-source data acquisition module, uses natural language processing technology to parse the risk clauses in the contract text, combines image recognition technology to extract the engineering quantity information in the design drawings, and standardizes the dynamic construction data through time series analysis to generate a multi-dimensional spatiotemporal data matrix. When the time series analysis standardizes the dynamic construction data, the original time series data is set as x t , t=1,2,…,n, normalized data y t The calculation formula is: in is the mean, is the standard deviation;

[0021] Deep learning prediction model module: A multi-task learning model is constructed based on the Transformer architecture. The data matrix is input and the predicted values of sub-item costs, total cost and risk confidence intervals are output. The model is pre-trained with historical project data through transfer learning and uses incremental learning to update model parameters in real time. The calculation formula of the self-attention mechanism in the Transformer architecture is: Where Q, K, and V are query, key, and value matrices respectively, and d k is the dimension of the key matrix;

[0022] Dynamic Optimization Engine Module: This module receives the calculation results of the deep learning prediction model module, dynamically adjusts the prediction results based on construction progress data and market fluctuation data, and generates multi-scenario cost optimization solutions, including material replacement suggestions and construction process adjustment strategies.

[0023] Collaborative management platform: Integrates blockchain technology to provide a visual interface that supports cross-departmental data sharing, change approval processes, and automatic report generation. It also records data operation logs throughout the entire life cycle for audit traceability.

[0024] Furthermore, the multi-source data acquisition module specifically includes:

[0025] BIM model parsing unit, used to extract geometric parameters, material properties and engineering quantity information of building components;

[0026] Market data interface unit, which obtains price fluctuation data of steel, cement, and labor rates in real time through API;

[0027] Construction monitoring unit, connected to IoT sensors, collects equipment operating status, construction progress, and environmental monitoring data;

[0028] Contract text analysis unit, which uses OCR technology to identify key clauses in paper contracts and converts them into structured data;

[0029] The design document unit is used to collect data related to the entire project design process, including:

[0030] Project feasibility study report: contains information such as project construction planning and economic evaluation, providing a basic framework and reference data for overall cost estimation;

[0031] Design drawings and bill of quantities: Design drawings are used to determine the size and structure of building components. The bill of quantities lists the measurement information of each sub-item and is the key data for cost calculation.

[0032] Bidding documents and contract information: Bidding documents specify project bidding and tendering requirements and quotations, while contract information specifies key terms such as project payment and change processing, which affect cost calculation and control;

[0033] Change records and actual cost data during construction: Change records reflect project change details and cost changes, while actual cost data reflects the actual costs of each construction stage, which is used for dynamic cost management and deviation adjustments;

[0034] Completion acceptance and settlement data: The completion acceptance report shows the project completion and quality acceptance status. The settlement data is the basis for final cost settlement and is used for the final calculation of project costs and benefit evaluation.

[0035] Furthermore, in the heterogeneous data fusion module:

[0036] The analysis of design drawings includes the following steps: identifying the component types in the drawings through convolutional neural networks (CNN), building component topological relationships in combination with graph neural networks (GNN), and calculating engineering quantities. In CNN, the calculation process of the convolution layer can be expressed as: in is the output of the lth convolutional layer at position (i, j), is the convolution kernel weight, b l is the bias, M and N are the convolution kernel sizes; in GNN, node v i The update formula for is: in For node v i In the representation of the kth layer, N(v i ) is the node v i The set of neighbor nodes, c i，j is the normalization constant, W k 、 is the weight matrix, σ is the activation function;

[0037] The parsing of the contract text includes the following steps: extracting the semantic features of the clauses through the pre-trained language model (BERT), classifying them into risk clauses, payment clauses, and liability clauses, and quantifying the risk weight coefficient. The output of the BERT model can be expressed as: H = BERT(T), where T is the input text sequence and H is the semantic feature vector output by the model. When quantifying the risk weight coefficient,

[0038] Assume that the risk clause set is R. For a certain risk clause r∈R, its risk weight coefficient w r It can be calculated by logistic regression model: Where y=1 means that the clause is a risk clause, w0, w i is the model parameter, f i (r) is the i-th feature extracted from risk term r.

[0039] Furthermore, the deep learning prediction model includes the following sub-models:

[0040] The material cost prediction sub-model inputs material price time series data and engineering quantity data, and outputs the material cost curve for the next N weeks. Let the material price time series data be p t , the engineering quantity data is q t , the material cost curve for the next N weeks can be predicted by a recurrent neural network (RNN), and the hidden layer state update formula is: h t =tanh(W hh ·h t-1 +W xh ·[p t ,q t ]+b h ), the output layer formula is: Where W hh 、W xh 、W oh is the weight matrix, b h 、b o For bias.

[0041] The labor cost prediction sub-model combines the construction schedule with labor market data to predict labor cost fluctuations. The construction schedule is represented by st and the labor market data is represented by l t , labor cost fluctuations can be predicted through the linear regression model: in is the predicted labor cost, β0, β1, β2 are regression coefficients, ∈ t is the error term;

[0042] The risk quantification sub-model uses Monte Carlo simulation to analyze the impact of construction delays, price fluctuations, and geological condition uncertainty on the total cost, and outputs the risk probability distribution. Assume that the total cost C is affected by multiple uncertain factors X1, X2, ..., X m Impact, through Monte Carlo simulation, N simulations are performed to calculate the total cost C (i) , i=1,…,N,the risk probability distribution can be calculated by C (i) The statistics are obtained, such as calculating the probability of exceeding a certain cost threshold C0: Among them, is the indicator function;

[0043] The estimation document model uses ensemble learning to combine decision trees, support vector machines, and neural network models to estimate project costs based on preliminary project design information, historical data from similar projects, and market price trends. It also leverages transfer learning to draw on experience from similar projects to quickly provide a rough estimate of project investment and provide an initial cost range reference for other sub-models.

[0044] The budget document model introduces an adaptive attention mechanism based on the Transformer architecture, combines semantic understanding technology to mine the text information of design documents, and uses multimodal data fusion technology to integrate design drawings and text information for cost calculation. This provides accurate budget estimates for projects in the preliminary design stage and collaborates with other sub-models to optimize cost forecasts.

[0045] The model training adopts a multi-task loss function to jointly optimize the itemized cost prediction error and risk quantification error. The itemized cost prediction error is set to L cost , the risk quantification error is L risk , multi-task loss function L = α·L cost +(1-α)·L risk , where α is the balance coefficient.

[0046] Furthermore, the workflow of the dynamic optimization engine includes:

[0047] Real-time monitoring of construction progress deviations and market price fluctuations. When the deviation exceeds the preset threshold, the model is retrained. The construction progress deviation rate is The market price deviation rate is When |δ s |>θ s or |δ p |>θ p Model retraining is triggered when θ s ,θ p is the preset threshold;

[0048] Generate multi-objective optimization solutions based on the Pareto optimality principle, including minimum cost solution, minimum risk solution and equilibrium solution;

[0049] Push optimization suggestions to the procurement and construction departments through the collaborative management platform, and track implementation feedback.

[0050] Furthermore, in the collaborative management platform:

[0051] The data sharing mechanism is implemented based on the alliance chain, and the participating nodes include the construction unit, design unit, construction unit and supervision unit;

[0052] The change approval process uses smart contracts to automatically verify data consistency, and the approval results are stored on the chain;

[0053] The visual interface provides a real-time data dashboard, including a cost deviation heat map, a risk warning radar map, and an optimization plan comparison chart.

[0054] Furthermore, a method for intelligent calculation of engineering cost based on multi-source heterogeneous data fusion is proposed, which includes the following steps:

[0055] Step S1: Obtain design documents, market data, construction data and contract text through a multi-source data acquisition module;

[0056] Step S2: Use the heterogeneous data fusion module to convert unstructured data into standardized feature vectors, associate them with structured data, and construct a spatiotemporal data cube;

[0057] Step S3: input the data cube into the deep learning prediction model, and output the itemized cost prediction value, the total cost prediction value and the risk confidence interval;

[0058] Step S4: Trigger the incremental learning of the model based on real-time construction data and market data, and update the prediction results. In the incremental learning, let the new data sample be (x new ,y new ), the model parameter update formula is: where θ t is the current model parameter, η is the learning rate, is the loss function L with respect to the parameter θ t Gradient on new examples;

[0059] Step S5: Generate a cost control strategy through the dynamic optimization engine and distribute it to relevant parties via the collaborative management platform.

[0060] Furthermore, the specific steps of data fusion in step S2 include:

[0061] Perform semantic segmentation on design drawings, identify component categories and calculate geometric parameters;

[0062] Perform entity recognition and relationship extraction on the contract text to build a risk-responsibility association map;

[0063] Align construction monitoring data with the BIM model to establish a 4D (3D+time) construction progress model with timestamp association.

[0064] Furthermore, the training method of the deep learning prediction model in step S3 includes:

[0065] Pre-training stage: Use historical project data to train the initial parameters of the model. The loss function is the mean square error (MSE). Assume that the historical project data is (xi, yi), i = 1, ..., n, and the MSE loss function is: in is the model prediction value;

[0066] Incremental learning stage: adopt online learning strategy, update training data in sliding window mode, and adjust learning rate dynamically with data freshness; let data freshness be f, and the adjustment formula of learning rate η is: η = η0 ·e -λ·f , where η0 is the initial learning rate and λ is the decay coefficient;

[0067] Model validation phase: The prediction error is evaluated through cross-validation. When the error exceeds 3%, the manual review process is triggered. Suppose the prediction error in the cross-validation is e. If e>0.03, the manual review process is triggered.

[0068] Furthermore, the generation of the cost control strategy in step S5 includes:

[0069] The optimal combination of material procurement and construction scheduling is searched based on the genetic algorithm. In the genetic algorithm, let individual x represent the combination of material procurement and construction scheduling. Its fitness function F(x) can be constructed according to the objectives such as cost and construction period, such as F(x) = w1·C(x)+w2·T(x), where C(x) is the cost of solution x, T(x) is the construction period of solution x, and w l , w2 is the weight coefficient;

[0070] Through reinforcement learning, we simulate the long-term cost impact of different decision paths and select the strategy with the highest cumulative reward. Let the state in reinforcement learning be s and the action set be A. In state s, we perform action a∈A to obtain reward r(s, a). The cumulative reward R=∑ t＝0 γ t ·r(s t , a t ), where γ is the discount factor and T is the time step;

[0071] Bind the optimization plan with the blockchain smart contract to ensure that the execution process is transparent and traceable.

[0072] Technical effects and advantages of the present invention:

[0073] The present invention uses a multi-source data acquisition module to widely collect data from various aspects such as design, market, construction and contracts, solving the data fragmentation and information island problems in traditional cost management and realizing the integration of multi-source heterogeneous data. The heterogeneous data fusion module uses advanced natural language processing, image recognition and other technologies to efficiently parse contract texts and design drawings, improving the processing ability of unstructured data and reducing the error rate and omission rate of manual interpretation.

[0074] The deep learning prediction model uses a multi-task learning model based on the Transformer architecture. Combining transfer learning and incremental learning, it can not only accurately predict itemized and total costs, but also quantify risks. This effectively addresses the disconnect between traditional static models and dynamic demand, and responds in real time to factors such as construction progress, market prices, and policy changes.

[0075] The dynamic optimization engine dynamically adjusts prediction results based on real-time data and generates multi-scenario cost optimization solutions, such as material replacement and construction process adjustment, to help achieve dynamic cost control and improve the flexibility and adaptability of cost management.

[0076] The collaborative management platform integrates blockchain technology to achieve cross-departmental data sharing and trusted collaboration, and provides a visual interface for easy operation and viewing. At the same time, it automatically verifies the consistency of changed data through smart contracts and records operation logs for easy audit traceability. It solves the problems of low collaboration efficiency and data tampering risks in traditional cost management, and realizes accurate, dynamic and transparent management of project costs, effectively reducing the probability of cost overruns, construction delays and disputes. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 This is a diagram of the system architecture of the present invention;

[0078] Figure 2 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0079] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0080] It will be understood that the terms "first," "second," etc. used herein may be used to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element.

[0081] In order to help those skilled in the art better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of this application.

[0082] Example

[0083] The present invention provides an intelligent calculation system for engineering cost based on multi-source heterogeneous data fusion, comprising:

[0084] Multi-source data acquisition module: used to collect structured and unstructured data from design documents, market databases, construction monitoring systems, contract documents and historical project libraries;

[0085] Heterogeneous data fusion module: connects to the multi-source data acquisition module, uses natural language processing technology to parse the risk clauses in the contract text, combines image recognition technology to extract the engineering quantity information in the design drawings, and standardizes the dynamic construction data through time series analysis to generate a multi-dimensional spatiotemporal data matrix. When the time series analysis standardizes the dynamic construction data, the original time series data is set as x t , t=1,2,…,n, normalized data y t The calculation formula is: in is the mean, is the standard deviation;

[0086] Deep learning prediction model: A multi-task learning model is constructed based on the Transformer architecture. The data matrix is input and the predicted values of sub-item costs, total cost and risk confidence intervals are output. The model is pre-trained with historical project data through transfer learning and uses incremental learning to update model parameters in real time. The calculation formula of the self-attention mechanism in the Transformer architecture is: Where Q, K, and V are query, key, and value matrices respectively, and d k is the dimension of the key matrix;

[0087] Dynamic Optimization Engine: Dynamically adjusts forecast results based on construction progress data and market fluctuations to generate multi-scenario cost optimization solutions, including material replacement suggestions and construction process adjustment strategies;

[0088] Collaborative management platform: Integrates blockchain technology to provide a visual interface that supports cross-departmental data sharing, change approval processes, and automatic report generation. It also records data operation logs throughout the entire life cycle for audit traceability.

[0089] Furthermore, the multi-source data acquisition module specifically includes:

[0090] The BIM model parsing unit is used to extract the geometric parameters, material properties, and engineering quantity information of building components. In large-scale commercial complex projects, the BIM model covers a vast amount of information. For example, the BIM model of a newly built commercial plaza records in detail the geometric parameters, material properties, and engineering quantity information of everything from the building's main frame structure to the various internal decorative components. The parsing unit uses advanced parsing algorithms to accurately extract this data, providing a key foundation for subsequent cost calculations. During the construction process, if it is discovered that some areas of the original design are not being used properly and internal partitions need to be adjusted, the BIM model parsing unit can quickly update the adjusted engineering quantity information to ensure the timeliness and accuracy of cost calculations.

[0091] The market data interface unit uses the API to obtain real-time price fluctuation data for steel, cement, and labor rates; it connects to professional market data platforms with the API to obtain real-time and accurate price fluctuation data. In actual applications, the prices of major construction materials such as steel and cement fluctuate frequently due to various factors such as market supply and demand and the international situation. For example, during a bridge construction project, international iron ore prices rose sharply. Through the market data interface unit, the system can obtain real-time changes in steel prices and update them to the cost calculation system in a timely manner, providing strong support for project cost control.

[0092] The construction monitoring unit is connected to IoT sensors to collect data on equipment operating status, construction progress, and environmental monitoring. In high-rise residential construction projects, the construction monitoring unit is connected to a large number of IoT sensors, which are distributed in construction equipment, construction site environment, and key parts of building structures. Through the sensors, equipment operating status data can be collected in real time, such as the operating time of tower cranes and the operating frequency of concrete mixers, which are used to analyze the equipment's usage cost and maintenance needs. Construction progress data is collected, accurate to the construction progress of each floor. Environmental monitoring data, such as temperature, humidity, and wind speed at the construction site, are collected to assess the potential impact of environmental factors on construction costs. For example, high temperatures may increase concrete maintenance costs.

[0093] The contract text analysis unit uses OCR technology to identify key clauses in paper contracts and convert them into structured data. In actual engineering projects, the content of contract texts is complex and diverse. For a large-scale infrastructure construction contract, the contract text analysis unit first uses OCR technology to identify key clauses in the paper contract and converts them into structured data. Then, using natural language processing technology, it conducts an in-depth analysis of the risk clauses, payment clauses, and liability clauses in the contract. For example, it identifies compensation clauses for construction delays and price adjustment clauses when material prices fluctuate by more than a certain percentage, providing important basis for project risk assessment and cost management.

[0094] The design document unit is used to collect data related to the entire project design process, including:

[0095] Project feasibility study report: contains information such as project construction planning and economic evaluation, providing a basic framework and reference data for overall cost estimation;

[0096] Design drawings and bill of quantities: Design drawings are used to determine the size and structure of building components. The bill of quantities lists the measurement information of each sub-item and is the key data for cost calculation.

[0097] Bidding documents and contract information: Bidding documents specify project bidding and tendering requirements and quotations, while contract information specifies key terms such as project payment and change processing, which affect cost calculation and control;

[0098] Change records and actual cost data during construction: Change records reflect project change details and cost changes, while actual cost data reflects the actual costs of each construction stage, which is used for dynamic cost management and deviation adjustments;

[0099] Completion acceptance and settlement data: The completion acceptance report shows the project completion and quality acceptance status. The settlement data is the basis for final cost settlement and is used for the final calculation of project costs and benefit evaluation.

[0100] Furthermore, in the heterogeneous data fusion module:

[0101] The analysis of design drawings includes the following steps: identifying the component types in the drawings through convolutional neural networks (CNN), building component topological relationships in combination with graph neural networks (GNN), and calculating engineering quantities. In CNN, the calculation process of the convolution layer can be expressed as: in is the output of the lth convolutional layer at position (i, j), is the convolution kernel weight, b 1 is the bias, M and N are the convolution kernel sizes; in GNN, node v i The update formula for is: in For node v i In the representation of the kth layer, N(v i ) is the node v i The set of neighbor nodes, c i，j is the normalization constant, W k 、 is the weight matrix, σ is the activation function;

[0102] In complex construction projects, such as large hospitals, design drawings contain numerous different types of components. Convolutional neural networks (CNNs) can accurately identify various component types, such as beams, slabs, columns, and walls, by learning from a large number of design drawing samples. Graph neural networks (GNNs) further calculate construction quantities by leveraging the topological relationships between components. When calculating the construction quantities of beam-column joints, GNNs accurately calculate the amount of concrete and steel required at each node based on the connection relationships and attribute information between the nodes. This approach significantly improves the accuracy and efficiency of construction quantity calculations compared to traditional manual calculations.

[0103] The parsing of the contract text includes the following steps: extracting the semantic features of the clauses through the pre-trained language model (BERT), classifying them into risk clauses, payment clauses, and liability clauses, and quantifying the risk weight coefficient. The output of the BERT model can be expressed as: H = BERT(T), where T is the input text sequence and H is the semantic feature vector output by the model. When quantifying the risk weight coefficient,

[0104] Assume that the risk clause set is R. For a certain risk clause r∈R, its risk weight coefficient w r It can be calculated by logistic regression model: Where y=1 means that the clause is a risk clause, W0, w i is the model parameter, fi(r) is the i-th feature extracted from the risk clause r; taking a real estate development project contract as an example, the pre-trained language model (BERT) conducts an in-depth analysis of the contract text, extracts the semantic features of the clauses, and classifies them into risk clauses, payment clauses, and liability clauses; for risk clauses, the risk weight coefficient is quantified through a logistic regression model; for example, the contract stipulates that if the construction party causes a delay of more than 30 days, a high penalty must be paid. The BERT model identifies this clause as a risk clause and calculates its risk weight coefficient through a logistic regression model, providing a quantitative basis for project risk assessment.

[0105] Furthermore, the deep learning prediction model includes the following sub-models:

[0106] The material cost prediction sub-model inputs material price time series data and engineering quantity data, and outputs the material cost curve for the next N weeks. Let the material price time series data be pt , the engineering quantity data is q t , the material cost curve for the next N weeks can be predicted by a recurrent neural network (RNN), and the hidden layer state update formula is: h t =tanh(W hh ·h t-1 +W xh ·[p t ,q t ]+b h ), the output layer formula is: Where W hh 、W xh 、W oh is the weight matrix, b h 、b o For a long-term municipal road construction project, the material cost prediction sub-model uses a recurrent neural network (RNN) to predict the material cost curve for the next N weeks based on the collected material price time series data and construction quantity data. Assuming the project construction period is 12 months, during the construction process, based on the material price change trends of the previous 6 months and the monthly construction quantity consumption data, the RNN model can predict the cost trends of different materials in the next few months. If the price of a major material is predicted to rise sharply, the project party can adjust the procurement plan in advance to reduce cost risks.

[0107] The labor cost prediction sub-model combines the construction schedule with labor market data to predict labor cost fluctuations. The construction schedule is represented by st and the labor market data is represented by l t , labor cost fluctuations can be predicted through the linear regression model: in is the predicted labor cost, β0, β1, β2 are regression coefficients, ∈ t The error term is the construction schedule. In construction projects, the construction schedule is continuously adjusted based on actual conditions, combining the construction schedule with labor market data. For example, in a certain office building construction project, the construction progress lagged due to complex geological conditions encountered during the early foundation construction. The labor cost prediction sub-model uses a linear regression model to re-predict labor cost fluctuations based on the adjusted construction schedule and real-time labor market data. This allows the project management team to prepare funds and deploy personnel in advance to avoid the impact of uncontrolled labor costs on project costs.

[0108] The risk quantification sub-model uses Monte Carlo simulation to analyze the impact of construction delays, price fluctuations, and geological condition uncertainty on the total cost, and outputs the risk probability distribution. Assume that the total cost C is affected by multiple uncertain factors X1, X2, ..., X m Impact, through Monte Carlo simulation, N simulations are performed to calculate the total cost C (i), i=1,…,N,the risk probability distribution can be calculated by C (i) The statistics are obtained, such as calculating the probability of exceeding a certain cost threshold C0: Among them, is an indicator function. In large-scale water conservancy project construction projects, the total cost is affected by many uncertain factors, such as construction delays, price fluctuations and changes in geological conditions. Through Monte Carlo simulation, a large number of simulations (such as 1,000 times) are performed to calculate the total cost. Assuming that the preset cost threshold of the project is C0, the risk quantification sub-model calculates the probability of exceeding the cost threshold. For example, if P(C>C0)=0.15 is calculated, it means that there is a 15% probability that the project will exceed the preset cost threshold. Project managers can assess project risks based on this probability and formulate corresponding risk response strategies.

[0109] The estimation document model uses ensemble learning to combine decision trees, support vector machines, and neural network models to estimate project costs based on preliminary project design information, historical data from similar projects, and market price trends. It also leverages transfer learning to draw on experience from similar projects to quickly provide a rough estimate of project investment and provide an initial cost range reference for other sub-models.

[0110] The budget document model introduces an adaptive attention mechanism based on the Transformer architecture, combines semantic understanding technology to mine the text information of design documents, and uses multimodal data fusion technology to integrate design drawings and text information for cost calculation. This provides accurate budget estimates for projects in the preliminary design stage and collaborates with other sub-models to optimize cost forecasts.

[0111] The model training adopts a multi-task loss function to jointly optimize the itemized cost prediction error and risk quantification error. The itemized cost prediction error is set to L cost , the risk quantification error is L risk , multi-task loss function L = α·L cost +(1-a)·L risk , where α is the balance coefficient. In actual projects, model training is crucial. A series of residential construction project data is used as historical project data for pre-training, and the mean squared error (MSE) is used as the loss function. During the incremental learning phase, new data is continuously generated as the project progresses. For example, in a large residential construction project, new construction data and market data are available every month. The training data is updated using a sliding window, and the learning rate is dynamically adjusted based on the freshness of the data. During the model validation phase, the prediction error is evaluated through cross-validation. If the prediction error exceeds 3% at a certain stage, a manual review process is triggered to ensure the accuracy of the model prediction.

[0112] Furthermore, the workflow of the dynamic optimization engine includes:

[0113] Real-time monitoring of construction progress deviations and market price fluctuations. When the deviation exceeds the preset threshold, the model is retrained. The construction progress deviation rate is The market price deviation rate is When |δ s |>θ s or |δ p |>θ p Model retraining is triggered when θ s ,θ p is the preset threshold; in the construction process of the construction project, the construction progress deviation and market price fluctuation are monitored in real time. Taking an industrial plant construction project as an example, the construction progress deviation rate threshold δ is set s =0.1, market price deviation rate threshold δ p =0.15. When the actual construction progress lags behind the planned progress by 15% (i.e. |δ s |=0.15>0.1), or the market price of a major material rises by 20% (i.e. |δ p =|=0.2>0.15), triggers model retraining. By retraining the model, the prediction results are more consistent with the actual situation, providing a more accurate basis for cost optimization.

[0114] Based on the Pareto optimality principle, a multi-objective optimization scheme is generated, including a minimum cost scheme, a minimum risk scheme, and a balanced scheme. In a commercial office building construction project, the dynamic optimization engine generated a multi-objective optimization scheme based on the Pareto optimality principle. For example, a minimum cost scheme might recommend the use of more cost-effective materials, but this might increase construction difficulty; a minimum risk scheme might prioritize materials with stable quality and reliable supply, but at a relatively high cost; and a balanced scheme seeks a balance between cost and risk. These schemes provide project managers with a variety of decision-making options to meet diverse project needs.

[0115] Optimization suggestions are pushed to the procurement and construction departments through the collaborative management platform, and implementation feedback is tracked. During the project implementation process, the collaborative management platform pushes optimization suggestions to the procurement and construction departments. For example, in a coal project, optimization suggestions may include changing a certain material supplier to reduce costs, or adjusting the construction process to improve construction efficiency. After the procurement and construction departments implement the suggestions, the implementation feedback information will be sent back through the collaborative management platform. If the procurement department finds that there are problems with the material quality after changing the supplier, or the construction department encounters difficulties in adjusting the construction process, these feedback information will be collected in a timely manner to adjust and improve the optimization plan

[0116] Furthermore, in the collaborative management platform:

[0117] The data sharing mechanism is implemented on a consortium blockchain, with participating nodes including the construction company, design company, construction company, and supervision company. In a large-scale urban rail transit project, these participating nodes include the construction company, design company, construction company, and supervision company. These parties share data through the consortium blockchain, ensuring that all parties have timely access to the latest project data. For example, if the design company adjusts the route of the line, the relevant design change data is synchronized to the consortium blockchain in real time. The construction company and supervision company can immediately access this data and adjust the construction plan and supervision plan accordingly, avoiding construction errors and increased costs caused by untimely information.

[0118] The change approval process uses smart contracts to automatically verify data consistency, and the approval results are stored on the chain. In construction projects, changes occur frequently. Taking a high-rise apartment construction project as an example, when the construction unit submits a change application, such as when the foundation construction plan needs to be adjusted due to changes in on-site geological conditions, the change approval process uses smart contracts to automatically verify data consistency. According to preset rules, the smart contract checks whether the data in the change application meets the requirements, such as whether the calculation of the changed engineering quantity is accurate, and whether the impact of the change on cost and construction period is reasonable. The approval results are stored on the chain to ensure the transparency and traceability of the approval process.

[0119] The visual interface provides a real-time data dashboard, including a cost deviation heat map, a risk warning radar map, and an optimization scheme comparison map. In actual project management, the visual interface provides project managers with intuitive decision-making support. In a large-scale sports stadium construction project, the cost deviation heat map of the visual interface uses different colors to intuitively display the cost deviation of each construction area. Red indicates serious cost overruns, and green indicates good cost control. The risk warning radar map displays in real time the various risks faced by the project, such as schedule risk, quality risk, cost risk, etc. The optimization scheme comparison map compares the key indicators of different optimization schemes, such as cost, schedule, risk, etc., to help project managers quickly select the most suitable scheme.

[0120] Furthermore, a method for intelligent calculation of engineering cost based on multi-source heterogeneous data fusion is proposed, which includes the following steps:

[0121] Step S1: The multi-source data acquisition module acquires design documents, market data, construction data, and contract documents. During the project launch phase, for example, a new airport construction project, the multi-source data acquisition module comprehensively collects design documents, including drawings of buildings like the airport terminal and runway; market data, such as prices for various building materials and labor rates; construction data, such as site conditions and equipment information; and contract documents, including project-related construction and procurement contracts. This data provides a wealth of foundational information for subsequent cost calculations and management.

[0122] Step S2: Utilizing the heterogeneous data fusion module, the unstructured data is converted into standardized feature vectors and associated with the structured data to construct a spatiotemporal data cube. During this data fusion process, semantic segmentation is performed on the design drawings to identify different building component categories. For example, in the airport terminal design drawings, steel and concrete components are accurately identified and their geometric parameters calculated. Entity recognition and relationship extraction are performed on the contract text to construct a risk-responsibility association map, for example, clarifying the responsibilities and risks of each party in the contract regarding project quality and construction delays. Construction monitoring data is aligned with the BlM model to establish a timestamp-linked 4D construction progress model, enabling dynamic tracking and management of construction progress.

[0123] Step S3: Input the data cube into the deep learning prediction model, and output the itemized cost prediction value, total cost prediction value and risk confidence interval; input the constructed data cube into the deep learning prediction model. In the airport construction project, the model outputs the itemized cost prediction value, such as the material cost and labor cost of runway construction; the total cost prediction value, that is, the estimated total cost of the entire airport construction project; and the risk confidence interval to help project managers understand the risk range of project costs.

[0124] Step S4: Trigger the incremental learning of the model based on real-time construction data and market data, and update the prediction results. In the incremental learning, let the new data sample be (x new ,y new ), the model parameter update formula is: where θ t is the current model parameter, η is the learning rate, is the loss function L with respect to the parameter θ t Gradients on new samples; During the airport construction project, real-time construction data and market data trigger incremental model learning. For example, when a new building material becomes available at a competitive price, or when construction progress deviates significantly due to unusual weather conditions, new data samples are fed into the model. The model updates parameters according to the rules of incremental learning, adjusting predictions for greater accuracy.

[0125] Step S5: Generate a cost control strategy using a dynamic optimization engine and distribute it to relevant parties via the collaborative management platform. This dynamic optimization engine generates a cost control strategy. For the airport construction project, a genetic algorithm is used to search for the optimal combination of material procurement and construction scheduling. For example, while ensuring project quality and progress, the optimal material procurement time and supplier combination, as well as the most reasonable construction scheduling plan, are sought. Reinforcement learning is used to simulate the long-term cost impact of different decision paths and select the strategy with the highest cumulative reward. The optimization solution is then tied to a blockchain smart contract to ensure transparency and traceability during execution. For example, during the material procurement process, every transaction is recorded on the blockchain, ensuring fairness and auditability in the procurement process.

[0126] Furthermore, the specific steps of data fusion in step S2 include:

[0127] Perform semantic segmentation on design drawings, identify component categories and calculate geometric parameters;

[0128] Perform entity recognition and relationship extraction on the contract text to build a risk-responsibility association map;

[0129] Align construction monitoring data with the BIM model to establish a 4D (3D+time) construction progress model with timestamp association.

[0130] Furthermore, the training method of the deep learning prediction model in step S3 includes:

[0131] Pre-training stage: Use historical project data to train the initial parameters of the model. The loss function is the mean square error (MSE). Assume that the historical project data is (xi, yi), i = 1, ..., n, and the MSF loss function is: in is the model prediction value;

[0132] Incremental learning stage: adopt online learning strategy, update training data in sliding window mode, and adjust learning rate dynamically with data freshness; let data freshness be f, and the adjustment formula of learning rate η is: η=η0·e -λ·f , where η0 is the initial learning rate and λ is the decay coefficient;

[0133] Model validation phase: The prediction error is evaluated through cross-validation. When the error exceeds 3%, the manual review process is triggered. Suppose the prediction error in the cross-validation is e. If e>0.03, the manual review process is triggered.

[0134] Furthermore, the generation of the cost control strategy in step S5 includes:

[0135] The optimal combination of material procurement and construction scheduling is searched based on the genetic algorithm. In the genetic algorithm, individual x represents the combination of material procurement and construction scheduling. Its fitness function F(x) can be constructed based on objectives such as cost and construction period, such as F(x) = w1·C(x)+w2·T(x), where C(x) is the cost of solution x, T(x) is the construction period of solution x, and w1 and w2 are weight coefficients.

[0136] Through reinforcement learning, we simulate the long-term cost impact of different decision paths and select the strategy with the highest cumulative reward. Let the state in reinforcement learning be s and the action set be A. In state s, we perform action a∈A to obtain reward r(s, a). The cumulative reward R=∑ t＝0 γ t ·r(s t , a t ), where γ is the discount factor and T is the time step;

[0137] Bind the optimization plan with the blockchain smart contract to ensure that the execution process is transparent and traceable.

[0138] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0139] The above embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

[0140] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An intelligent engineering cost calculation system based on multi-source heterogeneous data fusion, characterized by: include: Multi-source data acquisition module: used to collect structured and unstructured data from design documents, market databases, construction monitoring systems, contract documents and historical project libraries; Heterogeneous data fusion module: receives the data from the multi-source data acquisition module, uses natural language processing technology to parse the risk clauses in the contract text, combines image recognition technology to extract the engineering quantity information in the design drawings, and standardizes the dynamic construction data through time series analysis to generate a multi-dimensional time-space correlation data matrix. When the time series analysis standardizes the dynamic construction data, the original time series data is set as x t , t=1,2,…,n, normalized data y t The calculation formula is: in is the mean, is the standard deviation; Deep learning prediction model module: A multi-task learning model is constructed based on the Transformer architecture. The data matrix is input and the predicted values of sub-item costs, total cost and risk confidence intervals are output. The model is pre-trained with historical project data through transfer learning and uses incremental learning to update model parameters in real time. The calculation formula of the self-attention mechanism in the Transformer architecture is: Where Q, K, and V are query, key, and value matrices respectively, and d k is the dimension of the key matrix; Dynamic Optimization Engine Module: This module receives the calculation results of the deep learning prediction model module, dynamically adjusts the prediction results based on construction progress data and market fluctuation data, and generates multi-scenario cost optimization solutions, including material replacement suggestions and construction process adjustment strategies. Collaborative management platform: Integrates blockchain technology to provide a visual interface to support cross-departmental data sharing, change approval processes and automatic report generation, and records data operation logs throughout the entire life cycle to achieve audit traceability.

2. The intelligent calculation system for construction cost based on multi-source heterogeneous data fusion according to claim 1 is characterized by: The multi-source data acquisition module specifically includes: BIM model parsing unit, used to extract geometric parameters, material properties and engineering quantity information of building components; Market data interface unit, which obtains price fluctuation data of steel, cement, and labor rates in real time through API; Construction monitoring unit, connected to IoT sensors, collects equipment operating status, construction progress, and environmental monitoring data; Contract text analysis unit uses OCR technology to identify key clauses in paper contracts and convert them into structured data The design document unit is used to collect data related to the entire project design process, including: Project feasibility study report: contains project construction planning and economic evaluation information, providing a basic framework and reference data for overall cost estimation; Design drawings and bill of quantities: Design drawings are used to determine the size and structure of building components. The bill of quantities lists the measurement information of each sub-project accordingly and is the key data for cost calculation; Bidding documents and contract information: Bidding documents specify project bidding and tendering requirements and quotations, while contract information specifies key terms such as project payment and change processing, which affect cost calculation and control; Change records and actual cost data during construction: Change records reflect project change details and cost changes, while actual cost data reflects the actual costs of each construction stage, which is used for dynamic cost management and deviation adjustments; Completion acceptance and settlement data: The completion acceptance report shows the project completion and quality acceptance status. The settlement data is the basis for final cost settlement and is used for the final calculation of project costs and benefit evaluation.

3. The intelligent calculation system for construction cost based on multi-source heterogeneous data fusion according to claim 1 is characterized by: In the heterogeneous data fusion module: Design drawing parsing module: Use convolutional neural networks to identify component types in drawings, combine graph neural networks to build component topological relationships, and calculate engineering quantities. In CNN, the calculation process of the convolution layer is expressed as: in is the output of the lth convolutional layer at position (i, j), is the convolution kernel weight, b l is the bias, M and N are the convolution kernel sizes; in GNN, node v i The update formula for is: in For node v i In the representation of the kth layer, N(v i ) is the node v i The set of neighbor nodes, c i，j is the normalization constant, W k 、 is the weight matrix, a is the activation function; Contract text parsing module: The semantic features of the clauses are extracted through the pre-trained language model, classified into risk clauses, payment clauses and liability clauses, and the risk weight coefficient is quantified. The output of the BERT model is expressed as: H = BERT(T), where T is the input text sequence and H is the semantic feature vector output by the model. When quantifying the risk weight coefficient, let the risk clause set be R. For a certain risk clause r∈R, its risk weight coefficient w r It can be calculated by logistic regression model: Where y=1 means that the clause is a risk clause, w0, w i is the model parameter, f i (r) is the i-th feature extracted from risk term r.

4. The intelligent calculation system for construction cost based on multi-source heterogeneous data fusion according to claim 1 is characterized in that: The deep learning prediction model module includes the following sub-models: The material cost prediction sub-model inputs material price time series data and engineering quantity data, and outputs the material cost curve for the next N weeks. Let the material price time series data be p t , the engineering quantity data is q t , the material cost curve for the next N weeks can be predicted by a recurrent neural network, and the hidden layer state update formula is: h t =tanh(W hh ·h t-1 +W xh ·[p t ,q t ]+b h ), the output layer formula is: Where W hh 、W xh 、W oh is the weight matrix, b h 、b o For bias. The labor cost prediction sub-model combines the construction schedule with labor market data to predict labor cost fluctuations. The construction schedule is represented by st and the labor market data is represented by l t , predict labor cost fluctuations through linear regression model: in is the predicted labor cost, β0, β1, β2 are regression coefficients, ∈ t is the error term; The risk quantification sub-model uses Monte Carlo simulation to analyze the impact of construction delays, price fluctuations, and geological condition uncertainty on the total cost, and outputs the risk probability distribution. Assume that the total cost C is affected by multiple uncertain factors X1, X2, ..., X m Impact, through Monte Carlo simulation, N simulations are performed to calculate the total cost C (i) , i=1,…,N,the risk probability distribution can be calculated by C (i) The statistics are obtained, such as calculating the probability of exceeding a certain cost threshold C0: Among them, is the indicator function; The estimation document model uses ensemble learning to combine decision trees, support vector machines, and neural network models to estimate project costs based on preliminary project design information, historical data from similar projects, and market price trends. It also leverages transfer learning to draw on experience from similar projects to provide a rough estimate of project investment and provide an initial cost range reference for other sub-models. The budget document model introduces an adaptive attention mechanism based on the Transformer architecture, combines semantic understanding technology to mine the text information of design documents, and uses multimodal data fusion technology to integrate design drawings and text information for cost calculation. This provides accurate budget estimates for projects in the preliminary design stage and collaborates with other sub-models to optimize cost forecasts. The model training adopts a multi-task loss function to jointly optimize the itemized cost prediction error and risk quantification error. The itemized cost prediction error is set to L cost , the risk quantification error is L risk , multi-task loss function L = α·L cost +(1-a)·L risk , where α is the balance coefficient.

5. The intelligent calculation system for construction cost based on multi-source heterogeneous data fusion according to claim 1 is characterized by: In the dynamic optimization engine module, dynamically adjusting the forecast results based on construction progress data and market fluctuation data refers to: Real-time monitoring of construction progress deviations and market price fluctuations. When the deviation exceeds the preset threshold, the model is retrained. The construction progress deviation rate is The market price deviation rate is When |δ s |>θ s or |δ p |>θ p Model retraining is triggered when θ s ,θ p is the preset threshold; Generate multi-objective optimization solutions based on the Pareto optimality principle, including minimum cost solution, minimum risk solution and equilibrium solution; Push optimization suggestions to the procurement and construction departments through the collaborative management platform, and track implementation feedback.

6. The intelligent calculation system for construction cost based on multi-source heterogeneous data fusion according to claim 1 is characterized by: In the collaborative management platform: The data sharing mechanism is implemented based on the alliance chain, and the participating nodes include the construction unit, design unit, construction unit and supervision unit; The change approval process uses smart contracts to automatically verify data consistency, and the approval results are stored on the chain; The visual interface provides a real-time data dashboard, including a cost deviation heat map, a risk warning radar map, and an optimization plan comparison chart.

7. An intelligent calculation method for construction cost based on multi-source heterogeneous data fusion, characterized by: The following steps are involved: Step S1: Obtain design documents, market data, construction data and contract text through a multi-source data acquisition module; Step S2: Use the heterogeneous data fusion module to convert unstructured data into standardized feature vectors, associate them with structured data, and construct a spatiotemporal data cube; Step S3: input the data cube into the deep learning prediction model, and output the itemized cost prediction value, the total cost prediction value and the risk confidence interval; Step S4: Trigger the incremental learning of the model based on real-time construction data and market data, and update the prediction results. In the incremental learning, let the new data sample be (x new ,y new ), the model parameter update formula is: θ t+1 =θ t -η· where θ t is the current model parameter, η is the learning rate, is the loss function L with respect to the parameter θ t Gradient on new examples; Step S5: Generate a cost control strategy through the dynamic optimization engine and distribute it to relevant parties via the collaborative management platform.

8. The method for intelligent calculation of construction cost based on multi-source heterogeneous data fusion according to claim 7 is characterized by: The specific steps of data fusion in step S2 include: Perform semantic segmentation on design drawings, identify component categories and calculate geometric parameters; Perform entity recognition and relationship extraction on the contract text to build a risk-responsibility association map; Align construction monitoring data with the BIM model to create a 4D construction progress model with timestamp association.

9. The method for intelligent calculation of construction cost based on multi-source heterogeneous data fusion according to claim 7 is characterized by: The training method of the deep learning prediction model in step S3 includes: Pre-training stage: Use historical project data to train the initial parameters of the model, the loss function is the mean square error, and the historical project data is (x i ,y i ), i = 1, ..., n, the MSE loss function is: in is the model prediction value; Incremental learning stage: adopt online learning strategy, update training data in sliding window mode, and adjust learning rate dynamically with data freshness; let data freshness be f, and the adjustment formula of learning rate η is: η=η0·e -λ·f , where η0 is the initial learning rate and λ is the decay coefficient; Model validation phase: The prediction error is evaluated through cross-validation. When the error exceeds 3%, the manual review process is triggered. Suppose the prediction error in the cross-validation is e. If e>0.03, the manual review process is triggered.

10. The method for intelligent calculation of construction cost based on multi-source heterogeneous data fusion according to claim 7, characterized in that: The generation of the cost control strategy in step S5 includes: The optimal combination of material procurement and construction scheduling is searched based on the genetic algorithm. In the genetic algorithm, let individual x represent the combination of material procurement and construction scheduling. Its fitness function F(x) can be constructed based on objectives such as cost and construction period. F(x) = w1·C(x) + w2·T(x), where C(x) is the cost of solution x, T(x) is the construction period of solution x, and w1 and w2 are weight coefficients. Through reinforcement learning, we simulate the long-term cost impact of different decision paths and select the strategy with the highest cumulative reward. Let the state in reinforcement learning be s and the action set be A. In state s, we perform action a∈A to obtain reward r(s, a). The cumulative reward R=∑ t＝0 γ t ·r(s t , a t ), where γ is the discount factor and T is the time step; Bind the optimization plan with the blockchain smart contract to ensure that the execution process is transparent and traceable.

Images