Civil construction project cost evaluation optimization system

Abstract

The invention relates to the technical field of cost assessment, in particular to a civil construction project cost assessment optimization system, which comprises a feature analysis module, a cost estimation module, a real-time monitoring module, a model training and tuning module, a risk analysis module, a decision support module, a financial simulation module and a reinforcement learning cost adjustment module. According to the invention, a convolutional neural network algorithm and a principal component analysis method in the feature analysis module excavate building information model data and screen key features, and the cost estimation module combines a multi-layer perceptron neural network and a linear regression model to enhance the cost mode analysis capability and calculate the project cost. The digital twinborn technology and the dynamic programming algorithm realize tracking of project progress and real-time monitoring of cost change in the real-time monitoring module, and the genetic algorithm and the machine learning algorithm realize continuous optimization and self-learning of model parameters in the model training and tuning module. And a Monte Carlo simulation method and a sensitivity analysis method are used for identifying and evaluating project risks in the risk analysis module.

Description

Technical Field

[0001] The present invention relates to the technical field of cost assessment, and particularly to an optimized system for civil engineering and construction project cost assessment. Background Art

[0002] The technical field of cost assessment focuses on estimating and controlling the costs of construction projects, from the initial design stage to the completion stage. Cost assessment techniques cover a wide range of methods and tools aimed at accurately predicting the overall project expenditure, including the costs of materials, labor, equipment, and other resources. Such assessment is crucial for budget preparation, cost control, ensuring financial efficiency, and supporting the decision-making process. With the advancement of technology, the cost assessment field has started to adopt more automated tools and software, improving the accuracy and efficiency of predictions.

[0003] An optimized system for civil engineering and construction project cost assessment is a computer system specifically designed for civil and construction projects, aiming to provide more accurate and efficient cost assessment and optimization solutions. The core objective of this system is to reduce the overall cost of construction projects through automation and optimization processes, while maintaining or improving project quality. The introduction of the system is intended to help project managers and decision-makers better understand and control project costs, thereby improving the effective use of funds, reducing waste, and supporting more sustainable construction practices.

[0004] Traditional systems have many deficiencies in actual operation. Data processing and analysis methods often lack the support of deep learning and big data analysis, resulting in low accuracy and efficiency of data analysis and being unable to fully utilize building information model data. Budget preparation methods lack effective cost model analysis and detailed item-by-item consideration, making cost estimation inaccurate and difficult to meet the requirements of complex projects. In terms of project monitoring, there is a lack of real-time dynamic tracking and cost change monitoring, resulting in slow response from project management and being unable to adjust and optimize project execution strategies in a timely manner. Risk management methods usually lack effective prediction and quantification tools, making risk assessment and control lack foresight and precision. These deficiencies lead to low project management efficiency, inaccurate cost control, and insufficient risk response capabilities, affecting the overall project success rate and return on investment. Summary of the Invention

[0005] The object of the present invention is to solve the drawbacks existing in the prior art and propose an optimized system for civil engineering and construction project cost assessment.

[0006] To achieve the above object, the present invention adopts the following technical solutions: The optimized system for civil engineering and construction project cost assessment includes a feature analysis module, a cost estimation module, a real-time monitoring module, a model training and tuning module, a risk analysis module, a decision support module, a financial simulation module, and a reinforcement learning cost adjustment module;

[0007] The feature analysis module extracts features using a convolutional neural network algorithm based on building information model data, screens key features in combination with the principal component analysis method, standardizes the feature data through data normalization processing, and generates feature analysis data;

[0008] The cost estimation module analyzes the cost pattern using a multi-layer perceptron neural network based on the feature analysis data, calculates the project cost in combination with a linear regression model, and refines the budget item by item using the analysis of variance method to generate an estimated cost value;

[0009] The real-time monitoring module constructs a virtual model of the project using digital twin technology based on the estimated cost value, monitors the project progress in combination with time series analysis, tracks cost changes using a dynamic programming algorithm, and generates real-time monitoring data;

[0010] The model training and optimization module evaluates the model using cross-validation technology based on the real-time monitoring data, optimizes and adjusts the model parameters using a genetic algorithm, performs self-learning and optimization of the model using machine learning algorithms, and generates an optimized model;

[0011] The risk analysis module analyzes cost risks using the Monte Carlo simulation method based on the optimized model, evaluates the potential risk impacts in combination with a decision tree model, and explores the contribution values of multiple factors to project risks using a sensitivity analysis method to generate a risk analysis overview;

[0012] The decision support module optimizes decisions using logistic regression analysis and linear programming methods in combination with the risk analysis overview and real-time monitoring data, provides project decision support using an expert system, and generates a decision-making plan;

[0013] The financial simulation module simulates different financial scenarios using financial modeling methods according to the decision-making plan, conducts comprehensive financial forecasting in combination with market trend analysis methods, analyzes the feasibility of the simulation results using probability theory, and generates a financial forecasting analysis;

[0014] The reinforcement learning cost adjustment module dynamically adjusts cost estimation using Q-learning and deep deterministic policy gradient algorithms based on the financial forecasting analysis, optimizes the learning process using a simulated annealing algorithm, conducts adaptive adjustment of cost estimation, and generates a dynamic cost adjustment strategy.

[0015] As a further solution of the present invention, the feature analysis data includes the geometric shape of the building, the material type, and the design parameters; the estimated cost value includes the total cost estimate and the itemized cost estimate; the real-time monitoring data includes the project progress tracking, the cost expenditure monitoring, and the resource usage; the optimization model includes the adjusted network parameters, the prediction accuracy, and the data processing flow; the risk analysis overview includes the cost overrun risk, the schedule delay risk, and the resource shortage risk; the decision-making plan includes the budget adjustment plan, the risk management strategy, and the schedule optimization plan; the financial prediction analysis includes the revenue prediction scenario, the cost estimate scenario, and the market change prediction; the dynamic cost adjustment strategy includes the adjusted cost budget, the market adaptability strategy, and the resource allocation optimization.

[0016] As a further solution of the present invention, the feature analysis module includes a data mining sub-module, a structure analysis sub-module, and a design evaluation sub-module;

[0017] Based on the building information model data, the data mining sub-module adopts a deep learning feature extraction algorithm. During the process, through the convolutional neural network, the stacked convolutional layers and pooling layers automatically learn the spatial hierarchical features, and through the unsupervised learning method of the autoencoder, the input data is reconstructed to mine the structure and pattern in the data, and the in-depth mining of features is carried out to generate feature extraction data;

[0018] Based on the feature extraction data, the structure analysis sub-module adopts a key feature recognition method. By linear transformation, the original features are transformed into a set of linearly uncorrelated principal components to reduce the data dimension, and features are extracted by maximizing the between-class difference and minimizing the within-class difference to generate key feature data;

[0019] Based on the key feature data, the design evaluation sub-module adopts data preprocessing and standardization methods. By data normalization, the range of feature values is adjusted to eliminate the influence between different magnitude features, and the errors and incomplete data in the data set are removed or corrected to conduct a comprehensive design evaluation and generate feature analysis data.

[0020] As a further solution of the present invention, the cost estimation module includes a cost modeling sub-module, a budget prediction sub-module, and an itemized analysis sub-module;

[0021] Based on the feature analysis data, the cost modeling sub-module adopts a multi-layer perceptron neural network algorithm to construct the cost model, including setting the number of network layers, the number of neurons, and the selection of activation functions to capture the non-linear relationship of the cost data, and then applying the principal component analysis to reduce the dimension of the extracted features to generate the cost model analysis result;

[0022] The budget forecasting sub-module forecasts the total cost using a linear regression model based on the cost pattern analysis results, including establishing the linear relationship between variables, calculating the regression coefficients, and conducting hypothesis tests. At the same time, it combines the time series analysis method to analyze the time series characteristics of historical data through the autoregressive model and the moving average model, providing trend and periodic information for cost forecasting and generating the total project cost forecast.

[0023] The sub-item analysis sub-module refines the budget item by item based on the total project cost forecast using variance analysis, including comparing the mean differences between multiple sub-item costs, identifying the impact degree of multiple sub-item costs on the total cost, and allocating costs accordingly for budget analysis and generating the estimated cost value.

[0024] As a further solution of the present invention, the real-time monitoring module includes a digital twin modeling sub-module, a data collection sub-module, and a progress tracking sub-module.

[0025] The digital twin modeling sub-module conducts three-dimensional modeling of the project using computer-aided design based on the estimated cost value, including using geometric modeling tools to draw the three-dimensional structure of the project, and at the same time integrating finite element analysis to simulate the response of the project under various environmental conditions to generate the project virtual model.

[0026] The data collection sub-module collects project progress data based on the project virtual model using time series analysis techniques, collects real-time data by deploying a variety of sensors, and uses data collection software to integrate and analyze the time series data to monitor the project progress and key indicators, generating project progress monitoring data.

[0027] The progress tracking sub-module tracks the cost changes based on the project progress monitoring data using the dynamic programming algorithm, including establishing a relationship model between cost and progress, optimizing resource allocation and cost control through continuous decision points, conducting project cost management, and generating real-time monitoring data.

[0028] As a further solution of the present invention, the model training and tuning module includes a training strategy sub-module, a verification and testing sub-module, and a parameter adjustment sub-module.

[0029] The training strategy sub-module conducts preliminary training of the neural network based on the real-time monitoring data using the backpropagation algorithm, by calculating the output error and propagating it back into the network to adjust the weights, and at the same time combining the gradient descent method to adjust the weights and biases of each neuron, reducing the prediction error and gradually improving the fitting degree of the model to the cost data, generating a preliminary prediction model.

[0030] The validation test sub-module, based on the preliminary prediction model, applies K-fold cross-validation to divide the dataset into K mutually exclusive subsets. Each time, one subset is used as the test set, and the rest are used as the training set. The model training and testing are repeated to evaluate the performance of the model on different data subsets and generate validation test results.

[0031] The parameter adjustment sub-module, based on the validation test results, uses a genetic algorithm to optimize the model parameters. By simulating the selection, crossover, and mutation mechanisms in biological evolution, it optimizes the structure and parameter configuration of the neural network, captures the optimal parameter combination, and generates an optimized model.

[0032] As a further solution of the present invention, the risk analysis module includes a scenario simulation sub-module, a probability analysis sub-module, and an impact assessment sub-module.

[0033] The scenario simulation sub-module, based on the optimized model, uses the Monte Carlo simulation method. By generating a batch of random variables, it reflects the impact of different risk factors on the cost, including setting parameter ranges, generating random data points, simulating different risk scenarios, analyzing the distribution characteristics of their impact on the cost, and generating a cost risk scenario analysis.

[0034] The probability analysis sub-module, based on the cost risk scenario analysis, uses a decision tree model. By constructing a risk decision tree, where the nodes represent risk events and the edges represent the likelihood of events occurring, it calculates the conditional probability and cumulative probability of each node to evaluate the risk impact under different decision paths and generates a risk probability and impact assessment.

[0035] The impact assessment sub-module, based on the risk probability and impact assessment, uses the sensitivity analysis method. By changing the values of key variables and observing the impact on the final risk assessment, including selecting key variables, adjusting the values one by one, recording the result changes, and analyzing the degree of influence of key factors on the risk assessment, it generates a risk analysis overview.

[0036] As a further solution of the present invention, the decision support module includes a strategy plan sub-module, a budget adjustment sub-module, and a risk management sub-module.

[0037] The strategy plan sub-module, based on the risk analysis overview and real-time monitoring data, uses the logistic regression analysis method to analyze multiple decision options, determines the logical relationship between the decision options and the project success rate, calculates the success probability of each decision option by encoding data features and assigning weights, and generates a strategy priority analysis result.

[0038] The budget adjustment sub-module, based on the result of policy priority analysis, reallocates and optimizes the budget using the linear programming method, sets the budget constraints, including cost ceilings and resource availability, and then applies the linear programming algorithm to capture the optimal budget plan under the conditions, generating a budget optimization plan;

[0039] The risk management sub-module, based on the budget optimization plan and combined with the recommendations of the expert system, comprehensively manages the potential risks of the project, including using decision trees and Bayesian networks to analyze risk factors, identifying key risk points, and formulating mitigation measures, supporting the decision-making process through expert knowledge and historical data, and generating a decision-making plan.

[0040] As a further solution of the present invention, the financial simulation module includes a financial modeling sub-module, a market analysis sub-module, and a scenario planning sub-module;

[0041] The financial modeling sub-module, based on the decision-making plan, uses cash flow analysis to build a financial model, estimates various types of income and expenditure flows, including projected sales revenue, direct costs, and indirect expenses, and at the same time evaluates the financial health of the project with reference to capital expenditures and operating expenditures, generating preliminary financial simulation results;

[0042] The market analysis sub-module, based on the preliminary financial simulation results, applies trend analysis, refers to market dynamics, including the impact of price fluctuations and supply-demand relationships on the project's finances, and reflects market changes by adjusting model parameters, including modifying price assumptions and cost forecasts, optimizing the financial forecasting model, and generating a market-adjusted financial forecast;

[0043] The scenario planning sub-module, based on the market-adjusted financial forecast, uses probability theory methods to construct differentiated financial scenarios, analyzes the financial risks and opportunities under different market and operating conditions, evaluates the financial feasibility of multiple scenarios, and generates a financial forecast analysis.

[0044] As a further solution of the present invention, the reinforcement learning cost adjustment module includes a policy learning sub-module, a data feedback sub-module, and a continuous optimization sub-module;

[0045] The policy learning sub-module, based on the financial forecast analysis, adopts the Q-learning algorithm, guides the preliminary dynamic adjustment of cost estimation by establishing and updating a state-action value table, gradually improves through a cyclic trial-and-error process, and conducts policy evaluation and correction, generating a preliminary cost adjustment policy;

[0046] The data feedback sub-module, based on the preliminary cost adjustment policy, adopts the deep deterministic policy gradient algorithm, learns the optimal policy by constructing a deep neural network, refines the policy in combination with historical data, optimizes the action output in the decision-making process, and conducts iterative adjustment of the policy, generating an optimized cost adjustment policy;

[0047] The continuous optimization sub-module, based on the optimization cost adjustment strategy, adopts the simulated annealing algorithm. By simulating the gradual decrease of temperature in the physical annealing process, it controls the randomness in the search process, captures the optimal solution of cost estimation in the global search space, conducts adaptive optimization of the strategy, and generates a dynamic cost adjustment strategy.

[0048] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0049] In the present invention, the convolutional neural network algorithm and the principal component analysis method in the feature analysis module deeply mine the building information model data, accurately screen key features, and improve the accuracy and efficiency of data processing. The cost estimation module combines the multi-layer perceptron neural network and the linear regression model, enhances the cost pattern analysis ability, accurately calculates the project cost, and improves the accuracy and detail of budget preparation. The digital twin technology and the dynamic programming algorithm in the real-time monitoring module achieve accurate tracking of project progress and real-time monitoring of cost changes, improving the transparency and response speed of project management. The genetic algorithm and the machine learning algorithm in the model training and tuning module achieve continuous optimization of model parameters and self-learning, ensuring the adaptability and stability of the evaluation system. The Monte Carlo simulation method and the sensitivity analysis method effectively identify and evaluate project risks in the risk analysis module, improving the predictability and accuracy of risk management. Description of the Drawings

[0050] Figure 1 is the system flow chart of the present invention;

[0051] Figure 2 is the schematic diagram of the system framework of the present invention;

[0052] Figure 3 is the flow chart of the feature analysis module of the present invention;

[0053] Figure 4 is the flow chart of the cost estimation module of the present invention;

[0054] Figure 5 is the flow chart of the real-time monitoring module of the present invention;

[0055] Figure 6 is the flow chart of the model training and tuning module of the present invention;

[0056] Figure 7 is the flow chart of the risk analysis module of the present invention;

[0057] Figure 8 is the flow chart of the decision support module of the present invention;

[0058] Figure 9 is the flow chart of the financial simulation module of the present invention;

[0059] Figure 10 Flow chart of the reinforcement learning cost adjustment module of the present invention. Specific implementation mode

[0060] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, 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 used to limit the present invention.

[0061] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. In addition, in the description of the present invention, the meaning of "a plurality" is two or more unless otherwise specifically defined.

[0062] Embodiment 1

[0063] Please refer to Figures 1 to 2 , the civil engineering construction cost evaluation and optimization system includes a feature analysis module, a cost estimation module, a real-time monitoring module, a model training and tuning module, a risk analysis module, a decision support module, a financial simulation module, and a reinforcement learning cost adjustment module;

[0064] The feature analysis module extracts features using a convolutional neural network algorithm based on building information model data, combines the principal component analysis method to screen key features, standardizes the feature data through data normalization processing, and generates feature analysis data;

[0065] The cost estimation module analyzes the cost model using a multi-layer perceptron neural network based on the feature analysis data, calculates the project cost in combination with a linear regression model, and applies the analysis of variance method to break down the budget item by item, thereby generating an estimated cost value;

[0066] The real-time monitoring module constructs a virtual model of the project using digital twin technology based on the estimated cost value, monitors the project progress in combination with time series analysis, applies a dynamic programming algorithm to track cost changes, and generates real-time monitoring data;

[0067] The model training and tuning module evaluates the model using cross-validation technology based on the real-time monitoring data, optimizes and adjusts the model parameters using a genetic algorithm, performs self-learning and optimization of the model using a machine learning algorithm, and generates an optimized model;

[0068] Based on the optimization model, the risk analysis module uses the Monte Carlo simulation method to analyze cost risks, combines the decision tree model to evaluate the potential risk impacts, applies the sensitivity analysis method to explore the contribution values of multiple factors to project risks, and generates a risk analysis overview;

[0069] Combining the risk analysis overview and real-time monitoring data, the decision support module uses the logistic regression analysis method and the linear programming method for decision optimization, utilizes the expert system to provide project decision support, and generates a decision-making plan;

[0070] According to the decision-making plan, the financial simulation module uses financial modeling methods to simulate differential financial scenarios, combines the market trend analysis method for comprehensive financial forecasting, applies probability theory to analyze the feasibility of the simulation results, and generates a financial forecasting analysis;

[0071] Based on the financial forecasting analysis, the reinforcement learning cost adjustment module uses the Q-learning and deep deterministic policy gradient algorithms for dynamic adjustment of cost estimation, applies the simulated annealing algorithm to optimize the learning process, conducts adaptive adjustment of cost estimation, and generates a dynamic cost adjustment strategy.

[0072] The feature analysis data includes the geometric shape, material type, and design parameters of the building. The estimated cost values include the total cost estimate and the itemized cost estimate. The real-time monitoring data includes project progress tracking, cost expenditure monitoring, and resource usage. The optimization model includes the adjusted network parameters, prediction accuracy, and data processing flow. The risk analysis overview includes cost overrun risk, schedule delay risk, and resource shortage risk. The decision-making plan includes the budget adjustment plan, risk management strategy, and schedule optimization plan. The financial forecasting analysis includes the revenue forecasting scenario, cost estimation scenario, and market change prediction. The dynamic cost adjustment strategy includes the adjusted cost budget, market adaptability strategy, and resource allocation optimization.

[0073] In the feature analysis module, through the building information model (BIM) data, the convolutional neural network (CNN) algorithm deeply extracts features such as the building's geometric shape, material type, and design parameters. The CNN automatically learns data features through multiple layers of filters, abstracting layer by layer to obtain deeper information. Then, the principal component analysis (PCA) method screens out the most critical features from numerous features. This process determines the importance of each feature through variance analysis and retains the components that best represent the data characteristics. Data normalization then standardizes these feature data so that they fall within the same numerical range, facilitating algorithm processing. This series of operations effectively extracts and streamlines the feature data, enabling subsequent modules to perform data analysis and model establishment more efficiently.

[0074] In the cost estimation module, using the feature analysis data, a multi-layer perceptron (MLP) neural network is adopted for cost pattern analysis. The MLP conducts in-depth learning on the input data through its hidden layer, thereby establishing a complex association between construction costs and features. Combining with the linear regression model, this module calculates the total cost and itemized costs of the project, providing accurate cost estimates. The analysis of variance method further refines the budget, making the cost estimation more detailed and specific. This method that combines neural networks and traditional statistical models not only utilizes the powerful fitting ability of in-depth learning but also maintains the interpretability of the model. Finally, the generated estimated cost value has high accuracy and practicality.

[0075] In the real-time monitoring module, based on the estimated cost value, the digital twin technology is used to construct a virtual model of the project to achieve real-time monitoring of the project progress. The digital twin technology enables the project progress to be updated and monitored in real time in the digital space by creating a virtual copy of the physical project. Time series analysis is used to monitor the project progress, and the project trends and patterns are analyzed through the collected progress data. The dynamic programming algorithm is used to track the cost changes, optimize the resource allocation, and adjust the budget. The data processing and analysis mechanism of this module not only improves the efficiency and transparency of project management but also can discover and solve problems in a timely manner.

[0076] In the model training and tuning module, the real-time monitoring data is used to evaluate and optimize the model performance. The cross-validation technique plays a key role here. By dividing the data into multiple parts and alternately using one part as the test set and the rest as the training set, the generalization ability of the model can be comprehensively evaluated. Then, the genetic algorithm is used to optimize the model parameters. Imitating the mechanism of natural selection, it searches for the optimal solution in the parameter space through operations such as selection, crossover, and mutation, so as to improve the prediction accuracy of the model. At the same time, machine learning algorithms are used for the self-learning and continuous optimization of the model, which enables the model to continuously adjust and improve itself according to new data, thereby achieving continuous improvement of the model performance. This series of training and tuning processes ensure the efficiency and accuracy of the model when processing real-time monitoring data.

[0077] In the risk analysis module, the optimized model is used to analyze the cost risks. The Monte Carlo simulation method predicts the cost risks through a large number of random samplings, providing the probability distribution of cost overruns for decision-makers to help them better understand and prepare for risks. At the same time, the decision tree model evaluates the impact of potential risks and intuitively shows the risk results under different decision paths through the tree diagram model. The sensitivity analysis method further explores the contribution degree of each factor to the project risks, helping decision-makers identify which factors have a greater impact on the project risks. These analysis results are aggregated into a risk analysis overview, providing comprehensive risk assessment and decision support for project management.

[0078] The decision support module combines the risk analysis overview and real-time monitoring data, and uses logistic regression analysis and linear programming methods for decision optimization. Logistic regression analysis predicts the success probabilities of different decision-making scenarios through learning historical data, while linear programming is used to find the optimal solution under given constraints to achieve the best allocation and utilization of resources. In addition, the application of the expert system provides project decision support based on historical data and industry knowledge. By simulating the decision-making process of experts, it provides suggestions on budget adjustment, risk management strategies, and schedule optimization plans. These decision-making scenarios integrate data analysis and industry expert knowledge, improving the accuracy and reliability of decisions.

[0079] In the financial simulation module, according to the decision-making scenario, financial modeling methods are used to simulate different financial scenarios. This process combines market trend analysis to predict financial performance under different market conditions, such as revenues, costs, and market changes. Probability theory is used to analyze the feasibility of these simulation results. By calculating the success probabilities under different scenarios, it helps decision-makers understand the risks and returns of various financial strategies. These financial forecast analyses provide a comprehensive financial perspective for project management, helping decision-makers make more informed financial decisions.

[0080] In the reinforcement learning cost adjustment module, based on the financial forecast analysis, Q-learning and Deep Deterministic Policy Gradient (DDPG) algorithms are used to dynamically adjust the cost estimation. These algorithms learn the optimal strategy through interaction with the environment to achieve adaptive adjustment of the cost budget. The simulated annealing algorithm further optimizes the learning process. By simulating the temperature drop in the metallurgical annealing process, it effectively avoids getting stuck in local optimal solutions and thus finds the global optimal strategy. The final output of this module is the dynamic cost adjustment strategy, including the adjusted cost budget, market adaptability strategy, and resource allocation optimization, ensuring that the project maintains the optimal cost efficiency in a changing environment.

[0081] Please refer to Figure 3 , the feature analysis module includes a data mining sub-module, a structure analysis sub-module, and a design evaluation sub-module;

[0082] The data mining sub-module is based on building information model data and uses deep learning feature extraction algorithms. In the process, through convolutional neural networks with cascaded convolutional layers and pooling layers, it automatically learns spatial hierarchical features, and through the unsupervised learning method of autoencoders, it reconstructs the input data to mine the structure and patterns in the data and conduct in-depth mining of features to generate feature extraction data;

[0083] The structure analysis sub-module is based on the feature extraction data and uses key feature recognition methods. It transforms the original features into a set of linearly uncorrelated principal components through linear transformation to reduce the data dimension, and extracts features by maximizing the between-class difference and minimizing the within-class difference to generate key feature data;

[0084] Based on the key feature data, the design evaluation sub-module uses data preprocessing and normalization methods to adjust the range of feature values through data normalization, eliminate the influence between differential magnitude features, remove or correct errors and incomplete data in the dataset, and conduct comprehensive design evaluation to generate feature analysis data.

[0085] In the data mining sub-module, the system uses Building Information Modeling (BIM) data and adopts deep learning feature extraction algorithms, with the core being Convolutional Neural Networks (CNNs) and autoencoders. This process involves multiple refinement steps, each based on specific algorithms and mathematical methods, aiming to deeply mine the structure and patterns of building data. First, the application of the convolutional neural network starts with the input layer receiving BIM data. This data usually exists in the form of a three-dimensional building model, containing information such as the geometric shape, physical properties, and spatial relationships of the building. The convolutional layer filters the input data using multiple convolutional kernels to extract local features. These convolutional kernels are automatically learned during the training process through the backpropagation algorithm, aiming to identify various spatial shapes and patterns. Next, the pooling layer acts on the output of the convolutional layer to reduce the feature dimension and enhance the invariance of features. Pooling operations usually include max pooling or average pooling, which reduce the computational amount and prevent overfitting by reducing the data volume while retaining important features. After that, the autoencoder, as an unsupervised learning model, is used to reconstruct the input data, that is, compress the input into a latent space representation through the encoder and then reconstruct the data through the decoder. This process helps to discover the deep structure and patterns in the data. The training of the autoencoder involves minimizing the reconstruction error between the input and the output, usually using mean squared error or cross-entropy as the loss function. Through these steps, the data mining sub-module can extract deep features from BIM data, laying a foundation for subsequent analysis and applications. This process ultimately generates feature extraction data, that is, a set of feature vectors that can represent the key information of the original BIM data. These feature vectors can be used for subsequent structural analysis and design evaluation, improving the efficiency and accuracy of the entire system.

[0086] In the structural analysis sub-module, the system applies key feature recognition methods, mainly principal component analysis (PCA) and linear discriminant analysis (LDA), based on the feature extraction data. Principal component analysis generates a set of linearly uncorrelated principal components through linear transformation of the original features. The aim is to reduce the dimensionality of the data and retain the most important information. In this process, the algorithm first calculates the covariance matrix of the data, and then finds the eigenvectors and eigenvalues of this matrix. These eigenvectors define a new coordinate system for the data, while the eigenvalues determine the importance of each principal component in the data variation. By selecting the eigenvectors corresponding to the largest several eigenvalues, a reduced-dimensional data representation can be obtained. Linear discriminant analysis then further works on the reduced-dimensional data, extracting key features by maximizing the between-class difference and minimizing the within-class difference. First, the mean of the data within each class and the overall mean are calculated, and then a linear combination is constructed such that when the data is projected by this linear combination, the data of different classes are separated as much as possible, and the data of the same class are clustered as much as possible. In this way, LDA can extract the features that best represent the classification characteristics of the data. After these steps are completed, the structural analysis sub-module generates key feature data, which, while reducing the dimensionality, retains the most discriminative features, providing strong data support for subsequent design evaluation.

[0087] In the design evaluation sub-module, the system performs data preprocessing and standardization operations based on the key feature data. Data preprocessing includes data cleaning and data transformation, aiming to remove or correct errors and incomplete data in the dataset, such as filling missing values through interpolation methods or removing outliers in the data through outlier detection and handling methods. Data standardization adjusts the range of feature values through data normalization, enabling features of different magnitudes to be compared on the same scale. Common methods include min-max normalization and Z-score normalization. In this process, each step is carefully designed to ensure the quality and consistency of the data, providing a solid foundation for comprehensive design evaluation. After these operations, the design evaluation sub-module finally generates feature analysis data, which not only reflects the multi-faceted characteristics of the architectural design but also provides a comprehensive and accurate evaluation basis for decision-makers.

[0088] Suppose in a civil engineering construction project cost evaluation and optimization system, a new construction project needs to be analyzed for costs and benefits. This analysis process first involves extracting key information from the project's BIM data. Suppose this BIM data includes information such as the building's dimensions, types and quantities of materials used, and design complexity. In the data mining sub-module, the system first uses a convolutional neural network to preliminarily process this data, extracting the main structural features of the building, such as the layout of different floors and the distribution of materials. Then, an autoencoder further processes these features to identify patterns and structures, such as which design elements are key cost-driving factors. This step generates a set of feature vectors, where each vector represents a key aspect of the building. Next, in the structural analysis sub-module, the system uses principal component analysis and linear discriminant analysis to further process these feature vectors. Through these two steps, the system can determine which features are the most important in cost prediction. For example, it is found that certain specific design elements (such as specific types of windows or floor heights) have the highest correlation with costs. Finally, in the design evaluation sub-module, the system standardizes these key features and conducts a comprehensive analysis in combination with other relevant data (such as historical building cost data). This process includes applying different statistical models and algorithms, and ultimately generates a detailed cost evaluation report. This report not only gives the overall cost prediction but also includes the specific impact of different design choices on costs, providing valuable reference information for decision-makers.

[0089] Please refer to Figure 4 , the cost estimation module includes a cost modeling sub-module, a budget forecasting sub-module, and a breakdown analysis sub-module;

[0090] The cost modeling sub-module constructs a cost model based on the feature analysis data using a multi-layer perceptron neural network algorithm, including setting the number of network layers, the number of neurons, and the selection of activation functions to capture the non-linear relationships in the cost data. Then, principal component analysis is applied to reduce the dimensionality of the extracted features, generating the cost model analysis results;

[0091] The budget forecasting sub-module predicts the total cost based on the cost model analysis results using a linear regression model, including establishing the linear relationships between variables, calculating the regression coefficients, and conducting hypothesis tests. At the same time, combined with time series analysis methods, the autoregressive model and moving average model are used to analyze the time series characteristics of historical data to provide trend and periodic information for cost prediction, generating the total project cost prediction;

[0092] The breakdown analysis sub-module refines the budget item by item based on the total project cost prediction using variance analysis, including comparing the mean differences between multiple breakdown costs, identifying the impact degree of multiple breakdown costs on the total cost, and accordingly conducting cost allocation and budget analysis to generate the estimated cost value.

[0093] In the cost modeling sub-module, the data format is structured data based on feature analysis, including but not limited to cost items, historical cost data, time, location, etc. This sub-module uses the multi-layer perceptron neural network (MLP) algorithm to construct the cost model. MLP is a feed-forward neural network where each node (i.e., neuron) is connected to all nodes in the previous and next layers, but there are no connections between nodes in the same layer. First, according to the characteristics and requirements of the cost data, appropriate network layers and the number of neurons are set. Generally, increasing the number of network layers and neurons can improve the learning ability of the model, but it also leads to overfitting. Therefore, methods such as cross-validation are needed to optimize these parameters. Next, the activation function is selected, which is a very crucial step in the neural network. Commonly used activation functions include ReLU, Sigmoid, and Tanh, etc. The choice of the activation function affects the learning speed and effect of the network. In the sub-module, the selection of the activation function aims to capture the non-linear relationship of the cost data, so as to more accurately simulate and predict the cost model. After the network structure and activation function are determined, the neural network is trained using the cost data. During the training process, the network adjusts the weights and biases to minimize the difference between the predicted cost and the actual cost. After training is completed, principal component analysis (PCA) is used to reduce the dimensionality of the features. PCA is a statistical method that transforms the data into a set of linearly independent variables, called principal components, through orthogonal transformation. In cost data, PCA helps to remove noise and redundancy and highlight the most important features. Through these steps, the final cost model analysis results can reflect the complex relationship between cost and various factors and provide a basis for subsequent budget prediction.

[0094] In the budget forecasting sub-module, based on the results of cost pattern analysis, a linear regression model is used to forecast the total cost. Linear regression is a statistical method used to establish the relationship between one or more independent variables (predictors) and a dependent variable (in this case, the total cost). First, the key variables affecting the total cost are determined. These variables are derived from the results of the cost modeling sub-module and include the historical costs of specific projects, market trends, material prices, etc. After determining these variables, relevant data is collected to construct the linear regression model. During the model building process, the linear relationship hypothesis of the variables is first made. This means assuming that the dependent variable (total cost) is a linear combination of the independent variables (such as historical costs, market trends, etc.). Then, the regression coefficients are calculated, and these coefficients represent the influence intensity of the independent variables on the dependent variable. The calculation of the regression coefficients is usually completed by minimizing the difference between the actual cost value and the cost value predicted by the model. Next, hypothesis testing is carried out, including testing the significance of each independent variable and the goodness of fit of the entire model. This step is crucial because it helps to identify which variables have a significant impact on cost forecasting, thus ensuring the accuracy and reliability of the model. In addition, the time series analysis method, especially the autoregressive model (AR) and the moving average model (MA), is combined to analyze the time series characteristics of historical data. In this way, the budget forecasting model can not only reflect the relationship between costs and various factors but also capture the changing trends and periodic patterns of costs over time. Finally, the project total cost forecast generated by this sub-module is not only a specific numerical value or cost range but a comprehensive forecast result that includes time dynamics and analysis of key influencing factors. This provides strong support for project management and budget control, enabling decision-makers to more accurately predict and manage project costs.

[0095] In the itemized analysis sub-module, based on the prediction of the total project cost, the budget is itemized and refined using analysis of variance (ANOVA). ANOVA is a statistical method used to compare whether there are significant differences in the means of two or more samples. In this sub-module, first, the cost items to be analyzed are determined, such as labor costs, material costs, equipment usage fees, etc. Each itemized cost is regarded as an independent sample, and the purpose is to compare whether there are significant differences between the means of these samples. Before conducting ANOVA, it is necessary to ensure that the data meets the basic assumptions of ANOVA, including sample independence, normality, and homoscedasticity. After meeting these assumptions, ANOVA calculations are performed to draw conclusions on whether there are significant differences between each cost item. If significant differences are found, post hoc multiple comparison tests are further conducted to determine which specific cost items are different. Through ANOVA, it is possible to identify which itemized costs have a significant impact on the total cost, which is very important for cost control and optimization. For example, if it is found that a specific cost item is significantly higher compared to other items, the project manager can focus on cost control and optimization strategies in this area. In addition, the results of ANOVA can also be used for cost allocation, that is, reasonable cost allocation is carried out according to the contribution degree of each itemized cost to the total cost. Finally, the estimated cost values generated by the itemized analysis sub-module are not only the specific values of each itemized cost, but a comprehensive budget analysis result including cost allocation and cost control guidance. This is crucial for the refined management of project budgets and cost optimization.

[0096] Suppose there is a civil engineering project cost evaluation and optimization system. The data items included in this system are historical cost data, market price indices, quantities of work, material consumption, etc. Suppose in a specific project, the simulated values of these data items are input into the system. The cost modeling sub-module analyzes these data through an MLP neural network to generate cost pattern analysis results. The budget prediction sub-module, based on these analysis results, uses linear regression and time series analysis to generate a prediction of the total project cost. Finally, the itemized analysis sub-module refines the budget through ANOVA to generate detailed estimated cost values.

[0097] Please refer to Figure 5 , the real-time monitoring module includes a digital twin modeling sub-module, a data acquisition sub-module, and a progress tracking sub-module;

[0098] The digital twin modeling sub-module, based on the estimated cost value, uses computer-aided design to perform 3D modeling of the project, including using geometric modeling tools to draw the 3D structure of the project, and at the same time integrating finite element analysis to simulate the response of the project under various environmental conditions to generate a virtual model of the project;

[0099] The data acquisition sub-module collects project progress data based on the project virtual model, using time series analysis techniques. It deploys a variety of sensors to collect real-time data, and uses data collection software to integrate and analyze the time series data, monitor the project progress and key indicators, and generate project progress monitoring data;

[0100] The progress tracking sub-module tracks cost changes based on the project progress monitoring data, applying dynamic programming algorithms, including establishing a relationship model between cost and progress, optimizing resource allocation and cost control through continuous decision points, conducting project cost management, and generating real-time monitoring data.

[0101] In the digital twin modeling sub-module, through computer-aided design and geometric modeling tools, combined with the estimated cost value, three-dimensional modeling of the project is achieved. This process first involves accurately depicting the physical characteristics of the project, including dimensions, shapes, material properties, etc. The data format is usually a 3D CAD file, supporting precise geometric description and attribute assignment. On this basis, through finite element analysis software such as ANSYS or ABAQUS, detailed simulation calculations are carried out. At this stage, the algorithm predicts the response of the project under different environmental conditions according to physical principles and mathematical models. This includes calculations of stress, deformation, heat flow, etc., and the algorithm refines the operations according to the input material properties, boundary conditions, and load conditions. Finally, the generated virtual model not only shows the three-dimensional structure of the project but also reflects its behavioral characteristics under various environmental conditions, providing a basis for subsequent design optimization and decision-making.

[0102] In the data acquisition sub-module, based on the project virtual model, time series analysis techniques are used to collect project progress data. During this process, a variety of sensors are deployed to collect real-time data such as temperature, humidity, vibration, etc., and the data format is usually a time-stamped numerical sequence. These data are integrated and analyzed through dedicated data collection software, and the software uses time series analysis algorithms such as autoregressive moving average (ARMA) or long short-term memory network (LSTM) to deeply process the data. The algorithm monitors the project progress and key indicators in real time by analyzing the characteristics of the time series, such as periodicity, trend, and randomness. The specific implementation process of these operations includes data preprocessing (such as filtering and standardization), feature extraction, model training, and prediction. The finally generated project progress monitoring data can reflect the real-time status and key indicator changes of the project in detail, helping the project management team adjust strategies and resource allocation in a timely manner.

[0103] In the progress tracking sub-module, based on the project progress monitoring data, the dynamic programming algorithm is used to track cost changes. This sub-module establishes a relationship model between cost and progress, and optimizes resource allocation and cost control through continuous decision points. The data format is usually tabular data containing multi-dimensional information such as time, cost, and resource allocation. The dynamic programming algorithm plays a core role in this process. By decomposing the problem into a series of decision steps, calculating the optimal solution for each step, and using these solutions to construct the global optimal solution. The specific implementation process includes defining the state and state transition equation, determining the boundary conditions, solving each sub-problem through recursive or iterative methods, and recording the solution path. This method is particularly effective in project cost management because it can continuously adjust the strategy throughout the project cycle to cope with uncertainty and changes. Through this method, the finally generated real-time monitoring data not only reflects the real-time status of cost and progress, but also provides predictions of future trends, helping the project team make more accurate decisions.

[0104] Take the civil engineering and construction project cost evaluation and optimization system as an example. First, in the digital twin modeling sub-module, assume there is a bridge project with an estimated cost of $50 million. Use software such as AutoCAD to create a 3D model of the bridge, and the data format is a DWG file. Then, use ANSYS for finite element analysis, input material properties, load conditions, etc., and simulate the structural stress and deformation. The simulation results show that under specific loads, the maximum stress point of the bridge is in the middle section, indicating that the designer needs to reinforce this area. In the data collection sub-module, assume the project progress is 6 months, and collect data through temperature and pressure sensors deployed on the bridge. The sensors record data every hour, and the format is a timestamp and the corresponding measured value. These data are transmitted to the central database and processed through a time series analysis script written in Python. The script first cleans and normalizes the data, and then applies the LSTM network model to analyze the long-term dependencies of the data. Through this analysis, the project team can monitor the performance of the bridge under different environmental conditions and predict potential structural problems. In the progress tracking sub-module, the project team uses the dynamic programming algorithm to track cost changes. Based on the data of the project budget and actual expenditure, a relationship model between cost and progress is established. For example, assume the initial cost budget of the project is $50 million, but due to design adjustments and rising material costs, the total cost is expected to increase to $55 million. The dynamic programming algorithm finds the cost-minimizing strategy by evaluating different resource allocation plans. The algorithm analyzes the cost and benefits at each decision point and gradually constructs the optimal resource allocation path. This method enables the project team to adjust resource allocation in a timely manner and effectively control costs.

[0105] Please refer to Figure 6 , the model training and tuning module includes a training strategy sub-module, a validation and testing sub-module, and a parameter adjustment sub-module;

[0106] Based on the real-time monitoring data, the training strategy sub-module adopts the backpropagation algorithm. By calculating the output error and propagating it back into the network to adjust the weights, it conducts the preliminary training of the neural network. Meanwhile, it combines the gradient descent method to adjust the weights and biases of each neuron, reduces the prediction error, gradually improves the fitting degree of the model to the cost data, and generates a preliminary prediction model.

[0107] Based on the preliminary prediction model, the validation test sub-module applies K-fold cross-validation. It divides the data set into K mutually exclusive subsets, uses one subset as the test set each time, and the rest as the training set. It repeats the model training and testing, evaluates the performance of the model on the differentiated data subsets, and generates the validation test results.

[0108] Based on the validation test results, the parameter adjustment sub-module adopts the genetic algorithm to optimize the model parameters. By simulating the selection, crossover, and mutation mechanisms in biological evolution, it optimizes the structure and parameter configuration of the neural network, captures the optimal parameter combination, and generates an optimized model.

[0109] In the training strategy sub-module, based on the real-time monitoring data, through the backpropagation algorithm and the gradient descent method, the neural network is trained and optimized. The data format used by this sub-module is usually structured numerical data, which contains multi-dimensional information such as cost, time, and resource allocation. These data are used to train a neural network model aiming to accurately predict the project cost and schedule. During the training process, first, the backpropagation algorithm is used to calculate the output error. The steps involve calculating the difference between the model output and the actual value, and propagating the error backward along the network to adjust the weights of each neuron in the network. Then, the gradient descent method is used to optimize the weights and biases of the neural network. This process involves calculating the gradient of the loss function with respect to each weight and adjusting the weights according to this gradient to reduce the overall prediction error. The weight adjustment is an iterative process aiming to find the minimum value of the loss function, thereby improving the fitting degree of the model to the cost data. In this process, the learning rate is a key parameter that determines the step size of the weight adjustment. Finally, the sub-module generates a preliminary prediction model that can more accurately predict the project cost and schedule.

[0110] In the validation test sub-module, based on the preliminary prediction model, the K-fold cross-validation method is applied to evaluate the generalization ability of the model. During this process, the dataset is divided into K non-overlapping subsets. In each iteration, one subset is selected as the test set, and the remaining K-1 subsets are used as the training set. In this way, each subset has the opportunity to be used as the test set, ensuring the comprehensiveness and fairness of the evaluation. The key in K-fold cross-validation is to select an appropriate value of K to achieve a balance between bias and variance. Generally, a larger value of K can provide a more accurate model evaluation, but it will also increase the computational cost. Through the repeated training and testing process, the performance of the model on different data subsets can be obtained, thereby evaluating its overall performance and reliability. The output of this sub-module is a series of model performance metrics, such as accuracy, recall, and F1-score, which comprehensively reflect the performance of the model under various data conditions. Through these test results, the project team can comprehensively evaluate the effectiveness of the model and identify areas that need improvement.

[0111] In the parameter tuning sub-module, based on the validation test results, the genetic algorithm is used to optimize the structure and parameters of the neural network. The genetic algorithm simulates the process of biological evolution, including steps such as selection, crossover, and mutation, to find the optimal parameter combination. In this process, an initial population is first defined, representing different network parameter configurations. Each configuration is regarded as an "individual", and its fitness is determined by the validation test results. Then the selection process is carried out, in which individuals with higher fitness are selected to produce offspring. In the crossover process, parameters are randomly selected from two "parent" individuals to generate "offspring" individuals, which introduces new parameter combinations. The mutation process randomly adjusts some of the parameters in the individual to introduce additional diversity. This iterative process is repeated until the optimal parameter combination is found. Through the genetic algorithm, the parameter space can be effectively explored to find a network configuration that is more suitable for a specific dataset. The finally generated optimized model not only has higher prediction accuracy but also can better adapt to data changes and improve the generalization ability of the model.

[0112] Taking the civil engineering project cost assessment and optimization system as an example, the system first collects real-time monitoring data of the project, including multi-dimensional information such as cost, progress, and resource allocation. In the training strategy sub-module, a neural network model is trained using this data, and the model parameters are adjusted using backpropagation and gradient descent methods to generate a preliminary prediction model. Then, in the validation and testing sub-module, the K-fold cross-validation method is applied to evaluate the performance of the model on different data sets to ensure the accuracy and reliability of the model. Finally, in the parameter adjustment sub-module, the model parameters are further optimized using the genetic algorithm to improve the prediction ability and adaptability of the model. For example, assume that the real-time monitoring data of the project includes weekly cost expenditures, project progress, and resource usage. In the training strategy sub-module, this data is used to train the neural network model. The model is trained through multiple iterations with a learning rate set to 0.01, and the error gradually decreases, and the prediction accuracy of the model gradually improves. In the validation and testing sub-module, the data set is divided into 5 subsets for K-fold cross-validation. Each time, one subset is used as the test set, and the rest are used as the training set. The test results show that the average accuracy of the model on each subset reaches 85%, indicating good generalization ability. Finally, in the parameter adjustment sub-module, the genetic algorithm optimizes the parameters of the model. Through several generations of selection, crossover, and mutation, a better parameter configuration is found, further improving the accuracy and adaptability of the model.

[0113] Please refer to Figure 7 , the risk analysis module includes a scenario simulation sub-module, a probability analysis sub-module, and an impact assessment sub-module;

[0114] Based on the optimization model, the scenario simulation sub-module uses the Monte Carlo simulation method. By generating a batch of random variables, it reflects the impact of different risk factors on the cost, including setting parameter ranges, generating random data points, simulating different risk scenarios, analyzing the distribution characteristics of their impact on the cost, and generating a cost risk scenario analysis;

[0115] Based on the cost risk scenario analysis, the probability analysis sub-module uses a decision tree model. By constructing a risk decision tree, where the nodes represent risk events and the edges represent the likelihood of events occurring, the conditional probability and cumulative probability of each node are calculated to evaluate the risk impact under different decision paths and generate a risk probability and impact assessment;

[0116] Based on the risk probability and impact assessment, the impact assessment sub-module uses the sensitivity analysis method. By changing the values of key variables and observing the impact on the final risk assessment, including selecting key variables, adjusting the values one by one, recording the result changes, and analyzing the degree of influence of key factors on the risk assessment, a risk analysis overview is generated.

[0117] In the scenario simulation sub-module, the system adopts the Monte Carlo simulation method for cost risk analysis. The data format of this process is mainly a randomly generated multi-dimensional data set, which reflects various risk factors. First, the range of parameters is set based on the specific requirements of the project and historical data analysis to ensure that the randomly generated data points can comprehensively cover the risk scenarios. Then, the system generates a large number of random data points through the Monte Carlo algorithm. These data points represent the fluctuations in project costs under different combinations of risk factors. Subsequently, the system uses these data points to simulate different risk scenarios and analyze the specific impacts on costs. During the process, each generated data point is used to construct a specific risk scenario to simulate different situations that occur in reality. In this way, the system can analyze the distribution characteristics of costs under different combinations of risk factors, providing decision-makers with an intuitive understanding of cost fluctuations. In addition, this sub-module also generates a cost risk scenario analysis report, which details how different risk factors affect project costs and provides an important basis for subsequent risk management and decision-making.

[0118] In the probability analysis sub-module, based on the cost risk scenario analysis, the system uses the decision tree model to evaluate risks. The data format of this sub-module is based on the data set generated by the scenario simulation sub-module to further construct a risk decision tree. Each node of the decision tree represents a specific risk event, and the edges represent the likelihood of that event occurring. By calculating the conditional probability and cumulative probability of each node, the system can evaluate the risk impacts under different decision paths. The key in this process is to accurately calculate and display the probability of each risk event and how these events are concatenated to form the overall risk picture. What the system finally generates is a risk probability and impact assessment report, which details the specific impacts of various risk events under different decision paths, providing decision-makers with a quantitative risk assessment to help them make more informed decisions.

[0119] In the impact assessment sub-module, based on the risk probability and impact assessment, the sensitivity analysis method is used to further evaluate risks. During this process, key variables are selected and their values are adjusted one by one to observe the impacts on the final risk assessment. The system records the changes in results by changing the values of these key variables and analyzes the degree of influence of these key factors on the risk assessment. This method allows decision-makers to see how changes in certain key parameters in the project affect the overall risk assessment. Finally, this sub-module generates a risk analysis overview report, which comprehensively shows how changes in key variables affect the overall risk assessment, providing decision-makers with an in-depth understanding of the sensitivity of key risk factors. This analysis not only helps identify which factors have the greatest impact on the risk assessment but also reveals how risks evolve under specific conditions, enabling decision-makers to more effectively manage and mitigate these risks.

[0120] For example, in a civil engineering project cost assessment and optimization system, the scenario simulation sub-module receives a dataset containing multiple building material costs, labor costs, and time limits. Through Monte Carlo simulation, the system generates a series of random data points that reflect the impact of factors such as material price fluctuations, labor cost changes, and project schedule delays on the total project cost. These data points are used to simulate different risk scenarios and generate a detailed cost risk scenario analysis report that shows the range of cost fluctuations under different risk combinations. In the probability analysis sub-module, the system constructs a risk decision tree based on the above cost risk scenario analysis results. For example, a node represents the risk of rising material costs, and the edge represents the likelihood of this event occurring. By calculating the conditional probabilities of these nodes, the system evaluates the impact of the total cost under different risk combinations and generates a risk probability and impact assessment report. Finally, in the impact assessment sub-module, the system conducts a sensitivity analysis of key variables such as material costs, labor costs, and project duration. By adjusting the values of these variables, the system observes the changes in the final risk assessment and generates a risk analysis overview report. This report helps project managers understand how project costs and risks change under different conditions.

[0121] Please refer to Figure 8 , the decision support module includes a strategy plan sub-module, a budget adjustment sub-module, and a risk management sub-module;

[0122] Based on the risk analysis overview and real-time monitoring data, the strategy plan sub-module analyzes multiple decision options using logistic regression analysis to determine the logical relationship between decision options and project success rates. By encoding data features and assigning weights, the system calculates the success probability of each decision option and generates a strategy priority analysis result;

[0123] Based on the strategy priority analysis result, the budget adjustment sub-module reallocates and optimizes the budget using linear programming. It sets budget constraints, including cost ceilings and resource availability, and then uses linear programming algorithms to capture the optimal budget plan under the conditions and generates a budget optimization plan;

[0124] Based on the budget optimization plan and combined with the recommendations of the expert system, the risk management sub-module comprehensively manages the potential risks of the project, including using decision trees and Bayesian networks to analyze risk factors, identifying key risk points, and formulating mitigation measures. It supports the decision-making process with expert knowledge and historical data and generates a decision-making plan.

[0125] In the strategy plan sub-module, the system deeply analyzes multiple decision options through logistic regression analysis. The data format processed by this sub-module is usually a structured data set, including historical decision results, project success rates, and various related factors. The core role of the logistic regression model here is to determine the logical relationship between various decision options and the project success rate. First, it is necessary to encode the data features and convert qualitative data into a numerical form recognizable by the model. For example, different aspects of decision options, such as budget allocation, time arrangement, etc., are assigned specific codes. Then, weight distribution is carried out for these features, and the weights are determined based on the influence degree of these features on the success rate in historical data. The logistic regression model calculates the success probability of each decision option by weighted combination of features. This process not only considers a single factor but also the combined influence of multiple factors. Finally, the system generates the strategy priority analysis result, which details the success probability rankings of different decision options, provides clear priority guidance for decision-makers, and helps them make more reasonable decisions among multiple choices.

[0126] In the budget adjustment sub-module, based on the strategy priority analysis result, the linear programming method is applied to reallocate and optimize the budget. The data format processed by this sub-module usually includes the cost details of the project, the costs and availabilities of various resources, etc. Here, the key of the linear programming method lies in setting reasonable budget limit conditions, such as cost ceilings and resource availabilities. These conditions serve as the constraint conditions of the linear programming model, and the model aims to find the optimal budget allocation plan under these constraints. By constructing the objective function and constraint conditions, the system uses the linear programming algorithm to solve. For example, the objective function may be to maximize the overall value of the project or minimize the cost, while the constraint conditions involve budget ceilings, resource availabilities, etc. Through iterative calculations, the model finds the optimal budget plan that meets all constraint conditions. This process takes into account various cost and resource limitations to ensure that the budget allocation not only meets the project requirements but is also financially feasible. The finally generated budget optimization plan not only provides specific budget allocation details but also explains how to achieve the most effective resource utilization under various constraint conditions. This plan is crucial for ensuring the efficient operation of the project with limited resources.

[0127] In the risk management sub-module, based on the budget optimization plan and combined with the recommendations of the expert system, the potential risks of the project are comprehensively managed. The data formats of this sub-module include the detailed information of the budget optimization plan, the project's historical risk records, expert opinions, etc. During the risk management process, the system first uses decision trees and Bayesian networks to analyze various risk factors. Decision trees are used to identify and present risk events and their consequences, while Bayesian networks are used to calculate the probabilities of these risk events occurring. For example, a decision tree can show the risks of delays and cost overruns caused by choosing a certain supplier, and a Bayesian network can evaluate the likelihood of these risks occurring under the current market conditions. In addition, combined with the recommended opinions of the expert system, the system can provide more comprehensive risk management suggestions for decision-makers based on rich expert knowledge and historical data support. Finally, the system generates a decision-making plan that details the identified key risk points and corresponding mitigation measures. This plan is crucial for enhancing the risk awareness of project management and preventing potential problems.

[0128] For example, in the civil engineering and construction project cost assessment and optimization system, in the strategy plan sub-module, data on the project's historical successful cases are received, including various decision options and the project success rate. Through logistic regression analysis, the system calculates the project success probabilities based on different decision options and generates a strategic priority analysis result that clearly indicates the decision options leading to project success. In the budget adjustment sub-module, based on these analysis results, the system uses linear programming to optimize the budget allocation and generates a detailed budget optimization plan that guides how to most reasonably allocate the budget under the constraints of the cost ceiling and resource availability. This plan provides the project with the most effective financial management method and ensures the optimal utilization of resources. In the risk management sub-module, the system uses the budget optimization plan and expert opinions to deeply analyze the various risks faced by the project. Through decision trees and Bayesian networks, the system identifies the key risk points and evaluates the probabilities of occurrence. For example, the decision tree reveals the risk of delays caused by choosing a certain material, and the Bayesian network evaluates the likelihood of this risk occurring under the current market conditions. Combining the suggestions of experts, the system formulates a series of risk mitigation measures and generates a comprehensive decision-making plan. This plan provides important risk management guidance for project management, helps decision-makers prevent potential problems, and ensures the smooth progress of the project.

[0129] Please refer to Figure 9 , the financial simulation module includes a financial modeling sub-module, a market analysis sub-module, and a scenario planning sub-module;

[0130] The financial modeling sub-module constructs a financial model based on the decision-making plan, using cash flow analysis to estimate various types of income and expenditure flows, including projected sales revenue, direct costs, and indirect expenses. At the same time, with reference to capital expenditure and operating expenditure, it evaluates the financial health of the project and generates preliminary financial simulation results;

[0131] The market analysis sub-module, based on the preliminary financial simulation results, applies trend analysis and refers to market dynamics, including the impact of price fluctuations and supply-demand relationships on the project's finances. By adjusting the model parameters to reflect market changes, including modifying price assumptions and cost forecasts, it optimizes the financial forecasting model and generates a market-adjusted financial forecast;

[0132] The scenario planning sub-module, based on the market-adjusted financial forecast, constructs differentiated financial scenarios using probability theory methods, analyzes financial risks and opportunities under different market and operating conditions, evaluates the financial feasibility of multiple scenarios, and generates a financial forecast analysis.

[0133] In the financial modeling sub-module, the system constructs a financial model using the cash flow analysis method. During this process, the data format used is usually in tabular form, containing various financial indicators of the project such as projected sales revenue, direct costs, and indirect expenses. For each financial indicator, detailed data entry and processing are carried out using Excel or other financial modeling software. Taking projected sales revenue as an example, first, the expected sales volume and unit price are input, and the software generates a forecast value of the total sales revenue through simple multiplication. Subsequently, for direct costs and indirect expenses, the estimated values of each cost item are input, and the total cost forecast value is obtained by summing them up. In addition, capital expenditure and operating expenditure also need to be considered, which are usually automatically calculated through preset formulas or model parameters. For example, the calculation of depreciation expenses can use the straight-line depreciation method or the double-declining balance method, etc. In this series of operations, the financial model can accurately reflect the financial health of the project, and the generated preliminary financial simulation results are a series of financial statements, such as the income statement, cash flow statement, and balance sheet, which provide important bases for subsequent decision-making.

[0134] In the market analysis sub-module, based on the preliminary financial simulation results, by considering trend analysis and market dynamics, the financial model parameters are adjusted. During this process, the market analysis software applies various algorithms, such as linear regression analysis, time series analysis, etc., to predict the impact of price fluctuations and supply-demand relationships on the project's finances. For example, when considering price fluctuations, the model uses historical price data to predict future price trends, which usually involves methods such as moving average method or exponential smoothing method in time series analysis. According to these analyses, the price assumptions and cost forecasts in the model will be adjusted accordingly. These adjustments are automatically carried out through the model's algorithms to ensure the accuracy of financial forecasts and adaptation to market changes. After such adjustments, the market-adjusted financial forecasts generated by the model are financial reports that are more in line with the current market situation, such as the adjusted income statement and cash flow statement. These reports are of great significance for the enterprise's strategy adjustment and risk management.

[0135] In the scenario planning sub-module, based on the market-adjusted financial forecasts, multiple financial scenarios are constructed using probability theory methods. In this sub-module, the data format is usually multi-dimensional, including market data, operation data, and financial data. Using probability theory and statistical methods, such as Monte Carlo simulation, the model can simulate the financial results under different market and operation conditions. When performing Monte Carlo simulation, first, the probability distribution of each variable is defined, such as normal distribution or uniform distribution. Then, the model generates a large number of scenarios through random sampling and calculates the financial indicators under each scenario, such as net profit, cash flow, etc. These calculations are usually automatically completed by specialized software and involve a large number of iterative operations. Finally, these different scenarios will be evaluated to determine the financial risks and opportunities under various scenarios. The results of scenario planning are usually reflected in a series of reports, showing the distribution of financial indicators under different scenarios, such as risk scatter plots or probability density plots. These reports provide comprehensive risk assessment and decision support for decision-makers, enabling them to make more informed decisions with a full understanding of potential risks.

[0136] Next, taking the civil engineering construction cost evaluation and optimization system as an example, assume that the system receives a new construction project, which includes detailed data items such as building area, material cost, labor cost, etc., and the simulated values of these data items. In the financial modeling sub-module, the system first calculates the total building cost based on the input data items, such as the building cost per square meter and the estimated building area. Subsequently, in the market analysis sub-module, the system adjusts the original cost forecast according to the current market price fluctuations and supply-demand relationships to reflect the real-time market situation. Finally, in the scenario planning sub-module, the system analyzes the cost fluctuations under different market and operation conditions through Monte Carlo simulation and generates a series of financial forecasts under different scenarios. These forecasts help decision-makers evaluate the risks and returns of the project under various market conditions, thus making more reasonable and well-founded decisions.

[0137] Please refer to Figure 10 , the reinforcement learning cost adjustment module includes a policy learning sub-module, a data feedback sub-module, and a continuous optimization sub-module;

[0138] Based on financial forecast analysis, the policy learning sub-module uses the Q-learning algorithm to guide the preliminary dynamic adjustment of cost estimation by establishing and updating a state-action value table. It is gradually improved through a cyclic trial-and-error process, and policy evaluation and correction are carried out to generate a preliminary cost adjustment policy;

[0139] Based on the preliminary cost adjustment policy, the data feedback sub-module uses the Deep Deterministic Policy Gradient algorithm to learn the optimal policy by constructing a deep neural network. At the same time, it refines the policy by combining historical data, optimizes the action output in the decision-making process, and performs iterative adjustment of the policy to generate an optimized cost adjustment policy;

[0140] Based on the optimized cost adjustment policy, the continuous optimization sub-module uses the simulated annealing algorithm to control the randomness in the search process by simulating the gradual decrease of temperature in the physical annealing process, captures the optimal solution of cost estimation in the global search space, performs adaptive optimization of the policy, and generates a dynamic cost adjustment policy.

[0141] In the policy learning sub-module, the dynamic adjustment of cost estimation is carried out through the Q-learning algorithm. Q-learning is a model-free reinforcement learning algorithm mainly used to estimate the expected utility of taking a certain action in a given state. The data format used in this sub-module is mainly state-action pairs and the associated reward values. The state is usually a vector composed of various indicators in financial forecast analysis (such as cost, revenue, etc.), and the action is a decision option that affects the cost. In the Q-learning process, first, a state-action value table is initialized, which assigns an initial estimated value to each pair of state-action. Then, the algorithm updates this table through continuous trial-and-error. In each iteration, the algorithm selects an action to execute and observes the result and the obtained reward. Based on this information, the algorithm updates the value of the corresponding item in the state-action value table. The update formula includes a learning rate and a discount factor, which control the learning rate and the degree of emphasis on future rewards. With continuous iteration, the Q-learning algorithm gradually finds the optimal policy, that is, the best action to take in each state. The preliminary cost adjustment policy generated by this process is a series of decision rules that guide enterprises on how to adjust the cost policy according to the current financial situation to maximize the expected utility.

[0142] In the data feedback sub-module, the Deep Deterministic Policy Gradient algorithm (DDPG) is adopted to further optimize the cost adjustment strategy. DDPG is an algorithm that combines deep learning and reinforcement learning, and learns the optimal strategy by constructing a deep neural network. In this sub-module, the input data includes historical cost data and the output of the preliminary cost adjustment strategy. The DDPG algorithm first approximates a policy function through a deep neural network, which outputs an action based on the current state. At the same time, there is another neural network used to estimate the expected return generated by this policy. During the training process, the algorithm continuously collects state-action-return samples from historical data and uses this data to update the neural network. The deep neural network can capture complex non-linear relationships, making policy learning more accurate and efficient. The key of the DDPG algorithm lies in using the policy gradient to update the policy network to ensure that the selected action can maximize the expected return. By repeatedly iterating this process, the algorithm can optimize the action output in the decision-making process and continuously adjust and improve the strategy. The generated optimized cost adjustment strategy is a finely tuned decision-making model that can provide more accurate and efficient cost adjustment suggestions based on the current financial situation and historical data.

[0143] In the continuous optimization sub-module, the simulated annealing algorithm is adopted to achieve the adaptive optimization of the strategy. The simulated annealing algorithm is a heuristic search algorithm inspired by the annealing process in physics. In this process, the algorithm controls the randomness in the search process by simulating the gradual decrease of temperature in the physical annealing process. Initially, the algorithm allows random search within a large search space, which helps to jump out of local optimal solutions and find the global optimal solution. As the temperature decreases, the search gradually becomes more focused on the area near the current solution. In this sub-module, the input data is the output result of the optimized cost adjustment strategy, as well as various cost-related parameters and constraints. The simulated annealing algorithm selects a candidate solution at each step and calculates its cost. If the cost of the new solution is lower than the current solution, or even if the cost is higher but is still accepted according to the probability at the current temperature, the algorithm updates the current solution. The temperature parameter plays a key role here, determining the probability of the algorithm accepting a worse solution. In this way, the algorithm can effectively explore the global search space and finally find the optimal solution for cost estimation. The generated dynamic cost adjustment strategy is a continuously optimized decision-making model that can automatically adjust the cost strategy according to changes in market and financial conditions to ensure the financial efficiency and success of the project.

[0144] Taking the civil engineering construction project cost evaluation and optimization system as an example, assume that the system receives detailed data items of a construction project, including material costs, labor costs, building area, etc., as well as the simulated values of these data items. In the policy learning sub-module, the Q-learning algorithm makes a preliminary cost estimate based on these data items and gradually optimizes the decision rule through a trial-and-error process. Subsequently, in the data feedback sub-module, the DDPG algorithm further refines and optimizes the cost adjustment policy through a deep neural network based on historical cost data and the output results of Q-learning. Finally, in the continuous optimization sub-module, the simulated annealing algorithm adaptively adjusts the policy according to changes in market and financial conditions to capture the optimal solution for cost estimation.

[0145] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still belong to the protection scope of the technical solution of the present invention.

Claims

1. Civil engineering construction cost evaluation and optimization system, characterized in that: The system includes a feature analysis module, a cost estimation module, a real-time monitoring module, a model training and tuning module, a risk analysis module, a decision support module, a financial simulation module, and a reinforcement learning cost adjustment module; The feature analysis module uses a convolutional neural network algorithm to extract features based on building information model data, combines principal component analysis to screen key features, standardizes feature data through data normalization, and generates feature analysis data; The cost estimation module uses a multi-layer perceptron neural network to analyze cost patterns based on feature analysis data, calculates project costs in combination with a linear regression model, and uses variance analysis to itemize the budget, thereby generating an estimated cost value; The real-time monitoring module uses digital twin technology to build a virtual model of the project based on the estimated cost value, combines time series analysis to monitor project progress, applies dynamic programming algorithms to track cost changes, and generates real-time monitoring data; The model training and tuning module uses cross-validation technology to evaluate the model based on real-time monitoring data, uses genetic algorithms to optimize model parameters, uses machine learning algorithms to perform model self-learning and optimization, and generates an optimized model; The risk analysis module is based on an optimization model, uses the Monte Carlo simulation method to analyze cost risks, combines the decision tree model to evaluate the impact of potential risks, uses the sensitivity analysis method to explore the contribution of multiple factors to project risks, and generates a risk analysis overview; The decision support module combines the risk analysis overview and real-time monitoring data, uses logistic regression analysis and linear programming to optimize decisions, uses expert systems to provide project decision support, and generates decision plans; The financial simulation module uses financial modeling methods to simulate differentiated financial scenarios based on decision-making plans, combines market trend analysis methods to conduct comprehensive financial forecasts, applies probability theory to analyze the feasibility of simulation results, and generates financial forecast analysis; The reinforcement learning cost adjustment module is based on financial forecast analysis, uses Q learning and deep deterministic policy gradient algorithm to dynamically adjust cost estimates, applies simulated annealing algorithm to optimize the learning process, performs adaptive adjustment of cost estimates, and generates dynamic cost adjustment strategies.

2. The civil engineering construction cost evaluation and optimization system according to claim 1, wherein: The feature analysis data includes the building's geometric shape, material type, and design parameters; the estimated cost value includes total cost estimation and itemized cost estimation; the real-time monitoring data includes project progress tracking, cost expenditure monitoring, and resource usage; the optimization model includes adjusted network parameters, prediction accuracy, and data processing flow; the risk analysis overview includes cost overrun risk, schedule delay risk, and resource shortage risk; the decision-making plan includes budget adjustment plan, risk management strategy, and schedule optimization plan; the financial forecast analysis includes revenue forecast scenarios, cost estimation scenarios, and market change forecasts; the dynamic cost adjustment strategy includes adjusted cost budget, market adaptability strategy, and resource allocation optimization.

3. The optimized system for civil engineering construction project cost evaluation according to claim 1, characterized in that: The feature analysis module includes a data mining submodule, a structure analysis submodule, and a design evaluation submodule; The data mining sub-module is based on building information model data and adopts a deep learning feature extraction algorithm. During the process, through a convolutional neural network, stacked convolutional layers and pooling layers are used to automatically learn spatial hierarchical features. Through the unsupervised learning method of an autoencoder, the input data is reconstructed to mine the structure and patterns in the data, perform in-depth feature mining, and generate feature extraction data. The structure analysis sub-module is based on the feature extraction data and adopts a key feature recognition method. Through linear transformation, the original features are transformed into a set of linearly uncorrelated principal components to reduce the data dimension. By maximizing the between-class difference and minimizing the within-class difference, features are extracted to generate key feature data. The design evaluation sub-module is based on the key feature data and adopts data preprocessing and standardization methods. By normalizing the data, the range of feature values is adjusted to eliminate the influence between differential magnitude features, and errors and incomplete data in the dataset are removed or corrected for comprehensive design evaluation to generate feature analysis data.

4. The civil engineering construction cost evaluation and optimization system according to claim 1, characterized in that: The cost estimation module includes a cost modeling sub-module, a budget prediction sub-module, and a breakdown analysis sub-module. The cost modeling sub-module is based on the feature analysis data and adopts a multi-layer perceptron neural network algorithm to construct a cost model, including setting the number of network layers, the number of neurons, and the selection of activation functions to capture the non-linear relationship of cost data. Then, principal component analysis is applied to reduce the dimension of the extracted features to generate the cost model analysis result. The budget prediction sub-module is based on the cost model analysis result and uses a linear regression model to predict the total cost, including establishing a linear relationship between variables, calculating regression coefficients, and performing hypothesis testing. At the same time, combined with time series analysis, the time series characteristics of historical data are analyzed through an autoregressive model and a moving average model to provide trend and periodic information for cost prediction and generate the total project cost prediction. The breakdown analysis sub-module is based on the total project cost prediction and uses variance analysis to break down the budget item by item, including comparing the mean differences between multiple breakdown costs, identifying the impact degree of multiple breakdown costs on the total cost, and allocating costs accordingly for budget analysis to generate the estimated cost value.

5. The civil engineering construction cost evaluation and optimization system according to claim 1, characterized in that: The real-time monitoring module includes a digital twin modeling sub-module, a data collection sub-module, and a progress tracking sub-module. The digital twin modeling sub-module is based on the estimated cost value and uses computer-aided design to perform 3D modeling of the project, including using geometric modeling tools to draw the 3D structure of the project and integrating finite element analysis to simulate the response of the project under various environmental conditions to generate a project virtual model. The data collection sub-module is based on the project virtual model and uses time series analysis technology to collect project progress data. By deploying multiple sensors to collect real-time data and using data collection software to integrate and analyze time series data, the project progress and key indicators are monitored to generate project progress monitoring data. The progress tracking sub-module is based on the project progress monitoring data and applies a dynamic programming algorithm to track cost changes, including establishing a relationship model between cost and progress, optimizing resource allocation and cost control through continuous decision points, performing project cost management, and generating real-time monitoring data.

6. The civil engineering construction cost evaluation and optimization system according to claim 1, wherein: The model training and tuning module includes a training strategy sub-module, a validation and testing sub-module, and a parameter adjustment sub-module; The training strategy sub-module is based on real-time monitoring data, adopts the backpropagation algorithm, adjusts the weights by calculating the output error and propagating it back into the network, conducts preliminary training of the neural network, and at the same time combines the gradient descent method to adjust the weights and biases of each neuron, reduces the prediction error, gradually improves the fitting degree of the model to the cost data, and generates a preliminary prediction model; The validation and testing sub-module is based on the preliminary prediction model, applies K-fold cross-validation, divides the data set into K mutually exclusive subsets, uses one subset as the test set each time, and the rest as the training set, repeats model training and testing, evaluates the performance of the model on different data subsets, and generates validation and testing results; The parameter adjustment sub-module is based on the validation and testing results, adopts the genetic algorithm to optimize the model parameters, optimizes the structure and parameter configuration of the neural network by simulating the selection, crossover, and mutation mechanisms in biological evolution, captures the optimal parameter combination, and generates an optimized model.

7. The civil engineering construction cost evaluation and optimization system according to claim 1, characterized in that: The risk analysis module includes a scenario simulation sub-module, a probability analysis sub-module, and an impact assessment sub-module; The scenario simulation sub-module is based on the optimized model, adopts the Monte Carlo simulation method, reflects the impact of different risk factors on the cost by generating a batch of random variables, including setting parameter ranges, generating random data points, simulating different risk scenarios, analyzing the distribution characteristics of their impact on the cost, and generating a cost risk scenario analysis; The probability analysis sub-module is based on the cost risk scenario analysis, adopts the decision tree model, constructs a risk decision tree where the nodes represent risk events and the edges represent the likelihood of events occurring, evaluates the risk impact under different decision paths by calculating the conditional probability and cumulative probability of each node, and generates a risk probability and impact assessment; The impact assessment sub-module is based on the risk probability and impact assessment, adopts the sensitivity analysis method, observes the impact on the final risk assessment by changing the values of key variables, including selecting key variables, adjusting the values one by one, recording the result changes, analyzing the degree of influence of key factors on the risk assessment, and generating a risk analysis overview.

8. The civil engineering construction cost evaluation and optimization system according to claim 1, wherein: The decision support module includes a strategy and solution sub-module, a budget adjustment sub-module, and a risk management sub-module; The strategy and solution sub-module is based on the risk analysis overview and real-time monitoring data, analyzes multiple decision options using the logistic regression analysis method, determines the logical relationship between decision options and the project success rate, calculates the success probability of each decision option by encoding data features and assigning weights, and generates a strategy priority analysis result; The budget adjustment sub-module is based on the strategy priority analysis result, reallocates and optimizes the budget using the linear programming method, sets the budget constraints, including the cost ceiling and resource availability, and then uses the linear programming algorithm to capture the optimal budget plan under the conditions, and generates a budget optimization plan; The risk management sub-module comprehensively manages the potential risks of the project based on the budget optimization plan and in combination with the recommendations of the expert system, including using decision trees and Bayesian networks to analyze risk factors, identifying key risk points, and formulating mitigation measures, and supporting the decision-making process with expert knowledge and historical data to generate decision-making plans.

9. The civil engineering construction cost evaluation and optimization system according to claim 1, wherein: The financial simulation module includes a financial modeling sub-module, a market analysis sub-module, and a scenario planning sub-module; The financial modeling sub-module constructs a financial model based on the decision-making plan using cash flow analysis to estimate multiple types of income and expenditure flows, including projected sales revenue, direct costs, and indirect expenses, and at the same time evaluates the financial health of the project with reference to capital expenditures and operating expenditures to generate preliminary financial simulation results; The market analysis sub-module, based on the preliminary financial simulation results, applies trend analysis and refers to market dynamics, including the impact of price fluctuations and supply-demand relationships on the project's finances, and reflects market changes by adjusting model parameters, including modifying price assumptions and cost forecasts, to optimize the financial forecasting model and generate market-adjusted financial forecasts; The scenario planning sub-module constructs differentiated financial scenarios using probability theory methods based on the market-adjusted financial forecasts, analyzes the financial risks and opportunities under differentiated market and operating conditions, evaluates the financial feasibility of multiple scenarios, and generates financial forecasting analyses.

10. The civil engineering construction cost evaluation and optimization system according to claim 1, wherein: The reinforcement learning cost adjustment module includes a policy learning sub-module, a data feedback sub-module, and a continuous optimization sub-module; The policy learning sub-module, based on the financial forecasting analysis, adopts the Q-learning algorithm to guide the preliminary dynamic adjustment of cost estimation by establishing and updating a state-action value table, gradually improves through a cyclic trial-and-error process, and conducts policy evaluation and correction to generate a preliminary cost adjustment policy; The data feedback sub-module, based on the preliminary cost adjustment policy, adopts the deep deterministic policy gradient algorithm to learn the optimal policy by constructing a deep neural network, simultaneously refines the policy with historical data, optimizes the action output in the decision-making process, and conducts iterative adjustment of the policy to generate an optimized cost adjustment policy; The continuous optimization sub-module, based on the optimized cost adjustment policy, adopts the simulated annealing algorithm to control the randomness in the search process by simulating the gradual decrease in temperature during the physical annealing process, captures the optimal solution of cost estimation in the global search space, conducts adaptive optimization of the policy, and generates a dynamic cost adjustment policy.

Images