Engineering cost index dynamic analysis and intelligent prediction system and method

Abstract

The invention discloses a project cost index dynamic analysis and intelligent prediction system and method. The system comprises a multi-dimensional data acquisition module, a data processing module, a dynamic analysis module and an intelligent prediction module. According to the method, the multi-source heterogeneous data is collected, the multi-dimensional feature vectors are extracted through preprocessing, the dynamic analysis model fusing the chaos theory and the game theory is constructed, the reinforcement learning algorithm and the risk adjustment reward function are combined in the extended state space to achieve prediction, and the generalization ability is improved through transfer learning. According to the invention, the problems of insufficient dynamic analysis, low prediction accuracy and weak data processing capability in the prior art are solved, dynamic analysis and accurate prediction can be carried out on the project cost index, comprehensive, timely and accurate decision support is provided for project cost management, and the adaptability of the system to a complex market environment is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of engineering cost management, and in particular to a system and method for dynamic analysis and intelligent prediction of engineering cost indicators. Background Art

[0002] In the current field of construction cost management, traditional methods for analyzing and predicting construction cost indicators primarily rely on manual experience and simple statistical analysis. Although some systems have introduced certain data analysis models, these methods and systems still suffer from numerous problems. First, existing analysis methods often rely on simple summaries based on historical static data. This makes it difficult to adapt to the impact of dynamic factors such as material price fluctuations, policy and regulatory changes, and technological innovation on construction costs in the construction market, and thus cannot achieve dynamic, real-time analysis of construction cost indicators. Second, existing prediction models mostly use conventional algorithms such as linear regression and time series analysis, which lack the ability to effectively handle the nonlinear relationships between complex influencing factors, resulting in low accuracy and reliability of prediction results. Third, existing systems lack the ability to integrate and effectively utilize multi-source heterogeneous data, and struggle to deeply mine and extract value from data, making it impossible to provide comprehensive, accurate, and timely decision support for construction cost management. Summary of the Invention

[0003] The purpose of the present invention is to provide a system and method for dynamic analysis and intelligent prediction of engineering cost indicators to solve the problems in the prior art of lack of dynamic analysis of engineering cost indicators, low prediction accuracy and insufficient data processing capabilities.

[0004] To achieve the above objectives, this application provides the following solutions: In a first aspect, the present application provides a system for dynamic analysis and intelligent prediction of construction cost indicators, including: A multi-dimensional data acquisition module configured to acquire multi-source data related to project cost indicators, wherein the multi-source data includes market transaction data, policy and regulation data, and project environment data; a data processing module, connected to the multidimensional data acquisition module, configured to preprocess the acquired data and extract multidimensional feature vectors; A dynamic analysis module, connected to the data processing module, configured to construct a dynamic analysis model integrating chaos theory and game theory, wherein the dynamic analysis model is used to simulate the dynamic changes of the construction cost system and realize multi-dimensional dynamic analysis of construction cost indicators; An intelligent prediction module is connected to the data processing module and the dynamic analysis module, and is configured to apply a reinforcement learning algorithm in an extended state space to realize the prediction of the project cost index.

[0005] Optionally, the data processing module is further configured to: a data cleaning unit configured to apply a time-delay Kalman filter algorithm to process missing data values; a feature extraction unit, connected to the data cleaning unit, configured to construct a multidimensional feature vector comprising time domain features, frequency domain features, and correlation features, wherein the time domain features reflect the changing trend of the data over time, the frequency domain features reflect the periodicity of the data, and the correlation features quantify the relationship between different factors; The feature optimization unit is connected to the feature extraction unit and is configured to perform multi-scale feature decomposition through wavelet transform to achieve separation and reconstruction of different frequency components of the data.

[0006] Optionally, the dynamic analysis module is further configured to: a chaos analysis unit configured to apply an improved time-delay Lorenz equation to construct a chaos sub-model, wherein the chaos sub-model simulates the dynamic evolution of the construction cost system by introducing a multivariable time-delay term and a characteristic influence coefficient; a game analysis unit configured to construct a game sub-model involving incomplete information of four parties, namely, the owner, the contractor, the supplier, and the regulatory authority, wherein the game sub-model analyzes the impact of the strategic interaction of each party on the construction cost by solving the Bayesian Nash equilibrium; The model fusion unit is configured to automatically adjust the weights of the chaos sub-model and the game sub-model based on an exponential function of the model fitness score to achieve dynamic fusion.

[0007] Optionally, the chaotic sub-model adopts an improved time-delay Lorenz equation:

[0008] in: 、 、 They are the chaotic system state variables that reflect the cost fluctuation dimension, market supply and demand change dimension, and policy impact dimension indicators in the engineering cost system; 、 、 System parameters calibrated based on historical data are used to adjust system dynamic characteristics; 、 、 is the time lag coefficient, which reflects the lag of the influence of factors in the system; 、 、 Indicates time lag The rate of change of state; 、 、 Influencing factors The weight coefficient is calculated by the entropy weight method, which reflects the relative importance of each factor to the change of system state.

[0009] Optionally, the profit function expression of the game sub-model is:

[0010] in: For participants In its own strategy and other participant strategy combinations The following income; For participants Adopt a strategy Basic income; Cost of implementing the strategy; is the competition coefficient, which reflects the degree of competition among participants; For participants The covariance between the strategy of and the strategies of other participants, measuring the mutual influence between strategies; is the risk aversion coefficient, which reflects the participant’s aversion to risk; For strategy Risk value is used to evaluate the risk level during strategy implementation.

[0011] Optionally, the model fusion unit integrates the outputs of the chaos sub-model and the game sub-model through a dynamic weight function, which satisfies the following formula:

[0012] in: is the analysis result after fusion; is the chaotic sub-model output, Output of the game sub-model; , , 、 The chaos model and the game model are The fitness score at the moment.

[0013] Optionally, the intelligent prediction module is further configured to: State space construction unit, which constructs an extended state space containing historical data, market data, policy data, environmental data and market sentiment indicators; A reward design unit performs risk value assessment and predictive stability constraint operations on the extended state space, and constructs a reward function to guide the model learning direction; The network training unit uses the reward function as feedback, adopts a dual deep Q network architecture combined with priority experience replay, optimizes network parameters, and realizes project cost prediction; The transfer learning unit, based on the trained network, reduces domain differences through feature mapping, transfers knowledge to new scenarios, and improves the model's generalization ability.

[0014] Optionally, the reward function expression is:

[0015] in: for Actual project cost index value at all times, is the predicted value; is the risk aversion coefficient; The conditional value at risk of the next period's forecast value; To expand the state space; is the stability coefficient; Constrain the range of changes in prediction strategies between adjacent states.

[0016] Optionally, the network training unit operates according to the following steps: Get the extended state space generated by the state space building unit and the reward value calculated by the reward design unit , as basic training data; A dual deep Q-network architecture is used to calculate the Q-value of the current state and / or action pair and the optimal Q-value of the next state respectively; Calculate the TD error according to the network update rule, and update the priority of samples in the experience replay buffer according to the priority calculation formula to give priority to samples with larger errors; According to the training algorithm with policy gradient constraints, the gradient of the main network parameters is calculated based on the TD error, and the main network parameters are updated using the optimization algorithm; Every fixed number of training steps, the main network parameters are copied to the target network parameters to maintain the stability of the target network; Repeat the above parameter updating and optimization steps until the network converges or reaches the preset number of training times, completing the training of the project cost index prediction model.

[0017] Secondly, this application provides a method for dynamic analysis and intelligent prediction of construction cost indicators, including the following steps: Collect market transaction data, policy and regulation data, and project environment data related to engineering cost indicators to construct a multi-source heterogeneous data set; Preprocess the collected data and extract multidimensional feature vectors containing time domain, frequency domain and correlation features; Construct a dynamic analysis model that integrates chaos theory and game theory, simulate the dynamic changes of the construction cost system through the dynamic analysis model, and realize multi-dimensional dynamic analysis of construction cost indicators; In an extended state space that includes historical data, market status, policy environment, and market sentiment, engineering cost indicator prediction is achieved based on a reinforcement learning algorithm combined with a risk-adjusted reward function, and the model generalization ability is improved through transfer learning.

[0018] Through the above technical solutions, the beneficial effects of the present invention are as follows: this application uses a multi-dimensional data acquisition module to capture market transactions, policies and regulations, and project environment data in real time, and combines distributed acquisition clusters with hybrid storage systems to achieve dynamic integration and integrity verification of multi-source heterogeneous data, solving the problem of insufficient data processing capabilities; the data processing module uses a time-delay Kalman filter to repair missing values ​​and a wavelet transform to optimize features, and constructs a multi-dimensional vector containing time domain, frequency domain and correlation features to provide high-quality data support for dynamic analysis; the dynamic analysis module integrates the improved time-delay Lorenz equation and the four-agent game model, and dynamically adjusts the model weights through an exponential function to effectively capture nonlinear factors such as material price fluctuations and policy changes and the impact of strategic interactions among participants; the intelligent prediction module constructs a reinforcement learning model based on the extended state space, combines a dual deep Q network with transfer learning, and achieves accurate prediction of risk perception and cross-scenario knowledge transfer, comprehensively improving prediction accuracy and system generalization capabilities, and effectively solving the core pain points of lack of dynamic analysis and low prediction reliability in the existing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention, and the embodiments in the drawings do not constitute any limitation to the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 A schematic diagram of the structure of a system for dynamic analysis and intelligent prediction of construction cost indicators provided in one embodiment of the present application; Figure 2 A schematic diagram of the structure of a data processing module provided in one embodiment of the present application; Figure 3 A schematic diagram of the structure of a dynamic analysis module provided in one embodiment of the present application; Figure 4 A schematic diagram of the structure of an intelligent prediction module provided in one embodiment of the present application; Figure 5A flowchart of a method for dynamic analysis and intelligent prediction of construction cost indicators provided in one embodiment of the present application; Figure 6 A schematic diagram of the structure of a computer device provided in one embodiment of the present application; The realization of the objectives, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0021] It should be understood that the specific embodiments described herein are merely for explaining the present application and are not intended to limit the present application. Rather, these embodiments are provided to make the present disclosure more thorough and complete and to fully convey the scope of the present disclosure to those skilled in the art.

[0022] The above and other technical contents, features and effects of the present invention are described below with reference to the attached Figure 1-6 The detailed description of the embodiments will clearly show that the structural contents mentioned in the following embodiments are all based on the accompanying drawings.

[0023] Various exemplary embodiments of the present invention will be described below with reference to the accompanying drawings.

[0024] In an exemplary embodiment, Figure 1 As shown, a system for dynamic analysis and intelligent prediction of engineering cost indicators is provided, which includes a multi-dimensional data acquisition module 110, a data processing module 120, a dynamic analysis module 130 and an intelligent prediction module 140.

[0025] The multi-dimensional data collection module 110 is responsible for collecting three types of core data: market transaction data, policy and regulatory data, and project environment data. It ensures the timeliness, accuracy, and completeness of the data through clear data collection sources, standardized collection methods and frequencies, and strict data storage rules.

[0026] In a specific embodiment, market transaction data includes a construction material price index, a labor market wage index, and an equipment rental price index. The construction material price index is data segmented by region, type, and brand, such as real-time price series for various building materials such as steel and cement in different regions and brands. The labor market wage index is hourly or monthly wage data by region and type of work, reflecting the labor costs of different types of work in different regions. The equipment rental price index includes rental cost information for construction machinery such as tower cranes and excavators. For this data, the multi-dimensional data collection module 110 uses the Scrapy framework and web crawler technology to collect the construction material price index every 15 minutes from professional building material price websites such as China Building Materials Network and Centennial Construction Network. It also obtains the labor market wage index through an API interface connected to the official website of the local Human Resources and Social Security Bureau. Furthermore, a hybrid method of web crawlers and API interfaces is used to collect the equipment rental price index from rental platforms such as Tiejia Second-hand and Gongpinhui.

[0027] Policy and regulatory data includes quota standard adjustment coefficients, tax rate change records, and financial credit interest rates. Quota standard adjustment coefficients involve quarterly updates on labor and material consumption standards. These data are extracted by parsing XML / JSON policy documents through API access to the official website of the Ministry of Housing and Urban-Rural Development's Construction Cost Management Agency. Tax rate change records, including adjustments to value-added tax and urban construction and maintenance taxes, are obtained by identifying tax rate adjustment clauses within policy texts on the State Administration of Taxation's official website using a combination of web crawlers and natural language processing. Financial credit interest rates include interest rates on special loans for construction projects and corporate financing costs. Benchmark interest rates and special loan interest rates are collected from the official websites of the People's Bank of China and major banks using a combination of web crawlers and APIs.

[0028] Project environmental data primarily includes the geographic climate risk index and the surrounding infrastructure integrity index. The geographic climate risk index includes indicators that affect construction difficulty, such as earthquake intensity and rainfall frequency. Through API interfaces connected to the official website of the China Meteorological Administration and the data platform of the China Earthquake Administration, historical meteorological and earthquake data are obtained to construct a risk assessment model. The integrity of surrounding infrastructure is measured by invoking the APIs of AutoNavi Maps and Baidu Maps to calculate indicators such as the road network density and the distance to water and power facilities around the project.

[0029] In terms of collection frequency and rule setting, the multi-dimensional data collection module 110 collects real-time market transaction data every 15 minutes, requires the data to be stored in JSON or CSV format, and contains information such as timestamp and data source identifier, and performs integrity verification by checking whether the price is within the historical fluctuation range of ±3σ; daily data such as some policy statistical information is collected at 2 a.m. in a structured table format to check the integrity of data items and their logical consistency with the previous day's data; weekly data such as the project environment weekly report is collected at 8 p.m. on Sunday and records the weekly statistical period, average value, etc. in XML or JSON format to verify its consistency with the monthly data trend; policy change data is collected in a triggered manner to ensure the acquisition of original documents and structured analysis results, and to check the integrity of key information such as policy document number and effective date.

[0030] During data collection and initial storage, the multidimensional data collection module 110 utilizes a master-slave architecture within a distributed collection cluster. The master node is responsible for task scheduling, while the slave nodes perform the specific collection tasks. A resumable data transfer mechanism is also included to prevent data loss. Collected data enters a real-time processing queue or batch processing job, undergoes format conversion, and undergoes integrity verification. Qualified data is then stored by type in a hybrid storage system. Specifically, real-time data is stored in a Redis cache, structured data is stored in a MySQL database, and unstructured data is saved in MongoDB. This provides reliable data support for subsequent data processing and model analysis.

[0031] The data processing module 120 cleans the collected raw data, repairs missing values, processes outliers, and extracts features closely related to model calculations, converting the raw data into effective information suitable for model analysis. Figure 2 As shown, the data processing module 120 specifically includes the following units: a data cleaning unit configured to apply a time-delay Kalman filter algorithm to process missing data values; a feature extraction unit connected to the data cleaning unit and configured to construct a multidimensional feature vector including time domain features, frequency domain features, and correlation features, wherein the time domain features reflect the changing trend of the data over time, the frequency domain features reflect the periodicity of the data, and the correlation features quantify the relationship between different factors; The feature optimization unit is connected to the feature extraction unit and is configured to perform multi-scale feature decomposition through wavelet transform to achieve separation and reconstruction of different frequency components of the data.

[0032] In the above, the data cleaning unit is the first step in data processing, and it uses the time-delay Kalman filter algorithm to deal with the problem of missing data. In the actual collection process, market transaction data, policy and regulatory data, etc. often have missing values ​​due to network delays and abnormal data source updates, which affects the accuracy of subsequent analysis. The time-delay Kalman filter algorithm is based on the state space model, and its core expression is ,in is the lag factor, By incorporating historical state change rates, this algorithm not only uses current observations and forecasts to correct missing data but also captures data trends, making it more effective for data with distinct time series characteristics, such as the building materials price index. Furthermore, the unit combines the Local Outlier Factor (LOF) algorithm with statistical process control (SPC) methods. By calculating the local density ratio of data points and the boundaries of control charts, it identifies and removes noise data, such as unusual price fluctuations and policy coefficient abrupt changes, ensuring the reliability of input data.

[0033] The feature extraction unit takes over the cleaned data and constructs a multi-dimensional feature vector containing time domain, frequency domain and correlation features. In terms of time domain features, for market transaction data, the month-on-month growth rate of building material prices is calculated. , intuitively reflects the changing trend of data over time. These characteristics directly affect the cost fluctuation dimension in the chaotic system. In frequency domain feature extraction, discrete Fourier transform (FFT) is used Convert data, analyze seasonal fluctuations in material prices, cyclical patterns in policy adjustments, etc., to reflect changes in market supply and demand. Provide data support. At the correlation feature level, by calculating the Pearson correlation coefficient , quantify the relationship between tax rate adjustments and project cost fluctuations in policy and regulatory data, and the relationship between geographical climate risk index and construction costs in project environmental data. These coefficients serve as key parameters in the policy impact dimension. 's construction.

[0034] The feature optimization unit realizes in-depth processing of data features through wavelet transform, and its formula is:

[0035] in, is the approximate coefficient, The data is separated into high-frequency and low-frequency components through multi-scale decomposition. For example, it can distinguish high-frequency policy adjustments from low-frequency policy trend changes in policy and regulatory data, and separate short-term price fluctuations from long-term price trends in market transaction data. This allows the chaos model to more accurately capture the impact of different frequency components on the construction cost system.

[0036] The dynamic analysis module 130, as the core analysis engine of the dynamic analysis and intelligent prediction system for engineering cost indicators, builds a dynamic analysis model that integrates chaos theory and game theory to deeply analyze the complex dynamic changes of the engineering cost system and the strategic interaction mechanism of each participant, providing a scientific decision-making basis for engineering cost management. Figure 3 As shown, the module is further configured as follows: The chaos analysis unit 131 is configured to apply the improved time-delay Lorenz equation to construct a chaos sub-model, wherein the chaos sub-model simulates the dynamic evolution of the construction cost system by introducing a multivariable time-delay term and a characteristic influence coefficient; The game analysis unit 132 is configured to construct a game sub-model with incomplete information involving four parties: the owner, the contractor, the supplier, and the regulatory authority. The game sub-model analyzes the impact of the strategic interaction of each party on the construction cost by solving the Bayesian Nash equilibrium. The model fusion unit 133 is configured to automatically adjust the weights of the chaos sub-model and the game sub-model based on an exponential function of the model fitness score to achieve dynamic fusion.

[0037] During implementation, Chaos Analysis Unit 131 uses an improved time-delay Lorenz equation as its core to construct a chaotic sub-model capable of accurately characterizing the dynamic evolution of the construction cost system. The traditional Lorenz equation struggles to account for the hysteresis and multi-factor coupling of factors when dealing with complex systems. This unit optimizes this by introducing multivariable time-delay terms and characteristic influence coefficients. The improved equation is expressed as:

[0038] in: 、 、 They are the chaotic system state variables that reflect the cost fluctuation dimension, market supply and demand change dimension, and policy impact dimension indicators in the engineering cost system; 、 、 System parameters calibrated based on historical data are used to adjust system dynamic characteristics; 、 、 is the time lag coefficient, which reflects the lag of the influence of factors in the system; 、 、 Indicates time lag The rate of change of state; 、 、 Influencing factors The weight coefficient is calculated by the entropy weight method, which reflects the relative importance of each factor to the change of system state.

[0039] The game analysis unit 132 constructs a four-party incomplete information game sub-model from a micro perspective, including the owner, contractor, supplier and regulatory department. In the actual construction cost scenario, the information held by each participant is asymmetric and the decisions affect each other. This unit constructs a profit function To analyze the impact of the strategic interaction of each participant on the cost. For participants In its own strategy and other participant strategy combinations The following income; For participants Adopt a strategy Basic income; Cost of implementing the strategy; is the competition coefficient, which reflects the degree of competition among participants; For participants The covariance between the strategy of and the strategies of other participants, measuring the mutual influence between strategies; is the risk aversion coefficient, which reflects the participant’s aversion to risk; For strategy The unit assesses the risk value of a strategy and the magnitude of the risk during its implementation. By solving the Bayesian Nash equilibrium, it can determine the optimal strategy for each participant under different information conditions and reveal the mechanisms by which actions such as contractors colluding with suppliers to raise prices and owners adjusting their bidding strategies due to policy changes affect project costs.

[0040] The model fusion unit 133 realizes the dynamic weight adjustment and organic fusion of the chaos sub-model and the game sub-model based on the exponential function of the model fitness score. In different market environments and project stages, the influence of chaos factors (such as market fluctuations) and game factors (such as participant strategies) on project costs varies. This unit uses the formula To integrate, is the analysis result after fusion; is the chaotic sub-model output, Output of the game sub-model.

[0041] Further, , , 、 The chaos model and the game model are The fitness score of the chaos sub-model is calculated by comparing the degree of fit between the model output and the historical actual data, the prediction accuracy and other indicators. When the market fluctuates violently, the fitness score of the chaos sub-model increases, and its weight Increase, the system focuses more on analyzing the impact of market dynamics on construction costs; in the stage of frequent policy adjustments, the weight of the game sub-model It increases to highlight the impact of the strategic game of each participant based on policy changes on the construction cost, thereby achieving a multi-dimensional, dynamic and accurate analysis of engineering cost indicators.

[0042] As the core decision-making layer of the dynamic analysis and intelligent prediction system for engineering cost indicators, the intelligent prediction module 140 integrates the multi-dimensional feature vectors output by the data processing module 120 and the system evolution law revealed by the dynamic analysis module 130, and constructs a reinforcement learning prediction model in the extended state space to achieve accurate prediction and knowledge transfer of engineering cost indicators. Figure 4 As shown, the module is further configured as follows: A state space construction unit 141 constructs an extended state space including historical data, market data, policy data, environmental data, and market sentiment indicators; The reward design unit 142 performs risk value assessment and predictive stability constraint operations on the extended state space, and constructs a reward function to guide the model learning direction; The network training unit 143 uses a reward function as feedback, adopts a dual deep Q network architecture combined with priority experience replay, optimizes network parameters, and realizes project cost prediction; The transfer learning unit 144 reduces domain differences through feature mapping based on the trained network, transfers knowledge to new scenarios, and improves the generalization ability of the model.

[0043] In a specific embodiment, the state space construction unit 141 constructs a five-dimensional extended state space containing historical data, market data, policy data, environmental data and market sentiment indicators based on the cleaned and feature-optimized data output by the data processing module 120. Among them, the historical data feature vector Contains historical engineering cost data processed by time-delay Kalman filtering, and the time domain and frequency domain features extracted by wavelet transform to reflect the historical evolution of project costs; market data feature vector It integrates real-time market transaction data such as the building materials price index and the labor market wage index, and quantifies their correlation with cost fluctuations through correlation analysis; policy data feature vector It integrates policy and regulatory data such as quota standard adjustment coefficients and tax rate change records, and reflects the impact of policies on construction costs through policy sensitivity coefficients; environmental data feature vectors Integrate project environmental data such as geographic climate risk index and surrounding infrastructure completeness, and convert them into environmental risk impact factors through risk assessment models; market sentiment indicators Using natural language processing technology, sentiment analysis of industry news text is performed to extract market participants' expectations and sentiment towards the industry, capturing the impact of irrational market factors on construction costs. This expanded state space comprehensively covers internal and external factors influencing project costs, providing multi-dimensional environmental awareness for reinforcement learning.

[0044] The reward design unit 142 performs risk value assessment and prediction stability constraint operations on the extended state space, and constructs a reward function with double regularization:

[0045] in, for Actual project cost index value at all times, is the predicted value. Specifically, the first is the prediction error penalty term, which amplifies the penalty for larger errors through an exponential function to encourage the model to improve prediction accuracy; the second term This is a value-at-risk assessment item that calculates the conditional value-at-risk of the next period's forecast value based on the price fluctuation probability distribution output by the chaos sub-model of the dynamic analysis module 130, and measures the potential loss of the forecast result at a certain confidence level; The risk aversion coefficient is dynamically adjusted through the feature importance evaluation algorithm to guide the model to learn risk aversion strategies; the third is the prediction stability constraint term, , by constraining the prediction strategy change range between adjacent states through Kullback-Leibler divergence, The reward function is a stabilization coefficient that prevents the model from overfitting to short-term fluctuations and improves long-term prediction stability. This reward function provides a clear optimization direction for the reinforcement learning algorithm by quantifying prediction error, risk level, and strategy stability.

[0046] The network training unit 143 uses the reward function as a feedback signal and adopts a dual deep Q network (DDQN) architecture combined with a priority experience replay mechanism to optimize network parameters to achieve accurate prediction. Specifically, the network training unit operates as follows: Get the extended state space generated by the state space building unit and the reward value calculated by the reward design unit , as basic training data; A dual deep Q-network architecture is used to calculate the Q-value of the current state and / or action pair and the optimal Q-value of the next state respectively; Calculate the TD error according to the network update rule, and update the priority of samples in the experience replay buffer according to the priority calculation formula to give priority to samples with larger errors; According to the training algorithm with policy gradient constraints, the gradient of the main network parameters is calculated based on the TD error, and the main network parameters are updated using the optimization algorithm; Every fixed number of training steps, the main network parameters are copied to the target network parameters to maintain the stability of the target network; Repeat the above parameter updating and optimization steps until the network converges or reaches the preset number of training times, completing the training of the project cost index prediction model.

[0047] In the specific operation of the dual deep Q network architecture, this unit alleviates the overestimation problem in traditional Q learning by separating the action selection and action evaluation mechanisms. Specifically, the main network Responsible for calculating the Q value of the current state-action pair, while the target network It is used to evaluate the optimal Q value of the next state, and its expression is This separation design effectively reduces the positive bias caused by using the same network for action selection and evaluation, and improves the accuracy of value function estimation.

[0048] The introduction of the priority experience replay mechanism further improves the training efficiency. The network training unit 143 firstly Calculate the priority of each experience sample ,in To prevent small constants with priority zero, This is a hyperparameter that controls the degree of influence of priority. In this way, samples in the experience replay buffer are given a sampling probability proportional to their importance, allowing the model to prioritize learning samples with larger errors and more valuable information, accelerating the convergence process.

[0049] The parameter update phase uses a training algorithm with policy gradient constraints, which introduces the Nash equilibrium policy constraints of the game sub-model based on the traditional DDQN update rules. Specifically, the gradient calculation expression is ,in is the Nash equilibrium strategy of the game sub-model, is the Kullback-Leibler divergence, This constraint mechanism ensures that the model not only minimizes the prediction error during learning but also considers the impact of the strategic interactions of all participants on the cost, thereby improving the real-world adaptability of the prediction.

[0050] In order to maintain training stability, this unit adopts a soft update mechanism, every fixed training step , the main network parameters Copy to target network parameters ,Right now This periodic synchronization mechanism keeps the target network unchanged for a period of time, avoiding training oscillations caused by frequent changes in target values ​​and helping the model converge smoothly.

[0051] By continuously repeating the parameter update and optimization steps described above, the network gradually learns the optimal action strategy under different expansion states—that is, how to accurately predict construction cost indicators based on multidimensional features. The training process continues until convergence conditions are met or the preset number of training cycles is reached, ultimately forming a construction cost prediction model that comprehensively considers multidimensional factors such as historical data, market dynamics, policy changes, environmental risks, and market sentiment. This training architecture, combining dual deep Q networks with prioritized experience replay, not only significantly improves prediction accuracy and model stability, but also integrates game theory analysis results into the reinforcement learning process through a policy gradient constraint mechanism, making the predictions more closely aligned with the dynamic changes in construction costs in real market environments.

[0052] The transfer learning unit is based on the network model trained by 144. It effectively reduces the difference between the source domain and the target domain through an innovative feature mapping mechanism, thereby efficiently transferring the knowledge learned in the existing scenario to the new scenario.

[0053] In the specific implementation process, the unit first designs a feature mapping function Establish the connection between the source domain and the target domain. The specific expression is:

[0054] in, and As the mapping parameter, by minimizing the domain difference loss Optimize learning, and Represent the mean of the feature vectors of the source domain and the target domain respectively; and This optimization process aims to align the data distribution of the source domain with that of the target domain as closely as possible. For example, the material price fluctuation and construction cost correlation features learned in residential projects can be adapted to the data feature system of commercial construction projects through mapping and adjustment.

[0055] In the actual migration process, the migration learning unit 144 is based on the time-delay correlation matrix and data quality assessment matrix Dynamically adjust migration weights , building a cross-domain knowledge transfer framework .in, To predict task loss, ensure that the model can still accurately predict the construction cost indicators in the target area; Used to measure the similarity of the time-lag correlation matrix between the source domain and the target domain, Dynamically adjust according to the data quality assessment results. When the quality of the target domain data is high, appropriately reduce the migration weight and rely more on the target domain data for model adjustment; otherwise, increase the migration weight and make full use of the source domain knowledge.

[0056] Taking the transfer of knowledge between bridge engineering and housing construction projects as an example, the transfer learning unit can apply knowledge accumulated in housing construction projects, such as the impact of policies and regulations on construction costs, and the relationship between market supply and demand and cost fluctuations, to bridge construction cost forecasting through feature mapping and weight adjustment. When faced with new bridge projects, the model does not need to be trained from scratch. Instead, it reuses key features and prediction logic learned in housing construction projects and fine-tunes them based on the data characteristics of bridge projects. This significantly reduces the amount of data and time required for training, while effectively avoiding model overfitting caused by insufficient data. Ultimately, the model achieves rapid adaptation and accurate predictions in different construction cost scenarios, significantly enhancing the system's practicality and generalization capabilities.

[0057] Based on the same inventive concept, the embodiments of the present application also provide a method for dynamic analysis and intelligent prediction of engineering cost indicators for implementing the aforementioned dynamic analysis and intelligent prediction system for engineering cost indicators. The implementation solution provided by this method is similar to the implementation solution described in the aforementioned system. Therefore, the specific limitations of one or more embodiments of the method for dynamic analysis and intelligent prediction of engineering cost indicators provided below can be found in the limitations of the dynamic analysis and intelligent prediction system for engineering cost indicators above, and will not be repeated here.

[0058] In an exemplary embodiment, Figure 5 As shown, a method for dynamic analysis and intelligent prediction of engineering cost indicators is provided, including the following steps: Step S201 : Collect market transaction data, policy and regulation data, and project environment data related to engineering cost indicators to construct a multi-source heterogeneous data set.

[0059] Specifically, market transaction data covers construction material price indices, labor market wage indices, and equipment rental price indices. This data is collected from professional building materials websites, government websites, and industry platforms through web crawlers and API integration. Policy and regulatory data includes quota standard adjustment coefficients, tax rate change records, and financial credit interest rates, primarily obtained through government official websites. Project environmental data, including geoclimatic risk indices and the completeness of surrounding infrastructure, is collected through meteorological data platforms and map APIs. After collection, these data are categorized and stored to form a basic dataset that comprehensively reflects the factors influencing project costs.

[0060] Step S202 : pre-processing the collected data to extract a multi-dimensional feature vector including time domain, frequency domain and correlation features.

[0061] During preprocessing, missing values ​​are corrected using a time-delay Kalman filter algorithm, and outliers are eliminated using a local outlier factor algorithm and statistical process control methods to ensure data quality. Subsequently, a multidimensional feature vector is extracted, comprising time-domain, frequency-domain, and correlation features. Time-domain features reflect data trends over time, such as the month-over-month growth rate of material prices; frequency-domain features exploit cyclical patterns in the data, analyzing price fluctuations through Fourier transforms; and correlation features quantify the interrelationships between different factors, calculating the correlation coefficient between policy adjustments and cost fluctuations. This transforms the raw data into effective information suitable for model analysis.

[0062] Step S203: construct a dynamic analysis model that integrates chaos theory and game theory, simulate the dynamic changes of the engineering cost system through the dynamic analysis model, and realize multi-dimensional dynamic analysis of engineering cost indicators.

[0063] The chaos sub-model employs a modified time-delayed Lorenz equation, introducing multivariable time-delay terms and characteristic influence coefficients to simulate the dynamic evolution of the construction cost system and capture the nonlinear effects of factors such as market fluctuations and policy changes. The game sub-model constructs a four-agent incomplete information game and, by solving the Bayesian Nash equilibrium, analyzes the impact of strategic interactions among stakeholders such as owners, contractors, suppliers, and regulators on construction costs. Finally, the model fusion unit automatically adjusts the weights of the chaos and game sub-models based on an exponential function of the model fitness scores, achieving dynamic fusion and providing in-depth, multi-dimensional analysis of the evolving patterns of the construction cost system.

[0064] In step S204, in an extended state space including historical data, market status, policy environment, and market sentiment, the project cost index prediction is realized based on the reinforcement learning algorithm combined with the risk-adjusted reward function, and the generalization ability of the model is improved through transfer learning.

[0065] Step S204 focuses on the intelligent prediction and model optimization of engineering cost indicators. By constructing an extended state space, designing a risk-adjusted reward function, applying a reinforcement learning algorithm, and implementing a transfer learning strategy, it achieves accurate prediction of engineering costs and improves model generalization capabilities.

[0066] When constructing the extended state space, the system integrates historical project cost data, real-time market transaction data, policy and regulatory data, project environment data, and market sentiment indicators. Historical data is filtered through a time-delay Kalman filter to address missing values, and wavelet transforms are used to extract time and frequency domain features, revealing cost trends. Market state data includes information such as the construction material price index and the labor wage index, and correlation analysis is used to quantify their impact on costs. Policy environment data includes quota standard adjustment coefficients and tax rate change records, with policy sensitivity coefficients used to reflect the intensity of policy effects. Project environment data integrates geographic climate risk indices and infrastructure completeness, and is converted into impact factors through risk assessment. Market sentiment indicators leverage natural language processing techniques to extract market participants' expectations and emotions from industry news text, capturing irrational factors. Together, these data form a multidimensional extended state space, providing comprehensive environmental information for reinforcement learning.

[0067] Based on this state space, the system designs a risk-adjusted reward function to guide model learning. This reward function penalizes prediction errors through an exponential function, amplifying the penalty for larger errors and encouraging the model to improve its prediction accuracy. It also introduces a conditional value-at-risk (CVA) to assess the potential losses of prediction results. Based on the price fluctuation probability distribution output by the chaotic sub-model and combined with a risk adjustment coefficient, it guides the model to avoid risk. Furthermore, a stability constraint term, using the Kullback-Leibler divergence to constrain the change in prediction strategy between adjacent states, prevents the model from overfitting to short-term fluctuations, and ensures the stability of long-term predictions.

[0068] During the prediction process, the system employs a dual Deep Q-Network architecture combined with a prioritized experience replay mechanism for reinforcement learning. DDQN mitigates the problem of Q-value overestimation by separating action selection and evaluation. Priority experience replay calculates sample priority based on the time-delay error (TD error), prioritizing samples with large errors and high value to accelerate model training. During training, the model takes the state space as input, selects actions, and obtains rewards. The gradients of the main network parameters are calculated based on the TD error, and the parameters are updated using an optimization algorithm. The main network parameters are periodically copied to the target network to maintain training stability and gradually learn the optimal prediction strategy under different states.

[0069] Finally, to improve the model's generalization capabilities across different scenarios, a feature mapping function is designed to minimize the difference in the mean and covariance matrices between the source and target domains. Mapping parameters are learned to achieve data distribution alignment. Simultaneously, migration weights are dynamically adjusted based on the time-lag association matrix and data quality assessment matrix, constructing a cross-domain knowledge transfer framework. When applied to new scenarios, the model reuses knowledge learned in the source domain and fine-tunes it with data from the target domain. This reduces training data and time costs, avoids overfitting, enables rapid adaptation, and provides accurate predictions, enhancing the model's practicality in various engineering cost management projects. In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, memory and input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store data for dynamic analysis and intelligent prediction of engineering cost indicators. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a system and method for dynamic analysis and intelligent prediction of engineering cost indicators is implemented.

[0070] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0071] In an exemplary embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0072] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0073] In an exemplary embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0074] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0075] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0076] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0077] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0078] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. The dynamic analysis and intelligent prediction system of engineering cost indicators is characterized by: include: A multi-dimensional data acquisition module configured to acquire multi-source data related to project cost indicators, wherein the multi-source data includes market transaction data, policy and regulation data, and project environment data; a data processing module, connected to the multidimensional data acquisition module, configured to preprocess the acquired data and extract multidimensional feature vectors; A dynamic analysis module, connected to the data processing module, configured to construct a dynamic analysis model integrating chaos theory and game theory, wherein the dynamic analysis model is used to simulate the dynamic changes of the construction cost system and realize multi-dimensional dynamic analysis of construction cost indicators; An intelligent prediction module is connected to the data processing module and the dynamic analysis module, and is configured to apply a reinforcement learning algorithm in an extended state space to realize the prediction of the project cost index.

2. The system for dynamic analysis and intelligent prediction of construction cost indicators according to claim 1 is characterized in that: The data processing module is further configured to: a data cleaning unit configured to apply a time-delay Kalman filter algorithm to process missing data values; a feature extraction unit, connected to the data cleaning unit, configured to construct a multidimensional feature vector comprising time domain features, frequency domain features, and correlation features, wherein the time domain features reflect the changing trend of the data over time, the frequency domain features reflect the periodicity of the data, and the correlation features quantify the relationship between different factors; The feature optimization unit is connected to the feature extraction unit and is configured to perform multi-scale feature decomposition through wavelet transform to achieve separation and reconstruction of different frequency components of the data.

3. The system for dynamic analysis and intelligent prediction of construction cost indicators according to claim 1 is characterized in that: The dynamic analysis module is further configured to: a chaos analysis unit configured to apply an improved time-delay Lorenz equation to construct a chaos sub-model, wherein the chaos sub-model simulates the dynamic evolution of the construction cost system by introducing a multivariable time-delay term and a characteristic influence coefficient; a game analysis unit configured to construct a game sub-model involving incomplete information of four parties, namely, the owner, the contractor, the supplier, and the regulatory authority, wherein the game sub-model analyzes the impact of the strategic interaction of each party on the construction cost by solving the Bayesian Nash equilibrium; The model fusion unit is configured to automatically adjust the weights of the chaos sub-model and the game sub-model based on an exponential function of the model fitness score to achieve dynamic fusion.

4. The system for dynamic analysis and intelligent prediction of construction cost indicators according to claim 3 is characterized in that: The chaotic sub-model adopts the improved time-delay Lorenz equation as follows: in: 、 、 They are the chaotic system state variables that reflect the cost fluctuation dimension, market supply and demand change dimension, and policy impact dimension indicators in the engineering cost system; 、 、 System parameters calibrated based on historical data are used to adjust system dynamic characteristics; 、 、 is the time lag coefficient, which reflects the lag of the influence of factors in the system; 、 、 Indicates time lag The rate of change of state; 、 、 Influencing factors The weight coefficient is calculated by the entropy weight method, which reflects the relative importance of each factor to the change of system state.

5. The system for dynamic analysis and intelligent prediction of construction cost indicators according to claim 4 is characterized in that: The profit function expression of the game sub-model is: in: For participants In its own strategy and other participant strategy combinations The following income; For participants Adopt a strategy Basic income; Cost of implementing the strategy; is the competition coefficient, which reflects the degree of competition among participants; For participants The covariance between the strategy of and the strategies of other participants, measuring the mutual influence between strategies; is the risk aversion coefficient, which reflects the participant’s aversion to risk; For strategy Risk value is used to evaluate the risk level during strategy implementation.

6. The system for dynamic analysis and intelligent prediction of construction cost indicators according to claim 5 is characterized in that: The model fusion unit integrates the outputs of the chaos sub-model and the game sub-model through a dynamic weight function, which satisfies the following formula: in: is the analysis result after fusion; is the chaotic sub-model output, Output of the game sub-model; , , 、 The chaos model and the game model are The fitness score at the moment.

7. The system for dynamic analysis and intelligent prediction of construction cost indicators according to claim 1 is characterized in that: The intelligent prediction module is further configured as follows: State space construction unit, which constructs an extended state space containing historical data, market data, policy data, environmental data and market sentiment indicators; A reward design unit performs risk value assessment and predictive stability constraint operations on the extended state space, and constructs a reward function to guide the model learning direction; The network training unit uses the reward function as feedback, adopts a dual deep Q network architecture combined with priority experience replay, optimizes network parameters, and realizes project cost prediction; The transfer learning unit, based on the trained network, reduces domain differences through feature mapping, transfers knowledge to new scenarios, and improves the model's generalization ability.

8. The system for dynamic analysis and intelligent prediction of construction cost indicators according to claim 7 is characterized in that: The reward function expression is: in: for Actual project cost index value at all times, is the predicted value; is the risk aversion coefficient; The conditional value at risk of the next period's forecast value; To expand the state space; is the stability coefficient; Constrain the range of changes in prediction strategies between adjacent states.

9. The system for dynamic analysis and intelligent prediction of construction cost indicators according to claim 1 is characterized in that: The network training unit operates according to the following steps: Get the extended state space generated by the state space building unit and the reward value calculated by the reward design unit , as basic training data; A dual deep Q-network architecture is used to calculate the Q-value of the current state and / or action pair and the optimal Q-value of the next state respectively; Calculate the TD error according to the network update rule, and update the priority of samples in the experience replay buffer according to the priority calculation formula to give priority to samples with larger errors; According to the training algorithm with policy gradient constraints, the gradient of the main network parameters is calculated based on the TD error, and the main network parameters are updated using the optimization algorithm; Every fixed number of training steps, the main network parameters are copied to the target network parameters to maintain the stability of the target network; Repeat the above parameter updating and optimization steps until the network converges or reaches the preset number of training times, completing the training of the project cost index prediction model.

10. A method for dynamic analysis and intelligent prediction of engineering cost indicators, characterized in that: The steps include: Collect market transaction data, policy and regulation data, and project environment data related to engineering cost indicators to construct a multi-source heterogeneous data set; Preprocess the collected data and extract multidimensional feature vectors containing time domain, frequency domain and correlation features; Construct a dynamic analysis model that integrates chaos theory and game theory, simulate the dynamic changes of the construction cost system through the dynamic analysis model, and realize multi-dimensional dynamic analysis of construction cost indicators; In an extended state space that includes historical data, market status, policy environment, and market sentiment, engineering cost indicator prediction is achieved based on a reinforcement learning algorithm combined with a risk-adjusted reward function, and the model generalization ability is improved through transfer learning.

Images