A method for engineering cost data management based on artificial intelligence

Abstract

The present invention relates to the field of artificial intelligence technology, and discloses an engineering cost data management method based on artificial intelligence, including: step 1, extracting text content from paper documents and images related to the project through optical character recognition technology, parsing the extracted text data in combination with natural language processing technology, extracting data containing contract terms, budget project information and related cost content, and at the same time, using Web crawler technology to obtain dynamic data related to material market prices, and then collecting the operating status of construction site equipment and material consumption information through Internet of Things sensors. By combining optical character recognition, natural language processing and Web crawler technology, efficient collection and parsing of multi-source heterogeneous data is achieved, and uniformly converted into a structured JSON format, significantly improving data processing efficiency and integration standardization, and laying a solid foundation for subsequent data cleaning, storage and analysis.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an engineering cost data management method based on artificial intelligence. Background Art

[0002] As the scale and complexity of engineering projects continue to grow, the importance of cost management in the entire project life cycle becomes increasingly prominent. The goal of engineering cost management is to accurately predict the cost of the project, and it is necessary to dynamically adjust budget allocation and optimize resource use to maximize economic benefits and project efficiency. However, traditional cost management methods face many challenges in data collection, processing, prediction, and decision support.

[0003] In the implementation process of engineering projects, cost management, as a core link, is directly related to the effectiveness and economic benefits of project cost control. However, the existing engineering cost management methods have technical bottlenecks:

[0004] Traditional construction cost management usually relies on manual processing of paper documents, budget forms and construction site record data, which is inefficient and prone to human errors. At the same time, the data sources are diverse and the formats are not uniform, making data integration complicated and time-consuming, and unable to meet the requirements of modern engineering management for data real-time and standardization.

[0005] The existing engineering cost forecasting model is built based on static data. For example, it uses simple linear regression or empirical formulas for forecasting. It cannot dynamically reflect the impact of external environmental factors such as market price fluctuations and changes in construction progress, resulting in deviations between the forecast results and the actual situation. It cannot adapt to the changing market conditions and implementation requirements in engineering projects.

[0006] In the process of budget allocation and procurement plan formulation, traditional methods mainly rely on managers' experience and decision-making, lacking scientific quantitative analysis tools. When faced with complex multivariable optimization problems, they show low efficiency and poor accuracy, and cannot quickly adjust optimization strategies under complex and changing market conditions.

[0007] The existing engineering cost data analysis results are usually presented in the form of static reports or simple two-dimensional charts, which lack interactivity and intuitiveness. Managers cannot quickly understand complex analysis results, which in turn has an adverse impact on decision-making efficiency. In addition, the system's operability and user experience are poor, and cannot meet the needs of modern engineering management for efficient decision-making support.

[0008] Therefore, those skilled in the art provide an engineering cost data management method based on artificial intelligence to solve the above-mentioned problems. Summary of the invention

[0009] In view of the deficiencies in the prior art, the present invention provides an engineering cost data management method based on artificial intelligence to solve the problems raised in the above background technology.

[0010] To achieve the above objectives, the present invention is implemented through the following technical solutions: an engineering cost data management method based on artificial intelligence, comprising:

[0011] Step 1: Use optical character recognition technology to extract text content from paper documents and images related to the project, and use natural language processing technology to parse the extracted text data to extract data containing contract terms, budget project information and related cost content. At the same time, use Web crawler technology to obtain dynamic data related to material market prices, and then use IoT sensors to collect the operating status of construction site equipment and material consumption information. After integrating the above multi-source data, uniformly convert it into JSON format to provide standardized data input for subsequent processing;

[0012] Step 2: Clean the multi-source JSON format data generated in step 1, detect outliers in the data through statistical methods, remove or correct the outliers, and organize the cleaned data in a standardized table structure. During the cleaning process, for missing data, use collaborative filtering methods to predict and fill in missing items based on similarity calculations to ensure data integrity and provide a reliable data foundation for subsequent storage;

[0013] Step 3: Store the cleaned and padded structured and unstructured data in the distributed data storage architecture, where static data is stored in the distributed file system to support efficient query, and dynamic data and time series data are stored in the corresponding database. At the same time, data access rights are classified and managed through role-based access control technology, and the data version and update status are recorded to track the data change history and ensure data security.

[0014] Step 4: Based on the data stored in step 3, the recursive feature elimination method is used to extract features from the data, extracting factors affecting the construction cost including project scale, bill of materials and market price fluctuations, forming a feature set, and then building a total cost prediction model based on the feature set. For data involving time series features, a dynamic trend prediction model is built using a long short-term memory network model to provide prediction results for subsequent optimization steps;

[0015] Step 5: Based on the prediction results generated in step 4, a reinforcement learning algorithm is used to optimize the project cost control scheme. The reinforcement learning algorithm performs iterative calculations based on the simulation of different budget allocations and procurement schemes, and evaluates the cost savings and construction period adjustment of each scheme in combination with the reward function, outputs the optimal scheme, and compares and analyzes the optimal scheme with the original budget scheme to provide users with difference information and optimization references.

[0016] Step 6: Visualize the optimization solutions and comparative analysis results output in step 5, use dynamic trend charts to graphically display the cost forecast data and risk assessment results, and generate a heat map of engineering costs to mark high-risk areas. Then, present the optimization suggestions and detailed reports in a visual way through the Web interactive interface, supporting users to further operate to adjust or confirm the optimization results.

[0017] Preferably, the optical character recognition technology in step 1 uses a multi-layer convolutional neural network model to extract text content, wherein the text recognition accuracy is measured by the following formula:

[0018]

[0019] Among them, N correct is the number of correctly recognized characters, N total is the total number of characters, and Accuracy represents the accuracy of the algorithm.

[0020] Preferably, the statistical method for eliminating outliers in step 2 adopts the 3-times standard deviation method, and the calculation formula is:

[0021]

[0022] Among them, X is the data value, μ is the data mean, σ is the data standard deviation, and if |Z|>3, it is judged as an outlier.

[0023] Preferably, the missing data in step 2 is filled using a collaborative filtering algorithm based on user similarity, and the similarity is calculated using the following formula:

[0024]

[0025] Among them, r u.i and r v.i represents the ratings of users u and v on item i, and is the average rating of users u and v, Sim(u, v) is the similarity between users u and v, and I is the set of items rated by users u and v.

[0026] Preferably, the scoring calculation formula for collaborative filtering prediction filling in step 2 is:

[0027]

[0028] in, is the predicted score, N(u) is the set of users similar to user u, is the average rating of user u, Sim(u,v) is the similarity between users u and v, r v.i is the rating of item i by user v, is the average rating of user v.

[0029] Preferably, the access control of the data in step 3 is implemented by the following access matrix A:

[0030] A={a ij}, a ij ∈{0, 1},

[0031] Among them, a ij =1 means user i has access rights to data resource j.

[0032] a ij =0 means no permission.

[0033] Preferably, in step 3, the distributed file system adopts redundant storage technology, and the redundancy factor R is calculated as follows:

[0034]

[0035] Among them, T data is the original data volume, T redundant The amount of data stored for redundancy.

[0036] Preferably, the feature extraction in step 4 is performed by a recursive feature elimination method, and the loss function for evaluating the accuracy of the model is:

[0037]

[0038] Among them, y i is the true value, which is the target output of the i-th sample;

[0039] is the predicted value, which is the output of the i-th sample predicted by the model;

[0040] N is the sample size.

[0041] Preferably, the objective function of the long short-term memory network model for time series data after feature extraction is:

[0042]

[0043] Among them, y t and represents the true value and predicted probability at the tth time point, T is the length of the time series, represents the logarithm of the predicted probability, and L represents the value of the loss function.

[0044] Preferably, the reinforcement learning algorithm in step 5 uses Q learning, and the update rule is:

[0045]

[0046] Among them, Q(s, a) is the value of taking action a in state s, r is the immediate reward, α is the learning rate, γ is the discount factor, s ′ Indicates the next state reached after action a is executed.

[0047] The reward function in the reinforcement learning algorithm is defined as:

[0048] Reward=λ1·Cost saving -λ2·Time delay ,

[0049] Among them, Cost saving For cost savings, Time delay is the construction period delay, λ1 and λ2 are weight coefficients.

[0050] The present invention provides an engineering cost data management method based on artificial intelligence, which has the following beneficial effects:

[0051] 1. The present invention combines optical character recognition, natural language processing and Web crawler technology to achieve efficient collection and analysis of multi-source heterogeneous data, and uniformly converts them into a structured JSON format, significantly improving data processing efficiency and integration standardization, and laying a solid foundation for subsequent data cleaning, storage and analysis.

[0052] 2. The present invention predicts time series data through a long short-term memory network model to capture dynamic market price fluctuations and construction progress change trends. At the same time, it optimizes and selects static features in combination with a recursive feature elimination method, so that the cost prediction model can dynamically adapt to changing market conditions, significantly improve prediction accuracy, and achieve higher prediction flexibility and accuracy.

[0053] 3. The present invention adopts reinforcement learning algorithm to optimize budget allocation and procurement schemes, scientifically evaluates the cost savings and shortened construction period of different schemes based on reward functions, and outputs the optimal scheme through multiple rounds of iterations, thus overcoming the drawbacks of traditional optimization methods that rely heavily on experience, and providing dynamic optimization decisions under complex multivariable conditions, providing users with a more scientific and efficient cost control scheme.

[0054] 4. The present invention generates dynamic trend charts and engineering cost heat maps to intuitively display complex forecast data and risk assessment results, and supports users to adjust analysis parameters and optimization plans in real time through a Web interactive interface. Compared with traditional result presentation methods, it greatly improves managers' understanding of data and decision-making efficiency. At the same time, it enhances the operability of the system and user experience to meet diverse engineering management needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0056] In order to make the technical personnel in the technical field understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in combination with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a partial embodiment of the present invention, not a complete embodiment. Based on the embodiment of the present invention, other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

[0057] The present invention is described in detail below in conjunction with the accompanying drawings:

[0058] Example:

[0059] Please see attached Figure 1 , an embodiment of the present invention provides an engineering cost data management method based on artificial intelligence, comprising:

[0060] Step 1: Use optical character recognition technology to extract text content from paper documents and images related to the project, and use natural language processing technology to parse the extracted text data to extract data containing contract terms, budget project information and related cost content. At the same time, use Web crawler technology to obtain dynamic data related to material market prices, and then use IoT sensors to collect the operating status of construction site equipment and material consumption information. After integrating the above multi-source data, uniformly convert it into JSON format to provide standardized data input for subsequent processing;

[0061] Step 2: Clean the multi-source JSON format data generated in step 1, detect outliers in the data through statistical methods, remove or correct the outliers, and organize the cleaned data in a standardized table structure. During the cleaning process, for missing data, use collaborative filtering methods to predict and fill in missing items based on similarity calculations to ensure data integrity and provide a reliable data foundation for subsequent storage;

[0062] Step 3: Store the cleaned and padded structured and unstructured data in the distributed data storage architecture, where static data is stored in the distributed file system to support efficient query, and dynamic data and time series data are stored in the corresponding database. At the same time, data access rights are classified and managed through role-based access control technology, and the data version and update status are recorded to track the data change history and ensure data security.

[0063] Step 4: Based on the data stored in step 3, the recursive feature elimination method is used to extract features from the data, extracting factors affecting the construction cost including project scale, bill of materials and market price fluctuations, forming a feature set, and then building a total cost prediction model based on the feature set. For data involving time series features, a dynamic trend prediction model is built using a long short-term memory network model to provide prediction results for subsequent optimization steps;

[0064] Step 5: Based on the prediction results generated in step 4, a reinforcement learning algorithm is used to optimize the project cost control scheme. The reinforcement learning algorithm performs iterative calculations based on the simulation of different budget allocations and procurement schemes, and evaluates the cost savings and construction period adjustment of each scheme in combination with the reward function, outputs the optimal scheme, and compares and analyzes the optimal scheme with the original budget scheme to provide users with difference information and optimization references.

[0065] Step 6: Visualize the optimization solutions and comparative analysis results output in step 5, use dynamic trend charts to graphically display the cost forecast data and risk assessment results, and generate a heat map of engineering costs to mark high-risk areas. Then, present the optimization suggestions and detailed reports in a visual way through the Web interactive interface, supporting users to further operate to adjust or confirm the optimization results.

[0066] Benefits of Step 1:

[0067] Combine optical character recognition, natural language processing, web crawlers and IoT technologies to obtain information from multiple data sources and ensure the breadth and integrity of data sources. By converting to a unified JSON format, the problem of inconsistent formats of different data sources is eliminated, providing standardized data input for subsequent processing. Dynamically obtain material market prices and construction site data, support real-time updates, and ensure the timeliness of data.

[0068] Benefits of Step 2:

[0069] By removing and correcting outliers, data noise can be reduced and data accuracy and reliability can be improved. Collaborative filtering can be used to fill in missing data to avoid model deviations caused by incomplete data. Standardized table structures can be organized to ensure compatibility with subsequent storage and analysis and improve data processing efficiency.

[0070] Benefits of Step 3:

[0071] Static data is stored in a distributed file system, supporting fast access and efficient query. Role-based access control technology classifies management permissions to ensure the security of sensitive data. Recording versions and update status can track data change history for easy backtracking and auditing.

[0072] Benefits of Step 4:

[0073] Through recursive feature elimination, the key factors affecting the construction cost are extracted, which helps to understand the cost drivers. The time series data is processed based on the long short-term memory network to improve the accuracy of dynamic trend prediction. The total cost prediction model is generated to provide quantitative support for subsequent optimization steps and avoid empirical decision-making.

[0074] Benefits of Step 5:

[0075] Through reinforcement learning, we simulate various budget allocation and procurement plans, explore the best plan, and effectively reduce the project cost. Combined with the reward function, we evaluate the adjustment of the construction period, weigh the cost and time, and improve the overall efficiency of the project. We compare and analyze the optimized plan with the original plan to provide users with decision-making references and enhance operability.

[0076] Benefits of Step 6:

[0077] The cost forecast and risk assessment results are displayed through dynamic trend charts and heat maps to make users understand. High-risk areas are marked to help users detect and avoid potential risks in a timely manner. Optimization suggestions and reports are presented through the web interface to support users to further adjust or confirm the results, improving user experience and control capabilities.

[0078] Summary: Through the refined processing of multi-source data acquisition, cleaning, storage, feature extraction, optimization and visualization, the accuracy, real-time, efficiency and security of engineering cost data management can be comprehensively improved, providing users with scientific decision-making support. At the same time, it optimizes costs and construction periods and significantly improves the intelligence level of engineering management.

[0079] The optical character recognition technology in step 1 uses a multi-layer convolutional neural network model to extract text content, where the text recognition accuracy is measured by the following formula:

[0080]

[0081] Among them, N correct is the number of correctly recognized characters, N total is the total number of characters, and Accuracy represents the accuracy of the algorithm.

[0082] Multi-layer convolutional neural networks can capture local and global features of characters in images and improve recognition accuracy. Clearly measure the recognition effect to ensure the evaluability and transparency of model performance.

[0083] The convolutional neural network model has good robustness to complex backgrounds, different fonts, and character spacing, and can handle a variety of paper documents and image content to improve adaptability.

[0084] Automatically complete text extraction tasks, significantly reduce the workload of manual input and proofreading, and improve efficiency. After accurately extracting text data, combine it with natural language processing technology to analyze and ensure the integrity of key data such as contract terms and budget information.

[0085] The accuracy formula provides a quantitative basis for performance optimization, and continuously adjusts the network structure or hyperparameters to further improve model performance. Accurate identification reduces the cost of subsequent data verification and correction, and improves the economic efficiency of overall data processing.

[0086] Summary: Using a multi-layer convolutional neural network model to extract text content and combining it with the accuracy formula for measurement can greatly improve the accuracy and efficiency of text recognition, provide high-quality basic data input for subsequent engineering cost data management, and further promote the intelligence and reliability of the process.

[0087] The statistical method used to eliminate outliers in step 2 adopts the 3-times standard deviation method, and the calculation formula is:

[0088]

[0089] Among them, X is the data value, μ is the data mean, σ is the data standard deviation, and if |Z|>3, it is judged as an outlier.

[0090] The formula is clear, intuitive, easy to implement and highly efficient. Measuring based on the mean and standard deviation of the data can effectively capture outliers that deviate from the overall data and avoid noise data affecting subsequent analysis.

[0091] It is applicable to various types of data sets, especially data that meet normal distribution or are close to normal distribution, and has wide applicability. It can remove outliers, reduce data noise, improve the accuracy and reliability of cleaned data, and provide a high-quality data foundation for subsequent storage, modeling and analysis.

[0092] After removing extreme outliers, the impact of outliers on model training and prediction can be avoided, and the stability and generalization ability of the model can be improved. Using standard deviation as a metric can provide a clear mathematical basis for the definition of outliers and enhance the scientificity and objectivity of decision-making.

[0093] Through extension, it can be applied to anomaly detection of multi-dimensional data, which is convenient for use in complex scenarios. Compared with complex anomaly detection algorithms, the 3x standard deviation method is suitable for use in scenarios where data needs to be cleaned quickly.

[0094] In summary, eliminating outliers by the 3-times standard deviation method can significantly improve data quality and reduce the complexity of subsequent processing and analysis. At the same time, it ensures the robustness and prediction accuracy of the model. It is an efficient and reliable outlier detection method.

[0095] In step 2, missing data is filled using collaborative filtering algorithm based on user similarity, and similarity is calculated using the following formula:

[0096]

[0097] Among them, r u.i and r v.i represents the ratings of users u and v on item i, and is the average rating of users u and v, Sim(u, v) is the similarity between users u and v, and I is the set of items rated by users u and v.

[0098] Use the similarity between users to fill in missing data, effectively reduce data null values, and improve data integrity. By considering the user's rating behavior, generate a filling value that conforms to the data characteristics, rather than simply filling in the mean or median.

[0099] The collaborative filtering filling method cannot change the data distribution characteristics, which helps to preserve the original characteristics of the data in subsequent analysis and modeling. It does not rely on specific domain knowledge and can work based on scoring data, which is applicable to various fields.

[0100] By calculating the similarity between users and using highly correlated user ratings to fill in missing items, noise interference is reduced and the accuracy of filling is improved.

[0101] The formula supports multi-dimensional rating data and can be extended to more complex scenarios, such as multi-attribute ratings or multi-category missing data filling. Compared with global average filling, collaborative filtering filling uses the similarity of local users to avoid filling bias caused by the global mean.

[0102] The collaborative filtering results are used as input features of other models and combined with statistical methods or machine learning models to further improve the filling effect. The similarity formula is clear and transparent, and the generation process of the filling results is highly interpretable, which is convenient for verification and adjustment.

[0103] In summary, the collaborative filtering algorithm fills in missing data based on user similarity, improves the integrity and quality of data through the advantages of personalization, accuracy, and flexibility, and provides a reliable data foundation for subsequent analysis and modeling.

[0104] The scoring formula for collaborative filtering prediction filling in step 2 is:

[0105]

[0106] in, is the predicted score, N(u) is the set of users similar to user u, is the average rating of user u, Sim(u,v) is the similarity between users u and v, r v.i is the rating of item i by user v, is the average rating of user v.

[0107] Prediction filling is based on the similarity between users, which can reflect the user's personalized preferences and generate a score that is more in line with the user's characteristics. , eliminating individual biases in scoring and making filling more fair and accurate.

[0108] Using the similarity Sim(u, v) as a weight, the influence of users with high similarity on the predicted score is increased, improving the reliability of the filling result. The filling formula relies on the scores of similar users, combined with the average score of the target user, to balance global and individual features in the prediction.

[0109] The formula clearly shows the calculation logic of the predicted score, and the filled results are explained by user similarity and score weights for verification and optimization.

[0110] It is used to fill in the gaps in the scoring matrix using multiple scoring modes.

[0111] The filled data can fully reflect user preferences, reduce model training bias caused by missing data, and improve the generalization performance of subsequent models.

[0112] The predicted rating is calculated based on a set of similar users. The method is less complex and takes both efficiency and accuracy into consideration. It can be combined with other collaborative filtering algorithms and expanded to complex scenarios, such as multi-dimensional rating or dynamic rating prediction.

[0113] The filled scoring matrix is ​​more complete and can support subsequent modeling, clustering and predictive analysis, thus improving data utilization.

[0114] In summary, the collaborative filtering prediction filling formula provides personalized, accurate and explainable rating filling results by combining user similarity, individual preferences and rating context, laying a high-quality foundation for data cleaning and subsequent processing, while improving the scientificity and practicality of the filling process.

[0115] The access control of the data in step 3 is implemented through the following access matrix A:

[0116] A={a ij}, a ij ∈{0, 1},

[0117] Among them, a ij =1 means user i has access rights to data resource j.

[0118] a ij =0 means no permission.

[0119] The access matrix clearly defines each user's access rights to each resource, making it easier to manage and review.

[0120] The permission status is intuitive and easy to understand, which helps to quickly identify permission configuration issues. Unauthorized users cannot access sensitive data, reducing the risk of data leakage and unauthorized operations. By limiting the scope of access, the confidentiality and integrity of data are guaranteed.

[0121] The access matrix is ​​mapped to the role-based access control model, supporting role-based permission classification management to meet complex permission requirements. The access matrix has a simple structure and is easy to implement in the database or system, and is directly used in the permission verification logic.

[0122] By expanding the rows and columns of the access matrix, new users or new resources can be easily added to meet the needs of system expansion. Dynamic update of permission settings is supported to flexibly adapt to permission changes.

[0123] The access matrix provides a complete record of permission allocation, which facilitates permission auditing and historical tracking to ensure that permission configuration complies with regulations. The matrix-shaped data structure facilitates the use of automated tools for permission checking, updating and analysis, reducing the complexity of manual management.

[0124] In the distributed data storage architecture, the access matrix uniformly manages the permissions of different nodes to ensure global consistency. The matrix query method is simple and efficient, reducing the computational burden of complex permission checking logic and improving system response speed.

[0125] Define fine-grained access permissions for different resources, such as read-only, modify, delete, etc., to meet more complex permission control needs.

[0126] In summary, access matrix A is a simple, efficient and flexible permission management mechanism. It improves the security and controllability of data management by clarifying the access relationship between users and resources. At the same time, it reduces the complexity of permission management and provides reliable protection for the distributed storage and use of engineering data.

[0127] In step 3, the distributed file system uses redundant storage technology, and the redundancy factor R is calculated as follows:

[0128]

[0129] Among them, T data is the original data volume, T redundant The amount of data stored for redundancy.

[0130] Redundant storage stores data in multiple copies or checksums, so that if some nodes or hard disks fail, the data can be fully restored, significantly improving the system's fault tolerance.

[0131] Redundant storage ensures that users can access data when part of the system fails, reducing the risk of business interruption. The existence of redundant data enables the lost part to be reconstructed through redundant data in the event of damage to the original data, ensuring data integrity.

[0132] By calculating the redundancy factor R, the storage cost increased by redundant storage can be clearly quantified, and a balance and optimization can be made between security and economy. Redundant storage is suitable for distributed file systems, which can disperse storage pressure among different nodes and prevent single point failures.

[0133] Redundant copies are distributed in multiple nodes to support load balancing, improve data reading performance and access speed. According to the fault tolerance requirements of different applications, the R value is dynamically adjusted to achieve flexible configuration.

[0134] In summary, the use of redundant storage technology and quantitative calculation of redundant factors in distributed file systems can improve data security, availability, and recovery capabilities, provide a tool for dynamic trade-offs between storage resources and system performance, and ensure the stability and efficiency of system operation.

[0135] The feature extraction in step 4 is done by recursive feature elimination method, and the loss function for evaluating the accuracy of the model is:

[0136]

[0137] Among them, y i is the true value, which is the target output of the i-th sample;

[0138] is the predicted value, which is the output of the i-th sample predicted by the model;

[0139] N is the sample size.

[0140] Remove features that contribute little to prediction performance, reduce the model's dimensions, and improve the model's training efficiency and prediction speed. By removing redundant or noisy features, reduce the risk of overfitting and improve the model's prediction performance on unknown data.

[0141] Using mean square error as the loss function and clarifying the optimization goal makes the feature extraction process more intuitive and easy to explain. After feature extraction, the retained key features reduce the complexity of the model, making it easier to interpret the results and analyze the relationship between the features and the target variable.

[0142] The loss function is based on the difference between the true value and the predicted value. It is applicable to regression problems and can be extended to other problems. This loss function can be used in combination with linear regression, decision tree, support vector machine and other models to meet different scenario requirements.

[0143] By calculating the error of the entire sample data, we can fully utilize the information of each sample and improve the reliability of the evaluation results.

[0144] In summary, through the combination of recursive feature elimination and mean square error loss function, the feature extraction process becomes more accurate and efficient, while providing strong support for model simplification, performance improvement and result interpretation, improving the model prediction effect and providing a scientific basis for the extraction of factors affecting engineering cost.

[0145] The objective function of the LSTM network model for time series data after feature extraction is:

[0146]

[0147] Among them, y t and represents the true value and predicted probability at the tth time point, T is the length of the time series, represents the logarithm of the predicted probability, and L represents the value of the loss function.

[0148] The objective function is based on logarithmic loss, which penalizes the deviation of the predicted probability and optimizes the model output more accurately. The objective function performs a logarithmic transformation on the probability value, which helps to avoid the gradient vanishing problem and accelerate the convergence of the model.

[0149] Uses logarithmic loss function, suitable for probability prediction, and supports multi-category classification tasks.

[0150] The logarithmic loss function penalizes inaccurate predictions to a degree that is related to the prediction confidence, which encourages the model to output more credible prediction results.

[0151] The loss function is calculated through T-terms of the entire sequence, making full use of the data at each time point to improve the model's ability to capture the overall trend of the time series.

[0152] The error accumulation in long time series prediction can effectively suppress the influence of error by using logarithmic loss function, thus improving the prediction stability. The loss function is normalized to be suitable for time series of any length, thus enhancing the versatility of the model.

[0153] In summary, the use of the logarithmic loss function L and the long short-term memory network model can efficiently process the dynamic features in time series data, generate accurate and stable prediction results, and improve the model's ability to capture long-term trends. It has significant advantages in the field of dynamic engineering cost forecasting, especially when dealing with complex time series and probability distribution tasks.

[0154] The reinforcement learning algorithm in step 5 uses Q learning, and the update rule is:

[0155]

[0156] Among them, Q(s, a) is the value of taking action a in state s, r is the immediate reward, α is the learning rate, γ is the discount factor, s ′ Indicates the next state reached after action a is executed.

[0157] The reward function in the reinforcement learning algorithm is defined as:

[0158] Reward=λ1·Cost saving -λ2·Time delay ,

[0159] Among them, Cost saving For cost savings, Time delay is the construction period delay, λ1 and λ2 are weight coefficients.

[0160] Reinforcement learning combined with Q-learning and reward functions can effectively explore various budget allocation and procurement plans and discover the optimal solution.

[0161] Based on algorithm learning and feedback, it does not rely on human experience, which improves the objectivity and scientificity of decision-making. It can adapt to the real-time changes in budget, market price and construction period constraints in engineering projects and generate more practical optimization solutions.

[0162] The weight of the reward function gives the user control over the optimization goal, allowing the user to understand and adjust the optimization strategy. Through continuous learning of Q-learning and data accumulation, the optimization effect will continue to improve and eventually approach the optimal solution.

[0163] In summary, Q-learning update rules combined with reward functions can achieve accurate optimization of project costs and construction periods. Its dynamic learning ability, global optimization characteristics and flexible goal setting provide users with scientific, efficient and controllable engineering cost control solutions.

[0164] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for engineering cost data management based on artificial intelligence, characterized in that: include: Step 1: Use optical character recognition technology to extract text content from paper documents and images related to the project, and use natural language processing technology to parse the extracted text data to extract data containing contract terms, budget project information and related cost content. At the same time, use Web crawler technology to obtain dynamic data related to material market prices, and then use IoT sensors to collect the operating status of construction site equipment and material consumption information. After integrating the above multi-source data, uniformly convert it into JSON format to provide standardized data input for subsequent processing; Step 2: Clean the multi-source JSON format data generated in step 1, detect outliers in the data through statistical methods, remove or correct the outliers, and organize the cleaned data in a standardized table structure. During the cleaning process, for missing data, use collaborative filtering methods to predict and fill in missing items based on similarity calculations to ensure data integrity and provide a reliable data foundation for subsequent storage; Step 3: Store the cleaned and padded structured and unstructured data in the distributed data storage architecture, where static data is stored in the distributed file system to support efficient query, and dynamic data and time series data are stored in the corresponding database. At the same time, data access rights are classified and managed through role-based access control technology, and the data version and update status are recorded to track the data change history and ensure data security. Step 4: Based on the data stored in step 3, the recursive feature elimination method is used to extract features from the data, extracting factors affecting the construction cost including project scale, bill of materials and market price fluctuations, forming a feature set, and then building a total cost prediction model based on the feature set. For data involving time series features, a dynamic trend prediction model is built using a long short-term memory network model to provide prediction results for subsequent optimization steps; Step 5: Based on the prediction results generated in step 4, a reinforcement learning algorithm is used to optimize the project cost control scheme. The reinforcement learning algorithm performs iterative calculations based on the simulation of different budget allocations and procurement schemes, and evaluates the cost savings and construction period adjustment of each scheme in combination with the reward function, outputs the optimal scheme, and compares and analyzes the optimal scheme with the original budget scheme to provide users with difference information and optimization references. Step 6: Visualize the optimization solutions and comparative analysis results output in step 5, use dynamic trend charts to graphically display the cost forecast data and risk assessment results, and generate a heat map of engineering costs to mark high-risk areas. Then, present the optimization suggestions and detailed reports in a visual way through the Web interactive interface, supporting users to further operate to adjust or confirm the optimization results.

2. The method for managing engineering cost data based on artificial intelligence according to claim 1, characterized in that: The optical character recognition technology in step 1 uses a multi-layer convolutional neural network model to extract text content, wherein the text recognition accuracy is measured by the following formula: Among them, N correct is the number of correctly recognized characters, N total is the total number of characters, and Accuracy represents the accuracy of the algorithm.

3. The method for managing engineering cost data based on artificial intelligence according to claim 1, characterized in that: The statistical method used to eliminate outliers in step 2 adopts the 3-times standard deviation method, and the calculation formula is: Among them, X is the data value, μ is the data mean, and σ is the data standard deviation. | Z | When >3, it is considered an abnormal value.

4. The method for managing engineering cost data based on artificial intelligence according to claim 1, characterized in that: The missing data in step 2 is filled using collaborative filtering algorithm based on user similarity, and the similarity is calculated using the following formula: Among them, r u.i and r v.i represents the ratings of users u and v on item i, and is the average rating of users u and v, Sim ( u,v ) is the similarity between users u and v, and I is the set of items rated by users u and v.

5. The method for managing engineering cost data based on artificial intelligence according to claim 1, characterized in that: The scoring calculation formula for collaborative filtering prediction filling in step 2 is: in, For the prediction score, N ( u ) is a set of users similar to user u, is the average rating of user u, Sim ( u,v ) is the similarity between users u and v, r v.i is the rating of user v on item i, is the average rating of user v.

6. The method for managing engineering cost data based on artificial intelligence according to claim 1, characterized in that: The access control of the data in step 3 is implemented by the following access matrix A: A={a ij },a ij ∈{0,1}, Among them, a ij =1 means user i has access rights to data resource j. a ij =0 means no permission.

7. The method for managing engineering cost data based on artificial intelligence according to claim 1, characterized in that: In step 3, the distributed file system adopts redundant storage technology, and the redundancy factor R is calculated as follows: Among them, T data is the original data volume, T redundant The amount of data stored for redundancy.

8. The method for managing engineering cost data based on artificial intelligence according to claim 1, characterized in that: The feature extraction in step 4 is performed by recursive feature elimination method, and the loss function for evaluating the accuracy of the model is: Among them, y i is the true value, which is the target output of the i-th sample; is the predicted value, which is the output of the i-th sample predicted by the model; N is the sample size.

9. The method for managing engineering cost data based on artificial intelligence according to claim 1, characterized in that: The objective function of the long short-term memory network model for time series data after feature extraction is: Among them, y t and represents the true value and predicted probability at the tth time point, T is the length of the time series, represents the logarithm of the predicted probability, and L represents the value of the loss function.

10. The method for managing engineering cost data based on artificial intelligence according to claim 1, characterized in that: The reinforcement learning algorithm in step 5 uses Q learning, and the update rule is: Among them, Q(s, a) is the value of taking action a in state s, r is the immediate reward, α is the learning rate, γ is the discount factor, and s′ represents the next state reached after action a is executed. The reward function in the reinforcement learning algorithm is defined as: Reward=λ1·Cost saving -λ2·Time delay , Among them, Cost saving For cost savings, Time delay is the construction period delay, λ1 and λ2 are weight coefficients.

Images