Project supervision data management method and system based on deep learning

Abstract

The invention relates to the technical field of risk analysis, in particular to an engineering supervision data management method and system based on deep learning, and the method comprises the following steps: obtaining an engineering project data set, obtaining a project name, a place, starting time and supervision team information according to the engineering project data set, generating a project metadata file, and storing the project metadata file in a database; and performing time sequence analysis on the project metadata file, identifying key time nodes, and generating time monitoring data. According to the invention, the accuracy of data processing and risk prediction is improved by integrating deep learning and graph neural network technologies. And through the graph feature vector extracted through the graph neural network, quantification of risk contribution to each entity is realized, so that the risk management is more dynamic and real-time, and the transparency and response speed of the supervision process are guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of risk analysis, and in particular to an engineering supervision data management method and system based on deep learning. Background Art

[0002] Risk analysis is a crucial technique in the field of engineering supervision, focusing on identifying, evaluating and managing potential risks that may be encountered during the implementation of construction projects.

[0003] In existing engineering supervision, risk analysis often relies on static data processing methods and lacks the ability to dynamically analyze complex data, making it difficult to effectively predict and manage potential risks. In addition, existing technologies have shortcomings in terms of automation and intelligence, and the risk identification and assessment process is cumbersome and time-consuming, making it difficult to support efficient risk decision-making. These technical limitations are particularly prominent when dealing with large-scale or changeable engineering projects, often leading to delays in risk response measures and affecting the overall progress and quality assurance of the project. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings existing in the prior art and to propose an engineering supervision data management method and system based on deep learning.

[0005] In order to achieve the above object, the present invention adopts the following technical solution, a method for engineering supervision data management based on deep learning, comprising the following steps:

[0006] Acquire a project data set, obtain the project name, location, start time and supervision team information according to the project data set, generate a project metadata file, perform time series analysis on the project metadata file, identify key time nodes, and generate time monitoring data;

[0007] Applying a deep learning model to the time monitoring data, analyzing the supervision regulations, construction standards and safety regulations in the text, generating an entity recognition file, using the entity recognition file to refine the logical relationships and interactions between entities, and constructing an entity relationship model;

[0008] Use graph neural networks to learn features of entity relationship models, extract interaction features and risk-related attributes between entities, generate graph feature vectors, determine the contribution of each entity to engineering risk based on the graph feature vectors, and establish a risk assessment model;

[0009] The output of the risk assessment model is compared with the industry standard, the risk deviation of each entity is calculated, the supervision risk level is determined, and the knowledge graph is adjusted according to the supervision risk level to reflect the current supervision data and risk relationship to generate an updated supervision graph.

[0010] Preferably, the steps of acquiring the time monitoring data are:

[0011] Obtain project name, location, start time and supervision team information based on the engineering project dataset to generate preliminary project metadata archives;

[0012] Based on the preliminary project metadata archive, time series analysis was performed to identify key time nodes. The calculation formula is:

[0013]

[0014] Among them, T k Indicates the key time node, t i represents the date or time of the i-th time point, Z represents the total number of time points, and i is the index of the time point;

[0015] Based on the key time nodes, time monitoring data is generated.

[0016] Preferably, the steps of obtaining the entity identification file are:

[0017] Processing the time monitoring data using a deep learning model, extracting text content associated with supervision regulations, construction standards, and safety regulations, and generating a set of text feature vectors;

[0018] According to the text feature vector set, entity recognition is performed and the entity matching score is calculated. The calculation formula is:

[0019]

[0020] Among them, E m represents the entity matching score in the text, α k is the feature weight, s k is the score of the kth text feature, K is the total number of features;

[0021] Based on the entity matching scores, an entity recognition profile is organized and formed.

[0022] Preferably, the steps of obtaining the entity relationship model are:

[0023] Extracting identified entities from the entity recognition archive to form an entity list;

[0024] Analyzing the occurrence frequency and context dependency of each entity in the entity list to identify interaction patterns between entities;

[0025] According to the interaction pattern, an entity relationship model is constructed to map and record the logical relationships and interactions between entities.

[0026] Preferably, the steps of obtaining the graph feature vector are:

[0027] Inputting the entity relationship model into a graph neural network for feature learning;

[0028] Execute the forward propagation and feature extraction process of the graph neural network to calculate the neighborhood aggregation information of each entity node. The calculation formula is:

[0029]

[0030] Among them, F i represents the aggregate feature vector of node i, N(i) represents the set of adjacent nodes of node i, and f j is the feature vector of adjacent node j, W is the transformation matrix;

[0031] Based on the aggregated feature vector, the last layer of the network is used to extract the interaction features and risk-related attributes between entities, and generate a graph feature vector for each entity.

[0032] Preferably, the steps of obtaining the risk assessment model are:

[0033] Collect the graph feature vector of each entity, analyze the graph feature vector of each entity, and calculate the contribution of the graph feature vector to the engineering risk. The calculation formula is:

[0034]

[0035] Among them, R i represents the contribution of entity i to the engineering risk, g h is the importance coefficient of feature h, X ih is the value of entity i on feature h, and H is the total number of features;

[0036] A risk assessment model is established based on the contribution degree.

[0037] Preferably, the steps for obtaining the updated supervision map are:

[0038] Comparing the output of the risk assessment model with industry standards, quantifying the risk level of each entity, and calculating the risk deviation of each entity;

[0039] Determine the supervision risk level based on the risk bias of each entity;

[0040] Based on the supervision risk level, the existing knowledge graph is adjusted and updated, the relationships and interactions between entities and the connections between entities and risk factors are redefined to generate an updated supervision graph.

[0041] The present invention provides a project supervision data management system, comprising:

[0042] The data collection module collects project name, location, start time and supervision team information, and generates project metadata files; performs time series analysis on the project metadata files, identifies key time nodes, and obtains time monitoring data;

[0043] The entity relationship construction module uses the time monitoring data and applies a deep learning model to analyze supervision regulations, construction standards and safety regulations; generates an entity recognition file through text analysis, refines the logical relationship between entities through the entity recognition file, constructs an entity relationship model, and obtains the entity interaction feature model;

[0044] The risk assessment module, based on the entity interaction feature model, uses a graph neural network to learn the interaction features between entities, extracts risk-related attributes, constructs a graph feature vector, evaluates the risk contribution of each entity according to the graph feature vector, establishes a risk assessment model, compares the output of the risk assessment model with the industry standard, calculates the risk deviation of each entity, and determines the supervision risk level.

[0045] Compared with the prior art, the advantages and positive effects of the present invention are:

[0046] This invention improves the accuracy of data processing and risk prediction by integrating deep learning and graph neural network technology. Time series analysis helps to discover potential project risks and delays in advance. The deep learning model's efficient analysis of supervision regulations and safety regulations enables the complex logical relationships between data to be clearly mapped, thereby optimizing the decision support system. In addition, the graph feature vectors extracted by the graph neural network realize the quantification of the risk contribution of each entity, making risk management more dynamic and real-time, thereby ensuring the transparency and responsiveness of the supervision process. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the steps of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0049] See also Figure 1 The present invention provides a technical solution, a method for engineering supervision data management based on deep learning, comprising the following steps:

[0050] Select the engineering project data set, obtain the project name, location, start time and supervision team information, generate the project metadata file, perform time series analysis on the project metadata file, identify key time nodes, and generate time monitoring data;

[0051] Apply deep learning models to time monitoring data, analyze supervision regulations, construction standards, and safety regulations in the text, generate entity recognition files, use entity recognition files to refine the logical relationships and interactions between entities, and build entity relationship models;

[0052] Use graph neural networks to learn features of entity relationship models, extract interaction features and risk-related attributes between entities, generate graph feature vectors, determine the contribution of each entity to engineering risk based on the graph feature vectors, and establish a risk assessment model;

[0053] Compare the output of the risk assessment model with industry standards, calculate the risk deviation of each entity, determine the supervision risk level, adjust the knowledge graph according to the supervision risk level to reflect the current supervision data and risk relationship, and generate an updated supervision graph.

[0054] The steps to obtain time monitoring data are as follows:

[0055] Obtain project name, location, start time and supervision team information based on the engineering project dataset to generate preliminary project metadata archives;

[0056] Based on the preliminary project metadata archive, time series analysis was performed to identify key time nodes. The calculation formula is:

[0057]

[0058] Among them, T k Indicates the key time node, t i represents the date or time of the i-th time point, Z represents the total number of time points, and i is the index of the time point;

[0059] Generate time monitoring data based on key time nodes.

[0060] Specifically, relevant information is selected from the engineering project data set. First, the project name, location, start time and supervision team information are obtained through database query and data screening process. The data fields involved in this process include but are not limited to project number, geographic location code, planned start date and team member list. These data are efficiently retrieved through structured query language to ensure the accuracy and real-time nature of the acquired data. Then, these data are combined into a project metadata file to ensure that the metadata file contains complete project description information and configuration details of its supervision team, which is crucial for subsequent project supervision and data analysis, and a preliminary project metadata file is generated.

[0061] formula The benefit of is that by multiplying each point in the time series by its sequence position and averaging it, the time points can be effectively weighted so that the time points later in the sequence have a greater impact on the calculation of the key time nodes, which reflects the cumulative effect of time and the increase in criticality in the progress of the project; parameter T k It is obtained by calculating the weighted average of the time points in the time series and their sequence positions; the parameter t i It is obtained from the date or time of each time point recorded in chronological order from the actual project data; the parameter i is the index of the time point in the time series, obtained from the beginning of the project to the current timing position; the parameter Z is the total number of time points, obtained by counting the total number of data points contained in the entire time series.

[0062] Calculation process: Specific data: t i = [1,2,3,4,5], sequence position i = [1,2,3,4,5], total number Z = 5; the calculation process is as follows:

[0063]

[0064] The result shows that the key time node is 11, which means that in the time series, the average time node position calculated by weight is 11, and this value points to a key stage or milestone in the project.

[0065] Based on the identified key time nodes, time monitoring data is generated. The process includes extracting data from key time node archives and applying data conversion and processing technology to convert these time points into monitoring indicators. These indicators reflect the progress of the project from its initiation to the current state. Through continuous monitoring of these time points, the project status and the completion of key dependencies can be tracked in real time to ensure the real-time and effectiveness of supervision work, thereby forming time monitoring data archives. These archives will provide support for project management and ensure that supervision activities can be adjusted according to actual project progress.

[0066] The steps to obtain the entity recognition file are:

[0067] Use deep learning models to process time monitoring data, extract text content related to supervision regulations, construction standards, and safety regulations, and generate a set of text feature vectors;

[0068] According to the text feature vector set, entity recognition is performed and the entity matching score is calculated. The calculation formula is:

[0069]

[0070] Among them, E mrepresents the entity matching score in the text, α k is the feature weight, s k is the score of the kth text feature, K is the total number of features;

[0071] Based on the entity matching scores, the entity recognition profile is organized and formed.

[0072] Specifically, a deep learning model is used to process time monitoring data, and automatic content extraction is performed on text data in engineering supervision projects, such as supervision regulations, construction standards, and safety regulations. This process relies on the model's pre-trained word embedding and context capture capabilities. The deep learning model uses a multi-layer neural network structure to identify and parse relevant text content. The model first converts the text input into a numerical feature vector. This step is achieved through an embedding layer, which maps each word to a vector of a fixed size. The numerical value in the vector is obtained through training. These vectors will serve as input to the neural network. The network captures deep semantic relationships and patterns in the text through layer-by-layer transmission and nonlinear transformation of activation functions to generate a preliminary set of text feature vectors. The entire process is automatically executed and relies on a large amount of supervision-related text data for training to ensure that the model can accurately process and parse professional texts in the field of engineering supervision.

[0073] formula The calculation process:

[0074] First, define the feature weight α k is the importance index of each feature. These weights are learned from the previous training data set. The feature score s k is the actual observed value of each feature in the current document. For example, a feature may be the TF-IDF value of a word or phrase in a document. Calculate the square of the score of each feature to amplify the impact on the result, then multiply by the corresponding weight, do this for all features, and sum all items. Finally, take the square root to get the entity matching score E. m ,This score is used to quantify the degree of match between a text entity and supervision regulations, construction standards, and safety regulations.

[0075] There are three features with weights of 0.5, 0.3, and 0.2, and feature scores of 1, 2, and 3, respectively. The calculation process is as follows:

[0076]

[0077] The results show that the score E m =1.87 indicates that the entities obtained from the extracted text features have a high degree of match with the supervision regulations and construction standards. m Values ​​represent stronger associations and consistency, further guiding the generation of entity recognition profiles.

[0078] Based on the entity matching scores, the entity recognition archive is organized and formed. This process involves linking each identified entity with its corresponding regulations, standards or procedures, and recording in detail the relationship model of each entity and its interaction. This step not only includes the entity score E m Confirmation with existing entities in the database requires manual verification and adjustment to ensure the accuracy and practicality of the file. The generated entity identification file will be used for subsequent supervision review and quality control to ensure that all construction activities are carried out in accordance with relevant regulations and standards, providing accurate guidance and reference for the supervision team.

[0079] The steps to obtain the entity relationship model are:

[0080] Extracting identified entities from the entity recognition archive to form an entity list;

[0081] Analyze the frequency of occurrence and contextual dependencies of each entity in the entity list and identify the interaction patterns between entities;

[0082] Based on the interaction pattern, an entity relationship model is constructed to map and record the logical relationships and interactions between entities.

[0083] Specifically, the data collected from the entity identification archive, including text information of supervision regulations, construction standards and safety regulations, is used as the starting point for analyzing entities. The data is initially screened and classified in order to accurately identify each independent entity. Each entity is assigned corresponding labels and attributes according to the context in which it appears and its connection with other entities. Through the comprehensive arrangement of these data, a detailed entity list is formed, which contains various attributes of the entity such as definition, scope of application and associated regulations or standards. No preset weights or complex data processing methods are used in this process. Instead, a simple classification and matching process is used to ensure the purity of the data and the traceability of the processing process, forming a direct mapping of the original monitoring data. These preparatory works lay the foundation for subsequent in-depth analysis.

[0084] Based on the formed entity list, an in-depth analysis of the interaction patterns between entities was conducted. By calculating the frequency of each entity appearing in the document and its contextual association with other entities, the main interaction patterns were analyzed. For example, some entities are frequently linked together because they appear together in specific supervision regulations or safety standards. This analysis reveals the dependency and influence relationships between entities. It does not involve complex mathematical models or algorithms, but instead identifies these patterns through simple statistics and logical reasoning. These relationship patterns provide data support for understanding how entities interact with each other in actual supervision work and determine a preliminary model of entity relationships.

[0085] An entity relationship model is constructed based on the identified interaction patterns, which maps the logical relationships and interactions between entities in detail. The position and connection of each entity in the model are strictly defined, ensuring the accuracy and practicality of the model. This model directly reflects how entities play a role in project supervision. No preset calculation formula or weight adjustment is used in the model construction process. Instead, it relies on the entity interaction data obtained from the previous analysis. This method ensures the practical application value of the model and the ease of operation, forming a complete entity relationship model.

[0086] The steps to obtain the graph feature vector are:

[0087] Input the entity relationship model into the graph neural network for feature learning;

[0088] Execute the forward propagation and feature extraction process of the graph neural network to calculate the neighborhood aggregation information of each entity node. The calculation formula is:

[0089]

[0090] Among them, F i represents the aggregate feature vector of node i, N(i) represents the set of adjacent nodes of node i, and f j is the feature vector of adjacent node j, W is the transformation matrix;

[0091] Based on the aggregated feature vector, the last layer of the network is used to extract the interaction features and risk-related attributes between entities to generate a graph feature vector for each entity.

[0092] Specifically, the entity relationship model is input into the graph neural network to ensure that the network configuration correctly receives the entity relationship as input and prepares for feature learning. The key to this step is to ensure that all entity relationship data formats match the input requirements of the graph neural network. The correctness of the network configuration directly affects the subsequent learning efficiency and accuracy. The structural design of the network must be able to handle complex entity relationship data, including different types of nodes and diverse edge connections. The input of entity relationships not only involves data reading, but also includes pre-processing steps such as data cleaning and format conversion to ensure that the data quality meets the learning requirements. In addition, the integrity and accuracy of the entity relationship data are also crucial to the results of model training.

[0093] formula The benefit of is that it can effectively aggregate the neighborhood information of entity nodes. Through the application of the weight matrix W, it not only emphasizes the importance of each adjacent node feature, but also allows the network to automatically learn the optimal information aggregation method, so as to better capture the complex relationship between entities.

[0094] Derivation process: In the graph neural network, the feature vector F of each node i i By aggregating the feature vectors f of all its neighboring nodes j j To update, the transformation matrix W used in this aggregation process is learned by the network during training. For example, node i has 3 adjacent nodes, and the feature vector of each node is f1 = [0.2, 0.3], f2 = [0.4, 0.5], f3 = [0.1, 0.1]. The transformation matrix W is [1.0, 0.5; 0.5, 1.0], and the contribution of each adjacent node is calculated as:

[0095] [0.2,0.3]·[1.0,0.5;0.5,1.0]=[0.25,0.35]

[0096] [0.4,0.5]·[1.0,0.5;0.5,1.0]=[0.45,0.65]

[0097] [0.1,0.1]·[1.0,0.5; 0.5,1.0]=[0.15,0.15]

[0098] Aggregate all contributions:

[0099]

[0100] Therefore, the updated feature vector F of node i is i is [0.28, 0.38]. This result shows that the feature vector of node i is the weighted average of the features of its neighboring nodes, reflecting the influence of the neighboring nodes on i, which helps to propagate and fuse information in the graph and provides a basis for subsequent feature learning and data analysis.

[0101] Based on the aggregated feature vector, the last layer of the network is used to extract the interaction features and risk-related attributes between entities, and generate the graph feature vector of each entity. The key to this step is how to effectively utilize the learned feature vector F i To capture deeper interactions and potential risk connections between entities, the last layer of the network can perform a final transformation and abstraction on the features of each entity, thereby obtaining high-level feature vectors that can represent the complex relationships between entities. These feature vectors not only contain the entity’s own information, but also integrate the relationship information with other entities, so that the model can more comprehensively understand the dynamic relationships and risk factors between entities when making predictions or classifications. The resulting graph feature vector provides strong data support for deep learning and analysis of entity relationships, ensuring the accuracy and reliability of the model output.

[0102] The steps to obtain the risk assessment model are:

[0103] Collect the graph feature vector of each entity, analyze the graph feature vector of each entity, and calculate the contribution of the graph feature vector to the engineering risk. The calculation formula is:

[0104]

[0105] Among them, R i represents the contribution of entity i to the engineering risk, g h is the importance coefficient of feature h, X ih is the value of entity i on feature h, and H is the total number of features;

[0106] According to the contribution degree, a risk assessment model is established.

[0107] Specifically, collect the graph feature vectors of each entity and ensure that these vectors contain all relevant feature information obtained from the previous graph neural network analysis. This step first requires extracting all data points related to each entity from the dataset, including but not limited to the entity's interaction data, historical behavior data, and possible risk indicator data. Through feature engineering and dimensionality reduction, representative feature vectors are extracted from complex data. These vectors will serve as input for graph neural network learning, dynamically evaluate the risk contribution of each entity, and provide accurate input for subsequent risk assessment models.

[0108] formula The benefit is that through the weight parameter g h , which can dynamically weight each feature according to its actual importance, ensuring the accuracy and adaptability of risk assessment.

[0109] Calculation process: The entity has three feature scores, namely X i1 =40,X i2 =30,X i3 =20, and the importance weights of the features are set to g1=0.5, g2=0.3, g3=0.2 respectively; then:

[0110] R i =(0.5×40)+(0.3×30)+(0.2×20)

[0111] R i =20+9+4

[0112] R i =33

[0113] The result shows that the total risk contribution of the entity is 33. The higher the value, the greater the contribution of the entity to the project risk and the more it needs to be monitored and managed.

[0114] Based on the calculated risk contribution of each entity, a risk assessment model is established, which can output the risk level of each entity. This step involves converting the previously calculated risk contribution into an actual risk level by setting different risk thresholds, such as low risk, medium risk and high risk. Each threshold corresponds to a different risk level. These levels will help project managers develop corresponding risk response strategies, such as strengthening supervision, adjusting resource allocation or implementing preventive measures, to ensure that engineering risks are effectively managed and reduced.

[0115] The steps to obtain the updated supervision map are as follows:

[0116] Compare the output of the risk assessment model with industry standards, quantify the risk level of each entity, and calculate the risk deviation of each entity;

[0117] Determine the supervision risk level based on the risk bias of each entity;

[0118] Based on the supervision risk level, the existing knowledge graph is adjusted and updated, and the relationships and interactions between entities and the connections between entities and risk factors are redefined to generate an updated supervision graph.

[0119] Specifically, when comparing the output of the risk assessment model with the industry standard, it is first necessary to extract specific risk level standards from the industry standard documents, which include the definitions and quantitative indicators of various risks. The data output by the risk assessment model is formatted through professional software, and then compared with the risk indicators of the industry standard one by one. The entity risk deviation is calculated by comparing the difference between the assessed risk value of each entity and the corresponding level of the industry standard. The risk level of each entity is calibrated by the specific numerical difference. For example, if the assessed value of an entity exceeds the safety range of the industry standard, the risk level of the entity will be increased. This calculation process requires that all data processing steps are strictly performed in accordance with industry specifications to ensure the accuracy and reliability of each assessment. Finally, risk deviation data for each entity is generated.

[0120] For the risk deviation data of each entity, the next step is to determine the supervision risk level. In this process, it is necessary to follow the pre-defined risk level classification standards, such as classifying entities with smaller risk deviations as low risk and entities with larger risk deviations as high risk. This classification process involves a detailed analysis and classification of the risk deviation data, and the classification of entities is achieved through mathematical models such as quantile analysis or cluster analysis. After the classification is completed, the system will adjust the subsequent risk management strategies and supervision activities based on the classification results to ensure that entities of each risk level can be properly managed and monitored. The core of this step is to accurately convert risk deviation data into actual and operational risk levels.

[0121] Once the supervision risk level is determined, the next step is to adjust and update the knowledge graph. In this process, the core task is to redefine the relationships and interactions between entities based on the new risk level. This includes adjusting the weights of nodes and edges in the graph, and updating the association rules between entities. These adjustments are based on the latest risk data obtained from the risk management database. Graph editing tools are used to reconstruct the connections between entities according to predetermined rules, such as priority rules based on risk levels. At the same time, the links between entities and risk factors need to be updated to ensure that the graph can accurately reflect the current supervision data and risk relationships. Finally, after this step is completed, a fully updated supervision graph is generated, which will be used to guide subsequent supervision activities and risk decisions.

[0122] The present invention provides a project supervision data management system, comprising:

[0123] The data collection module collects project name, location, start time and supervision team information, and generates project metadata archives; performs time series analysis on the project metadata archives, identifies key time nodes, and obtains time monitoring data;

[0124] The entity relationship construction module uses time monitoring data and deep learning models to analyze supervision regulations, construction standards, and safety regulations. It generates entity recognition files through text analysis, refines the logical relationship between entities through the entity recognition files, constructs an entity relationship model, and obtains an entity interaction feature model.

[0125] The risk assessment module is based on the entity interaction feature model and uses graph neural networks to learn the interaction features between entities, extract risk-related attributes, construct graph feature vectors, evaluate the risk contribution of each entity based on the graph feature vectors, establish a risk assessment model, compare the output of the risk assessment model with industry standards, calculate the risk deviation of each entity, and determine the supervision risk level.

[0126] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for engineering supervision data management based on deep learning, characterized in that: The following steps are involved: Acquire a project data set, obtain the project name, location, start time and supervision team information according to the project data set, generate a project metadata file, perform time series analysis on the project metadata file, identify key time nodes, and generate time monitoring data; Applying a deep learning model to the time monitoring data, analyzing the supervision regulations, construction standards and safety regulations in the text, generating an entity recognition file, using the entity recognition file to refine the logical relationships and interactions between entities, and constructing an entity relationship model; Use graph neural networks to learn features of entity relationship models, extract interaction features and risk-related attributes between entities, generate graph feature vectors, determine the contribution of each entity to engineering risk based on the graph feature vectors, and establish a risk assessment model; The output of the risk assessment model is compared with the industry standard, the risk deviation of each entity is calculated, the supervision risk level is determined, and the knowledge graph is adjusted according to the supervision risk level to reflect the current supervision data and risk relationship to generate an updated supervision graph.

2. The engineering supervision data management method based on deep learning according to claim 1 is characterized in that: The steps for obtaining the time monitoring data are as follows: Obtain project name, location, start time and supervision team information based on the engineering project dataset to generate preliminary project metadata archives; Based on the preliminary project metadata archive, time series analysis was performed to identify key time nodes. The calculation formula is: Among them, T k Indicates the key time node, t i represents the date or time of the i-th time point, Z represents the total number of time points, and i is the index of the time point; Based on the key time nodes, time monitoring data is generated.

3. The engineering supervision data management method based on deep learning according to claim 1 is characterized in that: The steps for obtaining the entity identification file are: Processing the time monitoring data using a deep learning model, extracting text content associated with supervision regulations, construction standards, and safety regulations, and generating a set of text feature vectors; According to the text feature vector set, entity recognition is performed and the entity matching score is calculated. The calculation formula is: Among them, E m represents the entity matching score in the text, α k is the feature weight, s k is the score of the kth text feature, K is the total number of features; Based on the entity matching scores, an entity recognition profile is organized and formed.

4. The engineering supervision data management method based on deep learning according to claim 1 is characterized in that: The steps for obtaining the entity relationship model are: Extracting identified entities from the entity recognition archive to form an entity list; Analyzing the occurrence frequency and context dependency of each entity in the entity list to identify interaction patterns between entities; According to the interaction pattern, an entity relationship model is constructed to map and record the logical relationships and interactions between entities.

5. The engineering supervision data management method based on deep learning according to claim 1 is characterized in that: The steps for obtaining the graph feature vector are as follows: Inputting the entity relationship model into a graph neural network for feature learning; Execute the forward propagation and feature extraction process of the graph neural network to calculate the neighborhood aggregation information of each entity node. The calculation formula is: Among them, F i represents the aggregate feature vector of node i, N(i) represents the set of adjacent nodes of node i, and f j is the feature vector of adjacent node j, W is the transformation matrix; Based on the aggregated feature vector, the last layer of the network is used to extract the interaction features and risk-related attributes between entities, and generate a graph feature vector for each entity.

6. The engineering supervision data management method based on deep learning according to claim 1 is characterized in that: The steps for obtaining the risk assessment model are: Collect the graph feature vector of each entity, analyze the graph feature vector of each entity, and calculate the contribution of the graph feature vector to the engineering risk. The calculation formula is: Among them, R i Indicates the contribution of entity i to the engineering risk, g h is the importance coefficient of feature h, X ih is the value of entity i on feature h, and H is the total number of features; A risk assessment model is established based on the contribution degree.

7. The engineering supervision data management method based on deep learning according to claim 1 is characterized in that: The steps for obtaining the updated supervision map are as follows: Comparing the output of the risk assessment model with industry standards, quantifying the risk level of each entity, and calculating the risk deviation of each entity; Determine the supervision risk level based on the risk bias of each entity described; Based on the supervision risk level, the existing knowledge graph is adjusted and updated, the relationships and interactions between entities and the connections between entities and risk factors are redefined to generate an updated supervision graph.

8. An engineering supervision data management system according to the engineering supervision data management method based on deep learning according to any one of claims 1 to 7, characterized in that: include: The data collection module collects project name, location, start time and supervision team information, and generates project metadata archives; Performing time series analysis on the project metadata archive to identify key time nodes and obtain time monitoring data; An entity relationship building module utilizes the time monitoring data and applies a deep learning model to analyze supervision regulations, construction standards, and safety regulations; Generate entity recognition files through text analysis, refine logical relationships between entities through entity recognition files, build entity relationship models, and obtain the entity interaction feature model; The risk assessment module, based on the entity interaction feature model, uses a graph neural network to learn the interaction features between entities, extracts risk-related attributes, constructs a graph feature vector, evaluates the risk contribution of each entity according to the graph feature vector, establishes a risk assessment model, compares the output of the risk assessment model with the industry standard, calculates the risk deviation of each entity, and determines the supervision risk level.

Images