A method and system for intelligent identification of dangerous areas in engineering construction based on generative large model

Abstract

The present invention relates to the field of construction site safety technology, and discloses a method and system for intelligently identifying dangerous areas in engineering construction based on a generative large model, including: obtaining real-time monitoring data and records of safety events in various areas of the construction site, and constructing a correlation matrix between safety events and monitoring data; for the correlation matrix, obtaining the correlation pattern and law between safety events and monitoring data, and determining the key monitoring data set that affects the safety level; regularly backtesting and verifying the safety level and control measures, and by comparing and analyzing the relevant data and actual effect data of the control measures under the historical safety level, using an incremental learning algorithm to iteratively optimize the setting of the safety level. The present invention realizes the intelligence and automation of construction site safety management, effectively reduces the risk of safety accidents, and improves the efficiency and accuracy of safety management.

Description

Technical Field

[0001] The present invention relates to the field of construction site safety technology, and in particular to a method and system for intelligently identifying dangerous areas in engineering construction based on a generative large model. Background Art

[0002] Under the background of my country's rapid economic development, infrastructure construction, especially high-speed rail construction, is becoming an important force to promote economic growth. As infrastructure projects such as high-speed rail are not only widely carried out in China, they are also gradually extending to complex mountainous environments. These projects often face the challenges of natural conditions such as complex terrain and changeable climate, and also need to deal with various safety risks and hidden dangers during the construction process. Safety in production, prevention first. Pre-hazard source identification is an important part of construction safety management. At present, hazard source identification still relies on human identification, which is limited to differences in personal ability and experience, and the level of hazard source identification varies greatly. In addition, similar engineering projects have similar hazard sources, so hazard source identification for similar projects is a repetitive task, resulting in a waste of labor costs.

[0003] On the other hand, with the development of global science and technology, artificial intelligence has become a major driving force for improving productivity. Countries around the world have accelerated their strategic layout, strategic layout, capital investment, technology research and development, and field application of artificial intelligence. The development of artificial intelligence has become a key competitive field for countries around the world. With the proposal of Digital China, my country's artificial intelligence has also developed rapidly. With the development of information technologies such as big data, cloud computing, the Internet, and the Internet of Things, artificial intelligence technology represented by deep neural networks has developed rapidly, and the gap between science and application in the field of artificial intelligence is being broken. General artificial intelligence will be an important source of power for global economic development, and the application of artificial intelligence in specialized vertical fields is also a necessary supplement.

[0004] The present invention forms a generative industry big model (i.e., a vertical field big model) that intelligently identifies safety risks and hidden dangers in construction projects, empowers business in multiple aspects such as planning and design, construction, operation and maintenance management, integrates the project management business platform, and realizes the intelligent and digital management of engineering project safety. The use of generative big models to intelligently identify and warn of safety risks and hidden dangers that may arise during the construction process will help to take corresponding measures in advance to prevent accidents. In addition, the integration of the project management business platform will make the collaboration between various departments closer and more efficient, and further improve the management efficiency of the entire project. Summary of the invention

[0005] The purpose of the present invention is to overcome one or more of the above-mentioned existing technical problems and provide a method for intelligently identifying dangerous areas in engineering construction based on a generative large model.

[0006] To achieve the above-mentioned purpose, the present invention provides a method for intelligently identifying dangerous areas in engineering construction based on a generative large model, comprising:

[0007] A 3D model of the project is established based on the BIM model, 3D satellite GIS terrain and flight laser point cloud model. The 3D model is segmented, and real-time monitoring data and safety event records of each area of ​​the construction site are obtained based on the segmented 3D model and drone aerial photography, and a correlation matrix between safety events and monitoring data is constructed;

[0008] Based on the correlation matrix between security events and monitoring data, the correlation patterns and rules between security events and monitoring data are obtained, and the key monitoring data set that affects the security level is determined;

[0009] Based on the key monitoring data set, the changing trends and abnormal conditions of each monitoring data are analyzed in real time. If the monitoring data exceeds the preset threshold or shows abnormal fluctuations, it is judged that there are safety hazards in the area and the safety level of the area is adjusted dynamically;

[0010] According to the security level of each area, the predefined security control measures library is automatically matched to obtain the control measures of the corresponding security level of each area, and the control measures are automatically sent to the security control system through the rule engine to trigger the corresponding control actions;

[0011] Feedback the security level of each area and the execution of control actions to the project management business platform in real time, generate security situation maps and control effect evaluation reports, and provide auxiliary decision support for managers;

[0012] Backtest and verify the security level and control measures. By comparing and analyzing the relevant data and actual effect data of the control measures under the historical security level, an incremental learning algorithm is used to iteratively optimize the security level setting.

[0013] According to one aspect of the present invention, real-time monitoring data and safety event records of each area of ​​the construction site are obtained based on the segmented three-dimensional model, and data cleaning is performed on the acquired real-time monitoring data of each area of ​​the construction site to remove missing values ​​and abnormal values;

[0014] Perform feature extraction based on the real-time monitoring data of each area of ​​the construction site after cleaning, and extract features related to safety events as well as operational features such as equipment operating status and personnel behavior;

[0015] Using a feature selection algorithm, a feature subset with the highest correlation with the security event in the security event record is selected from the extracted features to reduce the feature dimension;

[0016] For the selected feature subset, the association rule mining algorithm is used to discover the association rules between security events and each feature, and the correlation between the feature and the event is obtained;

[0017] According to the mined association rules, a correlation matrix between security events and monitoring data is constructed. Each element in the matrix represents the correlation between the feature and the event.

[0018] Perform cluster analysis on the correlation matrix to aggregate features and events with similar correlations and identify different security event patterns;

[0019] According to the clustering results, a security incident prediction model is established to predict the security incidents that occur and issue early warnings through real-time monitoring data.

[0020] According to one aspect of the present invention, a correlation matrix between security events and monitoring data is constructed, and the elements in the matrix represent the correlation between the two;

[0021] Adopting association rule mining algorithm, based on support and confidence thresholds, the frequent item sets and association rules between security events and monitoring data are mined from the association matrix;

[0022] By mining the association rules, we can summarize the association patterns between security events and monitoring data, and determine the security events caused by abnormal monitoring data.

[0023] Evaluate the mined association rules, calculate the support, confidence and lift of each rule, and sort and filter the association rules;

[0024] The monitoring data involved in the screened strong association rules are used as the preliminary key monitoring data, and the final key monitoring data set is determined from them in combination with expert knowledge and practical experience;

[0025] Based on the determined key monitoring data set, a real-time monitoring and early warning mechanism is established to warn of possible security incidents when abnormalities are detected in key monitoring data;

[0026] Continuously collect security events and monitoring data, update association matrices and mine association rules, and dynamically adjust key monitoring data sets.

[0027] According to one aspect of the present invention, a key monitoring data set is obtained, and for each monitoring data, a time series data model is constructed to record data values ​​at different time points;

[0028] Perform trend prediction and anomaly detection on the time series data of each monitoring data, and judge whether the monitoring data exceeds the normal range according to the preset threshold. If it exceeds the threshold, mark the time point as an abnormal point;

[0029] Further analyze the abnormal situation of the monitoring data to determine whether the abnormal point is random noise or systematic abnormality. If the abnormal point of the monitoring data is systematic abnormality, it is determined that there is a safety hazard in the area. According to the severity of the abnormal situation, the safety level of the area is dynamically adjusted;

[0030] Continuously perform time series analysis and anomaly detection in real-time data streams, process newly generated monitoring data in real time, and visualize anomalies and changes in security levels of each monitoring data.

[0031] According to one aspect of the present invention, real-time monitoring data of various areas of the construction site are obtained based on drone aerial photography, the data are preprocessed, key characteristic parameters are extracted, and the safety level is calculated based on the key characteristic parameters to obtain the safety level of each area;

[0032] According to the security level, a set of control measures corresponding to the current security level is automatically matched in the predefined security control measures library;

[0033] The matched control measures are converted into executable control instructions, and logical judgment is performed through the pre-configured rule engine. If the rule conditions are met, the control instructions are automatically sent to the security control system;

[0034] After receiving the control command issued, the safety control system parses the command content and extracts the specific control actions that need to be triggered;

[0035] If the control action is to issue a warning signal, the preset signal generator is called to generate warning signals such as sound and light, and the warning information is pushed to the terminal of relevant personnel to prompt the dangerous situation;

[0036] While executing control actions, the regional security status, control measures, control action type and execution results are associated and stored to form a security control log.

[0037] According to one aspect of the present invention, the security level of each area and the execution status data of the control actions are obtained, and the data is transmitted in real time to the database of the visual project management business platform;

[0038] Analyze and process the security level and execution status data of control actions in the database to obtain security situation image data, and analyze the security situation image data to obtain the control effect evaluation result;

[0039] Output security situation image data and control effect evaluation results to generate a control effect evaluation report;

[0040] Determine whether the management and control effect evaluation result meets the preset auxiliary decision threshold. If so, push the management and control effect evaluation report to the management terminal;

[0041] If the control effect evaluation result does not meet the preset auxiliary decision threshold, the preset control measures are optimized, the execution data of the control actions are analyzed, and the optimized control measures are obtained;

[0042] The optimized control measures will be issued to the control systems of various regions, the control measures of each region will be adjusted, and the latest security level and execution status data of control actions will continue to be obtained to form a closed-loop control.

[0043] According to one aspect of the present invention, relevant data and actual effect data of control measures under historical security levels are obtained, and data cleaning is performed for each data to remove noise data and extract key features;

[0044] Compare and analyze the relevant data of the control measures under the processed historical safety level and the actual effect data, calculate the difference and correlation of the two sets of data, and obtain the difference data and correlation data;

[0045] Based on the difference data and correlation data, the effectiveness and accuracy of the current security level and control actions are judged. If the data is lower than the preset threshold, the optimization process of setting the security level is triggered;

[0046] Adopt incremental learning algorithm to fine-tune parameters and optimize structure of security level to obtain optimized security level;

[0047] Apply the optimized security level to the control action, regenerate the control action through the security control rule engine, and send it to the security control system;

[0048] Continuously monitor real-time data during system operation, use anomaly detection algorithms to identify potential security threats and abnormal behaviors, and form a security management and control knowledge base.

[0049] To achieve the above object, the present invention provides an intelligent identification system for engineering construction hazardous areas based on a generative large model, comprising:

[0050] Association matrix construction module: Build a 3D model of the project based on the BIM model, 3D satellite GIS terrain and flight laser point cloud model, segment the 3D model, obtain real-time monitoring data and safety incident records of each area of ​​the construction site based on the segmented 3D model and drone aerial photography, and build an association matrix between safety incidents and monitoring data;

[0051] Key monitoring data set acquisition module: Based on the association matrix between security events and monitoring data, the association pattern and law between security events and monitoring data are obtained, and the key monitoring data set that affects the security level is determined;

[0052] Security level adjustment module: Based on the key monitoring data set, it analyzes the changing trends and abnormal conditions of each monitoring data in real time. If the monitoring data exceeds the preset threshold or shows abnormal fluctuations, it is judged that there are security risks in the area and the security level of the area is adjusted dynamically.

[0053] Control action setting mode: According to the security level of each area, the predefined security control measures library is automatically matched to obtain the control measures of the security level of each area, and the control measures are automatically sent to the security control system through the rule engine to trigger the corresponding control action;

[0054] Platform feedback module: Feedback the security level of each area and the execution of control actions to the project management business platform in real time, generate security situation maps and control effect evaluation reports, and provide auxiliary decision support for managers;

[0055] Security level update module: backtest and verify the security level and control measures, compare and analyze the relevant data and actual effect data of the control measures under the historical security level, and use the incremental learning algorithm to iteratively optimize the security level setting.

[0056] Based on this, the beneficial effects of the present invention are: by constructing an association matrix between safety events and monitoring data, an association rule mining algorithm is used to identify key monitoring data, and time series analysis and anomaly detection algorithms are used to evaluate the safety level of each area in real time. According to the evaluation results, the present invention automatically matches and issues corresponding safety management and control measures. At the same time, the present invention visualizes the evaluation results and control effects to provide decision support for managers. Through regular backtesting and incremental learning, the present invention continuously optimizes the evaluation model and control actions to improve the overall safety management level. This method realizes the intelligent and automated management of construction site safety, effectively reduces the risk of safety accidents, and improves the efficiency and accuracy of safety management;

[0057] The use of AI + safety risk management can significantly reduce the workload and pressure of safety managers, realize intelligent identification of dangerous sources, improve safety management efficiency and risk identification accuracy, strengthen safety management and control in the process, reduce safety risks, and avoid major safety accidents. It can reduce safety workload, reduce the number of safety personnel, avoid safety accidents, and increase economic and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a flow chart of a method for intelligently identifying dangerous areas in engineering construction based on a generative large model according to an exemplary embodiment;

[0059] Figure 2 The present invention is a flow chart of an intelligent identification system for dangerous areas in engineering construction based on a generative large model according to an exemplary embodiment. DETAILED DESCRIPTION

[0060] The present invention will now be discussed with reference to exemplary embodiments. It should be understood that the embodiments discussed are only for enabling those skilled in the art to better understand and thereby implement the present invention, rather than implying any limitation on the scope of the present invention.

[0061] As used herein, the term “including” and variations thereof are to be interpreted as open-ended terms meaning “including but not limited to.” The term “based on” is to be interpreted as “based, at least in part, on,” and the terms “one embodiment” and “an embodiment” are to be interpreted as “at least one embodiment.”

[0062] According to one embodiment of the present invention, Figure 1 is a flow chart of a method for intelligently identifying dangerous areas in engineering construction based on a generative large model according to an exemplary embodiment, such as Figure 1 As shown, to achieve the above purpose, the present invention provides an intelligent identification method for engineering construction hazardous areas based on a generative large model, comprising:

[0063] A 3D model of the project is established based on the BIM model, 3D satellite GIS terrain and flight laser point cloud model. The 3D model is segmented, and real-time monitoring data and safety event records of each area of ​​the construction site are obtained based on the segmented 3D model and drone aerial photography, and a correlation matrix between safety events and monitoring data is constructed;

[0064] Based on the correlation matrix between security events and monitoring data, the correlation patterns and rules between security events and monitoring data are obtained, and the key monitoring data set that affects the security level is determined;

[0065] Based on the key monitoring data set, the changing trends and abnormal conditions of each monitoring data are analyzed in real time. If the monitoring data exceeds the preset threshold or shows abnormal fluctuations, it is judged that there are safety hazards in the area and the safety level of the area is adjusted dynamically;

[0066] According to the security level of each area, the predefined security control measures library is automatically matched to obtain the control measures of the corresponding security level of each area, and the control measures are automatically sent to the security control system through the rule engine to trigger the corresponding control actions;

[0067] Feedback the security level of each area and the execution of control actions to the project management business platform in real time, generate security situation maps and control effect evaluation reports, and provide auxiliary decision support for managers;

[0068] Backtest and verify the security level and control measures. By comparing and analyzing the relevant data and actual effect data of the control measures under the historical security level, an incremental learning algorithm is used to iteratively optimize the security level setting.

[0069] According to one embodiment of the present invention, real-time monitoring data and safety event records of each area of ​​the construction site are obtained based on the segmented three-dimensional model, and data cleaning is performed on the acquired real-time monitoring data of each area of ​​the construction site to remove missing values ​​and abnormal values;

[0070] Perform feature extraction based on the real-time monitoring data of each area of ​​the construction site after cleaning, and extract features related to safety events as well as operational features such as equipment operating status and personnel behavior;

[0071] Using a feature selection algorithm, a feature subset with the highest correlation with the security event in the security event record is selected from the extracted features to reduce the feature dimension;

[0072] For the selected feature subset, the association rule mining algorithm is used to discover the association rules between security events and each feature, and the correlation between the feature and the event is obtained;

[0073] According to the mined association rules, a correlation matrix between security events and monitoring data is constructed. Each element in the matrix represents the correlation between the feature and the event.

[0074] Perform cluster analysis on the correlation matrix to aggregate features and events with similar correlations and identify different security event patterns;

[0075] According to the clustering results, a security incident prediction model is established to predict the security incidents that occur and issue early warnings through real-time monitoring data.

[0076] According to one embodiment of the present invention, a correlation matrix between security events and monitoring data is constructed, and the elements in the matrix represent the correlation between the two;

[0077] Adopting association rule mining algorithm, based on support and confidence thresholds, the frequent item sets and association rules between security events and monitoring data are mined from the association matrix;

[0078] By mining the association rules, we can summarize the association patterns between security events and monitoring data, and determine the security events caused by abnormal monitoring data.

[0079] Evaluate the mined association rules, calculate the support, confidence and lift of each rule, and sort and filter the association rules;

[0080] The monitoring data involved in the screened strong association rules are used as the preliminary key monitoring data, and the final key monitoring data set is determined from them in combination with expert knowledge and practical experience;

[0081] Based on the determined key monitoring data set, a real-time monitoring and early warning mechanism is established to warn of possible security incidents when abnormalities are detected in key monitoring data;

[0082] Continuously collect security events and monitoring data, update association matrices and mine association rules, and dynamically adjust key monitoring data sets.

[0083] According to one embodiment of the present invention, a key monitoring data set is obtained, and for each monitoring data, a time series data model is constructed to record data values ​​at different time points;

[0084] Perform trend prediction and anomaly detection on the time series data of each monitoring data, and judge whether the monitoring data exceeds the normal range according to the preset threshold. If it exceeds the threshold, mark the time point as an abnormal point;

[0085] Further analyze the abnormal situation of the monitoring data to determine whether the abnormal point is random noise or systematic abnormality. If the abnormal point of the monitoring data is systematic abnormality, it is determined that there is a safety hazard in the area. According to the severity of the abnormal situation, the safety level of the area is dynamically adjusted;

[0086] Continuously perform time series analysis and anomaly detection in real-time data streams, process newly generated monitoring data in real time, and visualize anomalies and changes in security levels of each monitoring data.

[0087] According to one embodiment of the present invention, real-time monitoring data of various areas of the construction site are obtained based on drone aerial photography, the data are preprocessed, key characteristic parameters are extracted, and the safety level is calculated based on the key characteristic parameters to obtain the safety level of each area;

[0088] According to the security level, a set of control measures corresponding to the current security level is automatically matched in the predefined security control measures library;

[0089] The matched control measures are converted into executable control instructions, and logical judgment is performed through the pre-configured rule engine. If the rule conditions are met, the control instructions are automatically sent to the security control system;

[0090] After receiving the control command issued, the safety control system parses the command content and extracts the specific control actions that need to be triggered;

[0091] If the control action is to issue a warning signal, the preset signal generator is called to generate warning signals such as sound and light, and the warning information is pushed to the terminal of relevant personnel to prompt the dangerous situation;

[0092] While executing control actions, the regional security status, control measures, control action type and execution results are associated and stored to form a security control log.

[0093] According to one embodiment of the present invention, the security level of each area and the execution status data of the control actions are obtained, and the data is transmitted in real time to the database of the visual project management business platform;

[0094] Analyze and process the security level and execution status data of control actions in the database to obtain security situation image data, and analyze the security situation image data to obtain the control effect evaluation result;

[0095] Output security situation image data and control effect evaluation results to generate a control effect evaluation report;

[0096] Determine whether the management and control effect evaluation result meets the preset auxiliary decision threshold. If so, push the management and control effect evaluation report to the management terminal;

[0097] If the control effect evaluation result does not meet the preset auxiliary decision threshold, the preset control measures are optimized, the execution data of the control actions are analyzed, and the optimized control measures are obtained;

[0098] The optimized control measures will be issued to the control systems of various regions, the control measures of each region will be adjusted, and the latest security level and execution status data of control actions will continue to be obtained to form a closed-loop control.

[0099] According to one embodiment of the present invention, relevant data and actual effect data of control measures under historical security levels are obtained, and data cleaning is performed on each data to remove noise data and extract key features;

[0100] Compare and analyze the relevant data of the control measures under the processed historical safety level and the actual effect data, calculate the difference and correlation of the two sets of data, and obtain the difference data and correlation data;

[0101] Based on the difference data and correlation data, the effectiveness and accuracy of the current security level and control actions are judged. If the data is lower than the preset threshold, the optimization process of setting the security level is triggered;

[0102] Adopt incremental learning algorithm to fine-tune parameters and optimize structure of security level to obtain optimized security level;

[0103] Apply the optimized security level to the control action, regenerate the control action through the security control rule engine, and send it to the security control system;

[0104] Continuously monitor real-time data during system operation, use anomaly detection algorithms to identify potential security threats and abnormal behaviors, and form a security management and control knowledge base.

[0105] According to one embodiment of the present invention, based on a pre-established safety incident prediction model combined with a three-dimensional model, before the project starts, each section of the project can be simulated based on the segmented three-dimensional model, and the project type and the risk characteristics of the project of this type can be predicted in advance through the safety incident prediction model, and the safety incidents that may occur in the risk area project with time and space characteristics can be identified, and the warning content of the safety incident, how to deal with the incident and other related content are all marked in the three-dimensional model. Construction personnel and relevant responsible personnel can learn all safety warning information about the construction project in advance by viewing the three-dimensional model, and determine the spatial position of the construction personnel through wearable terminal devices (such as helmets and bracelets with positioning and voice functions) during construction. In combination with the current construction progress, the construction personnel are promptly reminded of the risks, precautions, preventive measures, etc. faced by the construction personnel. In combination with the project management business platform, the progress management function is used to provide the project progress time node, intelligently analyze the safety risks and precautions faced by the current construction progress, and remind the person in charge and construction personnel.

[0106] According to one embodiment of the present invention, for open long line projects, the content can be fed back to the safety event prediction model through drone aerial photography to assist the safety event prediction model, and the model can be updated by adding new parameters so that it can intelligently identify safety hazards and automatically notify the responsible person to make rectifications through the project management business platform.

[0107] According to one embodiment of the present invention, when a safety manager discovers a safety hazard on site and takes a photo (including location information) and submits it to the project management business platform, the safety event prediction model intelligently identifies the current hazard type and adds the safety event to the training set, so that the model can be updated to better identify the safety accident, the person responsible, the measures to be taken, the rectification deadline, and automatically notify the person responsible.

[0108] According to one embodiment of the present invention, a 100×50 security event and monitoring data association matrix is ​​constructed, and the matrix element value 0-1 represents the association degree. Set the support threshold to 05, the confidence threshold to 8, and use the Apriori algorithm to mine 20 frequent item sets and 10 association rules with a confidence of 9. Calculate the support, confidence, and lift of the rules, and screen out 5 strong association rules with a confidence greater than 85. The 10 monitoring data involved are used as preliminary data, and then the final 5 key data are determined by expert review. Monitor these 5 data in real time, trigger an early warning when an abnormality is detected, and update the matrix and rules at any time.

[0109] According to one embodiment of the present invention, when obtaining a key monitoring data set, 10 key data such as temperature, humidity, and pressure can be selected. For each monitoring data, a time series data model is constructed using Python's Pandas library, and data values ​​are recorded at intervals of 1 minute. The ARIMA model is used to predict the trend of time series data, and the prediction step is set to 10, that is, the data trend of the next 10 minutes is predicted. At the same time, the abnormal threshold is set using the 3-sigma principle, that is, the value exceeding the mean ± 3 times the standard deviation is judged as abnormal. The abnormal points are clustered and analyzed by the isolation forest algorithm. If the number of abnormal points exceeds 5% of the total data points, it is judged as a systematic abnormality. According to the number and degree of abnormal data, the regional safety level is dynamically adjusted, which is divided into three levels: safety, warning, and danger. Spark Streaming is used to process the newly generated monitoring data in real time, and the safety level is updated every 30 seconds. Finally, ECharts is used for visual display, and the historical trend of each data is displayed with a line chart, and the current safety level is displayed with a dashboard, which is convenient for safety managers to monitor in real time.

[0110] According to one embodiment of the present invention, environmental parameters such as temperature, humidity, pressure, and equipment operating status data are collected in real time, and data collection is performed every 5 seconds. The collected raw data is subjected to a digital filtering algorithm to remove noise interference, and 10 key characteristic parameters such as average temperature, maximum pressure, and equipment vibration frequency are extracted. The extracted characteristic parameters are input into a pre-trained neural network model for security level setting. The model can map the multi-dimensional characteristic parameters to security levels of 1-5. The higher the level, the greater the security risk. When the security level of a certain area is assessed to be level 4 (high risk), the system automatically matches the corresponding measures in the predefined management and control measures library. The measures are converted into control instructions and conditional judgment is performed through the rule engine. If the current time is the working period and the number of people in the area exceeds 10, the warning signal is triggered. After receiving the warning instruction, the security control system calls the sound and light alarm to send out an 85dB siren and a red flash signal, and at the same time pushes a text warning message to the mobile phone of the person in charge of the area through the instant messaging system. After the control action is completed, the system stores the current time, area number, security level, control measures, execution results and other information in a fixed format in the MySQL database to form a complete security control log, which provides data support for subsequent statistical analysis and algorithm optimization. Through a series of processes such as real-time monitoring, intelligent evaluation, automatic control, and log traces, the system can effectively prevent and control regional safety risks and improve the overall safety management level.

[0111] According to one embodiment of the present invention, the security level and the execution status data of the control action are collected in real time, such as the security level of area A is level 3, the patrol frequency is once per hour; the security level of area B is level 2, and the patrol frequency is once every 2 hours. These data are transmitted to the MongoDB database of the visual project management business platform in real time through the 5G network. The visual project management business platform uses the YOLO algorithm based on convolutional neural network to analyze and process the data in the database, extract key features, generate a security situation heat map, and intuitively display the distribution of security risk levels in each area. Then, using the control effect evaluation model based on support vector machine, by analyzing the data such as the area ratio of risk areas and the number of high-risk areas in the security situation image, the control effect evaluation score is 85 points, indicating that the control action is well executed, but there is still room for optimization. The control effect evaluation report is automatically generated. If the evaluation score is higher than the preset threshold of 90 points, the report will be pushed to the management terminal to provide auxiliary decision support for it; if the evaluation score is lower than the threshold, the control measure optimization process is triggered. The optimization algorithm is based on the Q-learning algorithm in reinforcement learning. By analyzing the correlation between the execution of historical control actions and the occurrence of security incidents, it learns to optimize control actions, such as adjusting the patrol frequency of area A to once every 30 minutes and the patrol frequency of area B to once every hour. The optimized control measures are sent to the control system of each area, and the control measures are adjusted by adjusting the rotation angle of the camera, adding patrol routes, etc. After the adjustment, the latest security level and execution data of control actions are continuously obtained, and the next analysis and optimization cycle is entered to form a closed-loop control and continuously improve the control effect.

[0112] According to one embodiment of the present invention, data cleaning is performed on the relevant data and actual effect data of the control measures under the historical security level of the security level and the control action, and the fields with missing values ​​exceeding 20% ​​are removed, and the outliers are corrected, and then the key features are extracted. Then, the Pearson correlation analysis is performed on the relevant data and actual effect data of the control measures under the processed historical security level. If the correlation coefficient is lower than 6, it is considered that the two sets of data are quite different, triggering optimization. The gradient boosting decision tree algorithm is used to optimize the model, and the model performance is evaluated using 10-fold cross validation. If the accuracy rate is improved by more than 5%, the optimization is considered to be effective. The setting of the optimized security level is applied to the control action, and the isolation forest algorithm is used for anomaly detection during the operation of the system. If the anomaly score exceeds 7, the corresponding control action is triggered. Finally, the K-means clustering algorithm and the Apriori association rule mining algorithm are used to collect the relevant data of the control measures under the historical security level and form a corresponding security control knowledge base for optimizing subsequent security levels and control measures.

[0113] According to one embodiment of the present invention, data sets for training and evaluation are manually collected and organized. These data sets should contain various types of tasks and scenarios so that the model can learn rich knowledge and skills. SFT supervised training data is usually provided in the form of text, which includes a question description part and an answer label part, which can be constructed as a structured json string. Generally, the amount of data to be annotated needs to be determined according to specific needs and tasks. Generally, for the optimization of a single task, 2000-10000 supervised data can be annotated, and then the SFT fine-tuning tool is used for optimization. If there are multiple tasks, it is possible to consider further increasing the annotated data to 5000-100000 supervised data. If there are many tasks to be optimized, the original data is very rich, and the effect requirements for the large model are very high, it is possible to consider annotating tens of thousands to hundreds of thousands or even millions of supervised data, and then use the SFT fine-tuning tool for optimization. By using the SFT tool, the annotated data is trained on the basis of the existing model. When uploading data based on the training platform, the user needs to create scene classification according to the actual task objectives. If there is no clear scenario, create a "general" level to manage all data. After data collection, data cleaning is required. Through data cleaning, a lot of low-quality data, duplicate data, and unsafe data can be eliminated. After data cleaning is completed, you can enter the formal model pre-training process.

[0114] According to an embodiment of the present invention, a user can create a data cleaning task, with the data to be cleaned as a unit, and each data to be cleaned is regarded as a data cleaning task. The cleaned data will be automatically stored in the same data set as the original data, and iteratively incremented on the existing version number. When creating a data cleaning task, different data cleaning configurations can be performed according to actual needs, that is, what cleaning operations the user needs to do, and the detailed requirements of the cleaning process are configured, which mainly include four steps: abnormal cleaning, filtering, deduplication and privacy removal.

[0115] Furthermore, in order to achieve the above-mentioned purpose of the invention, the present invention also provides an intelligent identification system for engineering construction dangerous areas based on a generative large model. Figure 2 is a flow chart of an intelligent identification system for engineering construction hazardous areas based on a generative large model according to an exemplary embodiment, such as Figure 2 As shown, an intelligent identification system for engineering construction hazardous areas based on a generative large model in the present invention includes:

[0116] Association matrix construction module: Build a 3D model of the project based on the BIM model, 3D satellite GIS terrain and flight laser point cloud model, segment the 3D model, obtain real-time monitoring data and safety incident records of each area of ​​the construction site based on the segmented 3D model and drone aerial photography, and build an association matrix between safety incidents and monitoring data;

[0117] Key monitoring data set acquisition module: Based on the association matrix between security events and monitoring data, the association pattern and law between security events and monitoring data are obtained, and the key monitoring data set that affects the security level is determined;

[0118] Security level adjustment module: Based on the key monitoring data set, it analyzes the changing trends and abnormal conditions of each monitoring data in real time. If the monitoring data exceeds the preset threshold or shows abnormal fluctuations, it is judged that there are security risks in the area and the security level of the area is adjusted dynamically.

[0119] Control action setting mode: According to the security level of each area, the predefined security control measures library is automatically matched to obtain the control measures of the security level of each area, and the control measures are automatically sent to the security control system through the rule engine to trigger the corresponding control action;

[0120] Platform feedback module: Feedback the security level of each area and the execution of control actions to the project management business platform in real time, generate security situation maps and control effect evaluation reports, and provide auxiliary decision support for managers;

[0121] Security level update module: backtest and verify the security level and control measures, compare and analyze the relevant data and actual effect data of the control measures under the historical security level, and use the incremental learning algorithm to iteratively optimize the security level setting.

[0122] Monitoring data Those of ordinary skill in the art can appreciate that the modules and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0123] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and equipment can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0124] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0125] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of the present invention.

[0126] In addition, each functional module in the embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0127] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the energy-saving signal sending / receiving method of each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.

[0128] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the present application.

[0129] It should be understood that the size of the serial numbers of each step in the content of the invention and the embodiments of the present invention does not absolutely mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

Claims

1. A method for intelligent identification of dangerous areas in engineering construction based on a generative large model, characterized in that: include: A 3D model of the project is established based on the BIM model, 3D satellite GIS terrain and flight laser point cloud model. The 3D model is segmented, and real-time monitoring data and safety event records of each area of ​​the construction site are obtained based on the segmented 3D model and drone aerial photography, and a correlation matrix between safety events and monitoring data is constructed; Based on the segmented 3D model, real-time monitoring data and safety event records of each area of ​​the construction site are obtained, and data cleaning is performed on the acquired real-time monitoring data of each area of ​​the construction site to remove missing values ​​and outliers; Perform feature extraction based on the real-time monitoring data of each area of ​​the construction site after cleaning, and extract features related to safety events, equipment operation status, and personnel behavior operation features; Using a feature selection algorithm, a feature subset with the highest correlation with the security event in the security event record is selected from the extracted features to reduce the feature dimension; For the selected feature subset, the association rule mining algorithm is used to discover the association rules between security events and each feature, and the correlation between the feature and the event is obtained; According to the mined association rules, a correlation matrix between security events and monitoring data is constructed. Each element in the matrix represents the correlation between the feature and the event. Perform cluster analysis on the correlation matrix to aggregate features and events with similar correlations and identify different security event patterns; According to the clustering results, a security event prediction model is established to predict the security events that occur and issue early warnings through real-time monitoring data; Based on the correlation matrix between security events and monitoring data, the correlation patterns and rules between security events and monitoring data are obtained, and the key monitoring data set that affects the security level is determined; Construct a correlation matrix between security events and monitoring data. The elements in the matrix represent the correlation between the two. Adopting association rule mining algorithm, based on support and confidence thresholds, the frequent item sets and association rules between security events and monitoring data are mined from the association matrix; By mining the association rules, we can summarize the association patterns between security events and monitoring data, and determine the security events caused by abnormal monitoring data. Evaluate the mined association rules, calculate the support, confidence and lift of each rule, and sort and filter the association rules; The monitoring data involved in the screened strong association rules are used as the preliminary key monitoring data, and the final key monitoring data set is determined from them in combination with expert knowledge and practical experience; Based on the determined key monitoring data set, a real-time monitoring and early warning mechanism is established to warn of possible security incidents when abnormalities are detected in key monitoring data; Continuously collect security events and monitoring data, update association matrices and mine association rules, and dynamically adjust key monitoring data sets; Based on the key monitoring data set, the changing trends and abnormal conditions of each monitoring data are analyzed in real time. If the monitoring data exceeds the preset threshold or shows abnormal fluctuations, it is judged that there are safety hazards in the area and the safety level of the area is adjusted dynamically; According to the security level of each area, the predefined security control measures library is automatically matched to obtain the control measures of the corresponding security level of each area, and the control measures are automatically sent to the security control system through the rule engine to trigger the corresponding control actions; Feedback the security level of each area and the execution of control actions to the project management business platform in real time, generate security situation maps and control effect evaluation reports, and provide auxiliary decision support for managers; Backtest and verify the security level and control measures. By comparing and analyzing the relevant data and actual effect data of the control measures under the historical security level, an incremental learning algorithm is used to iteratively optimize the security level setting.

2. The method for intelligently identifying dangerous areas in engineering construction based on a generative large model as claimed in claim 1, characterized in that: Obtain key monitoring data sets, build a time series data model for each monitoring data, and record data values ​​at different time points; Perform trend prediction and anomaly detection on the time series data of each monitoring data, and judge whether the monitoring data exceeds the normal range according to the preset threshold. If it exceeds the threshold, mark the time point as an abnormal point; Further analyze the abnormal situation of the monitoring data to determine whether the abnormal point is random noise or systematic abnormality. If the abnormal point of the monitoring data is systematic abnormality, it is determined that there is a safety hazard in the area. According to the severity of the abnormal situation, the safety level of the area is dynamically adjusted; Continuously perform time series analysis and anomaly detection in real-time data streams, process newly generated monitoring data in real time, and visualize anomalies and changes in security levels of each monitoring data.

3. The method for intelligently identifying dangerous areas in engineering construction based on a generative large model as claimed in claim 2, characterized in that: Based on drone aerial photography, real-time monitoring data of various areas of the construction site is obtained, and the data is pre-processed to extract key characteristic parameters. The safety level is calculated based on the key characteristic parameters to obtain the safety level of each area; According to the security level, a set of control measures corresponding to the current security level is automatically matched in the predefined security control measures library; The matched control measures are converted into executable control instructions, and logical judgment is performed through the pre-configured rule engine. If the rule conditions are met, the control instructions are automatically sent to the security control system; After receiving the control command issued, the safety control system parses the command content and extracts the specific control actions that need to be triggered; If the control action is to issue a warning signal, the preset signal generator is called to generate sound and light warning signals, and the warning information is pushed to the terminal of relevant personnel to indicate the dangerous situation; While executing control actions, the regional security status, control measures, control action type and execution results are associated and stored to form a security control log.

4. The method for intelligently identifying dangerous areas in engineering construction based on a generative large model as claimed in claim 3, characterized in that: Obtain the security level of each area and the execution status of control actions, and transmit the data to the database of the visual project management business platform in real time; Analyze and process the security level and execution status data of control actions in the database to obtain security situation image data, analyze the security situation image data, and obtain the control effect evaluation results; Output security situation image data and control effect evaluation results to generate a control effect evaluation report; Determine whether the management and control effect evaluation result meets the preset auxiliary decision threshold. If so, push the management and control effect evaluation report to the management terminal; If the control effect evaluation result does not meet the preset auxiliary decision threshold, the preset control measures are optimized, the execution data of the control actions are analyzed, and the optimized control measures are obtained; The optimized control measures will be issued to the control systems of various regions, the control measures of each region will be adjusted, and the latest security level and execution status data of control actions will continue to be obtained to form a closed-loop control.

5. The method for intelligently identifying dangerous areas in engineering construction based on a generative large model as claimed in claim 4, characterized in that: Obtain relevant data and actual effect data of control measures under historical safety levels, perform data cleaning on each data, remove noise data, and extract key features; Compare and analyze the relevant data of the control measures under the processed historical safety level and the actual effect data, calculate the difference and correlation of the two sets of data, and obtain the difference data and correlation data; Based on the difference data and correlation data, the effectiveness and accuracy of the current security level and control actions are judged. If the data is lower than the preset threshold, the optimization process of setting the security level is triggered; An incremental learning algorithm is used to fine-tune the parameters and optimize the structure of the security level setting to obtain the optimized security level; Apply the optimized security level to the control action, regenerate the control action through the security control rule engine, and send it to the security control system; Continuously monitor real-time data during system operation, use anomaly detection algorithms to identify potential security threats and abnormal behaviors, and form a security management and control knowledge base.

6. A system for intelligently identifying dangerous areas in engineering construction based on a generative large model, using a method for intelligently identifying dangerous areas in engineering construction based on a generative large model as described in any one of claims 1 to 5, characterized in that: include: Association matrix construction module: Build a 3D model of the project based on the BIM model, 3D satellite GIS terrain and flight laser point cloud model, segment the 3D model, obtain real-time monitoring data and safety incident records of each area of ​​the construction site based on the segmented 3D model and drone aerial photography, and build an association matrix between safety incidents and monitoring data; Key monitoring data set acquisition module: Based on the association matrix between security events and monitoring data, the association pattern and law between security events and monitoring data are obtained, and the key monitoring data set that affects the security level is determined; Security level adjustment module: Based on the key monitoring data set, it analyzes the changing trends and abnormal conditions of each monitoring data in real time. If the monitoring data exceeds the preset threshold or shows abnormal fluctuations, it is judged that there are security risks in the area and the security level of the area is adjusted dynamically. Control action setting mode: According to the security level of each area, the predefined security control measures library is automatically matched to obtain the control measures of the security level of each area, and the control measures are automatically sent to the security control system through the rule engine to trigger the corresponding control action; Platform feedback module: Feedback the security level of each area and the execution of control actions to the project management business platform in real time, generate security situation maps and control effect evaluation reports, and provide auxiliary decision support for managers; Security level update module: backtest and verify the security level and control measures, compare and analyze the relevant data and actual effect data of the control measures under the historical security level, and use the incremental learning algorithm to iteratively optimize the security level setting.

Images