Construction risk situation monitoring method and system based on deep neural network

Abstract

The invention discloses a construction risk situation monitoring method and system based on a deep neural network, and relates to the technical field of construction risk monitoring, a construction safety evaluation model is used to evaluate the construction safety in a construction area, and if the safety is lower than expectation, a suitable alarm mode is recommended for construction workers from an alarm mode library, and the construction risk situation is monitored. According to the construction state data and the environmental condition data of the construction workers, corresponding operation reminding is matched from the operation reminding library, and various adjusting devices in the construction area are automatically adjusted; inspecting the interior of the construction area according to the inspection time nodes meeting the constraint conditions, and matching a corresponding emergency plan according to the emergency features by an emergency plan library; and after key factors influencing construction safety in the construction area are determined, a corresponding construction optimization strategy is output by the construction scheme optimization knowledge graph. The inspection frequency adapts to the current state of the construction area, the inspection efficiency is improved, and the construction risk possibly existing in the construction area is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of construction risk monitoring, and specifically to a construction risk situation monitoring method and system based on a deep neural network. Background Technique

[0002] Construction risk monitoring refers to the systematic and real-time monitoring and tracking activities of various potential or existing risks in a construction project (such as structural safety, construction operations, environmental impacts, etc.). By setting up monitoring points at key positions on the construction site and using advanced instruments and equipment such as sensors, surveillance cameras, and data analysis software, key parameters during the construction process (such as structural stress, deformation conditions, environmental quality, etc.) are collected, recorded, and analyzed in real time. The purpose of this process is to promptly discover and accurately evaluate potential quality defects, safety hazards, or environmental risks, provide timely decision-making support for construction managers, so as to quickly take targeted control and response measures, effectively contain the further development of risks, ensure the quality, progress, and safety of the construction project, and at the same time ensure the safety of the surrounding environment and personnel.

[0003] In the Chinese invention patent with the authorization announcement number CN118313670B, a method for monitoring and evaluating risks in a smart construction site is disclosed, which relates to the technical field of construction site risk monitoring, specifically as follows: receiving construction site information and transmitting the received construction site information to an information analysis module; analyzing the construction site information to obtain environmental data and construction site data, and transmitting the environmental data and construction site data to a data calculation module for calculation to obtain construction site risk reference data and environmental risk reference data, receiving the construction site risk reference data for risk level division to obtain a construction site risk level threshold, and the information analysis module receiving the environmental risk reference data for risk level division to obtain an environmental risk level threshold; transmitting the distinguished risk level thresholds to a risk judgment module for risk judgment through the risk level thresholds; the present invention obtains and analyzes the construction site information during the construction process of the construction site, and judges the safety of the construction site by obtaining the construction site information in real time.

[0004] Combined with the above application and the content in the prior art:

[0005] When construction needs to be carried out indoors, especially when working through scaffolding and carrying out interior decoration, if the construction area is large, there are many construction workers, and there are many construction tasks, construction safety becomes particularly important, and continuous safety management of the construction area is required, such as real-time video monitoring and environmental monitoring in the construction area.

[0006] The existing method for monitoring the construction risk situation of deep neural networks mainly relies on real-time collection of construction data in the construction area, uses the construction data of construction workers as input, and uses deep neural networks for safety evaluation. When construction safety risks are identified, alarms and responses are made to ensure construction safety. However, when there are many drilling and cutting operations in the construction area, the dust concentration in the air is relatively high and the noise is also relatively high. The hearing and vision of construction workers will be disturbed to a certain extent. In such a construction scenario, when there is a safety risk in the construction area, construction workers may not hear the alarm sound, which may lead to the failure of the alarm. At the same time, the existing risk situation perception methods mainly focus on safety alarms. When an actual alarm occurs, there are also lack of corresponding emergency management measures, resulting in an incomplete risk situation perception effect.

[0007] Therefore, the present invention provides a method and system for monitoring the construction risk situation based on a deep neural network. Summary of the Invention

[0008] (1) Technical problems to be solved

[0009] In view of the deficiencies of the prior art, the present invention provides a method and system for monitoring the construction risk situation based on a deep neural network. By using a construction safety assessment model to evaluate the construction safety in the construction area, if the safety level is lower than expected, a suitable alarm method is recommended for construction workers from the alarm method library. According to the construction status data and environmental condition data of construction workers, corresponding operation reminders are matched from the operation reminder library, and various adjustment devices in the construction area are automatically adjusted. The construction area is patrolled according to the inspection time nodes that meet the constraint conditions, and the emergency plan library matches corresponding emergency plans according to the emergency characteristics. The construction optimization strategy is output by the construction plan optimization knowledge graph. The inspection frequency is adapted to the current state of the construction area, improving the inspection efficiency and reducing the possible construction risks in the construction area, thereby solving the technical problems recorded in the background art.

[0010] (2) Technical solutions

[0011] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0012] The method for monitoring the construction risk situation based on a deep neural network includes: after collecting the environmental condition data in the construction area, performing an environmental risk assessment, generating an environmental risk degree Hjx from a number of continuously obtained risk values. If the environmental risk degree Hjx exceeds the risk threshold, a data collection instruction is sent to the outside.

[0013] Use the construction safety assessment model to evaluate the construction safety within the construction area. If the safety level is lower than expected, recommend a suitable alarm method for construction workers from the alarm method library. After switching the alarm method based on the alarm feedback data, send a first-level alarm instruction to the outside;

[0014] After receiving the first-level alarm instruction, based on the construction status data and environmental condition data of the construction workers, match the corresponding operation reminders from the operation reminder library, and use the conditional automatic control model to automatically adjust the various adjustment devices within the construction area;

[0015] Conduct inspections within the construction area according to the inspection time nodes that meet the constraint conditions. Generate a warning value Jzp from the status data of continuously sending the first-level alarm instruction. If the warning value Jzp exceeds the expectation, match the corresponding emergency plan from the emergency plan library according to the emergency characteristics;

[0016] Automatically generate a construction safety report after the inspection cycle ends. After determining the key factors affecting the construction safety within the construction area, according to the correspondence between the key factors and the construction optimization plan, output the corresponding construction optimization strategy by the construction plan optimization knowledge graph.

[0017] Furthermore, continuously collect environmental condition data by the sensor network within the construction area, and generate multi-modal environmental condition data after fusing the collected multi-source environmental condition data;

[0018] Use the trained risk assessment deep model to conduct environmental risk assessment with the multi-modal environmental condition data as the input, and obtain the risk values at a continuous number of time nodes.

[0019] Furthermore, generate the environmental risk degree from a continuous number of risk values, and the method is as follows:

[0020]

[0021] In the formula, Xz(t) is the risk value at time t, λ is the time decay coefficient, β is the trend influence factor, and [a, b] is the time interval of the risk value distribution.

[0022] Furthermore, collect image data by the image acquisition device arranged within the construction area, including the action and posture data of construction workers during construction, and summarize them as construction status data;

[0023] Use the trained construction safety assessment model to conduct safety assessment with the construction status data as the input. If the obtained safety score is lower than expected, send a reminder instruction to the outside.

[0024] Furthermore, collect several alarm methods in advance, summarize them to generate an alarm method library; after receiving a reminder instruction, determine the location information of the worker, and based on the distribution status of dust and noise and the reminder preference data in the area where the construction worker is located, recommend an alarm method for the construction worker based on a filtered recommendation algorithm.

[0025] Furthermore, perform an alarm for the construction worker using a personalized alarm method, collect the response time of each construction worker, and construct a switching value xH from several response times as follows:

[0026]

[0027] In the formula: xT i is the response time of the i-th construction worker, xT a is the mean value of the response times, IQR is the interquartile range of several response times; F 1 、F 2 and F 3 are weight coefficients.

[0028] Furthermore, collect the environmental risk level Hjx and alarm data in the construction area within several consecutive inspection cycles, constrain the inspection frequency Cv in the next inspection cycle based on the environmental risk level Hjx and alarm data, conduct inspections in the construction area according to the inspection time nodes that meet the constraint conditions, and obtain inspection feedback data.

[0029] Furthermore, the constraint conditions for constraining the inspection frequency Cv in the next inspection cycle are as follows:

[0030]

[0031] Weight coefficients: 0 ≤ ρ ≤ 1, 0 ≤ ζ ≤ 1, and ρ + ζ = 1; n is the number of first-level alarm instructions, U(i,j) is the time interval from the i-th first-level alarm instruction to the j-th first-level alarm instruction, Ut a is the average value of the time intervals.

[0032] Furthermore, if the number of first-level alarm instructions received in the current inspection cycle exceeds the corresponding number threshold, generate a warning value Jzp based on the status data of the continuously issued first-level alarm instructions. If the warning value Jzp exceeds the pre-set warning threshold, send a second-level alarm instruction to the outside.

[0033] Furthermore, the method for generating the warning value Jzp based on the status data of the continuously issued first-level alarm instructions is as follows:

[0034]

[0035] Where: Pf(t) is the average safety score of construction workers at time t, and Δt(t) is the time interval between two first-level alarm instructions. is the change rate of the time interval, and F 1 , F 2 and F 3 are weight coefficients.

[0036] Furthermore, a construction safety report is automatically generated after the inspection cycle ends, and the construction safety report is sent to the inspection personnel; combining the construction safety report, the key factors affecting construction safety in the construction area are determined through principal component analysis, and each key factor is ranked according to the degree of influence. Taking construction plan optimization as the target word, a construction plan optimization knowledge graph is pre-constructed.

[0037] Furthermore, the warning threshold is adjusted according to the optimized inspection frequency, and the adjustment method is as follows:

[0038]

[0039] Among them, m is the number of inspection nodes within the inspection cycle, Xj(i,j) is the time interval from the i-th inspection node to the j-th inspection node, and Xj a is the average value of the time interval, and Pa is the warning threshold adjustment ratio.

[0040] The construction risk situation monitoring system based on a deep neural network includes an environmental risk analysis unit that collects environmental condition data in the construction area and conducts environmental risk assessment. An environmental risk degree Hjx is generated from a number of continuously obtained risk values. If the environmental risk degree Hjx exceeds the risk threshold, a data collection instruction is sent to the outside.

[0041] The alarm mode matching unit uses a construction safety assessment model to evaluate the construction safety in the construction area. If the safety is lower than expected, a suitable alarm mode is recommended for construction workers from the alarm mode library. After switching the alarm mode based on the alarm feedback data, a first-level alarm instruction is sent to the outside.

[0042] The automatic control unit, after receiving the first-level alarm instruction, matches the corresponding operation reminder from the operation reminder library according to the construction status data and environmental condition data of the construction workers, and automatically adjusts various adjustment devices in the construction area using a conditional automatic control model.

[0043] The multi-level alarm unit patrols the construction area according to the inspection time nodes that meet the constraint conditions. A warning value Jzp is generated from the status data of continuously sending first-level alarm instructions. If the warning value Jzp exceeds the expectation, the corresponding emergency plan is matched from the emergency plan library according to the emergency characteristics.

[0044] The scheme optimization unit automatically generates a construction safety report after the inspection cycle ends. After determining the key factors affecting construction safety in the construction area, the construction scheme optimization knowledge graph outputs the corresponding construction optimization strategy based on the correspondence between the key factors and the construction optimization scheme.

[0045] (III) Beneficial effects

[0046] The present invention provides a construction risk situation monitoring method and system based on a deep neural network, which has the following beneficial effects:

[0047] 1. Conduct a comprehensive risk assessment of the construction area to verify whether the construction environment conditions will affect the construction process; the environmental risk degree Hjx can be used to comprehensively evaluate the possible construction environment risks in the construction area, and when the construction environment conditions are not conducive to the current construction, targeted treatment can be carried out in a timely manner.

[0048] 2. Evaluate the safety of the current construction operation, determine whether the current construction operation is compliant, and ensure that the current construction process can be carried out safely; consider the user's environment and personal preferences, and match the user with an appropriate alarm method to make the alarm method more adaptable to the current scene and reduce the risk of user neglect.

[0049] 3. Feedback on the alarm effect at the current stage can be formed based on the switching value xH. When the current alarm mode fails to achieve full effect, the alarm mode can be switched to provide a comprehensive alarm in the construction area and improve the effectiveness of the alarm. By issuing reminders or guidance to users, protection can be provided against possible construction risks. When there are abnormalities in the current construction environment, adaptive modifications can be made to the environmental conditions in the construction area, which can improve the environmental conditions in the construction area.

[0050] 4. Adaptively adjust the patrol frequency to match it with the current status of the construction area, avoid too many or too few patrols in the construction area, improve patrol efficiency and reduce possible construction risks in the construction area; construct a warning value Jzp based on the status of receiving the first-level alarm command, and enable a higher-level alarm when the current alarm fails to achieve the expected effect, so as to facilitate the adoption of corresponding measures in a targeted manner.

[0051] 5. As feedback and response to the secondary alarm command, when the current risk level in the construction area is relatively high, a targeted emergency response plan is matched based on the construction data in the construction area to form a targeted response to the construction risks in the construction area and ensure the safety of construction personnel.

[0052] 6. By generating a construction safety report, a summary of the current stage of construction can be formed; by adjusting the patrol frequency within the construction area, the warning threshold can be adaptively adjusted, reducing the warning threshold and increasing the patrol frequency, increasing the sensitivity to construction risks and the risk perception ability within the construction area.

[0053] 7. By identifying the corresponding influencing factors and outputting the corresponding construction optimization strategies by the construction plan optimization knowledge graph, the current construction and management plan can be optimized, improving the current construction state at a deeper level and reducing the subsequent construction risks. Description of the Drawings

[0054] Figure 1 It is a schematic flow chart of the construction risk situation monitoring method of the present invention;

[0055] Figure 2 It is a schematic structural diagram of the construction risk situation monitoring system of the present invention. Detailed Embodiments

[0056] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0057] Please refer to Figure 1 , the present invention provides a construction risk situation monitoring method based on a deep neural network, including,

[0058] Step 1: After collecting the environmental condition data in the construction area, conduct an environmental risk assessment, generate an environmental risk degree Hjx from a number of continuously obtained risk values. If the environmental risk degree Hjx exceeds the risk threshold, send a data collection instruction to the outside;

[0059] The content of the above Step 1 includes the following:

[0060] Step 101: After determining the current construction area, arrange a sensor network within the construction area, including dust sensors, acoustic sensors, gas monitoring sensors, etc. Continuously collect environmental condition data by the sensor network within the construction area, such as dust, noise, and air quality data, etc. After fusing the collected multi-source environmental condition data, generate multi-modal environmental condition data, and thus generate a multi-source environmental condition data set after summarization;

[0061] Step 102: Combine neural networks such as CNN, RNN, Transformer, and GNN to generate a deep neural network. After training the deep neural network with the labeled sample data, obtain the trained risk assessment deep model;

[0062] Use the multi-modal environmental condition data as input, and use the trained risk assessment deep model to conduct environmental risk assessment to obtain risk values at a continuous number of time nodes, and mark the risk values with the corresponding time nodes;

[0063] During use, when the construction area is in a construction state, in order to ensure the construction safety of the construction area, collect various data within the construction area, and use the trained multi-modal evaluation model to comprehensively evaluate the risk of the construction area to verify whether the current construction environmental conditions will affect the construction process;

[0064] Step 103: Generate an environmental risk degree from a continuous number of risk values, and use the obtained environmental risk degree to evaluate the construction environment within the construction area. The method for generating the environmental risk degree is as follows:

[0065]

[0066] In the formula, Xz(t) is the risk value at time t, λ is the time decay coefficient, with a value between 0.1 and 1, β is the trend influence factor, with a value between 0 and 1, e can take the value of 2.713, and [a, b] is the time interval of the risk value distribution;

[0067] According to historical data and the safety management of the environmental conditions within the construction area, preset a risk threshold; if the obtained environmental risk degree Hjx exceeds the risk threshold, it indicates that the current environmental conditions within the construction area may have a certain impact on the physical health and construction safety of the construction personnel. At this time, send a data collection instruction to the outside;

[0068] During use, combine the content in Steps 101 to 103:

[0069] As a further content, after continuously obtaining a number of risk values along the time axis, construct an environmental risk degree Hjx from the risk values. The environmental risk degree Hjx can comprehensively evaluate the possible construction environmental risks within the construction area. When the construction environmental conditions are not conducive to the current construction, targeted treatment can be carried out in a timely manner.

[0070] The existing methods for monitoring the construction risk situation of deep neural networks mainly rely on real-time collection of construction data in the construction area, using the construction data of construction workers as input, and using deep neural networks for safety evaluation. When construction safety risks are identified, alarms and responses are made to ensure construction safety. However, when there are many drilling and cutting operations in the construction area, the dust concentration in the air is relatively high, and the noise is also relatively high. The hearing and vision of construction workers will be disturbed to a certain extent. In such a construction scenario, when there is a safety risk in the construction area, construction workers may not hear the alarm sound, which may lead to the failure of the alarm. At the same time, the existing risk situation perception methods mainly focus on safety alarms. When an actual alarm occurs, there are also lack of corresponding emergency management measures, resulting in an incomplete risk situation perception effect.

[0071] Step 2: Use the construction safety assessment model to evaluate the construction safety in the construction area. If the safety level is lower than expected, recommend a suitable alarm method for the construction workers from the alarm method library. After switching the alarm method based on the alarm feedback data, send a first-level alarm instruction to the outside.

[0072] The above step 2 includes the following contents:

[0073] Step 201: Image data collection is carried out by the image acquisition device arranged in the construction area, including the action and posture data of construction workers during construction, etc. After summarization, it is used as construction status data.

[0074] Train the machine learning algorithm with the labeled sample data to obtain the trained construction safety assessment model; use the trained construction safety assessment model for safety assessment with the construction status data as input. If the obtained safety score is lower than expected, send a reminder instruction to the outside.

[0075] In use, as a further content, starting from the construction status of construction workers, evaluate the safety of the current construction operation, judge whether the current construction operation is compliant and meets the safety requirement specifications. For example, send a reminder in a timely manner when in an open operation state to ensure the safety of the current construction process.

[0076] Step 202: Collect several alarm methods in advance, such as using the personal devices of workers for visual signal alarms and vibration prompts. After summarizing the obtained alarm methods, generate an alarm method library.

[0077] After receiving the reminder instruction, determine the location information of the worker. The distribution status of dust and noise and the reminder preference data in the area where the construction worker is located are summarized as feedback data; use the filtering-based recommendation algorithm to recommend a suitable alarm method for the construction worker from the alarm method library according to the feedback data.

[0078] During use, as further content, considering that construction workers may not be able to effectively receive alarm instructions in the current construction scenario, when it is necessary to send an alarm instruction externally, considering the environment and personal preferences of the user, an appropriate alarm method is matched for the user to make the adaptability of the alarm method to the current scenario higher, reduce the risk of the user ignoring it, and enable the user to handle the alarm in a timely manner when receiving it.

[0079] Step 203: Alarm the construction workers using the personalized alarm method, collect the response times of each construction worker, and construct a switching value xH from several response times, as follows:

[0080]

[0081] In the formula: xT i is the response time of the i-th construction worker, xT a is the mean value of the response times, IQR is the interquartile range of several response times; F 1 、F 2 and F 3 are weight coefficients, and their values all fall between 0 and 1, and the sum of the three is 1;

[0082] Based on historical data and management expectations for the alarm method, a switching threshold is set in advance; if the obtained switching value xH exceeds the pre-set switching threshold, it indicates that the current alarm method may not achieve the expected notification effect, switch the alarm method, and send a first-level alarm instruction externally;

[0083] During use, combine the content in Steps 201 to 203:

[0084] As a further processing method, collect the response and feedback data of construction workers when receiving the alarm, thereby constructing a switching value. Based on the switching value xH, feedback on the alarm effect in the current stage can be formed, and the alarm method can be switched when the current alarm method fails to achieve full effect. For example, activate audible and visual alarms covering the entire area within the construction area, etc., and the personalized alarm can be switched to a full-area alarm within the construction area to improve the effectiveness of the alarm.

[0085] Step Three: After receiving the first-level alarm instruction, match the corresponding operation reminder from the operation reminder library according to the construction status data and environmental condition data of the construction workers, and use the conditional automatic control model to automatically adjust the various adjustment devices within the construction area;

[0086] The above Step Three includes the following content:

[0087] Step 301: Set up an operation reminder library in advance, such as wearing protective devices, starting noise reduction, lighting or ventilation equipment, etc.;

[0088] Based on the construction status data of construction workers and environmental condition data, corresponding operation reminders can be matched from the operation reminder library, providing customized risk warnings and suggestions. For example, workers can be reminded to pay attention to environmental changes and start wearing basic protective equipment such as masks and earplugs. For some difficult operations, voice guidance can also be provided through a smart voice assistant to help workers evacuate safely or adjust their work.

[0089] Step 302: Train a machine learning algorithm with the labeled sample data to obtain a trained conditional automatic control model.

[0090] After setting corresponding qualified intervals for various environmental condition parameters in the construction area, use the trained conditional automatic control model to control various adjustment devices in the construction area, such as turning on or automatically adjusting ventilation devices or noise reduction devices, etc.

[0091] When in use, combine the content in Steps 301 and 302:

[0092] As a feedback to the issuance of a first-level alarm instruction, when the current construction method or operation of construction workers is unqualified, by sending a reminder or guidance to the user, it provides a guarantee for the possible construction risks currently. As a further measure, when the current construction environment is abnormal, the environmental conditions in the construction area can be adaptively modified to improve the environmental conditions in the construction area.

[0093] Step Four: Patrol the construction area according to the inspection time nodes that meet the constraint conditions, generate a warning value Jzp from the status data of continuously issuing first-level alarm instructions. If the warning value Jzp exceeds the expectation, match the corresponding emergency plan from the emergency plan library according to the emergency characteristics.

[0094] The above Step Four includes the following content:

[0095] Step 401: Collect the environmental risk level Hjx and alarm data in the construction area within a continuous number of patrol cycles, and constrain the patrol frequency Cv in the next patrol cycle based on the environmental risk level Hjx and alarm data. The constraint conditions are as follows:

[0096]

[0097] Weight coefficients: 0 ≤ ρ ≤ 1, 0 ≤ ζ ≤ 1, and ρ + ζ = 1; the weight coefficients can be obtained by referring to the analytic hierarchy process.

[0098] Among them, n is the number of first-level alarm instructions, U(i,j) is the time interval from the i-th first-level alarm instruction to the j-th first-level alarm instruction, and Ut a is the average value of the time intervals;

[0099] Conduct inspections in the construction area according to the inspection time nodes that meet the constraints and obtain inspection feedback data;

[0100] When in use, as feedback on the possible risks in the current construction environment and construction status, when the construction safety officer conducts safety inspections in the construction area, the inspection frequency is adaptively adjusted to make it compatible with the current status of the construction area, avoiding too many or too few inspections in the construction area, improving inspection efficiency and reducing possible construction risks in the construction area.

[0101] Step 402: If the number of first-level alarm instructions received in the current patrol cycle exceeds the corresponding number threshold, it means that the environmental conditions in the area before construction are poor. At this time, under dimensionless conditions, the warning value Jzp is generated according to the status data of the first-level alarm instructions issued continuously, in the following manner:

[0102]

[0103] Where: Pf(t) is the mean safety score of the construction workers at time t, Δt(t) is the time interval between two first-level alarm instructions, is the rate of change of the time interval, F 1 、F 2 and F 3 is the weight coefficient, and its value is consistent with the previous value;

[0104] If the warning value Jzp exceeds the preset warning threshold, a secondary alarm command is issued to the outside;

[0105] When in use, as a response to the first-level alarm command, the warning value Jzp is constructed according to the status of the received first-level alarm command, and this is used as the starting point to activate the multi-level alarm mechanism. When the current alarm fails to achieve the expected effect, a higher-level alarm is enabled to facilitate targeted measures.

[0106] Step 403: Develop a detailed emergency plan, including emergency handling plans, evacuation routes, assembly locations, and emergency contacts, etc., and aggregate the developed emergency plans to generate an emergency plan library;

[0107] After receiving the second-level alarm command, the construction data in the construction area, such as construction environment data and construction status data, are extracted to obtain the corresponding emergency features. The emergency plan library matches the corresponding emergency plan based on the emergency features and pushes the emergency plan to the construction workers.

[0108] When used, combine the contents in steps 401 to 403:

[0109] As a feedback and response to the secondary alarm instruction, when the current risk level in the construction area is relatively high, a targeted emergency response plan is matched based on the construction data in the construction area. For example, arranging for the rapid evacuation of construction personnel can form a targeted response to the construction risks in the construction area and ensure the safety of construction personnel.

[0110] Step Five: Automatically generate a construction safety report after the inspection cycle ends. After determining the key factors affecting construction safety in the construction area, according to the correspondence between the key factors and the construction optimization plan, the construction plan optimization knowledge graph outputs the corresponding construction optimization strategy.

[0111] The above Step Five includes the following content:

[0112] Step 501: Automatically generate a construction safety report after the inspection cycle ends, including the alarm time and location, the reasons for alarm triggering, the implemented emergency response plan, environmental condition data, video records, etc.; send the construction safety report to the inspection personnel.

[0113] The warning threshold can be adjusted according to the optimized inspection frequency to update the current alarm mechanism. The adjustment method is as follows:

[0114]

[0115] Among them, m is the number of inspection nodes in the inspection cycle, Xj(i,j) is the time interval from the i-th inspection node to the j-th inspection node, and Xj a is the average value of the time intervals, and Pa is the adjustment ratio of the warning threshold.

[0116] When in use, as a feedback to the secondary alarm instruction, after the current construction stage ends, by generating a construction safety report, a summary of the current stage of construction can be formed. As another step of feedback, by adjusting the inspection frequency in the construction area, the warning threshold is adaptively adjusted, reducing the warning threshold and increasing the inspection frequency, increasing the sensitivity to construction risks and the risk perception ability in the construction area.

[0117] Step 502: Combine the construction safety report, determine the key factors affecting construction safety in the construction area through principal component analysis, and sort each key factor according to the degree of influence.

[0118] Taking construction plan optimization as the target word, after in-depth retrieval and entity relationship construction, a construction plan optimization knowledge graph is pre-constructed. According to the correspondence between the key factors and the construction optimization plan, the construction plan optimization knowledge graph outputs the corresponding construction optimization strategy.

[0119] When in use, combine the content in Steps 501 and 502:

[0120] As further feedback content, when there are significant construction risks in the construction area, such as construction environment risks and construction operation risks, by identifying the corresponding influencing factors and having the construction plan optimization knowledge graph output the corresponding construction optimization strategies, the current construction and management plans can be optimized, the current construction status can be improved at a deeper level, and the subsequent construction risks can be reduced.

[0121] The Analytic Hierarchy Process (AHP) is a structured and systematic decision-making support technique. It decomposes complex multi-criteria decision-making problems into a series of interrelated and hierarchical sub-problems or factors, uses expert judgment to compare and quantitatively evaluate the importance of factors at each level, and finally obtains the relative ranking of the pros and cons of each decision-making scheme through synthetic calculation, providing clear and quantitative decision-making basis for decision-makers. This method combines qualitative and quantitative analysis, effectively improving the scientificity and accuracy of decision-making.

[0122] The process of constructing a knowledge graph with "construction plan optimization" as the target word is as follows:

[0123] Clarify the construction goal - First, it is necessary to clarify the construction goal of the knowledge graph, that is, to collect and organize relevant knowledge, concepts, relationships, etc. around the theme of "construction plan optimization".

[0124] Select the construction method - According to the scale, complexity and available resources of the knowledge graph, one or more of the following construction methods can be selected: Manual construction method: Suitable for small-scale and specific-domain knowledge graphs, collecting, organizing and annotating data manually. Automatic extraction method: Using natural language processing technology to automatically extract entities, relationships, attributes and other information from a large amount of text. Semi-automatic construction method: Combining manual construction and automatic extraction methods, and manually reviewing and correcting the results of automatic extraction. Ontology-based construction method: Using ontology to model domain knowledge to improve the consistency and scalability of the knowledge graph.

[0125] Data collection and preprocessing - Collect data: Collect data related to "construction plan optimization" from relevant books, academic papers, engineering practice reports and other channels. Data cleaning: Remove duplicate, invalid or incorrect data to ensure the accuracy and consistency of the data. Data annotation: Annotate the collected data, including entity annotation, relationship annotation, etc., providing a basis for subsequent automatic extraction and knowledge graph construction.

[0126] Constructing a Knowledge Graph - Defining Entities and Relationships: Based on the domain characteristics of construction plan optimization, define relevant entities (such as construction methods, construction sequences, construction organizations, etc.) and relationships (such as optimization methods, influencing factors, etc.). Constructing the Schema Layer: Based on the defined entities and relationships, construct the schema layer of the knowledge graph, including entity types, relationship types, etc. Entity Filling: Use automatic extraction or manual annotation methods to fill the collected data into the knowledge graph to form entities and relationships. Relationship Linking: According to the relationship information in the data, establish links between entities to form a complete knowledge graph.

[0127] Verification and Optimization - Verifying the Knowledge Graph: Verify the accuracy and integrity of the knowledge graph through methods such as manual inspection and logical reasoning. Optimizing the Knowledge Graph: According to the verification results, correct and optimize the knowledge graph to improve its quality and usability.

[0128] Application Scenarios and Expansion - Application Scenarios: Apply the constructed knowledge graph to scenarios such as decision-making support and knowledge query for construction plan optimization. Continuous Expansion: As new knowledge emerges, continuously update and expand the knowledge graph to maintain its timeliness and integrity.

[0129] Specific Step Content - Defining Key Elements of Construction Plan Optimization: Include construction methods, construction sequences, construction organizations, construction labor force organizations, construction machinery organizations, etc. Collecting and Organizing Relevant Cases: Collect successful cases of construction plan optimization from actual engineering projects, and extract key information and optimization strategies. Establishing an Entity-Relationship Model: According to the collected cases and key elements, establish an entity-relationship model to clarify the relationships between various elements. Filling the Knowledge Graph: Fill the collected cases and optimization strategies into the knowledge graph to form a complete knowledge graph for construction plan optimization. Verifying and Optimizing the Knowledge Graph: Verify the effectiveness of the knowledge graph through actual project applications and optimize it according to the feedback.

[0130] Please refer to Figure 2 , the present invention provides a construction risk situation monitoring system based on a deep neural network, including,

[0131] An environmental risk analysis unit, which collects environmental condition data in the construction area and then conducts an environmental risk assessment, generates an environmental risk degree Hjx from a number of continuously obtained risk values. If the environmental risk degree Hjx exceeds the risk threshold, it sends a data collection instruction to the outside;

[0132] An alarm method matching unit, which uses a construction safety assessment model to evaluate the construction safety in the construction area. If the safety is lower than expected, it recommends a suitable alarm method for construction workers from the alarm method library. After switching the alarm method based on the alarm feedback data, it sends a first-level alarm instruction to the outside;

[0133] The automatic control unit, after receiving a first-level alarm instruction, matches corresponding operation reminders from the operation reminder library based on the construction status data of construction workers and the environmental condition data, and automatically adjusts various regulating devices in the construction area using a conditional automatic control model;

[0134] The multi-level alarm unit patrols the construction area according to the inspection time nodes that meet the constraint conditions, generates a warning value Jzp from the status data that continuously issues first-level alarm instructions, and if the warning value Jzp exceeds the expectation, the emergency plan library matches the corresponding emergency plan according to the emergency characteristics;

[0135] The plan optimization unit automatically generates a construction safety report after the inspection cycle ends. After determining the key factors affecting construction safety in the construction area, according to the correspondence between the key factors and the construction optimization plan, the construction plan optimization knowledge graph outputs the corresponding construction optimization strategy.

[0136] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0137] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0138] In several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only for some logical function divisions. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some interfaces. The indirect coupling or communication connection of the devices or units can be in an electrical, mechanical, or other form.

[0139] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or they can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0140] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claimed rights.

Claims

1. A construction risk situation monitoring method based on a deep neural network, characterized by: include, After collecting environmental condition data in the construction area, an environmental risk assessment is conducted. The environmental risk level Hjx is generated from a number of continuously acquired risk values. If the environmental risk level Hjx exceeds the risk threshold, a data collection instruction is issued to the outside. Use the construction safety assessment model to evaluate the construction safety in the construction area. If the safety is lower than expected, recommend appropriate alarm methods to the construction workers from the alarm method library. After switching the alarm method based on the alarm feedback data, issue a first-level alarm command to the outside. After receiving the first-level alarm command, the corresponding operation reminder is matched from the operation reminder library according to the construction status data and environmental condition data of the construction workers, and the conditional automatic control model is used to automatically adjust various regulating equipment in the construction area; According to the inspection time nodes that meet the constraints, the construction area is inspected, and the warning value Jzp is generated by the status data of the continuous issuance of the first-level alarm command. If the warning value Jzp exceeds the expectation, the emergency plan library matches the corresponding emergency plan according to the emergency characteristics; A construction safety report is automatically generated after the inspection cycle ends. After determining the key factors affecting construction safety in the construction area, the construction plan optimization knowledge graph outputs the corresponding construction optimization strategy based on the correspondence between the key factors and the construction optimization plan.

2. The construction risk situation monitoring method based on deep neural network according to claim 1, characterized in that: The sensor network in the construction area continuously collects environmental condition data, and the collected multi-source environmental condition data are integrated to generate multi-modal environmental condition data; Taking multimodal environmental condition data as input, the trained risk assessment deep model is used to perform environmental risk assessment and obtain risk values ​​at several consecutive time nodes.

3. The construction risk situation monitoring method based on deep neural network as claimed in claim 2 is characterized by: The environmental risk Hjx is generated by a number of consecutive risk values ​​in the following way: Where Xz(t) is the risk value at time t, λ is the time attenuation coefficient, β is the trend influencing factor, and [a, b] is the time interval of the risk value distribution.

4. The construction risk situation monitoring method based on deep neural network as claimed in claim 3 is characterized by: The image acquisition device arranged in the construction area acquires image data, including the action and posture data of the construction workers during construction, which are summarized as construction status data; Taking construction status data as input, the trained construction safety assessment model is used to perform safety assessment. If the safety score obtained is lower than expected, a reminder instruction is issued to the outside.

5. The construction risk situation monitoring method based on deep neural network as claimed in claim 4 is characterized by: Collect several alarm methods in advance and generate an alarm method library after summarizing them; After receiving the reminder instruction, the worker's location information is determined, and the distribution status of dust and noise in the construction worker's area and his reminder preference data are used to recommend an alarm method for the construction worker based on the filtered recommendation algorithm.

6. The construction risk situation monitoring method based on deep neural network according to claim 5, characterized in that: Execute personalized alarm mode to alarm construction workers, collect the response time of each construction worker, and construct the switching value xH from several response times, as follows: Where: xT i is the response time of the i-th construction worker, xT a is the mean of the response time, IQR is the interquartile range of several response times; F1, F2 and F3 are weight coefficients.

7. The construction risk situation monitoring method based on deep neural network according to claim 6, characterized in that: The environmental risk degree Hjx and alarm data in the construction area are collected in several consecutive patrol cycles, and the patrol frequency Cv in the next patrol cycle is constrained according to the environmental risk degree Hjx and the alarm data. The construction area is patrolled according to the patrol nodes that meet the constraints, and the patrol feedback data is obtained.

8. The construction risk situation monitoring method based on deep neural network according to claim 7, characterized in that: The constraints for constraining the patrol frequency Cv in the next patrol cycle are as follows: Weight coefficient: 0≤ρ≤1, 0≤ζ≤1, and ρ+(=1; n is the number of first-level alarm instructions, U(i, j) is the time interval from the i-th first-level alarm instruction to the j-th first-level alarm instruction, and Ut a is the average value of the time interval.

9. The construction risk situation monitoring method based on deep neural network according to claim 8, characterized in that: If the number of first-level alarm instructions received during the current patrol cycle exceeds the corresponding number threshold, a warning value Jzp is generated based on the status data of the continuously issued first-level alarm instructions. If the warning 1 value Jzp exceeds the preset warning threshold, a second-level alarm instruction is issued to the outside.

10. The construction risk situation monitoring method based on deep neural network according to claim 9, characterized in that: The method of generating the warning value Jzp based on the status data of the first-level alarm instructions issued continuously is as follows: Where: Pf(t) is the mean safety score of the construction workers at time t, Δt(t) is the time interval between two first-level alarm instructions, is the rate of change of the time interval, and F1, F2 and F3 are weight coefficients.

11. The construction risk situation monitoring method based on deep neural network according to claim 10, characterized in that: Automatically generate a construction safety report after the inspection cycle is over and send it to the inspectors; Combined with the construction safety report, principal component analysis is used to determine the key factors affecting construction safety in the construction area, and each key factor is ranked according to the degree of influence. Construction plan optimization is used as the target word, and a construction plan optimization knowledge map is pre-constructed.

12. The construction risk situation monitoring method based on deep neural network according to claim 11, characterized in that: The warning threshold is adjusted according to the optimized patrol frequency. The adjustment method is as follows: Where m is the number of inspection nodes in the inspection cycle, Xj(i, j) is the time interval from the i-th inspection node to the j-th inspection node, and Xj a is the average value of the time interval, and Pa is the adjustment ratio of the warning threshold.

13. A construction risk situation monitoring system based on a deep neural network, characterized by: include, The environmental risk analysis unit collects environmental condition data in the construction area and then conducts environmental risk assessment. It generates an environmental risk level Hjx from a number of continuously acquired risk values. If the environmental risk level Hjx exceeds the risk threshold, it issues a data collection instruction to the outside. The alarm mode matching unit uses the construction safety assessment model to evaluate the construction safety in the construction area. If the safety is lower than expected, it recommends a suitable alarm mode for the construction workers from the alarm mode library. After switching the alarm mode according to the alarm feedback data, it issues a first-level alarm command to the outside. After receiving the first-level alarm command, the automatic control unit matches the corresponding operation reminder from the operation reminder library according to the construction status data and environmental condition data of the construction workers, and uses the conditional automatic control model to automatically adjust various regulating equipment in the construction area; The multi-level alarm unit patrols the construction area according to the inspection time nodes that meet the constraints, and generates a warning value Jzp from the status data of the continuous issuance of the first-level alarm command. If the warning value Jzp exceeds the expectation, the emergency plan library matches the corresponding emergency plan according to the emergency characteristics; The scheme optimization unit automatically generates a construction safety report after the inspection cycle ends. After determining the key factors affecting construction safety in the construction area, the construction scheme optimization knowledge graph outputs the corresponding construction optimization strategy based on the correspondence between the key factors and the construction optimization scheme.

Images