A construction safety hazard management method and system

Abstract

The present invention proposes a method and system for managing building safety hazards, which relate to the technical field of construction safety management. The method comprises: dividing a plurality of sub-areas to collect data, classifying known safety hazard events, and obtaining a first classification; selecting corresponding data features and fusion features through comparative analysis of baseline features and control features, and establishing risk assessment models for the first classification respectively; obtaining a first risk probability and a comprehensive risk index of each sub-area according to real-time monitoring data and the risk assessment model; issuing an early warning to the sub-area according to the first risk probability, the comprehensive risk index and / or the risk change; adjusting the early warning threshold and / or optimizing the risk assessment model according to the early warning processing result and on-site feedback; and realizing real-time monitoring, early warning and management of safety hazards at the construction site, which can not only improve the efficiency and accuracy of safety hazard management, but also reduce labor costs and improve the safety level of the construction site.

Description

Technical Field

[0001] The present application relates to the technical field of construction safety management, and in particular to a method and system for managing potential safety hazards in construction. Background Art

[0002] Safety management is a crucial part of the construction process. Due to the complex environment, large number of personnel, various types of equipment, and various potential safety hazards at the construction site, such as unsafe behavior of personnel, equipment failure, abnormal environmental factors, etc., these may lead to serious safety accidents. Therefore, how to effectively identify, evaluate and manage these safety hazards has become an important issue in construction safety management.

[0003] Traditional methods of managing construction safety hazards mainly rely on manual inspections and on-site monitoring, which has many limitations, such as high labor costs, low efficiency, and strong subjectivity. With the development of technologies such as the Internet of Things, big data, and artificial intelligence, new ideas and methods have been provided for the management of construction safety hazards; by deploying a variety of sensors at the construction site, such as temperature and humidity sensors, gas detectors, video surveillance, and RFID personnel positioning systems, it is possible to collect data such as environmental parameters, equipment status, and personnel behavior at the construction site in real time, providing data support for the identification, evaluation, and management of safety hazards.

[0004] However, relying solely on data collected by sensors is not enough to achieve effective safety hazard management. First, the construction site is vast and the work activities are diverse. How to effectively divide the monitoring area and accurately deploy sensors is a major challenge; second, multi-source heterogeneous monitoring data (such as environmental parameters, equipment operating status, personnel behavior, etc.) requires efficient pre-processing and fusion analysis methods to extract key information; third, the classification of safety hazards, the construction of risk assessment models and dynamic adjustment strategies need to be more refined and intelligent to adapt to the complexity and dynamics of the construction site; finally, how to monitor and warn the safety status of the construction site in real time, timely discover potential safety hazards, and take corresponding measures to intervene and deal with them, so as to avoid the occurrence of safety accidents; the early warning mechanism needs to have a high degree of sensitivity and accuracy, and at the same time be able to dynamically adjust the warning threshold according to the actual situation to avoid false alarms and missed alarms. Summary of the invention

[0005] Based on the above problems, the present application provides a method and system for managing safety hazards in construction. The method divides the construction area into multiple sub-areas, and uses a variety of sensors to collect monitoring data of the sub-areas. The data are then preprocessed and feature extracted. Based on the features and fusion features, safety hazard event classification modeling is combined, and different early warning mechanisms are set up to ultimately achieve real-time monitoring, early warning and management of safety hazards at the construction site. This can not only improve the efficiency and accuracy of safety hazard management, but also reduce labor costs and improve the safety level of the construction site.

[0006] The purpose of this application is achieved by the following technical solutions:

[0007] In a first aspect, the present application provides a method for managing safety hazards in construction, the method comprising:

[0008] S1. Divide the construction area into multiple sub-areas; the sub-areas include a first sub-area and a second sub-area; collect monitoring data of the multiple sub-areas through multiple sensors; the monitoring data include environmental parameters, equipment status and personnel behavior;

[0009] S2. Preprocessing the monitoring data of different sensors, wherein the preprocessing includes feature extraction and alignment of data features of different sensors at the same time and / or space;

[0010] S3. Classify known safety hazard events to obtain a first classification; select corresponding data features and fusion features through comparative analysis of baseline features and control features, and establish risk assessment models for the first classification respectively; wherein the baseline features are data features of each sensor in a safe state; and the control features are data features of each sensor of a safety hazard event under the same first classification;

[0011] S4. Obtaining a first risk probability and a comprehensive risk index for each sub-area based on the real-time monitoring data and the risk assessment model;

[0012] S5. Issue early warning to the sub-area based on the first risk probability, the comprehensive risk index and / or the risk change;

[0013] S6. Collect early warning processing results and on-site feedback; adjust the early warning threshold and / or optimize the risk assessment model based on the early warning processing results and on-site feedback.

[0014] The present application provides a construction safety hazard management system, the system comprising:

[0015] A data acquisition module is used to divide the construction area into multiple sub-areas; the sub-areas include a first sub-area and a second sub-area; and to collect monitoring data of the multiple sub-areas through multiple sensors; the monitoring data includes environmental parameters, equipment status, and personnel behavior;

[0016] A data preprocessing module, used for preprocessing the monitoring data of different sensors, wherein the preprocessing includes feature extraction and alignment of data features of different sensors at the same time and / or space;

[0017] The model building module is used to classify known safety hazard events to obtain a first classification; by comparing and analyzing the baseline features and the control features, the corresponding data features and fusion features are selected to respectively establish a risk assessment model for the first classification; wherein the baseline features are the data features of each sensor in a safe state; and the control features are the data features of each sensor of the safety hazard event under the same first classification;

[0018] A monitoring module, used to obtain the first risk probability and comprehensive risk index of each sub-area according to the real-time monitoring data and the risk assessment model;

[0019] An early warning module, used for issuing an early warning to a sub-region according to the first risk probability, the comprehensive risk index and / or the risk change;

[0020] The optimization module is used to collect early warning processing results and on-site feedback; adjust the early warning threshold and / or optimize the risk assessment model based on the early warning processing results and on-site feedback.

[0021] The beneficial effects of the present invention include: by dividing the construction area into multiple sub-areas, and using multiple sensors to monitor environmental parameters, equipment status and personnel behavior, the micro-monitoring of the safety status of the construction site is realized, which is helpful to discover and prevent safety hazards as early as possible. Based on the comparative analysis of the baseline characteristics and the control characteristics, combined with the fusion of data characteristics, the established risk assessment model can intelligently identify potential safety risks, not only considering a single factor, but also considering the composite influence of multiple factors, thereby improving the accuracy and comprehensiveness of risk identification. By combining real-time monitoring data with the risk assessment model, the first risk probability and the comprehensive risk index of each sub-area are dynamically calculated to realize a hierarchical early warning mechanism, which is helpful to take countermeasures in time and effectively prevent the occurrence of safety accidents. Through the closed-loop feedback system, not only timely response is made according to the early warning results, but also by collecting processing results and on-site feedback, the early warning threshold is continuously adjusted and the risk assessment model is optimized, ensuring the self-learning and continuous optimization capabilities of the system, and improving management efficiency and effect. Through accurate risk identification and early warning, safety management resources, such as human resources, material allocation, etc., can be more reasonably allocated, problems in high-risk areas can be solved first, waste of resources can be avoided, and the economy and efficiency of safety management can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of a construction safety hazard management method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] Below, the present application is further described in conjunction with the accompanying drawings and specific implementation methods. It should be noted that, under the premise of no conflict, the various embodiments or technical features described below can be arbitrarily combined to form a new embodiment.

[0024] Reference Figure 1 Some embodiments of the present application provide a method for managing safety hazards in construction, the method comprising:

[0025] S1. Divide the construction area into multiple sub-areas; the sub-areas include a first sub-area and a second sub-area; collect monitoring data of the multiple sub-areas through multiple sensors; the monitoring data include environmental parameters, equipment status and personnel behavior;

[0026] S2. Preprocessing the monitoring data of different sensors, wherein the preprocessing includes feature extraction and alignment of data features of different sensors at the same time and / or space;

[0027] S3. Classify known safety hazard events to obtain a first classification; select corresponding data features and fusion features through comparative analysis of baseline features and control features, and establish risk assessment models for the first classification respectively; wherein the baseline features are data features of each sensor in a safe state; and the control features are data features of each sensor of a safety hazard event under the same first classification;

[0028] S4. Obtaining a first risk probability and a comprehensive risk index for each sub-area based on the real-time monitoring data and the risk assessment model;

[0029] S5. Issue early warning to the sub-area based on the first risk probability, the comprehensive risk index and / or the risk change;

[0030] S6. Collect early warning processing results and on-site feedback; adjust the early warning threshold and / or optimize the risk assessment model based on the early warning processing results and on-site feedback.

[0031] The working principle of the above technical solution is:

[0032] This method mainly focuses on the core steps of data collection, preprocessing, model building, risk assessment and early warning feedback, specifically:

[0033] Data collection (S1):

[0034] The construction area is first divided into multiple sub-areas, which can be further subdivided into first sub-areas and second sub-areas based on factors such as construction type, operation complexity and personnel density.

[0035] A variety of sensors (such as temperature and humidity sensors, gas detectors, video surveillance, RFID personnel positioning systems, etc.) are used to collect real-time data on environmental parameters (such as temperature and humidity, gas concentration), equipment status (such as equipment operating status, aging degree) and personnel behavior (such as personnel location, movement trajectory, working status) in these sub-areas.

[0036] Data preprocessing (S2):

[0037] The collected raw data needs to be preprocessed to eliminate noise, outliers, missing values, etc. to ensure the accuracy and consistency of the data.

[0038] Feature extraction is a key step in preprocessing, which extracts features that can characterize safety hazards from raw data.

[0039] Since data from different sensors may have different temporal and spatial scales, these data features need to be aligned at the same time and / or space for subsequent analysis and modeling.

[0040] Model building (S3):

[0041] Classify known safety hazards to form the first category, including personnel safety hazards, equipment safety hazards, and environmental safety hazards. For each category, select data features under safe conditions (baseline features) and safety hazard events (control features) for comparison to identify features with strong risk indicative properties; analyze the data features of each sensor and the fusion features (i.e., the correlation between multiple features), and based on these features, establish an assessment model for each type of risk through machine learning techniques (such as logistic regression, random forest, neural network).

[0042] Risk Assessment (S4):

[0043] Using the preprocessed real-time monitoring data as input, the first risk probability of each sub-area under various safety hazards is calculated through the established risk assessment model.

[0044] Combining various first risk probabilities, the comprehensive risk index of each sub-region is calculated, which reflects the overall safety status of the sub-region.

[0045] Early warning and feedback (S5-S6):

[0046] Based on the first risk probability, comprehensive risk index and / or risk change, early warning is issued to sub-areas with potential safety hazards; the early warning is divided into multiple levels and modes.

[0047] Collect early warning processing results and on-site feedback to understand the effectiveness of early warning measures and the actual conditions on site.

[0048] Based on this feedback, the warning threshold can be adjusted to more accurately reflect the actual security risks. At the same time, the risk assessment model can also be optimized to improve the model's prediction accuracy and generalization ability.

[0049] Overall, the construction safety hazard management method forms a closed-loop safety management system through steps such as data collection, preprocessing, model building, risk assessment, and early warning feedback. It can realize real-time monitoring, early warning, and management of safety hazards on construction sites, thereby improving the safety of construction.

[0050] The effects of the above technical solution are: by subdividing the construction area into multiple sub-areas and implementing targeted monitoring for each sub-area, refined management of construction safety is achieved; comprehensive monitoring of environmental parameters, equipment status and personnel behavior enables safety hazards to be identified in the embryonic stage and effectively prevent accidents. The real-time monitoring data combined with the pre-established risk assessment model can quickly calculate the risk probability and comprehensive risk index of each sub-area, so that managers can immediately understand the safety status of the site. Once a risk is discovered, the system immediately issues an early warning to ensure a rapid response and take measures to avoid the expansion of the accident. According to the probability and comprehensive risk index of different risks, the system can implement graded early warnings, distinguish the degree of urgency, help the rational allocation of resources, give priority to high-risk areas, and improve management efficiency and emergency response capabilities. By collecting early warning processing results and on-site feedback, the method can continue to learn and improve; the dynamic adjustment of the early warning threshold ensures the sensitivity and accuracy of the early warning system, and the continuous optimization of the risk assessment model improves the predictive ability and adaptability of the model, so that it can better match the changes in the construction environment.

[0051] In some embodiments, the S1 includes:

[0052] According to the construction type, the construction area is initially divided into different first sub-areas; based on the overall planning drawings of the construction project, combined with the construction process and special construction plans, through 3D modeling, the operation areas of different construction types are identified. For example, deep foundation pit operations, tower crane operation areas, prefabricated component installation areas, etc. are regarded as independent first sub-areas, and each area has its own unique safety control points and protection requirements;

[0053] Divide the first sub-area into multiple second sub-areas according to the complexity of the operation and the density of personnel; adjust the boundary between the first sub-area and the second sub-area according to the construction progress; including:

[0054] Set complexity scores for different operations in each first sub-area; quantitative indicators may include the number of types of work, frequency of cross-operation, proportion of special operations, and density of personnel; based on the scoring results, merge or subdivide areas of similar complexity to form the second sub-area; among them, use shift arrangements, construction plans and other information to predict the distribution of personnel in each area at different time periods; establish a dynamic monitoring system to regularly (such as weekly or biweekly) evaluate changes in construction progress, staffing, and types of operations; including reviewing information such as construction logs, on-site inspection reports, and staff attendance records, as well as using Internet of Things technology to track the movement of equipment and personnel in real time. According to the evaluation results, adjust the boundaries of the sub-areas in a timely manner; factors to be considered during the adjustment include newly launched construction activities, personnel flow trends, and the arrival of new equipment. For areas that are about to be completed, the monitoring scope can be appropriately narrowed or the monitoring intensity can be reduced; while for areas that are about to start complex operations, the monitoring area should be expanded in advance and monitoring resources should be increased.

[0055] Monitoring data of multiple sub-areas is collected through a variety of sensors, including temperature and humidity sensors, gas detectors, video surveillance, RFID personnel positioning systems, etc.

[0056] The working principle and effect of the above technical solution are:

[0057] According to the specific types of construction projects (such as residential buildings, bridges, tunnels, etc.), and taking into account the characteristics and needs of different types of construction, the entire construction area is preliminarily divided into several first sub-areas; this division is intended to adapt to the unique safety risk distribution of different construction types, such as deep foundation pits, high-altitude work areas, heavy machinery operation areas, etc., to facilitate subsequent more detailed management and monitoring.

[0058] On the basis of the first sub-area, further subdivision is performed based on the complexity of the operation and the density of personnel to form multiple second sub-areas. Areas with high operation complexity are usually accompanied by more safety hazards, such as complex cross-operations, difficult installation operations, etc., while areas with dense personnel need to pay special attention to the mutual influence between personnel and the smooth flow of evacuation routes. In addition, places with high population density are often the focus of safety management. The division at this stage is intended to improve the pertinence of safety monitoring and ensure key monitoring of high-risk areas.

[0059] The division of the second sub-area includes:

[0060] First, we need to clarify several key dimensions of complexity scoring, including:

[0061] Number of types of work: The more types of work involved in an area, the greater the diversity of coordination and professional skills required, and therefore the greater the complexity.

[0062] Frequency of cross-operation: The degree of mutual interference between different work activities in the same space or time. The higher the frequency, the greater the safety risks and coordination difficulties.

[0063] Proportion of special operations: Special operations (such as high-altitude operations, welding, blasting, etc.) often require special safety measures and professional qualifications. An increase in their proportion will significantly increase the complexity of the operations.

[0064] Crowd density: Areas with dense crowds are more likely to be crowded and chaotic, increasing the risk of accidents.

[0065] Set specific scoring criteria for each indicator, which can be a numerical range or level (such as low, medium, high) to ensure that the scoring is comparable and objective. For example:

[0066] Number of job types: For each additional job type, the complexity score increases by 1.

[0067] Frequency of cross-operation: Based on the number of cross-operations per day, less than 5 times is low complexity (1 point), 5-10 times is medium complexity (2 points), and more than 10 times is high complexity (3 points).

[0068] Proportion of special operations: 1 point for a ratio below 10%, 2 points for a ratio between 10% and 30%, and 3 points for a ratio above 30%.

[0069] Population density: Based on the number of people per square meter, less than 1 person is low density (1 point), 1-5 people are medium density (2 points), and more than 5 people are high density (3 points).

[0070] Calculate the total complexity score for each first sub-area based on the above indicators. For example, if the number of types of work in an area is 4, the frequency of cross-operation is 7 times a day, the proportion of special operations is 20%, and the personnel density is 3 people per square meter, then its total complexity score is:

[0071] Number of jobs: 4 points;

[0072] Frequency of cross-assignment: 2 points;

[0073] Special operation ratio: 2 points;

[0074] Population density: 2 points;

[0075] Total score = 4+2+2+2=10 points;

[0076] Merge similar complexity areas: If the complexity scores of adjacent or close first sub-areas are close and the nature of the work activities is similar, you can consider merging them into a larger second sub-area for unified management and resource allocation.

[0077] According to the actual operation progress, personnel flow and other factors, the complexity score of each sub-area is reviewed regularly, and the division of the second sub-area is adjusted accordingly to adapt to the dynamic changes of the construction site. Through the above steps, the complexity score can be set scientifically and reasonably for each first sub-area, and a more appropriate second sub-area division can be formed accordingly, thereby improving the efficiency and effectiveness of safety management.

[0078] Various types of sensors are deployed in each sub-area, including but not limited to temperature and humidity sensors to monitor environmental conditions to ensure a suitable working environment; gas detectors to monitor the concentration of harmful gases to prevent poisoning and explosion accidents; video surveillance to capture on-site conditions in real time, monitor personnel operation compliance and on-site order; and RFID personnel positioning systems to track workers' real-time locations to facilitate rapid evacuation and rescue in emergency situations. These sensors work together to comprehensively and continuously collect environmental parameters, equipment status, and personnel behavior data in each sub-area, providing basic information for subsequent analysis and early warning.

[0079] In some embodiments, the S2 includes:

[0080] Performing a first processing on the monitoring data to obtain first processed data; the first processing includes data cleaning and outlier detection;

[0081] Performing standardization on the first processed data of different sensors; obtaining the first data after the standardization;

[0082] Features are extracted from the standardized first data and time alignment and / or spatial alignment is performed.

[0083] The working principle and effect of the above technical solution are as follows: Through this step, irrelevant information (such as noise, duplicate records, etc.) and erroneous data (such as sensor false alarms, erroneous readings caused by communication failures, etc.) in the original monitoring data are removed. By setting data quality standards, filtering, deduplication, missing value processing and other methods are used to ensure the purity and reliability of the data set. Statistical methods (such as box plot analysis, Z-score test, IQR interval method, etc.) or machine learning algorithms (such as cluster analysis, isolation forest, etc.) are used to identify and process outliers in the data. These outliers may be caused by sensor failures, extreme environmental conditions or other unexpected factors; removing or correcting these outliers can prevent them from causing deviations in subsequent analysis.

[0084] After data cleaning and outlier processing, the data generated by each sensor is standardized to eliminate the dimension effect so that data from different sensors or at different time points can be compared on the same scale. Standardization methods can include Min-Max Scaling, Z-score standardization, etc. Through standardization, the numerical range of each feature in the data set is unified, which facilitates subsequent feature extraction and model construction.

[0085] Extract key features that reflect safety hazards from the standardized data. These features may include the extreme values ​​of environmental parameters, the rate of change of equipment status, the pattern of personnel behavior, etc. Feature selection is performed through feature importance evaluation algorithms (such as correlation analysis, principal component analysis PCA, etc.).

[0086] Time-space alignment: Since the timestamps of data collected by different sensors may be different, and the data in spatial locations also need to be integrated, the data collection time points are adjusted through time series analysis to ensure that the data from different sensors at the same or similar time points can be compared and integrated. Spatial alignment refers to the integration of data from different geographical locations based on their spatial relationships to ensure that spatial correlation is taken into account during analysis, which is crucial for understanding the overall safety status within the construction area.

[0087] Through the above steps, it is ensured that the raw data collected from various sensors undergoes strict quality control and format conversion, and is transformed into high-quality, easy-to-analyze data sets that can be used for risk assessment, providing a solid data foundation for subsequent risk model construction and early warning mechanisms.

[0088] In some embodiments, the S3 includes:

[0089] Classify the known safety hazard events to obtain a first classification; the first classification includes personnel safety hazards, equipment safety hazards and environmental safety hazards; for example, personnel safety hazards may include unsafe conditions of personnel (such as fatigue, failure to wear protective equipment) and unsafe behaviors (such as illegal operations, ignoring warnings), equipment safety hazards may include equipment failures (such as mechanical failure, electrical short circuit), high-altitude fall risks, excessive gas, etc., and environmental safety hazards include foundation collapse, adverse weather conditions, etc.; historical data may be used for statistical analysis and pattern recognition, and machine learning algorithms (such as cluster analysis, decision trees, etc.) may be used to automatically discover patterns and categories of safety hazards;

[0090] Classify the risk levels of the events in each first category according to their severity; a five-level or more risk level system may be adopted, and the specific indicators and thresholds for each level of risk should be clearly defined, such as mild, moderate, severe, and extremely high risk;

[0091] The risk assessment models of the first category are established through the data characteristics and fusion characteristics of each sensor. The risk level is divided not only based on the static event type and severity, but also the influence of dynamic variables such as time series analysis, seasonal factors, special weather conditions, etc., so that the risk assessment can be adjusted with time and environmental changes.

[0092] The working principle and effect of the above technical solution are as follows: First, the system collects a large amount of information about known safety hazard events from historical databases, accident reports, monitoring records and other sources; and uses machine learning algorithms, such as cluster analysis and decision trees, to conduct in-depth analysis of the collected data. These algorithms can automatically identify the patterns and associations between different safety hazard events, thereby classifying them into three categories: personnel safety hazards, equipment safety hazards, and environmental safety hazards, and further subdividing them, such as specifying unsafe conditions and behaviors of personnel into subcategories such as fatigue, failure to wear protective equipment, and illegal operations; based on factors such as the potential impact, probability of occurrence, and controllability of safety hazards, a detailed risk level system is designed, such as a five-level division from mild to extremely high risk. Integrate real-time data from different sensors (such as temperature and humidity sensors, vibration sensors, video surveillance, etc.) and static features in historical data to form a comprehensive feature set. In addition to static event types and severity, the model will also incorporate time series analysis, consider dynamic variables such as seasonal changes and special weather conditions, and enhance the model's time sensitivity and environmental adaptability.

[0093] For each category, a risk assessment model is established using machine learning or deep learning models (such as support vector machines and neural networks). These models can dynamically predict the probability of occurrence and possible severity of various safety hazards based on current environmental conditions, equipment status, and personnel behavior.

[0094] In some embodiments, the risk assessment model of the first category is established respectively by using the data features and fusion features of each sensor; including:

[0095] Obtaining data features of each sensor in a safe state as reference features;

[0096] Obtaining data features of each sensor of a potential safety hazard event under the same first category as a control feature;

[0097] Compare each baseline feature with the control feature; obtain the difference or similarity between each feature; if the feature is a numerical value, obtain the difference between the baseline feature and the control feature; if the feature is a picture, text, audio, etc.; calculate the similarity between the baseline feature and the control feature;

[0098] The similarity is:

[0099]

[0100]

[0101] is the similarity between the control feature and the benchmark feature; is the number of sub-features contained in the control feature; This is the first The sub-features and the base features The maximum similarity among the sub-features; is the number of sub-features contained in the reference feature; This is the first The sub-features and the base features Similarity of individual features; To obtain the maximum value; To take the minimum value;

[0102] If the difference is greater than a preset difference threshold or the similarity is less than a preset similarity threshold; the corresponding feature is used as the first input of the risk assessment model; wherein the risk assessment model can be any one of logistic regression, random forest, and neural network;

[0103] Perform correlation analysis on each control feature; use the feature whose correlation coefficient with the first input feature is greater than a preset coefficient threshold as the second input of the risk assessment model, and use the fusion feature of the second input and the first input as the candidate third input; the first input and the second input may be fused by weighted average, or a deep learning model (such as a multi-layer perceptron, an attention mechanism, etc.) may be used to perform a high-level integration of features from different modalities;

[0104] Obtaining a score of the fusion feature of the second input and the first input according to a ratio of a preset difference of the first input to a difference threshold or a ratio of a preset similarity to the first input similarity, and a correlation coefficient between the first input and the second input;

[0105] The scores of the fused features are:

[0106]

[0107] in, For the The score of the fusion features; For the The ratio of the first input preset difference corresponding to the fusion feature to the difference threshold or the ratio of the preset similarity to the first input similarity; For the The correlation coefficient between the first input and the second input corresponding to the fusion feature; , is the weight coefficient;

[0108] ranking the candidate third inputs according to the scores of the fused features;

[0109] According to the sorting results, the third alternative input is gradually added, and the risk assessment model is trained through the first input and the second input; the degree of equipment aging is used as the input for equipment safety hazard assessment;

[0110] The model performance was evaluated through cross-validation to obtain the final first risk assessment model.

[0111] The working principle of the above technical solution is as follows: First, extract features from the sensor data collected in a safe state as baseline features, which represent the normal state when there are no safety hazards. At the same time, for each type of safety hazard event, collect the sensor data features at the corresponding time or conditions as reference features, which reflect the specific state of the safety hazard; compare the baseline features with the reference features one-to-one; for numerical features, directly calculate the difference; for non-numerical features (such as images, text, audio), calculate the similarity between features through a specific algorithm.

[0112] Among them, the similarity formula takes into account the matching of sub-features and measures the overall similarity by combining the maximum similarity and the total number of sub-features, ensuring the comprehensiveness and accuracy of the evaluation.

[0113] For example, there are two pictures as the baseline feature and the comparison feature, and both pictures contain several visual elements (such as color distribution, texture, shape, etc.). The sub-features here can be:

[0114] Color Histogram: Quantifies the frequency with which each color occurs.

[0115] Edge detection: extract edge contour information from images.

[0116] Texture features: features extracted by texture analysis methods (such as Gabor filters and gray-level co-occurrence matrices).

[0117] Key point matching: such as SIFT feature points, which are used to describe invariant local features in an image.

[0118] For image similarity calculation, It can be the key point matching score of two images at the same position or feature space, the overlap of color histograms, or the correlation of texture features.

[0119] Or in text analysis, the text can be segmented into words, n-grams (n consecutive words), topics, etc. as sub-features:

[0120] Term Frequency-Inverse Document Frequency (TF-IDF): measures the importance of each word in a document.

[0121] Topic model (such as LDA): Represent documents as topic vectors, with each topic considered as a sub-feature.

[0122] Sentiment score: The sentiment polarity of each sentence or paragraph can be used as a sub-feature.

[0123] Here, it can be understood as the similarity between two documents on the same topic, the proportion of shared vocabulary, or the consistency of sentiment tendency.

[0124] Based on the comparison of the difference or similarity with the preset threshold, the control features that significantly deviate from the baseline features are screened out as the first input of the risk assessment model; this step aims to identify feature changes that are directly related to safety hazards.

[0125] Perform correlation analysis on all control features, find out the features that are highly correlated with the first input features, use them as the second input, and try to fuse the features of the first input with the second input to generate an alternative third input. The first input and the second input can be fused by weighted average, or a deep learning model (such as multi-layer perceptron, attention mechanism, etc.) can be used to perform a high-level integration of features from different modalities; this is not specifically limited here and depends on the specific situation;

[0126] The scoring mechanism of fused features comprehensively considers the deviation degree, correlation and weight coefficient of the features to ensure that the fused features are both sensitive and have practical predictive value.

[0127] The candidate third input features are sorted according to the scores, and then the risk assessment model is gradually built and trained based on the first input, the second input, and the sorted third input features. This process is implemented through machine learning algorithms (such as logistic regression, random forest, neural network), which aims to learn the complex relationship between features and safety hazards. When performing cross-validation, if the cross-validation score of the model (such as mean square error, mean absolute error, etc.) changes gradually after adding features several times in a row, it means that the model performance has reached saturation, and you can stop adding the features of the candidate third input.

[0128] In the equipment safety hazard assessment, the aging factor of the equipment is considered as an additional input because it directly affects the performance and safety of the equipment. Finally, the prediction ability of the model is evaluated through cross-validation to ensure that the model can perform well on unseen data, and finally the first risk assessment model is obtained.

[0129] The whole process reflects a systematic approach from feature extraction to model building. It not only focuses on the importance of a single feature, but also emphasizes the importance of interaction and fusion between features, aiming to build an efficient and accurate risk prediction model.

[0130] The effect of the above technical solution is that by comparing the sensor data features in a safe state and under a potential safety event, the subtle differences in state changes can be accurately captured. In particular, by setting the difference threshold and similarity threshold, non-critical changes can be effectively filtered out, focusing on the features that may actually indicate risks, thereby improving the accuracy of risk identification.

[0131] Not only does it take into account the changes in individual features, but it also achieves multi-dimensional risk assessment by calculating the similarity and correlation coefficient between features. This comprehensive analysis method can more comprehensively evaluate the safety status of equipment or systems and reduce the probability of false alarms and missed alarms.

[0132] The model can automatically adjust according to preset thresholds, which means it can adapt to risk assessment needs under different environmental or equipment conditions, increasing the practicality and flexibility of the model.

[0133] By fusing the first input and the second input through weighted averaging or deep learning models, it is possible to effectively integrate information from different sources, explore potential complex associations, and enhance the model's predictive and generalization capabilities. In particular, the application of deep learning models can handle nonlinear relationships and better understand and simulate complex patterns in the real world.

[0134] The calculation of the fusion feature score gives each feature a quantitative risk level by combining the key indicators of the first input and the second input, as well as the correlation between them. This quantitative method makes it easier for decision makers to intuitively understand the urgency and importance of risks, and guide the setting of priorities and the allocation of resources.

[0135] The model performance is evaluated through cross-validation, which ensures the effectiveness and reliability of the risk assessment model. At the same time, it allows the model to be continuously adjusted and optimized to achieve the best prediction effect and adapt to new challenges brought about by factors such as equipment aging and environmental changes.

[0136] In some embodiments, the S4 includes:

[0137] According to the real-time monitoring data and the risk assessment model, a first risk probability and a first risk level of each second sub-area under each first category are obtained;

[0138] Obtaining a comprehensive risk index of the second sub-region according to the plurality of first risk probabilities and the first risk levels of the second sub-region;

[0139] Among them, the comprehensive risk index is:

[0140]

[0141]

[0142] in, For the A second sub-regional comprehensive risk index; For the The second sub-region The first risk probability of the first category; For the The second sub-region The weight of the first risk probability of the first category; For the A first risk level score for the first category; For the The frequency score of the first risk in the first category; , is the coefficient, ; ; ; For the The number of first risk probabilities obtained for the second sub-areas;

[0143] According to the multiple first risk levels of the second sub-area, the comprehensive risk level of the second sub-area is obtained; it can be a weighted average, or it can be the maximum value among multiple levels, or it can be the maximum value plus the average of other levels, depending on the specific situation, and is not specifically limited here.

[0144] Conduct time series analysis on the comprehensive risk index to identify trends and cyclical characteristics of risk changes.

[0145] The working principle of the above technical solution is as follows: First, the system continuously receives real-time monitoring data from various sensors in the second sub-area. These data cover multiple dimensions such as environmental parameters, equipment status, and personnel behavior. Subsequently, these data are input into the risk assessment model established for each first category (such as personnel safety hazards, equipment safety hazards, and environmental safety hazards). The model predicts the risk probability of each category under the current state based on the patterns and relationships learned from historical data. According to the risk probability, combined with professional knowledge or historical data, a risk level is assigned to each risk event under the first category to reflect its severity. At the same time, the frequency of risk occurrence is evaluated to comprehensively consider the urgency of the risk. The weight of the risk probability of each first category is calculated using the risk probability, risk level score, and risk occurrence frequency score. The weight reflects the relative importance of each category risk, and It is a coefficient that adjusts the influence of risk level and frequency of occurrence. The comprehensive risk index is the weighted sum of all first-class risk probabilities and their corresponding weights, which represents the overall risk level of the second sub-area.

[0146] After obtaining the comprehensive risk index, the comprehensive risk index of each second sub-area can be converted into an easy-to-understand comprehensive risk level, such as low risk, medium risk, high risk or extremely high risk, according to preset standards or thresholds, so as to intuitively show the severity of the risk.

[0147] Time series analysis of the comprehensive risk index of each second sub-region is conducted to identify the trend, periodic characteristics or abnormal fluctuations of risk changes. By analyzing the changes in the risk index over time, future risk trends can be predicted and a basis for taking preventive measures in advance can be provided. This step may involve statistical methods such as moving average, trend analysis, and periodic pattern recognition.

[0148] The effect of the above technical solution is: through the combination of real-time monitoring data and risk assessment models, the risks of each second sub-area in different dimensions such as personnel, equipment, and environment can be accurately quantified, that is, the first risk probability, which greatly improves the accuracy and pertinence of risk identification and makes safety management more refined. The calculated comprehensive risk index integrates the risk probabilities of each first category and considers the risk level and frequency of occurrence through weights. This comprehensive perspective ensures a comprehensive grasp of the overall risk of the sub-area, helps to prioritize areas with higher risks, and effectively allocate resources. The weight calculation formula and The coefficient can be adjusted according to actual management needs to balance the impact of the severity and frequency of risk levels, which makes risk assessment more flexible and able to adapt to the risk management focus at different project stages or under specific conditions.

[0149] Converting the comprehensive risk index into a comprehensive risk level provides decision makers with an intuitive indication of the severity of the risk, which is conducive to rapid response measures and is easy for on-site personnel to understand and implement. By conducting time series analysis on the comprehensive risk index, the long-term trend and cyclical characteristics of risk changes can be identified, which not only helps to understand the law of risk evolution, but also predicts possible risk peaks in the future, achieves forward-looking management, takes preventive measures in advance, and reduces the occurrence of accidents. By continuously monitoring the changes in the risk index and combining it with actual event feedback, the risk assessment model and management strategy can be continuously adjusted to form a closed loop of continuous improvement and continuously improve the efficiency and effectiveness of safety management.

[0150] In some embodiments, the S5 includes:

[0151] If any first risk probability is greater than a preset probability threshold or the comprehensive risk index is greater than or equal to a preset index threshold, and the corresponding risk level is less than or equal to a preset level, a first warning is issued; the risk level includes the first risk level or the comprehensive risk level;

[0152] If any of the first risk probabilities is greater than the preset probability threshold or the comprehensive risk index is greater than or equal to the preset index threshold, and the corresponding risk level is greater than or equal to the preset level, a second warning is issued;

[0153] If any two sets of data among the multiple first risk probabilities and comprehensive risk indexes are greater than or equal to the corresponding preset thresholds, a third warning is issued;

[0154] If any first risk probability is greater than the preset probability threshold or the comprehensive risk index is less than the preset index threshold, but when the first risk probability or comprehensive risk index predicted for multiple consecutive times shows an upward trend; or, any first risk probability or comprehensive risk index conforms to the periodic law and enters the warning cycle within the preset time, a fourth warning will be issued.

[0155] The working principle and effect of the above technical solution are: by setting different early warning rules, the risk probability, comprehensive risk index and risk level calculated according to real-time monitoring data are used to trigger early warning mechanisms at different levels, so as to achieve multi-level and forward-looking management of construction safety hazards. The details are as follows:

[0156] First warning triggering condition: When the system detects that the risk probability of any first category exceeds the preset probability threshold, or the comprehensive risk index reaches or exceeds the preset index threshold, and the corresponding risk level assessment is low (less than or equal to the preset level), the system will initiate the first warning.

[0157] Second warning triggering condition: If any of the first risk probabilities exceeds the preset probability threshold, or the comprehensive risk index also exceeds the preset index threshold, but the corresponding risk level assessment is higher (greater than or equal to the preset level), it indicates that there are more serious safety hazards and immediate action may be required to avoid accidents. The system will issue a second warning.

[0158] The third warning trigger condition: When multiple groups (at least two groups) of risk indicators are detected to exceed their corresponding preset thresholds at the same time, even if a single indicator may not reach the warning standard, this also indicates the accumulation of risks in multiple dimensions, and we need to be vigilant about potential chain reactions. At this time, the system executes the third warning.

[0159] Fourth warning triggering conditions: Even if the current risk probability or comprehensive risk index does not directly reach the warning threshold, the system will activate the fourth warning if one of the following two situations occurs:

[0160] If monitoring data shows that although the current risk level seems stable or slightly below the threshold, the risk probability or comprehensive risk index shows a continuous upward trend through time series analysis, this suggests the possibility of increased risk and needs to be prepared in advance. If the risk shows obvious cyclical patterns and the forecast model indicates that a historically high risk period is about to begin, even if the current indicators seem safe, an early warning should be issued to preventively control the risk.

[0161] Through the above-mentioned early warning mechanism, S5 realizes dynamic monitoring and graded response to safety hazards, ensures appropriate and timely response measures for safety issues of different severity and development trends, and effectively improves the predictability and effectiveness of construction safety management. By setting different early warning levels, the corresponding level of alarm is triggered according to the severity and urgency of the risk, ensuring that the response measures match the actual risk. Not only the probability of a single risk is considered, but also the risk index and level, as well as the interaction between them, to ensure the comprehensiveness and comprehensiveness of risk assessment. This multi-dimensional consideration can more accurately reflect the actual safety situation and avoid the one-sidedness that may be caused by a single indicator assessment. By monitoring the trend of risk probability and index changes, even if the current risk level does not reach the threshold, early warning can be given, realizing the transition from passive response to active prevention, and effectively reducing the occurrence of safety accidents. Through a precise early warning mechanism, resources can be mobilized in a targeted manner to avoid unnecessary panic and waste of resources. In the case of limited resources, ensure that key areas and high-risk periods receive priority attention and protection.

[0162] In some embodiments, the S6 includes:

[0163] Automatically learn and optimize threshold settings based on historical warning results and actual events;

[0164] Risk assessment models are regularly retrained with new data.

[0165] The working principle and effect of the above technical solution are as follows: First, the system will collect and organize historical warning records and the corresponding actual event occurrences. These data include but are not limited to warning time, warning level, warning parameters (such as vibration frequency, pressure value, etc.), actual event type, specific time of event occurrence, etc.

[0166] Use statistical methods and machine learning models, such as confusion matrix and ROC curve, to evaluate the historical performance of the early warning model and identify whether the early warning threshold setting is reasonable, that is, whether the sensitivity (true positive rate) and specificity (true negative rate) of the early warning meet expectations. Apply algorithms such as gradient boosting, random forest or deep learning to learn the matching of historical warnings and actual events, and identify the threshold interval that can best distinguish "false positives" from "false negatives". Through iterative optimization, such as gradient descent method to find the minimum value of the loss function, the threshold parameters of the early warning model are automatically adjusted, so that the model can better balance false positives and false negatives in future predictions and improve the effectiveness of early warnings.

[0167] Periodically or when a new data set is available, the system automatically integrates the latest monitoring data into the training set. Retrain the warning model using the latest data set. This may involve full training (training on all historical data) or incremental learning (fine-tuning using only new data or combining some historical data), depending on the model type and resource constraints. Continue to track the warning effect of the model in actual applications, collect new warning results and actual event feedback to form a closed loop. Based on the feedback from real-time monitoring, the system can automatically trigger the optimization of thresholds or model parameters to achieve continuous self-learning and adaptive improvement.

[0168] The present application provides a construction safety hazard management system, the system comprising:

[0169] A data acquisition module is used to divide the construction area into multiple sub-areas; the sub-areas include a first sub-area and a second sub-area; and to collect monitoring data of the multiple sub-areas through multiple sensors; the monitoring data includes environmental parameters, equipment status, and personnel behavior;

[0170] A data preprocessing module, used for preprocessing the monitoring data of different sensors, wherein the preprocessing includes feature extraction and alignment of data features of different sensors at the same time and / or space;

[0171] The model building module is used to classify known safety hazard events to obtain a first classification; by comparing and analyzing the baseline features and the control features, the corresponding data features and fusion features are selected to respectively establish a risk assessment model for the first classification; wherein the baseline features are the data features of each sensor in a safe state; and the control features are the data features of each sensor of the safety hazard event under the same first classification;

[0172] A monitoring module, used to obtain the first risk probability and comprehensive risk index of each sub-area according to the real-time monitoring data and the risk assessment model;

[0173] An early warning module, used for issuing an early warning to a sub-region according to the first risk probability, the comprehensive risk index and / or the risk change;

[0174] The optimization module is used to collect early warning processing results and on-site feedback; adjust the early warning threshold and / or optimize the risk assessment model based on the early warning processing results and on-site feedback.

[0175] In some embodiments, the data acquisition module includes:

[0176] A first division unit is used to preliminarily divide the construction area into different first sub-areas according to the construction type;

[0177] The second division unit is used to divide the first sub-area into a plurality of second sub-areas according to the complexity of the operation and the density of personnel; and adjust the boundary between the first sub-area and the second sub-area according to the construction progress;

[0178] The collection unit is used to collect monitoring data of multiple sub-areas through multiple sensors.

[0179] In some embodiments, the data preprocessing module includes:

[0180] A first processing unit, configured to perform a first processing on the monitoring data to obtain first processed data; the first processing includes data cleaning and outlier detection;

[0181] A standardization unit, used for performing standardization processing on the first processed data of different sensors to obtain the first data after the standardization processing;

[0182] The feature extraction unit is used to extract features from the first data after the standardization process and perform time alignment and / or space alignment.

[0183] In some embodiments, the model building module includes:

[0184] An event classification unit, used to classify the known safety hazard events to obtain a first classification; the first classification includes personnel safety hazards, equipment safety hazards and environmental safety hazards;

[0185] A grading unit, used to classify the risk level of each event in the first category according to the severity;

[0186] The modeling unit is used to establish a risk assessment model of the first category respectively through the data features and fusion features of each sensor.

[0187] In some embodiments, the modeling unit includes:

[0188] A reference feature acquisition subunit, used to obtain data features of each sensor in a safe state as reference features;

[0189] The comparison feature acquisition subunit obtains data features of each sensor of the potential safety hazard event under the same first category as comparison features;

[0190] The feature comparison subunit is used to compare each reference feature with the control feature; obtain the difference or similarity between each feature; if the feature is a numerical value, obtain the difference between the reference feature and the control feature; if the feature is a picture, text, audio, etc.; calculate the similarity between the reference feature and the control feature;

[0191] The similarity is:

[0192]

[0193]

[0194] is the similarity between the control feature and the benchmark feature; is the number of sub-features contained in the control feature; This is the first The sub-features and the base features The maximum similarity among the sub-features;

[0195] is the number of sub-features contained in the reference feature; This is the first The sub-features and the base features Similarity of individual features; To obtain the maximum value; To take the minimum value;

[0196] A first selection subunit is used for, if the difference is greater than a preset difference threshold or the similarity is less than a preset similarity threshold, using the corresponding feature as the first input of the risk assessment model; wherein the risk assessment model can be any one of logistic regression, random forest, and neural network;

[0197] The second selection subunit is used to perform correlation analysis on each comparison feature; the feature whose correlation coefficient with the first input feature is greater than a preset coefficient threshold is used as the second input of the risk assessment model, and the fusion feature of the second input and the first input is used as the candidate third input; the first input and the second input can be fused by weighted average, or a deep learning model (such as a multi-layer perceptron, an attention mechanism, etc.) can be used to perform a high-level integration of features from different modalities; no specific limitation is made here, and it depends on the specific situation;

[0198] A fusion feature scoring unit, used to obtain a score of the fusion feature of the second input and the first input according to a ratio of a preset difference of the first input to a difference threshold or a ratio of a preset similarity to a similarity of the first input, and a correlation coefficient between the first input and the second input;

[0199] The scores of the fused features are:

[0200]

[0201] in, For the The score of the fusion features; For the The ratio of the first input preset difference corresponding to the fusion feature to the difference threshold or the ratio of the preset similarity to the first input similarity; For the The correlation coefficient between the first input and the second input corresponding to the fusion feature; , is the weight coefficient;

[0202] A sorting subunit, used for sorting the candidate third inputs according to the scores of the fused features;

[0203] A modeling subunit is used to gradually add a third candidate input according to the sorting result and to train the risk assessment model through the first input and the second input; and to use the equipment aging degree as an input for the equipment safety hazard assessment;

[0204] The validation subunit is used to evaluate the model performance through cross-validation to obtain the final first risk assessment model.

[0205] In some embodiments, the monitoring module includes:

[0206] A first risk level determination unit is used to obtain a first risk probability and a first risk level of each second sub-area under each first category according to the real-time monitoring data and the risk assessment model;

[0207] A comprehensive risk index determination unit, configured to obtain a comprehensive risk index of the second sub-region according to a plurality of first risk probabilities and first risk levels of the second sub-region;

[0208] Among them, the comprehensive risk index is:

[0209]

[0210]

[0211] in, For the A second sub-regional comprehensive risk index; For the The second sub-region The first risk probability of the first category; For the The second sub-region The weight of the first risk probability of the first category; For the A first risk level score for the first category; For the The frequency score of the first risk in the first category; , is the coefficient, ; ; ; For the The number of first risk probabilities obtained for the second sub-areas;

[0212] The comprehensive risk level determination unit obtains the comprehensive risk level of the second sub-area based on multiple first risk levels of the second sub-area; it can be a weighted average, or the maximum value among multiple levels, or the maximum value plus the average of other levels, depending on the specific situation and is not specifically limited here.

[0213] The trend cycle analysis unit is used to conduct time series analysis on the comprehensive risk index and identify the trend and periodic characteristics of risk changes.

[0214] In some embodiments, the early warning module includes:

[0215] A first warning unit, configured to issue a first warning if any first risk probability is greater than a preset probability threshold or a comprehensive risk index is greater than or equal to a preset index threshold, and the corresponding risk level is less than or equal to a preset level; the risk level includes the first risk level or the comprehensive risk level;

[0216] A second warning unit is used to issue a second warning if any first risk probability is greater than a preset probability threshold or the comprehensive risk index is greater than or equal to a preset index threshold, and the corresponding risk level is greater than or equal to a preset level;

[0217] A third warning unit, configured to issue a third warning if any two sets of data among the plurality of first risk probabilities and comprehensive risk indexes are greater than or equal to corresponding preset thresholds;

[0218] The fourth warning unit is used to issue a fourth warning if any first risk probability is greater than a preset probability threshold or the comprehensive risk index is less than a preset index threshold, but when the first risk probability or the comprehensive risk index predicted for multiple consecutive times shows an upward trend; or, any first risk probability or the comprehensive risk index conforms to a periodic law and enters a warning cycle within a preset time.

[0219] In some embodiments, the optimization module includes:

[0220] The threshold optimization unit automatically learns and optimizes the threshold settings based on historical warning results and actual events;

[0221] The periodic training unit is used to periodically retrain the risk assessment model using new data.

[0222] The principles and effects of the above technical solution are consistent with the principles and effects of a construction safety hazard management method in an embodiment of the present application, and will not be elaborated here.

[0223] This application is explained from the perspectives of purpose of use, effectiveness, progress and novelty, and has met the functional enhancement and usage requirements emphasized by the Patent Law. The above description and drawings of this application are only the preferred embodiments of this application, and are not intended to limit this application. Therefore, all structures, devices, features, etc. that are similar or identical to this application, that is, all equivalent replacements or modifications made in accordance with the scope of the patent application of this application, should fall within the scope of protection of the patent application of this application.

Claims

1. A method for managing potential safety hazards in construction, characterized in that: The method comprises: S1. Divide the construction area into multiple sub-areas; the sub-areas include a first sub-area and a second sub-area; collect monitoring data of the multiple sub-areas through multiple sensors; the monitoring data include environmental parameters, equipment status and personnel behavior; S2. Preprocessing the monitoring data of different sensors, wherein the preprocessing includes feature extraction and alignment of data features of different sensors at the same time and / or space; S3. Classify known safety hazard events to obtain a first classification; select corresponding data features and fusion features through comparative analysis of baseline features and control features, and establish risk assessment models for the first classification respectively; wherein the baseline features are data features of each sensor in a safe state; and the control features are data features of each sensor of a safety hazard event under the same first classification; S4. Obtaining a first risk probability and a comprehensive risk index for each sub-area based on the real-time monitoring data and the risk assessment model; S5. Issue early warning to the sub-area based on the first risk probability, the comprehensive risk index and / or the risk change; S6. Collect early warning processing results and on-site feedback; adjust the early warning threshold and / or optimize the risk assessment model based on the early warning processing results and on-site feedback; The S1 includes: According to the construction type, the construction area is initially divided into different first sub-areas; Divide the first sub-area into multiple second sub-areas according to the complexity of the operation and the density of personnel; adjust the boundary between the first sub-area and the second sub-area according to the construction progress; Collect monitoring data of multiple sub-areas through various sensors; The S3 includes: Classifying the known potential safety hazard events to obtain a first classification; the first classification includes personnel safety hazards, equipment safety hazards and environmental safety hazards; Classify the events in each first category into risk levels according to their severity; Through the data features and fusion features of each sensor, a risk assessment model for the first category is established respectively; The risk assessment model of the first category is established respectively through the data features and fusion features of each sensor; including: Obtaining data features of each sensor in a safe state as reference features; Obtaining data features of each sensor of a potential safety hazard event under the same first category as a control feature; Compare each reference feature with the control feature; obtain the difference or similarity between each feature; If the difference is greater than the preset difference threshold or the similarity is less than the preset similarity threshold; the corresponding feature is used as the first input of the risk assessment model; Perform correlation analysis on each control feature; use the feature whose correlation coefficient with the first input feature is greater than a preset coefficient threshold as the second input of the risk assessment model, and use the fusion feature of the second input and the first input as the alternative third input; Obtaining a score of the fusion feature of the second input and the first input according to a ratio of a preset difference of the first input to a difference threshold or a ratio of a preset similarity to the first input similarity, and a correlation coefficient between the first input and the second input; ranking the candidate third inputs according to the scores of the fused features; According to the sorting results, the third candidate input is gradually added to train the risk assessment model; The model performance was evaluated through cross-validation to obtain the final first risk assessment model.

2. The construction safety hazard management method according to claim 1 is characterized in that: The S2 includes: Performing a first processing on the monitoring data to obtain first processed data; the first processing includes data cleaning and outlier detection; Performing standardization on the first processed data of different sensors; obtaining the first data after the standardization; Features are extracted from the standardized first data and time alignment and / or spatial alignment is performed.

3. The construction safety hazard management method according to claim 1, characterized in that: The S4 includes: According to the real-time monitoring data and the risk assessment model, a first risk probability and a first risk level of each second sub-area under each first category are obtained; Obtaining a comprehensive risk index of the second sub-region according to the multiple first risk probabilities and the first risk levels of the second sub-region; Among them, the comprehensive risk index is: Among them, PZ j is the comprehensive risk index of the jth second sub-region; P ij is the first risk probability of the i-th first category in the j-th second sub-region; w ij is the weight of the first risk probability of the i-th first category in the j-th second sub-region; Di is the first risk level score of the i-th first category; Fi is the first risk occurrence frequency score of the i-th first category; α and β are coefficients, 0<α<1; 0<β<1; α+β=1; n is the number of first risk probabilities obtained in the j-th second sub-region; Conduct time series analysis on the comprehensive risk index to identify trends and cyclical characteristics of risk changes.

4. The construction safety hazard management method according to claim 1, characterized in that: The S5 includes: If any first risk probability is greater than the preset probability threshold or the comprehensive risk index is greater than or equal to the preset index threshold, and the corresponding risk level is less than or equal to the preset level, a first warning is issued; If any of the first risk probabilities is greater than the preset probability threshold or the comprehensive risk index is greater than or equal to the preset index threshold, and the corresponding risk level is greater than or equal to the preset level, a second warning is issued; If any two sets of data among the multiple first risk probabilities and comprehensive risk indexes are greater than or equal to the corresponding preset thresholds, a third warning is issued; If any first risk probability is greater than the preset probability threshold or the comprehensive risk index is less than the preset index threshold, but when the first risk probability or comprehensive risk index predicted for multiple consecutive times shows an upward trend; or, any first risk probability or comprehensive risk index conforms to the periodic law and enters the warning cycle within the preset time, a fourth warning will be issued.

5. The construction safety hazard management method according to claim 1, characterized in that: The S6 includes: Automatically learn and optimize threshold settings based on historical warning results and actual events; Risk assessment models are regularly retrained with new data.

6. A construction safety hazard management system, characterized in that: The system comprises: A data acquisition module is used to divide the construction area into multiple sub-areas; the sub-areas include a first sub-area and a second sub-area; and to collect monitoring data of the multiple sub-areas through multiple sensors; the monitoring data includes environmental parameters, equipment status, and personnel behavior; The sub-areas are divided as follows: According to the construction type, the construction area is initially divided into different first sub-areas; Divide the first sub-area into multiple second sub-areas according to the complexity of the operation and the density of personnel; adjust the boundary between the first sub-area and the second sub-area according to the construction progress; A data preprocessing module, used for preprocessing the monitoring data of different sensors, wherein the preprocessing includes feature extraction and alignment of data features of different sensors at the same time and / or space; The model building module is used to classify known safety hazard events to obtain a first classification; by comparing and analyzing the baseline features and the control features, the corresponding data features and fusion features are selected to respectively establish a risk assessment model for the first classification; wherein the baseline features are the data features of each sensor in a safe state; and the control features are the data features of each sensor of the safety hazard event under the same first classification; A monitoring module, used to obtain the first risk probability and comprehensive risk index of each sub-area according to the real-time monitoring data and the risk assessment model; An early warning module, used for issuing an early warning to a sub-region according to the first risk probability, the comprehensive risk index and / or the risk change; An optimization module is used to collect early warning processing results and on-site feedback; adjust the early warning threshold and / or optimize the risk assessment model based on the early warning processing results and on-site feedback; The first category includes personnel safety hazards, equipment safety hazards and environmental safety hazards; Classify the events in each first category into risk levels according to their severity; Through the data features and fusion features of each sensor, a risk assessment model for the first category is established respectively; The risk assessment model of the first category is established respectively through the data features and fusion features of each sensor; including: Obtaining data features of each sensor in a safe state as reference features; Obtaining data features of each sensor of a potential safety hazard event under the same first category as a control feature; Compare each reference feature with the control feature; obtain the difference or similarity between each feature; If the difference is greater than the preset difference threshold or the similarity is less than the preset similarity threshold; the corresponding feature is used as the first input of the risk assessment model; Perform correlation analysis on each control feature; use the feature whose correlation coefficient with the first input feature is greater than a preset coefficient threshold as the second input of the risk assessment model, and use the fusion feature of the second input and the first input as the alternative third input; Obtaining a score of the fusion feature of the second input and the first input according to a ratio of a preset difference of the first input to a difference threshold or a ratio of a preset similarity to the first input similarity, and a correlation coefficient between the first input and the second input; ranking the candidate third inputs according to the scores of the fused features; According to the sorting results, the third candidate input is gradually added to train the risk assessment model; The model performance was evaluated through cross-validation to obtain the final first risk assessment model.

Images