Real-time safety early warning method for high support frame engineering

Abstract

The invention provides a real-time safety early warning method for a high support frame project, and relates to the field of project safety management, and the method comprises the steps: interacting a high support frame project BIM model, extracting high support frame basic information including a structure material label, a structure size label and a support frame structure label, activating a load prediction channel for processing, and generating a prediction load threshold value, the load threshold value represents the deformation external force strength of any position of the high support frame; real-time image information of the high supporting frame is obtained through bottom layer sensing equipment; carrying out load carrier identification according to the real-time image information of the high support frame, and generating a real-time load monitoring value; and when the real-time load monitoring value does not meet the predicted load threshold value, generating real-time early warning information, and sending the real-time early warning information to a safety management terminal. The technical problems that in the prior art, when a safety accident occurs, alarm hysteresis exists, and the safety state of a specific position cannot be reminded in real time are solved.

Description

Technical Field

[0001] The invention relates to the technical field of engineering safety management, and in particular to a real-time safety early warning method for high-support frame engineering. Background Art

[0002] High support frame is an important component used for supporting in the construction process. The safety management of the high support frame construction process is related to the personal safety of construction workers. Therefore, timely monitoring and alarming of safety behaviors in the high support frame construction process are important construction management contents.

[0003] Currently, the safety management of high support frames usually pre-sets the load-bearing strength of specific locations based on experience, and then sets warning signs to prevent safety accidents.

[0004] The existing management solutions have the problem of delayed warning when a safety accident occurs and the inability to provide real-time reminders of the safety status of a specific location. Summary of the invention

[0005] The present application provides a real-time safety warning method for high-support frame projects, which is used to solve the technical problem that when a safety accident occurs, the existing technology has an alarm lag and cannot provide real-time reminders of the safety status of a specific location.

[0006] In view of the above problems, the present application provides a real-time safety early warning method for high support frame projects.

[0007] The first aspect of the present application provides a real-time safety warning method for high-support frame projects, including: interacting with the BIM model of the high-support frame project, extracting basic information of the high-support frame, wherein the basic information of the high-support frame includes structural material labels, structural dimension labels and support frame structure labels; activating the load prediction channel, processing the structural material labels, the structural dimension labels and the support frame structure labels, and generating a predicted load threshold, wherein the load threshold represents the strength of the deformation external force at any position of the high-support frame; obtaining real-time image information of the high-support frame through the underlying sensing device; identifying the load carrier according to the real-time image information of the high-support frame, and generating a real-time load monitoring value; when the real-time load monitoring value does not meet the predicted load threshold, generating real-time warning information and sending it to the safety management terminal.

[0008] The second aspect of the present application provides a real-time safety warning system for high-support frame projects, including: a first interaction unit, used to interact with the BIM model of the high-support frame project and extract basic information of the high-support frame, wherein the basic information of the high-support frame includes structural material labels, structural dimension labels and support frame structure labels; a first prediction unit, used to activate the load prediction channel, process the structural material labels, the structural dimension labels and the support frame structure labels, and generate a predicted load threshold, wherein the load threshold represents the strength of the deformation external force at any position of the high-support frame; a first perception unit, used to obtain real-time image information of the high-support frame through an underlying perception device; a first identification unit, used to identify the load carrier according to the real-time image information of the high-support frame and generate a real-time load monitoring value; a real-time warning unit, used to generate real-time warning information and send it to a safety management terminal when the real-time load monitoring value does not meet the predicted load threshold.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] This application uses the BIM model of the high-support frame project to collect the basic information of the high-support frame, including the structural material label, structural dimension label and support frame structure label; uses the load prediction channel to process the structural material label, structural dimension label and support frame structure label to obtain the deformation external force strength of any position of the high-support frame, and stores it as the predicted load threshold; further, through the underlying sensing device, obtain the real-time image information of the high-support frame to identify and determine the real-time load monitoring value; when the real-time load monitoring value does not meet the predicted load threshold, generate real-time warning information and send it to the security management terminal. Using the intelligent model, configure the load intensity threshold of each position, and then use image recognition to determine the real-time load intensity of each position; warn the position where the real-time load intensity does not meet the load intensity threshold, so as to achieve the technical effect of improving the timeliness of the high-support frame construction warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 A schematic flow chart of a real-time safety warning method for a high-support frame project provided in this application;

[0012] Figure 2 A schematic diagram of a process for executing discrete value deletion in a real-time safety warning method for a high-support frame project provided in this application;

[0013] Figure 3 A structural schematic diagram of a real-time safety warning system for high-support frame engineering provided in this application.

[0014] Explanation of the accompanying drawings: first interaction unit 100, first prediction unit 200, first perception unit 300, first recognition unit 400, real-time warning unit 500. DETAILED DESCRIPTION

[0015] This application provides a real-time safety warning method for high-support frame projects, which is used to solve the technical problem that the existing technology has a lag in warning when a safety accident occurs, and cannot remind the safety status of a specific location in real time. Using an intelligent model, configure the load intensity threshold of each location, and then use image recognition to determine the real-time load intensity of each location; warn the location where the real-time load intensity does not meet the load intensity threshold, and achieve the technical effect of improving the timeliness of the high-support frame construction warning.

[0016] Embodiment 1

[0017] like Figure 1 As shown, the present application provides a real-time safety early warning method for high support frame engineering, comprising:

[0018] Interact with the high support frame engineering BIM model to extract the high support frame basic information, wherein the high support frame basic information includes structural material labels, structural dimension labels and support frame structure labels;

[0019] In a preferred embodiment, the high support frame engineering BIM model is an engineering model constructed based on BIM simulation technology. It stores at least the structural data, material data and dimension data of each position of the construction building; the support performance of the high support frame, that is, the load strength threshold of each position, is mainly related to the structural data, material data and dimension data, so the structural material label, structural dimension label and support frame structure label of the high support frame are extracted from the high support frame engineering BIM model as the basic data for the subsequent evaluation of the load strength threshold.

[0020] activating a load prediction channel, processing the structural material label, the structural dimension label and the support frame structure label, and generating a predicted load threshold, wherein the load threshold represents the strength of the deformation external force at any position of the high support frame;

[0021] In a preferred embodiment, the load prediction channel is used to configure the load strength threshold of each position of the high support frame. The load threshold refers to the deformation external force strength of any position of the high support frame. The structural material label, structural dimension label and support frame structure label are input into the load prediction channel to calibrate the load threshold of each position of the high support frame to generate a predicted load threshold. It can be used as a reference load for the subsequent real-time warning, that is, if the real-time load of the corresponding position exceeds the predicted load threshold, a warning is issued.

[0022] Furthermore, the preferred construction method and working principle of the load forecasting channel are as follows:

[0023] Further, activating the load prediction channel, processing the structural material label, the structural dimension label and the support frame structure label to generate a predicted load threshold, including:

[0024] Collect structural material record data, structural dimension record data and support frame structure record data, and identify load characteristic thresholds;

[0025] According to the structural material record data, the structural dimension record data, the support frame structure record data and the load characteristic threshold, based on the regression decision forest, the load prediction channel is trained;

[0026] The load prediction channel is activated, and the structural material label, the structural dimension label and the support frame structure label are processed to generate the predicted load threshold.

[0027] In a preferred embodiment, the load forecasting channel is a machine learning model, such as, for example, a neural network, a support vector machine, a decision tree, etc. Since the decision tree has a fast convergence speed and the load forecasting of the present application is essentially a classification problem, not a linear problem, the embodiment of the present application preferably constructs a load forecasting channel based on a decision tree.

[0028] Among them, the structural material record data characterizes the type of constituent materials at each position in the structural topology, such as, for example, data such as alloy model and steel pipe model; the structural dimension record data refers to the dimension information of the constituent materials at each position in the structural topology, such as, for example, the length, thickness, height, etc. of the alloy. The support frame structure record data characterizes the structural characteristics of the structural topology, including the distribution position of each size of each material, shape data, etc. The load characteristic threshold refers to the maximum value of the external force that can be calibrated to cause deformation at each position of the structural topology, constrained by the structural material record data, structural dimension record data, and support frame structure record data.

[0029] The load characteristic threshold is used as the output supervision data, the structural material record data, the structural dimension record data, and the support frame structure record data are used as the output data, and the load prediction channel is trained based on the regression decision forest. Preferably, a training loss function is constructed:

[0030]

[0031] Among them, LOSS represents the loss value of each M training, F ji Characterizes the predicted load threshold of the i-th position during the j-th training, F ji0 represents the load characteristic threshold of the i-th position during the j-th training, N represents the total number of positions marked by the load characteristic threshold, and M represents the number of training times for setting the statistical loss value.

[0032] Configure the loss value threshold. When LOSS is less than the loss value threshold, retrieve several groups of data for verification. If the ratio of the number of groups whose LOSS is less than the loss value threshold is greater than the preset ratio of the number of groups, preferably 0.85, the model is considered to have converged, and the load prediction channel is obtained. Regression decision forest refers to multiple decision trees trained in the same way, and then taking their mean as the actual output each time, and improving the output accuracy of the load prediction channel through the cluster model. Supervised training is a relatively mature training method, which is not the protection content of the embodiments of the present application, and will not affect the implementation of those skilled in the art, so it will not be elaborated on. Activate the load prediction channel, process the structural material label, the structural dimension label and the support frame structure label, and generate the predicted load threshold.

[0033] Furthermore, the training process of regression decision forest is as follows:

[0034] Further, according to the structural material record data, the structural dimension record data, the support frame structure record data and the load characteristic threshold, the load prediction channel is trained based on the regression decision forest, including:

[0035] Divide the structural material record data, the structural dimension record data, the support frame structure record data and the load characteristic threshold into a training data set and a verification data set according to a ratio of 8:2;

[0036] Perform k extractions with replacement on the training data set, and repeat M times to generate M sets of training data;

[0037] Retrieving the M groups of training data, combining them with the verification data set, and training M load prediction base channels based on a regression decision tree;

[0038] The output mean of the M load prediction base channels is used as the output of the load prediction channel, and the M load prediction base channels are fused to generate the load prediction channel.

[0039] In a preferred embodiment, the structural material record data, the structural dimension record data, the support frame structure record data and the load characteristic threshold are first divided into a training data set and a verification data set according to a ratio of 8:2. The training data set is extracted k times with replacement and repeated M times to generate M groups of training data, that is, the training data set is divided into k parts, k is an integer, and the total number of training data sets is ≥ k ≥ 0.2*the total number of training data sets. Then the k parts of training data are extracted and combined with replacement to obtain the first group of training data. Repeat M times to obtain M groups of training data. M is an integer. The more M there are, the more accurate the output result is, but it should be considered that the more M there are, the higher the training cost is, so M is determined based on the training cost that the user can afford, but M is at least equal to 5.

[0040] Furthermore, the load characteristic threshold of the first set of training data is used as the supervision data, the structural material record data, the structural dimension record data, and the support frame structure record data of the first set of training data are used as the input data, and the loss function and the validation data set are combined to perform supervised training, configure the decision tree, and obtain the first load prediction base channel; the load characteristic threshold of the second set of training data is used as the supervision data, the structural material record data, the structural dimension record data, and the support frame structure record data of the second set of training data are used as the input data, and the loss function and the validation data set are combined to perform supervised training, configure the decision tree, and obtain the second load prediction base channel; until the load characteristic threshold of the Mth set of training data is used as the supervision data, the structural material record data, the structural dimension record data, and the support frame structure record data of the Mth set of training data are used as the input data, and the loss function and the validation data set are combined to perform supervised training, configure the decision tree, and obtain the Mth load prediction base channel. Integrate into M load prediction base channels. Take the output mean of the M load prediction base channels as the output of the load prediction channel, and fuse the output layers of the M load prediction base channels to generate the load prediction channel.

[0041] Furthermore, in order to ensure the accuracy of the training data, the present embodiment proposes an algorithm flow for calibrating the load characteristic threshold based on different materials, structures and sizes, as follows:

[0042] Furthermore, the structural material record data, the structural dimension record data and the support frame structure record data are collected to identify the load characteristic threshold, including:

[0043] Taking the structural material record data, the structural dimension record data and the support frame structure record data as constraints, based on the Internet of Things, deformation load detection data is collected, wherein the deformation load detection data includes a load threshold record value set;

[0044] Deleting discrete values ​​of the load threshold record value set to generate a load threshold concentrated value;

[0045] The load characteristic threshold is generated by performing mean statistics on the concentrated values ​​of the load threshold.

[0046] Further, such as Figure 2 As shown, the discrete value of the load threshold record value set is deleted to generate a load threshold concentrated value, including:

[0047] According to the preset load deviation, cluster analysis is performed on the load threshold record value set to generate a plurality of cluster load threshold record values;

[0048] Counting the total number of recorded values ​​of the plurality of cluster load threshold recorded values;

[0049] Counting the number of record values ​​of the plurality of cluster load threshold record values;

[0050] Calculate the proportion of the number of the plurality of record values ​​in the total number of the record values ​​to generate a plurality of concentrated support degrees;

[0051] The clusters whose concentrated supports are less than or equal to the concentrated support threshold are deleted to generate the load threshold concentrated value.

[0052] In a preferred embodiment, the deformation load detection data is collected based on the Internet of Things, taking the structural material record data, the structural dimension record data and the support frame structure record data as constraints, and the test or experimental data of each company for the high support frame is collected to determine the deformation load detection data of the maximum external force that reaches the deformation under the constraints of different structural material record data, structural dimension record data and support frame structure record data. Since a set of deformation load detection data under the constraints of structural material record data, structural dimension record data and support frame structure record data has been tested many times and has multiple load values, in order to obtain a more accurate load characteristic threshold, it is necessary to perform a centralized trend analysis on the multiple load values. The details are as follows:

[0053] The discrete value of the load threshold record value set is deleted, and the remaining load threshold record values ​​are set as the load threshold concentration value, wherein the detailed process of the discrete value deletion is as follows:

[0054] The user configures a preset load deviation to characterize the maximum deviation of whether the load can be considered consistent. When the two load deviations are greater than the preset load deviation, they are considered different loads, otherwise they are considered the same load. The set of load threshold record values ​​is clustered and analyzed to generate several clusters of load threshold record values, and the deviations between the load threshold record values ​​in each cluster are less than or equal to the preset load deviation. The deviations between any two load threshold record values ​​between each cluster are greater than the preset load deviation. The total number of record values ​​of the several clusters of load threshold record values ​​is counted, that is, the total number of record values ​​of all clusters; the number of record values ​​of the several clusters of load threshold record values ​​is counted, that is, the number of record values ​​of each cluster; the proportion of the several record values ​​in the total number of record values ​​is calculated to generate several concentrated support degrees; the clusters whose concentrated support degrees are less than or equal to the concentrated support threshold are deleted, and the remaining record values ​​are set as the load threshold concentrated value. The load threshold concentrated value is averaged and the load characteristic threshold is generated. A central trend analysis algorithm is constructed based on the principle of data statistical analysis to obtain a more accurate load characteristic threshold to ensure the accuracy of the training data. Further, the stability of real-time warning is guaranteed to achieve the technical effect of reducing the false alarm rate.

[0055] Obtain real-time image information of high support frames through underlying sensing equipment;

[0056] Perform load carrier identification based on the real-time image information of the high support frame to generate a real-time load monitoring value;

[0057] When the real-time load monitoring value does not meet the predicted load threshold, real-time warning information is generated and sent to the security management terminal.

[0058] In a preferred embodiment, the underlying sensing device includes an image sensor, such as a camera, a millimeter-wave radar and other devices, to achieve real-time monitoring of the construction site. The real-time image information of the high support frame is uploaded in real time by the underlying sensing device, and the target type and quantity of the applied load in the real-time image information of the high support frame are counted by image recognition. Preferably, a data table is pre-configured, and the data table has a one-to-one correspondence between the target type and the target load value. The data table is customized by the user, and the load value is calculated based on the target type and quantity, and is set as the real-time load monitoring value. The target to which the load is applied is the load carrier, such as construction personnel, objects, building materials, etc. can all be regarded as load carriers. When the real-time load monitoring value does not meet the predicted load threshold, that is, the real-time load monitoring value of a certain position is greater than or equal to the predicted load threshold, the corresponding position is marked with real-time warning information, which is sent to the security management terminal where the management personnel is located to remind the management and control of the targets at the relevant positions.

[0059] Further, load carrier identification is performed according to the real-time image information of the high support frame to generate a real-time load monitoring value, including:

[0060] Collect high-support frame operation image dataset;

[0061] Performing load carrier type identification on the high support frame operation image data set to generate a load carrier image identification data set, wherein the load carrier includes a human body or an object;

[0062] Using the load carrier image identification dataset as supervision data and the high support frame operation image dataset as input data, training a convolution classification model to generate a load carrier identification channel;

[0063] According to the real-time image information of the high support frame, the load carrier identification channel is activated to perform load carrier identification, and the load carrier type and load carrier quantity are generated;

[0064] The real-time load monitoring value is calibrated according to the load carrier type and the load carrier quantity.

[0065] In a preferred embodiment, the load carrier identification can be implemented by a mature image recognition algorithm or by machine learning. The embodiment of the present application is implemented by a convolutional neural network of machine learning. For example, the recorded image data of the past work of the high support frame with the same structure, material and size constraints are collected and set as the high support frame operation image data set. The staff identifies the load carrier type of the high support frame operation image data set to generate a load carrier image identification data set, wherein the load carrier includes a human body or an object. That is, the types of different load carriers distributed in the operation image can be determined. The load carrier image identification data set is used as the supervision data, and the high support frame operation image data set is used as the input data to train the convolution classification model, that is, to train the convolutional neural network, and generate a load carrier identification channel, so that the load carrier in the real-time image information of the high support frame can be identified, and the load carrier type and the number of load carriers can be obtained. Then, based on the data table, the load value of each load carrier type at a position is calibrated, and then multiplied by the number of load carriers, that is, the real-time load monitoring value of a certain position = the load value of the first load carrier type * the number of first load carriers + ... + the load value of the Qth load carrier type * the number of Qth load carriers. It is convenient to carry out real-time safety monitoring of high support frames.

[0066] In summary, the embodiments of the present application have at least the following technical effects:

[0067] The embodiment of the present application provides a real-time safety warning method for high-support frame projects, which is used to solve the technical problem that the existing technology has a lag in warning when a safety accident occurs, and cannot remind the safety status of a specific location in real time. Using an intelligent model, the load intensity threshold of each location is configured, and then image recognition is used to determine the real-time load intensity of each location; warnings are issued for locations where the real-time load intensity does not meet the load intensity threshold, achieving the technical effect of improving the timeliness of warnings for high-support frame construction.

[0068] Embodiment 2

[0069] Based on the same inventive concept as the real-time safety warning method for high support frame engineering in the aforementioned embodiment, Figure 3 As shown, the embodiment of the present application provides a real-time safety warning system for high-support frame engineering, including:

[0070] The first interaction unit 100 is used to interact with the BIM model of the high support frame project and extract the basic information of the high support frame, wherein the basic information of the high support frame includes a structural material label, a structural dimension label and a support frame structure label;

[0071] The first prediction unit 200 is used to activate the load prediction channel, process the structural material label, the structural size label and the support frame structure label, and generate a predicted load threshold, wherein the load threshold represents the deformation external force strength at any position of the high support frame;

[0072] The first sensing unit 300 is used to obtain real-time image information of the high support frame through the underlying sensing device;

[0073] The first identification unit 400 is used to identify the load carrier according to the real-time image information of the high support frame and generate a real-time load monitoring value;

[0074] The real-time warning unit 500 is used to generate real-time warning information and send it to the security management terminal when the real-time load monitoring value does not meet the predicted load threshold.

[0075] Furthermore, the first prediction unit 200 executes the steps including:

[0076] Collect structural material record data, structural dimension record data and support frame structure record data, and identify load characteristic thresholds;

[0077] According to the structural material record data, the structural dimension record data, the support frame structure record data and the load characteristic threshold, based on the regression decision forest, the load prediction channel is trained;

[0078] The load prediction channel is activated, and the structural material label, the structural dimension label and the support frame structure label are processed to generate the predicted load threshold.

[0079] Furthermore, the first prediction unit 200 executes the steps including:

[0080] Taking the structural material record data, the structural dimension record data and the support frame structure record data as constraints, based on the Internet of Things, deformation load detection data is collected, wherein the deformation load detection data includes a load threshold record value set;

[0081] Deleting discrete values ​​of the load threshold record value set to generate a load threshold concentrated value;

[0082] The load characteristic threshold is generated by performing mean statistics on the concentrated values ​​of the load threshold.

[0083] Furthermore, the first prediction unit 200 executes the steps including:

[0084] Divide the structural material record data, the structural dimension record data, the support frame structure record data and the load characteristic threshold into a training data set and a verification data set according to a ratio of 8:2;

[0085] Perform k extractions with replacement on the training data set, and repeat M times to generate M sets of training data;

[0086] Retrieving the M groups of training data, combining them with the verification data set, and training M load prediction base channels based on a regression decision tree;

[0087] The output mean of the M load prediction base channels is used as the output of the load prediction channel, and the M load prediction base channels are fused to generate the load prediction channel.

[0088] Furthermore, the first prediction unit 200 executes the steps including:

[0089] According to the preset load deviation, cluster analysis is performed on the load threshold record value set to generate a plurality of cluster load threshold record values;

[0090] Counting the total number of recorded values ​​of the plurality of cluster load threshold recorded values;

[0091] Counting the number of record values ​​of the plurality of cluster load threshold record values;

[0092] Calculate the proportion of the number of the plurality of record values ​​in the total number of the record values ​​to generate a plurality of concentrated support degrees;

[0093] The clusters whose concentrated supports are less than or equal to the concentrated support threshold are deleted to generate the load threshold concentrated value.

[0094] Further, the first identification unit 400 performs the following steps:

[0095] Collect high-support frame operation image dataset;

[0096] Performing load carrier type identification on the high support frame operation image data set to generate a load carrier image identification data set, wherein the load carrier includes a human body or an object;

[0097] Using the load carrier image identification dataset as supervision data and the high support frame operation image dataset as input data, training a convolution classification model to generate a load carrier identification channel;

[0098] According to the real-time image information of the high support frame, the load carrier identification channel is activated to perform load carrier identification, and the load carrier type and load carrier quantity are generated;

[0099] The real-time load monitoring value is calibrated according to the load carrier type and the load carrier quantity.

[0100] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A real-time safety warning method for high-support frame engineering, characterized in that: include: Interact with the high support frame engineering BIM model to extract the high support frame basic information, wherein the high support frame basic information includes structural material labels, structural dimension labels and support frame structure labels; activating a load prediction channel, processing the structural material label, the structural dimension label and the support frame structure label, and generating a predicted load threshold, wherein the load threshold represents the strength of the deformation external force at any position of the high support frame; Obtain real-time image information of high support frames through underlying sensing equipment; Perform load carrier identification based on the real-time image information of the high support frame to generate a real-time load monitoring value; When the real-time load monitoring value does not meet the predicted load threshold, real-time warning information is generated and sent to the security management terminal.

2. The method according to claim 1, characterized in that Activating the load prediction channel, processing the structural material label, the structural dimension label and the support frame structure label, and generating a predicted load threshold, including: Collect structural material record data, structural dimension record data and support frame structure record data, and identify load characteristic thresholds; According to the structural material record data, the structural dimension record data, the support frame structure record data and the load characteristic threshold, based on the regression decision forest, the load prediction channel is trained; The load prediction channel is activated, and the structural material label, the structural dimension label and the support frame structure label are processed to generate the predicted load threshold.

3. The method according to claim 2, characterized in that Collect structural material record data, structural dimension record data and support frame structure record data, and identify load characteristic thresholds, including: Taking the structural material record data, the structural dimension record data and the support frame structure record data as constraints, based on the Internet of Things, deformation load detection data is collected, wherein the deformation load detection data includes a load threshold record value set; Deleting discrete values ​​of the load threshold record value set to generate a load threshold concentrated value; The load characteristic threshold is generated by performing mean statistics on the concentrated values ​​of the load threshold.

4. The method according to claim 2, characterized in that According to the structural material record data, the structural dimension record data, the support frame structure record data and the load characteristic threshold, the load prediction channel is trained based on the regression decision forest, including: Divide the structural material record data, the structural dimension record data, the support frame structure record data and the load characteristic threshold into a training data set and a verification data set according to a ratio of 8:2; Perform k extractions with replacement on the training data set, and repeat M times to generate M sets of training data; Retrieving the M groups of training data, combining with the verification data set, and training M load prediction base channels based on a regression decision tree; The output mean of the M load prediction base channels is used as the output of the load prediction channel, and the M load prediction base channels are fused to generate the load prediction channel.

5. The method according to claim 3, characterized in that Deleting discrete values ​​of the load threshold record value set to generate a load threshold concentrated value includes: According to a preset load deviation, cluster analysis is performed on the load threshold record value set to generate a plurality of cluster load threshold record values; Counting the total number of recorded values ​​of the plurality of cluster load threshold recorded values; Counting the number of record values ​​of the plurality of cluster load threshold record values; Calculate the proportion of the number of the plurality of record values ​​in the total number of the record values ​​to generate a plurality of concentrated support degrees; The clusters whose concentrated supports are less than or equal to the concentrated support threshold are deleted to generate the load threshold concentrated value.

6. The method according to claim 1, characterized in that The load carrier is identified according to the real-time image information of the high support frame to generate a real-time load monitoring value, including: Collect high-support frame operation image dataset; Performing load carrier type identification on the high support frame operation image data set to generate a load carrier image identification data set, wherein the load carrier includes a human body or an object; Using the load carrier image identification dataset as supervision data and the high support frame operation image dataset as input data, training a convolution classification model to generate a load carrier identification channel; According to the real-time image information of the high support frame, the load carrier identification channel is activated to perform load carrier identification, and the load carrier type and load carrier quantity are generated; The real-time load monitoring value is calibrated according to the load carrier type and the load carrier quantity.

7. A real-time safety warning system for high-support frame engineering, characterized in that: include: The first interaction unit is used to interact with the BIM model of the high support frame project and extract the basic information of the high support frame, wherein the basic information of the high support frame includes a structural material label, a structural dimension label and a support frame structure label; A first prediction unit is used to activate a load prediction channel, process the structural material label, the structural dimension label and the support frame structure label, and generate a predicted load threshold, wherein the load threshold represents the strength of the deformation external force at any position of the high support frame; The first sensing unit is used to obtain real-time image information of the high support frame through the underlying sensing device; A first identification unit is used to identify the load carrier according to the real-time image information of the high support frame and generate a real-time load monitoring value; The real-time warning unit is used to generate real-time warning information and send it to the security management terminal when the real-time load monitoring value does not meet the predicted load threshold.

Images