Intelligent construction site cloud platform based on Internet of Things

Abstract

The invention discloses an intelligent construction site cloud platform based on the Internet of Things, and relates to the technical field of construction site management, and the platform employs a multi-mode data fusion technology in combination with RFID, GPS, video monitoring and wearable equipment, monitors the position, behavior and physiological state of a worker, and automatically recognizes the violation behavior of the worker in combination with a deep learning model. Meanwhile, the regional risk coefficient is adopted to dynamically adjust the safety threshold values of the environment and the behavior, early warning adjustment is carried out in different regions according to the risk level, false alarm and missing alarm are greatly reduced, meanwhile, the working environment is monitored in real time through the temperature, humidity and dust environment sensors, and the working efficiency is improved. Early warning is triggered once the environmental condition exceeds the safety range, whether work meets the safety standard or not is judged in combination with environmental factors and worker behaviors, the weight of recent high-risk behaviors is increased through time weighting based on past behaviors of workers, and the current safety state of the workers is preferentially concerned.

Description

Technical Field

[0001] The present invention relates to the technical field of construction site management, and in particular to an intelligent construction site cloud platform based on the Internet of Things. Background Art

[0002] One of the core objectives of construction site management is to ensure the safety of workers and improve construction efficiency. In traditional construction site management, the behavior management and safety supervision of workers mainly rely on manual inspections and regular trainings, which are inefficient and prone to negligence, and it is difficult to achieve real-time monitoring of large-scale construction sites.

[0003] Managers check whether workers wear protective equipment or violate regulations through regular inspections. However, this method often fails to detect potential safety hazards in a timely manner. Secondly, most of the existing technologies use means such as RFID and GPS for location tracking. Although it can ensure that workers operate within the designated area, it is difficult to accurately identify their behaviors, such as whether they correctly wear safety equipment or operate in high-risk areas. In addition, some solutions use video surveillance. Although video surveillance can record events, it is mostly used for retrospective review after the event, and it is difficult to achieve timely intervention and prevention, increasing the risk of accidents. To address such problems, some construction site management solutions adopt location tracking and real-name management, tracking the locations of workers through positioning technologies such as RFID and GPS, preventing workers from entering high-risk areas, and recording identity information and attendance data, thereby strengthening personnel management to a certain extent. In addition, video surveillance is generally used in construction sites to cover the main areas, and workers' behaviors or abnormal situations are viewed remotely by humans. However, these means mostly rely on the workers' self-awareness to execute, and the effects are difficult to continuously monitor and difficult to cope with the rapidly changing construction environment.

[0004] It can be seen that although the above solutions have improved the construction site management efficiency to a certain extent, they lack real-time monitoring of workers' behaviors. The positioning technology can only monitor the locations of workers and cannot detect specific operation behaviors. Video surveillance also mainly relies on manual viewing of videos after the event and cannot automatically identify violations. In addition, there is a lack of integration among various data sources, such as video surveillance, RFID, and environmental sensors, making it difficult to conduct comprehensive analysis. Therefore, there is an urgent need for an intelligent construction site cloud platform based on the Internet of Things to solve such problems. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides an intelligent construction site cloud platform based on the Internet of Things to solve the problems of lack of real-time monitoring of workers' behaviors, the positioning technology can only monitor the locations of workers and cannot detect specific operation behaviors, and video surveillance also mainly relies on manual viewing of videos after the event and cannot automatically identify violations.

[0007] To solve the above technical problems, the present invention provides the following technical solutions:

[0008] The present invention provides an Internet of Things-based intelligent construction site cloud platform, which includes,

[0009] A safety monitoring module for real-time location tracking of workers, physiological state monitoring, wearing state monitoring, behavior recognition and safety warning,

[0010] An environmental monitoring module for monitoring construction site environmental parameters, including temperature, humidity, and dust concentration, and issuing a warning when the detected environmental conditions exceed the safety threshold,

[0011] An intelligent analysis module that integrates the data of the safety monitoring module and the environmental monitoring module, including RFID, GPS positioning, video monitoring, environmental sensor, and worker wearable device data, and based on the fused data, uses a deep learning model to identify worker behaviors and automatically detect whether workers have potential safety hazard behaviors,

[0012] A personalized safety management module that generates a safety score based on the recognition results of the intelligent analysis module, as well as the analysis of workers' daily operation habits and high-risk behaviors,

[0013] In the personalized safety management module, scoring is performed based on the frequency of workers' violation behaviors and operation patterns, and workers with low scores are regarded as training targets,

[0014] And a personalized safety training plan is generated according to workers' operation habits and risk behaviors.

[0015] Furthermore, the management steps of the Internet of Things-based intelligent construction site cloud platform include:

[0016] Step S1, real-time monitoring of multi-modal data,

[0017] Adopt multi-modal sensor fusion technology to collect and integrate construction site data to generate fused data;

[0018] Step S2, intelligent behavior recognition and warning,

[0019] Based on the fused data in Step S1, construct a deep learning model, combine the convolutional neural network CNN and the long short-term memory network LSTM to identify worker behaviors and perform risk detection: this model analyzes the real-time video monitoring of the construction site and the data of wearable devices to identify potential safety hazard behaviors of workers,

[0020] Step S3, personalized behavior analysis,

[0021] Based on the worker behavior data identified in Step S2 and the risk detection results, that is, the identified potential safety hazard behaviors, analyze the operation habits of workers and generate personalized safety training suggestions.

[0022] Further, in step S1, the construction site data includes RFID, GPS positioning, video surveillance, environmental sensors, and smart devices worn by workers (such as smart bracelets and smart safety helmets).

[0023] In step S1, a warning is issued when the detected environmental conditions exceed the safety threshold.

[0024] Among the construction site data, RFID and GPS data are used to track the positions of workers, environmental sensor data is used to monitor temperature, humidity, and dust environmental data, wearable device data is used to monitor the physiological status of workers, and smart safety helmets can detect whether they are worn correctly.

[0025] Further, in step S1, multi-modal data fusion is performed. Specifically:

[0026] Define different types of data sources: p RFID represents the worker position information provided by the RFID sensor, p GPS represents the worker position information provided by GPS, e temp , e humid , e dust are the environmental sensor data of temperature, humidity, and dust respectively, h heart represents the worker heart rate information detected by the smart bracelet, s helmet represents the data of the smart safety helmet, detecting the wearing situation of the safety helmet, 0 means not worn, 1 means correctly worn. The weighted linear fusion method is used to fuse different data:

[0027] F fusion = w1·p RFID + w2·p GPS + w3·e temp + w4·e humid + w5·e dust + w6·h heart + w7·s helmet , where F fusion represents the fused data, w1, w2,..., w7 represent the weight coefficients of different data sources, and the weights are determined based on the confidence of sensor data and the influence of historical data:

[0028] Among them, α i represents the confidence of the i-th type of sensor data,

[0029] Based on the fused data, real-time monitoring is carried out to judge whether the environmental conditions exceed the safety threshold, set the safety interval of the environmental variable, and compare the real-time monitoring value with this interval to trigger a warning. Set the safety threshold of the environmental variable: T safe The safety interval of temperature is [T min , Tmax , H safe The safe range of humidity is [H min , H max , D safe The safe range of dust concentration is [D min , D max ,

[0030] For each moment t, if or or then P alert (t) = 1, otherwise, P alert (t) = 0, where P alert (t)) represents the environmental warning trigger function. If the value is 1, it means that the current environmental conditions have exceeded the safety threshold and a warning is triggered. If the value is 0, it means that the environmental conditions are normal, e temp (t), e humid (t), e dust (t) represent the actual values of temperature, humidity, and dust concentration collected by the environmental sensor at moment t;

[0031] In step S1, perform dynamic adjustment of the fusion data threshold:

[0032] According to the work area or operation risk level, dynamically adjust the warning threshold, and define the regional risk coefficient r zone , dynamically adjust the warning threshold:

[0033] T adjusted = r zone ·T safe , H adjusted = r zone ·H safe , D adjusted = r zone ·D safe , where T adjusted , H adjusted , D adjusted represent the safe thresholds of temperature, humidity, and dust concentration after dynamic adjustment, and T safe , H safe , D safe represent the safe thresholds of temperature, humidity, and dust concentration before adjustment respectively. Adjust according to the risk level of different areas of the construction site. r zone represents the risk coefficient of the current operation area. For areas with higher risks, r zone > 1, and for areas with lower risks, r zone < 1.

[0034] Furthermore, in step S2, the potential safety hazard behaviors identified include:

[0035] Workers do not wear safety helmets or wear them incorrectly.

[0036] Workers enter high-risk areas.

[0037] Workers do not wear safety ropes during high-risk operations.

[0038] In step S2, after identifying potential safety hazards, the warning threshold is dynamically adjusted according to the risk levels of different operation areas, and an alarm is issued.

[0039] Furthermore, in step S2, based on the fusion data of step S1, a model combining a convolutional neural network (CNN) and a long short-term memory network (LSTM) is constructed to identify workers' behaviors and perform potential risk detection. The specific steps include:

[0040] Let the data fusion result X obtained in step S1 be the temporal representation of multi-modal data, that is:

[0041] X = {x1, x2, …, x T}, x t ∈R d , where X is a temporal data sequence containing T time steps, the data dimension of each time step is d, and x t represents the fusion data of the t-th time step, including data from video surveillance, wearable devices, and environmental sensor data sources.

[0042] Extract the spatial features in the video frame. Let the video input be V t , the video frame corresponding to the moment t, and use a 2D convolutional layer to extract spatial features:

[0043] h t = ReLU(W conv *V t + b conv ), where h t is the feature map extracted from the video frame V t at the moment t, W conv is the weight matrix of the convolution kernel, * represents the convolution operation, b conv is the bias term, ReLU is a non-linear activation function, ReLU(x) = max(0, x), and the convolution feature h t is fused with other sensor data as video features to form new temporal data. For each time step t, the data x t from different sensors is concatenated with the video feature h t to obtain the temporal feature: z t = [x t ; h t , where z t is the comprehensive feature at the moment t, including the video feature h tConcatenation with other sensor data x t of [x t ; h t represents the concatenation operation of vectors.

[0044] Furthermore, in step S2, a long short-term memory network LSTM is used for time series modeling:

[0045] Input the concatenated feature sequence {z1, z2,..., z T} into the LSTM network. The LSTM framework is:

[0046] Forget gate, f t = σ(W f z t + U f h t-1 + b f ),

[0047] Input gate, i t = σ(W i z t + U i h t-1 + b i ),

[0048] Candidate memory unit,

[0049] Updated memory unit,

[0050] Output gate, o t = σ(W o z t + U o h t-1 + b o ),

[0051] Hidden state update where f t , i t , o t are the activation values of the forget gate, input gate, and output gate respectively. W f , W i , W c , W o are the weight matrices of different gates, multiplied by the current input z t , U f , U i , U c , U o are the weight matrices of each gate for the hidden state h t-1 at the previous time step, b f , b i , b c , bo is the bias term of each gate, is the candidate memory cell, c t is the updated memory cell, c t-1 is the memory cell of the previous time step, ° represents element-wise multiplication,

[0052] Based on the LSTM network modeling, the hidden state sequence {h1, h2, …, h T} is obtained, which contains the behavior information of the worker during the entire time period. Then, the hidden state is classified through a fully connected layer to identify the potential safety hazard behaviors of the worker. For each time step t, the probability distribution of the behavior classification is output as:

[0053] y t = Softmax(W y h t + b y ), where y t is the behavior prediction result at time step t, indicating whether the worker has potential safety hazard behaviors (such as not wearing a safety helmet, entering a high-risk area, etc.). W y is the weight matrix of the behavior classification, b y is the bias term of the classification layer. The Softmax function converts the output into a probability distribution, which is defined as:

[0054] According to the output y t of the model, a risk detection mechanism for the worker's behavior is defined. When the model predicts a certain high-risk behavior, a warning is triggered. The risk detection is defined as:

[0055] r t = I(y t ∈ HighRiskActions), where r t is the risk detection result at time step t. If the value is 1, it indicates that there is a high-risk behavior currently. I is the indicator function. If y t belongs to the predefined high-risk behaviors, it outputs 1, otherwise it outputs 0. Once a high-risk behavior r t = 1 is detected, a warning signal is immediately sent to remind the management to take measures.

[0056] Furthermore, in step S3, the way to analyze the operation habits of workers and generate personalized safety training suggestions is:

[0057] Generate a safety score for the worker according to the frequency of the worker's operations and the proportion of high-risk behaviors. Workers with frequent high-risk behaviors get a low score.

[0058] Combined with the score data, generate a personalized safety training plan for each worker, and recommend corresponding training content for different violation behaviors.

[0059] For example:

[0060] For workers who frequently do not wear protective equipment, emphasize the correct use of PPE (Personal Protective Equipment).

[0061] For workers who operate equipment improperly during the use of specific equipment, the system will push training on equipment usage specifications.

[0062] Furthermore, in step S3, a safety score is generated based on the operation frequency and the proportion of high-risk behaviors of the worker. Specifically:

[0063] Set the total number of all operations completed by worker i during the monitoring period as N i , classify the different operation behaviors of the worker into m types of behavior categories, and record the number of times n of each behavior i,j , where j = 1, 2,..., m represents different behavior types, and the frequency of the worker's behavior is expressed as: Among them, f i,j represents the frequency of the j-th operation behavior of worker i, reflecting the proportion of this behavior in his total operation behaviors, n i,j represents the number of times of the j-th operation completed by worker i, N i is the total number of all operations of worker i.

[0064] Let the risk level of the j-th operation behavior be r j , and the value of the risk level depends on the danger of this operation. Larger r j values are assigned to higher-risk behaviors. The proportion R of high-risk behaviors of worker i i is expressed as:

[0065] Among them, R i represents the proportion of high-risk behaviors of worker i, reflecting the proportion of high-risk operations in his operation behaviors, r j is the risk level of the j-th operation behavior, n i,j is the number of times of the j-th operation completed by worker i, N i is the total number of operations of worker i.

[0066] Define a time-weighted function w(t) to assign greater weight to recently occurring high-risk behaviors. Let the high-risk behavior of the worker at time t be r j (t), then the time-weighted high-risk behavior within the time period T is expressed as:

[0067] Among them, is the time - weighted high - risk behavior ratio of worker i, w(t) is the time - weighted function, w(t)=exp(-λ(T - t)), where λ is the time - decay coefficient, indicating the degree to which the impact of the behavior decreases over time, r j (t) is the risk level at time t, n i,j (t) is the number of the j - th operation behaviors of worker i at time t, N i (t) is the total number of all operations of the worker at time t,

[0068] Based on the operation frequency f of the worker i,j and the time - weighted high - risk behavior ratio define the safety score S of worker i i :

[0069] where S i is the safety score of worker i, S max is the preset maximum safety score, representing the score when the worker operates completely without risk, is the time - weighted high - risk behavior ratio of the worker, reflecting the proportion of high - risk behaviors in the worker's past operations, f i,j is the operation frequency of the j - th operation behavior of worker i, r j is the risk level of the j - th operation behavior, α and β are adjustment coefficients, balancing the impacts of high - risk behaviors and operation frequencies on the safety score, set the threshold S threshold , when the safety score S of worker i i is lower than this threshold, mark this worker as a high - risk worker: I(S i <S threshold ), where, I(S i <S threshold ) is an indicator function. When the safety score of the worker is lower than the threshold, this value is 1, indicating that this worker is a high - risk worker, otherwise it is 0. Calculate the safety score according to the operation frequency and high - risk behavior ratio of the worker, reflecting the overall safety level of the worker.

[0070] Furthermore, in step S3, the way to generate a personalized safety training plan for each worker is as follows:

[0071] Use the safety score S i to classify the workers, define the training priority. According to the safety score of each worker, divide them into different risk levels, and also use the threshold S threshold to classify:

[0072] I(S i <S threshold ), where, S i is the safety score of worker i, S thresholdis the preset safety score threshold. Workers with scores below this threshold are given priority for safety training. When S i is less than S threshold , the value is 1, indicating that the worker needs additional safety training.

[0073] Detect the specific violations of the worker and classify them according to different types of violations b j . Assume there are m types of violations, and the risk weight for each type is r j . The cumulative number of the j-th type of violation for worker i is v i,j . Then the severity of worker i's violations is:

[0074] Among them, B i represents the comprehensive severity of worker i's violations, r j is the risk weight of the j-th type of violation, v i,j is the cumulative number of worker i for the j-th type of violation. Based on the severity of the worker's violations B i and their specific behavior b j , generate a personalized training plan. For each type of violation b j , define the corresponding training P j , and determine the training content to be pushed according to the frequency of the worker's specific violations. The recommended training P i is expressed as:

[0075] Among them, P i is the personalized training for worker i. I(v i,j >v threshold,j ) is an indicator function. When the number v i,j of worker i for the j-th type of violation exceeds the set threshold v threshold,j , push the corresponding training P j , and P j is the training for the j-th type of violation. For example, if the behavior b j is not wearing a safety helmet, then P j is the training on the use of PPE personal protective equipment.

[0076] Adjust the weight of the training plan according to the overall performance of the worker. Define the weight coefficient ω i , and adjust it dynamically according to the overall violation severity B i of the worker and the safety score S i . The formula for adjusting the weight of the training plan is: The final personalized training content is: Among them, ω i is the training weight of worker i, reflecting the importance of the training content. is the final personalized training plan for worker i,

[0077] Finally, based on the behavior performance of the worker after training, the training plan is dynamically adjusted, and the feedback score F is defined i , which is based on the change rate ΔS of the worker's performance after completing the training i and the new violation frequency v' i,j Definition:

[0078] where F i is the training feedback score of worker i, ΔS i is the change value of the safety score of worker i after training, v' i,j is the latest number of the jth type of violation of the worker after training, and γ and δ are adjustment coefficients. Based on the feedback score F i , the training plan is further adjusted

[0079] The beneficial effects of the present invention are as follows:

[0080] In the present invention, the multi-modal data fusion technology is combined with RFID, GPS, video monitoring and wearable devices to realize the monitoring of the worker's position, behavior and physiological state. The deep learning model is combined to automatically identify the worker's violation behavior, improve the recognition effect of the violation behavior. Once a violation behavior or potential safety hazard is detected, an alarm is immediately issued, and measures can be taken in the first time to reduce the accident risk.

[0081] In the present invention, the safety thresholds of the environment and behavior are dynamically adjusted by the regional risk coefficient. Workers in high-risk areas are monitored more strictly to avoid over-alarm. In different regions, early warning adjustments are made according to the risk level, significantly reducing false alarms and missed alarms. At the same time, the working environment is monitored in real time through temperature, humidity and dust environment sensors. Once the environmental conditions exceed the safety range, an early warning is immediately triggered. Whether the operation meets the safety standards is judged by combining environmental factors and worker behavior, and based on the worker's past behavior, the weight of recent high-risk behaviors is increased by time weighting, and the current safety state of the worker is given priority attention.

[0082] In the present invention, according to the safety score of the worker and the specific type of violation behavior, a personalized training plan is customized for each worker to improve the effectiveness of the training. At the same time, the performance of the worker after training is continuously tracked, and the training effect is feedback-adjusted by the change of the safety score and the new violation frequency, so as to improve the training effect.

[0083] The present invention uses a deep learning model to automatically analyze video surveillance and sensor data, reducing the dependence on manual intervention, enhancing the management efficiency and the intelligence of decision-making. Once abnormal situations of the environment, equipment or workers are detected, it immediately issues an early warning automatically without waiting for manual judgment, and processes automatically before the occurrence of danger, effectively reducing the accident rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following-described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0085] Figure 1 It is a schematic structural diagram of the intelligent construction site cloud platform based on the Internet of Things of the present invention;

[0086] Figure 2 It is a schematic flow diagram of the management method of the intelligent construction site cloud platform based on the Internet of Things of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0087] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention in conjunction with the drawings of the specification.

[0088] Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0089] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure or characteristic that can be included in at least one implementation manner of the present invention. The "in one embodiment" appearing in different places in this specification does not all refer to the same embodiment, nor is it a separate or alternative embodiment that excludes other embodiments.

[0090] Embodiment 1, referring to Figure 1 This embodiment provides an intelligent construction site cloud platform based on the Internet of Things, including:

[0091] A safety monitoring module for real-time location tracking of workers, physiological state monitoring, wearing state monitoring, and behavior recognition and safety warning.

[0092] An environment monitoring module for monitoring construction site environmental parameters, including temperature, humidity, and dust concentration, and issuing a warning when it detects that the environmental conditions exceed the safety threshold.

[0093] An intelligent analysis module integrates the data from the security monitoring module and the environmental monitoring module, including RFID, GPS positioning, video monitoring, environmental sensor, and worker wearable device data. Based on the fused data, it uses a deep learning model to identify worker behaviors and automatically detect whether there are any potential safety hazard behaviors of workers.

[0094] A personalized safety management module generates a safety score based on the recognition results of the intelligent analysis module, as well as the analysis of workers' daily operation habits and high-risk behaviors.

[0095] In the personalized safety management module, scoring is carried out based on the frequency of workers' violation behaviors and operation patterns, and workers with low scores are regarded as training targets.

[0096] And according to the operation habits and risk behaviors of workers, a personalized safety training plan is generated.

[0097] Example 2, referring to Figure 2 , this example provides an Internet of Things-based intelligent construction site cloud platform, including the following steps:

[0098] Step S1, real-time monitoring of multi-modal data

[0099] Adopt multi-modal sensor fusion technology to collect and integrate construction site data, generating fused data;

[0100] In step S1, the construction site data includes RFID, GPS positioning, video monitoring, environmental sensors, and intelligent devices worn by workers (such as intelligent bracelets and intelligent safety helmets).

[0101] In step S1, when it is detected that the environmental conditions exceed the safety threshold, a warning is issued.

[0102] In the construction site data, RFID and GPS data are used to track the positions of workers, environmental sensor data is used to monitor temperature, humidity, and dust environment data, wearable device data is used to monitor the physiological states of workers, and the intelligent safety helmet can detect whether it is worn correctly;

[0103] In step S1, multi-modal data fusion is carried out. Specifically:

[0104] Define different types of data sources: p RFID represents the worker location information provided by the RFID sensor, p GPS represents the worker location information provided by GPS, e temp 、e humid 、e dust are the environmental sensor data of temperature, humidity, and dust respectively, h heart represents the worker heart rate information detected by the intelligent bracelet, s helmetData representing the intelligent safety helmet, detecting the wearing situation of the safety helmet, 0 indicates not worn, 1 indicates correctly worn, and a weighted linear fusion method is used to fuse different data:

[0105] F fusion = w1·p RFID + w2·p GPS + w3·e temp + w4·e humid + w5·e dust + w6·h heart + w7·s helmet , where F fusion represents the fused data, w1, w2,..., w7 represent the weight coefficients of different data sources, and the weights are determined based on the confidence of sensor data and the influence of historical data:

[0106] where α i represents the confidence of the i-th type of sensor data,

[0107] Based on the fused data, real-time monitoring is carried out to determine whether the environmental conditions exceed the safety threshold, set the safety interval of the environmental variable, and compare it with the real-time monitoring value in this interval to trigger an alarm, and set the safety threshold of the environmental variable: T safe The safety interval of temperature is [T min , T max , H safe The safety interval of humidity is [H min , H max , D safe The safety interval of dust concentration is [D min , D max ,

[0108] For each moment t, if or or then P alert (t) = 1, otherwise, P alert (t) = 0, where P alert (t) represents the environmental warning trigger function. If the value is 1, it means that the current environmental conditions have exceeded the safety threshold and trigger an alarm. If the value is 0, it means that the environmental conditions are normal, e temp (t), e humid (t), e dust (t) represent the actual values of temperature, humidity, and dust concentration collected by the environmental sensor at moment t;

[0109] In step S1, dynamic adjustment of the fused data threshold is performed:

[0110] Dynamically adjust the warning threshold according to the construction area or operation risk level, and define the regional risk coefficient r zone , dynamically adjust the warning threshold:

[0111] T adjusted = r zone ·T safe , H adjusted = r zone ·H safe , D adjusted = r zone ·D safe , where T adjusted , H adjusted , D adjusted represent the safety thresholds of temperature, humidity and dust concentration after dynamic adjustment, and T safe , H safe , D safe represent the safety thresholds of temperature, humidity and dust concentration before adjustment respectively, and are adjusted according to the risk levels of different areas of the construction site. r zone represents the risk coefficient of the current operation area. For areas with higher risks, r zone > 1, and for areas with lower risks, r zone < 1.

[0112] Specifically, integrate the data of RFID, GPS, environmental sensors and intelligent wearable devices to realize the real-time monitoring of the location of workers, the working environment and their health status. In terms of environmental monitoring, it can immediately issue a warning when the environmental conditions, temperature, humidity and dust concentration exceed the set safety thresholds, avoiding worker health problems or safety accidents caused by environmental factors. In addition, the warning threshold is dynamically adjusted according to the risk levels of different areas, and safety precautions are enhanced in high-risk areas.

[0113] Step S2, intelligent behavior recognition and warning,

[0114] Based on the fused data in step S1, construct a deep learning model, and combine the convolutional neural network CNN and the long short-term memory network LSTM to identify workers' behaviors and conduct risk detection: This model analyzes the real-time video monitoring of the construction site and the data of wearable devices to identify potential safety hazard behaviors of workers.

[0115] In step S2, the identified potential safety hazard behaviors include:

[0116] Workers not wearing or wearing safety helmets incorrectly,

[0117] Workers entering high-risk areas,

[0118] Workers not wearing safety ropes during high-risk operations,

[0119] In step S2, after identifying potential safety hazards, the warning threshold is dynamically adjusted according to the risk levels of different operation areas, and an alarm is issued.

[0120] In step S2, based on the fusion data of step S1, a model combining convolutional neural network (CNN) and long short-term memory network (LSTM) is constructed to identify the behaviors of workers and detect potential risks. The specific steps include:

[0121] Let the data fusion result x obtained in step S1 be the temporal representation of multi-modal data, that is:

[0122] x = {x1, x2,..., x T}, x t ∈R d , where X is a temporal data sequence containing T time steps, the data dimension of each time step is d, and x t represents the fusion data at the t-th time step, including data from video surveillance, wearable devices, and environmental sensor data sources.

[0123] Extract the spatial features in the video frame. Let the video input be V t , the video frame corresponding to the moment t, and use a 2D convolutional layer to extract spatial features:

[0124] h t = ReLU(W conv *V t + b conv ), where h t is the feature map extracted from the video frame V t at the moment t, W conv is the weight matrix of the convolution kernel, * represents the convolution operation, b conv is the bias term, ReLU is a non-linear activation function, ReLU(x) = max(0, x), and the convolution feature h t is fused with other sensor data as video features to form new temporal data. For each time step t, the data x t from different sensors is concatenated with the video feature h t to obtain the temporal feature: z t = [x t ; h t , where z t is the comprehensive feature at the moment t, including the concatenation of the video feature h t and other sensor data x t , and [x t ; h t represents the vector concatenation operation.

[0125] In step S2, a long short-term memory network (LSTM) is used for temporal modeling:

[0126] Input the spliced feature sequence {z1, z2, …, z T} into the LSTM network. The LSTM framework is as follows:

[0127] Forget gate, f t = σ(W f z t + U f h t-1 + b f ),

[0128] Input gate, i t = σ(W i z t + U i h t-1 + b i ),

[0129] Candidate memory cell,

[0130] Updated memory cell,

[0131] Output gate, o t = σ(W o z t + U o h t-1 + b o ),

[0132] Hidden state update where f t , i t , o t are the activation values of the forget gate, input gate, and output gate respectively. W f , W i , W c , W o are the weight matrices of different gates, multiplied by the current input z t , U f , U i , U c , U o are the weight matrices of each gate for the hidden state h t-1 at the previous time step. b f , b i , b c , b o are the bias terms of each gate. is the candidate memory cell, c t is the updated memory cell, c t-1 is the memory cell at the previous time step. ° represents element-wise multiplication.

[0133] Based on the LSTM network modeling, the hidden state sequence {h1, h2, …, h T} is obtained. This sequence contains the behavior information of the worker during the entire time period. Then, the hidden state is classified through a fully connected layer to identify the potential safety hazard behaviors of the worker. For each time step t, the probability distribution of the behavior classification is output as:

[0134] y t = Softmax(W y h t + b y ), where y t is the behavior prediction result at time step t, indicating whether the worker has potential safety hazard behaviors (such as not wearing a safety helmet, entering a high-risk area, etc.). W y is the weight matrix for behavior classification, and b y is the bias term of the classification layer. The Softmax function converts the output into a probability distribution, which is defined as:

[0135] According to the output y t of the model, a risk detection mechanism for the worker's behavior is defined. When the model predicts a certain high-risk behavior, a warning is triggered. The risk detection is defined as:

[0136] r t = I(y t ∈ HighRiskActions), where r t is the risk detection result at time step t. If the value is 1, it means that there is a high-risk behavior currently. I is the indicator function. If y t belongs to the predefined high-risk behaviors, it outputs 1; otherwise, it outputs 0. Once a high-risk behavior r t = 1 is detected, a warning signal is immediately sent out to remind the management to take measures.

[0137] Specifically, by combining the convolutional neural network CNN and the long short-term memory network LSTM, the real-time video surveillance data of the construction site and the data of wearable devices are automatically analyzed to identify the workers' illegal operations: not wearing a safety helmet, entering a high-risk area, or not wearing a safety rope during high-risk operations, effectively identifying potential safety hazard behaviors, and adjusting the warning threshold according to the risk level of the operation area to ensure the accuracy and timeliness of the warning information, greatly reducing the probability of accidents.

[0138] Step S3, personalized behavior analysis

[0139] Based on the worker behavior data identified in step S2 and the risk detection results, that is, the identified potential safety hazard behaviors, the operation habits of the worker are analyzed, and personalized safety training suggestions are generated.

[0140] In step S3, the way to analyze the operation habits of workers and generate personalized safety training suggestions is as follows:

[0141] Generate a safety score for the worker based on the frequency of their operations and the proportion of high-risk behaviors. Workers with frequent high-risk behaviors receive a low score.

[0142] Combine the score data to generate a personalized safety training plan for each worker, and recommend corresponding training content for different violations.

[0143] For example:

[0144] For workers who frequently do not wear protective equipment, emphasize the correct use of PPE (Personal Protective Equipment).

[0145] For workers who operate improperly during the use of specific equipment, the system will push training on equipment usage specifications.

[0146] Specifically, based on the frequency of workers' operations and the proportion of high-risk behaviors, automatically generate a safety score, mark workers with lower scores as high-risk workers, analyze the workers' past violations, and develop a personalized training plan for them. Different violations correspond to different training contents, significantly improving the pertinence and effectiveness of training.

[0147] In step S3, generate a safety score based on the frequency of workers' operations and the proportion of high-risk behaviors. Specifically:

[0148] Set the total number of all operations completed by worker i during the monitoring period as N i , classify the different operation behaviors of the worker into m types of behavior, and record the number of times n of each behavior i,j , where j = 1, 2,..., m represents different types of behaviors, and the frequency of the worker's behavior is expressed as: Among them, f i,j represents the frequency of the j-th operation behavior of worker i, reflecting the proportion of this behavior in their total operation behaviors, n i,j represents the number of times of the j-th operation completed by worker i, and N i is the total number of all operations of worker i.

[0149] Let the risk level of the j-th operation behavior be r j , and the value of the risk level depends on the danger of this operation. Higher-risk behaviors are given a larger r j value, and the proportion R of high-risk behaviors of worker i i is expressed as:

[0150] Among them, R iRepresents the proportion of high-risk behaviors of worker i, reflecting the proportion of high-risk operations in their operation behaviors, r j Is the risk level of the j-th operation behavior, n i,j Is the number of times worker i has completed the j-th operation, N i Is the total number of operations of worker i,

[0151] Define the time-weighting function w(t) to assign greater weight to the most recently occurring high-risk behaviors. Let the high-risk behavior of the worker at time t be r j (t), then the time-weighted high-risk behavior within the time period T is expressed as:

[0152] Among them, Is the time-weighted high-risk behavior proportion of worker i, w(t) is the time-weighting function, w(t) = exp(-λ(T - t)), where λ is the time decay coefficient, indicating the degree to which the impact of the behavior decreases over time, r j (t) is the risk level at time t, n i,j (t) is the number of times of the j-th operation behavior of worker i at time t, N i (t) is the total number of operations of the worker at time t,

[0153] Based on the operation frequency f of the worker i,j And the time-weighted high-risk behavior proportion Define the safety score S of worker i i :

[0154] Among them, S i Is the safety score of worker i, S max Is the preset maximum safety score, representing the score when the worker has completely risk-free operations, Is the time-weighted high-risk behavior proportion of the worker, reflecting the proportion of high-risk behaviors in the worker's past operations, f i,j Is the operation frequency of the j-th operation behavior of worker i, r j Is the risk level of the j-th operation behavior, α and β are adjustment coefficients to balance the impact of high-risk behaviors and operation frequencies on the safety score, set the threshold S threshold , when the safety score S of worker i i Is lower than this threshold, mark this worker as a high-risk worker: I(S i < S threshold ), among them, I(S i < S threshold) is an indicator function. When the safety score of a worker is lower than the threshold, the value is 1, indicating that the worker is a high-risk worker; otherwise, it is 0. The safety score is calculated based on the operation frequency and the proportion of high-risk behaviors of the worker, reflecting the overall safety level of the worker.

[0155] In step S3, the way to generate a personalized safety training plan for each worker is as follows:

[0156] Use the safety score S i Classify the workers and define the training priorities. According to the safety score of each worker, divide them into different risk levels. The same threshold S threshold is used for classification:

[0157] I(S i < S threshold ), where S i is the safety score of worker i, and S threshold is the preset safety score threshold. Workers with scores lower than this threshold are given priority for safety training. When S i is less than S threshold , the value is 1, indicating that the worker needs additional safety training.

[0158] Detect the specific violations of the worker and classify them according to different violation types b j . Assume there are m types of violations, and the risk weight of each type is r j , and the cumulative number of times of the j-th violation of worker i is v i,j . Then the severity of the violations of worker i is:

[0159] Among them, B i represents the comprehensive severity of the violations of worker i, r j is the risk weight of the j-th violation, and v i,j is the cumulative number of times of worker i for the j-th violation. According to the severity of the worker's violations B i and their specific behavior b j , generate a personalized training plan. For each type of violation b j , define the corresponding training P j , and determine the training content to be pushed according to the specific violation frequency of the worker. The recommended training P i is expressed as:

[0160] Among them, P i is the personalized training of worker i, and I(v i,j > v threshold,j ) is an indicator function. When the number of times v i,j of worker i for the j-th violation exceeds the set threshold vthreshold,j At this time, push the corresponding training P j , where P j is the training for the j-th type of violation. For example, if the behavior b j is not wearing a safety helmet, then P j is the training on the use of PPE personal protective equipment.

[0161] Adjust the weight of the training plan according to the overall performance of the workers, and define the weight coefficient ω i , according to the overall violation severity B i of the workers and the safety score S i for dynamic adjustment. The training plan weight adjustment formula is: The final personalized training content is: where ω i is the training weight of worker i, reflecting the importance of the training content. is the final personalized training plan of worker i.

[0162] Finally, based on the behavior performance of the workers after training, dynamically adjust the training plan, and define the feedback score F i , which is based on the change rate ΔS i of the workers' performance after completing the training and the new violation frequency v' i,j is defined as:

[0163] where F i is the training feedback score of worker i, ΔS i is the change value of the safety score of worker i after training, v' i,j is the latest number of the j-th type of violation of the workers after training, and γ and δ are adjustment coefficients. Based on the feedback score F i , further adjust the training plan Specifically, analyze the performance changes of the workers after training for real-time adjustment to long-term reduce the incidence of violation operations.

[0164] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An IoT-based intelligent construction site cloud platform, characterized in that: including, a safety monitoring module for obtaining construction site data; an environmental monitoring module that, based on the construction site data, determines whether the environmental conditions exceed the safety threshold, with a corresponding threshold set for each environmental factor, and the threshold is dynamically adjusted based on the risk coefficient of the operation area; an intelligent analysis module that fuses the data obtained by the safety monitoring module and the environmental monitoring module, and uses a deep learning model to classify workers' behaviors and detect risks; a personalized safety management module that scores each worker based on the analysis results of the intelligent analysis module, and provides targeted training for the workers based on the scores, and the scores are dynamically adjusted according to the high-risk behaviors of the corresponding workers; In the personalized safety management module, the training is dynamically adjusted according to the severity and frequency of each worker's violation behaviors.

2. The intelligent construction site cloud platform based on the Internet of Things according to claim 1, characterized in that, The management method of the intelligent construction site cloud platform is as follows: Step S1, multi-modal data real-time monitoring: Adopt multi-modal sensor fusion technology to collect and integrate construction site data to generate fusion data; Step S2, intelligent behavior recognition and early warning: Based on the fusion data in Step S1, construct a deep learning model, and combine the convolutional neural network CNN and the long short-term memory network LSTM to identify workers' behaviors and detect risks: This model analyzes the real-time video monitoring of the construction site and the data of wearable devices to identify potential safety hazard behaviors of workers, Step S3, personalized behavior analysis: Based on the worker behavior data identified in Step S2 and the risk detection results, that is, the identified potential safety hazard behaviors, analyze the operation habits of workers and generate personalized safety training suggestions.

3. The intelligent construction site cloud platform based on the Internet of Things according to claim 2, characterized in that In Step S1, the construction site data includes RFID, GPS positioning, video monitoring, environmental sensors, and intelligent devices worn by workers, In Step S1, an early warning is issued when it is detected that the environmental conditions exceed the safety threshold.

4. The intelligent construction site cloud platform based on the Internet of Things according to claim 2, characterized in that In Step S1, multi-modal data fusion is performed. Specifically: Define different types of data sources: p RFID Indicates the worker location information provided by the RFID sensor, p GPS Indicates the worker location information provided by the GPS, e temp e humid e dust Are the environmental sensor data of temperature, humidity, and dust respectively, h heart Indicates the worker heart rate information detected by the smart bracelet, s helmet Indicates the data of the smart safety helmet, detects the wearing situation of the safety helmet, 0 means not worn, 1 means correctly worn, and uses the weighted linear fusion method to fuse different data: F fusion = w1·p RFID + w2·p GPS + w3·e temp + w4·e humid + w5·e dust + w6·h heart + w7·s helmet where F fusion represents the fused data, and w1, w2, …, w7 represent the weight coefficients of different data sources. The weights are determined based on the confidence of sensor data and the influence of historical data: where α i represents the confidence level of the i-th type of sensor data Based on the fusion data, conduct real-time monitoring to determine whether the environmental conditions exceed the safety threshold, set the safety range of the environmental variables, compare the real-time monitoring values with this range, trigger an early warning, and set the safety threshold of the environmental variable: T safe The safety range of temperature is [T min , T max , H safe The safety range of humidity is [H min , H max , D safe The safety range of dust concentration is [D min , D max , For each moment t, if or or then P alert (t) = 1, otherwise, P alert (t) = 0, where P alert (t) represents the environmental warning trigger function. If the value is 1, it means that the current environmental conditions have exceeded the safety threshold and the warning is triggered. If the value is 0, it means that the environmental conditions are normal, e temp (t), e humid (t), e dust (t) represent the actual values of temperature, humidity, and dust concentration collected by the environmental sensor at moment t; In Step S1, dynamic adjustment of the fusion data threshold: Dynamically adjust the warning threshold according to the construction area or operation risk level, and define the regional risk coefficient r zone , and dynamically adjust the warning threshold: T adjusted = r zone ·T safe , H adjusted = r zone ·H safe , D adjusted = r zone ·D safe , where T adjusted , H adjusted , D adjusted represent the safety thresholds of temperature, humidity, and dust concentration after dynamic adjustment, and T safe , H safe , D safe represent the safety thresholds of temperature, humidity, and dust concentration before adjustment respectively, and are adjusted according to the risk levels of different areas of the construction site. r zone represents the risk coefficient of the current working area.

5. The intelligent construction site cloud platform based on the Internet of Things according to claim 2, characterized in that, In Step S2, the identified potential safety hazard behaviors include: Workers not wearing or wearing helmets incorrectly, Workers entering high-risk areas, Workers not wearing safety ropes during high-risk operations, In Step S2, after identifying potential safety hazards, according to the risk levels of different operation areas, dynamically adjust the early warning threshold and issue an alarm.

6. The intelligent construction site cloud platform based on the Internet of Things according to claim 5, characterized in that In Step S2, based on the fusion data in Step S1, construct a model combining the convolutional neural network CNN and the long short-term memory network LSTM to identify workers' behaviors and detect potential risks. The specific steps include: Let the data fusion result X obtained in Step S1, where X is the temporal representation of multi-modal data, that is: X = {x1, x2, …, x T}, x t ∈ R d , where X is a time-series data sequence containing T time steps, the data dimension of each time step is d, and x t represents the fused data at the t-th time step, including data from video surveillance, wearable devices, and environmental sensor data sources. Extract the spatial features in the video frames. Let the video input be V t , for the video frame at the corresponding time t, use a 2D convolutional layer to extract the spatial features: h t = ReLU(W conv * V t + b conv ), where h t is the feature map extracted from the video frame V t at time t, W conv is the weight matrix of the convolutional kernel, * represents the convolution operation, b conv is the bias term, ReLU is a non - linear activation function, ReLU(x) = max(0, x), and the convolutional feature h t is fused with other sensor data as video features to form new time - series data. For each time step t, the data x t from different sensors is concatenated with the video feature h t to obtain the time - series feature: z t = [x t ; h t , where z t is the comprehensive feature at time t, including the concatenation of the video feature h t and other sensor data x t [x t ; h t represents the concatenation operation of vectors.

7. The intelligent construction site cloud platform based on the Internet of Things according to claim 6, characterized in that In Step S2, use the long short-term memory network LSTM for temporal modeling: Input the concatenated feature sequence {z1, z2, …, z T} into the LSTM network. The LSTM framework is as follows: Forgotten gate, f t = σ(W f z t + U f h t-1 + b f ) Input gate, i t = σ(W i z t + U i h t-1 + b i ), Candidate memory unit, Update memory unit, Output gate, o t = σ(W o z t + U o h t-1 + b o ) Hidden state update where f t , i t , o t are the activation values of the forget gate, input gate, and output gate respectively, W f , W i , W c , W o are the weight matrices of different gates, multiplied by the current input z t , U f , U i , U c , U o are the weight matrices of each gate for the hidden state h t-1 at the previous time step, b f , b i , b c , b o are the bias terms of each gate, is the candidate memory cell, c t is the updated memory cell, c t-1 is the memory cell at the previous time step, ° represents element-wise multiplication, Based on the LSTM network modeling, the hidden state sequence {h1, h2, …, h T} is obtained. This sequence contains the behavior information of the worker during the entire time period. Then, the hidden state is classified through a fully connected layer to identify the potential safety hazard behaviors of the worker. For each time step t, the probability distribution of the behavior classification is output as: y t = Softmax(W y h t + b y ), where y t is the behavior prediction result at time t, indicating whether there is a potential safety hazard behavior of the worker, W y is the weight matrix for behavior classification, b y is the bias term of the classification layer, and the Softmax function converts the output into a probability distribution, defined as: Based on the output y of the model t Define a risk detection mechanism for workers' behaviors. When the model predicts a certain high-risk behavior, an early warning is triggered. The risk detection is defined as: r t = I(y t ∈ HighRiskActions), where r t is the risk detection result at time t. If the value is 1, it means that there is a high-risk behavior currently. I is the indicator function. If y t belongs to the predefined high-risk actions, it outputs 1; otherwise, it outputs 0. Once the high-risk behavior r t = 1 is detected, a warning signal is immediately sent to remind the management to take measures.

8. The intelligent construction site cloud platform based on the Internet of Things according to claim 2, characterized in that In Step S3, the method of analyzing the operation habits of workers and generating personalized safety training suggestions is: Generate a safety score for workers according to the frequency of their operations and the proportion of high-risk behaviors. Workers with frequent high-risk behaviors receive low scores, Combined with the score data, generate a personalized safety training plan for each worker, and recommend corresponding training content for different violation behaviors.

9. The intelligent construction site cloud platform based on the Internet of Things according to claim 8, characterized in that, In Step S3, generate a safety score according to the frequency of workers' operations and the proportion of high-risk behaviors. Specifically: During the monitoring period, the total number of all operations completed by worker i is set to N i , the different operation behaviors of the worker are classified into m types of behavior, and the number of times n of each behavior is recorded i,j , where j = 1, 2, …, m represents different types of behavior, and the frequency of the worker's behavior is expressed as: Among them, f i,j represents the frequency of the j-th operation behavior of worker i, reflecting the proportion of this behavior in his total operation behaviors, n i,j represents the number of times of the j-th operation completed by worker i, N i is the total number of all operations of worker i Let the risk level of the j-th operation behavior be r j , and the high-risk behavior ratio R of worker i i is expressed as: Among them, R i represents the proportion of high-risk behaviors of worker i, reflecting the proportion of high-risk operations in his operation behaviors, and r j is the risk level of the jth operation behavior, and n i,j is the number of times worker i has completed the jth operation, and N i is the total number of operations of worker i. Define the time-weighted function \(w(t)\) to assign a greater weight to the most recent high-risk behaviors. Let the high-risk behavior of the worker at time \(t\) be \(r(t)\). Then the time-weighted high-risk behavior within the time period \(T\) is expressed as: j (t), then the time-weighted high-risk behavior within the time period \(T\) is expressed as: Among them, is the time-weighted high-risk behavior ratio of worker i, w(t) is the time-weighted function, w(t) = exp(-λ(T - t)), where λ is the time decay coefficient, indicating the degree to which the impact of the behavior decreases over time, r j (t) is the risk level at time t, n i,j (t) is the number of the jth operation behavior of worker i at time t, N i (t) is the total number of all operations of the worker at time t, Based on the operation frequency f of the worker i,j and the time-weighted proportion of high-risk behaviors Define the safety score S of worker i i : where S i is the safety score of worker i, and S max is the preset maximum safety score, representing the score when the worker operates without any risk, is the time-weighted high-risk behavior ratio of the worker, reflecting the proportion of high-risk behaviors in the worker's past operations, f i,j is the frequency of the jth operation behavior of worker i, r j is the risk level of the jth operation behavior, and α and β are adjustment coefficients to balance the impacts of high-risk behaviors and operation frequencies on the safety score. Set the threshold S threshold . When the safety score S i of worker i is lower than this threshold, mark this worker as a high-risk worker: I(S i < S threshold ), where I(S i < S threshold ) is an indicator function. When the safety score of the worker is lower than the threshold, this value is 1, indicating that the worker is a high-risk worker; otherwise, it is 0.

10. A smart construction site cloud platform based on the Internet of Things according to claim 9, characterized in that, In step S3, the way to generate a personalized safety training plan for each worker is as follows: Using safety score S i Classify workers and define training priorities. Based on the safety score of each worker, divide them into different risk levels, also using the threshold S threshold Conduct classification: I(S i <S threshold ), where S i is the safety score of worker i, and S threshold is the preset safety score threshold. Workers with scores lower than this threshold are given priority for safety training. When S i is less than S threshold , the value is 1, indicating that the worker requires additional safety training. Detect the specific violations of the workers, and classify them according to different types of violations b j For classification, assume that there are m types of violations, and the risk weight of each type of behavior is r j , and the cumulative number of the j-th type of violation of worker i is v i,j , then the severity of the violations of worker i is as follows: Among them, B i represents the comprehensive severity of the violation behavior of worker i, r j is the risk weight of the jth type of violation behavior, v i,j is the cumulative number of times worker i has committed the jth type of violation behavior. Based on the violation behavior severity B i of the worker and his specific behavior b j , a personalized training plan is generated. For each type of violation behavior b j , the corresponding training P j is defined, and the training content to be pushed is determined according to the specific violation behavior frequency of the worker. The recommended training P i is expressed as: Among them, P i is the personalized training for worker i, and I(v i,j >v threshold,j ) is an indicator function. When the number of times v i,j of worker i for the j-th type of violation exceeds the set threshold v threshold,j , the corresponding training P j is pushed. P j is the training for the j-th type of violation. Adjust the weight of the training plan according to the overall performance of the workers, and define the weight coefficient ω i , according to the overall violation severity B of the workers i and the safety score S i for dynamic adjustment. The adjustment formula for the training plan weight is as follows: The final personalized training content is: Among them, ω i is the training weight of worker i, reflecting the importance of the training content, is the final personalized training plan for worker i Finally, based on the worker's behavior performance after training, the training plan is dynamically adjusted, and the feedback score F is defined i , which is based on the change rate of the worker's performance ΔS after completing the training i and the frequency of new violations v' i,j Definition: Among them, F i is the training feedback score of worker i, and ΔS i is the change value of the safety score of worker i after training, and v' i,j is the latest number of the j-th type of violation behavior of the worker after training. γ and δ are adjustment coefficients, and based on the feedback score F i , further adjust the training plan

Images