Construction safety grid management optimization method and system based on artificial intelligence

Abstract

The present invention discloses an artificial intelligence-based construction safety grid management optimization method and system, belonging to the field of construction safety management technology. The method includes grid area division, data preprocessing, environmental risk prediction, accident risk prediction, and construction safety management optimization. The present invention uses a self-attention multi-scale convolutional long short-term memory model for environmental risk prediction, comprehensively capturing information of different scales in environmental-related data and enhancing attention to critical moments, thereby accurately capturing important risk signals; uses a dual-path multimodal deep neural network model for accident risk prediction, comprehensively capturing multiple factors affecting accident risk, enhancing the accuracy of the prediction results, and improving the model's computational efficiency by processing different types of data through dual paths; and uses two steps, environmental risk prediction and accident risk prediction, to assist in grid-based construction safety management optimization, improving the accuracy, response speed, and visualization of construction safety management.

Description

Technical Field

[0001] The present invention belongs to the technical field of construction safety management, and specifically refers to an artificial intelligence-based construction safety grid management optimization method and system. Background Art

[0002] Construction safety grid management optimization combines artificial intelligence technology and grid management concepts to conduct dynamic risk analysis of construction sites, aiming to provide accurate risk predictions for construction sites, improve the scientific nature and efficiency of construction safety management, thereby promoting the refined and intelligent development of construction safety management and effectively reducing the accident rate.

[0003] However, in the existing environmental risk prediction process, changes in data such as temperature, humidity, and dust concentration in the environment often have trends on different time scales. Traditional environmental prediction models only consider single-scale characteristics, ignore the information complementarity between different scales, and cannot effectively capture multi-scale environmental changes, resulting in technical problems such as poor prediction accuracy; in the existing accident risk prediction process, accident risks at construction sites need to be analyzed by comprehensively considering environmental changes, historical safety accidents, and real-time monitoring, while traditional accident risk prediction usually relies on a single data source and cannot fully consider information from multiple dimensions, resulting in technical problems such as reduced prediction accuracy; in the existing construction safety grid management optimization process, there is the existing construction site management process, which makes it difficult to accurately predict and intervene in potential risks in different areas of the construction site, and the main safety hazards at the construction site come from environmental changes and construction site accidents, resulting in the lack of regionalization, precise prediction and real-time adjustment of traditional safety management. Summary of the Invention

[0004] In view of the above situation, in order to overcome the defects of the existing technology, the present invention provides an artificial intelligence-based construction safety grid management optimization method and system. In view of the fact that in the existing environmental risk prediction process, changes in data such as temperature, humidity, and dust concentration in the environment often have trends on different time scales, and traditional environmental prediction models only consider single-scale features, ignore the information complementarity between different scales, and cannot effectively capture multi-scale environmental changes, resulting in poor prediction accuracy. This solution creatively adopts a self-attention multi-scale convolutional long short-term memory model for environmental risk prediction, comprehensively captures different-scale information in environmental-related data, effectively handles long-term dependencies, and enhances attention to critical moments, thereby accurately capturing important risk signals, improving prediction accuracy and sensitivity, and helping to improve the overall effectiveness and reliability of the method; In view of the fact that in the existing accident risk prediction process, there are accident risks at the construction site that need to be analyzed by comprehensively considering environmental changes, historical safety accidents, and real-time monitoring, while traditional accident risk prediction usually relies on a single data source and cannot fully consider information from multiple dimensions, resulting in reduced prediction accuracy, this solution creatively adopts A dual-path multimodal deep neural network model is used to predict accident risks, effectively processing multimodal data related to accident risks and comprehensively capturing the various factors that influence accident risks. This enhances the accuracy of prediction results. Furthermore, by processing different types of data through dual paths, the model's computational efficiency is improved, facilitating the timely identification of potential safety hazards at construction sites. In the current optimization process of grid-based construction safety management, the existing construction site management system is unable to accurately predict and intervene in potential risks in different areas of the construction site. The main safety hazards at construction sites come from environmental changes and construction site accidents, resulting in a lack of regionalized, precise predictions and real-time adjustments in traditional safety management. This solution creatively employs two steps, environmental risk prediction and accident risk prediction, to assist in the optimization of grid-based construction safety management. By predicting both environmental and accident risks, dynamic safety assessments and targeted inspections can be implemented for each area within the grid management framework, thereby optimizing resource allocation and inspection efficiency, reducing safety hazards, and improving the accuracy, response speed, and visualization of construction safety management, ensuring more scientific and efficient safety management in each grid area.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides an artificial intelligence-based construction safety grid management optimization method, which includes the following steps:

[0006] Step S1: grid area division;

[0007] Step S2: data preprocessing;

[0008] Step S3: Environmental risk prediction;

[0009] Step S4: accident risk prediction;

[0010] Step S5: Optimize construction safety management.

[0011] Furthermore, in step S1, the grid area division is used to divide the grid area and perform data collection, specifically dividing the construction site into N grid areas according to the distribution of the work area, and collecting historical safety accident data of the construction site, and collecting environmental data and video surveillance data in the grid area through sensors;

[0012] The historical safety accident data, including accident date, location, category and level;

[0013] The environmental data includes temperature, humidity, light, noise and dust concentration.

[0014] Furthermore, in step S2, the data preprocessing is used to optimize the data quality, including the following steps:

[0015] Step S21: pre-processing the historical safety accident data, specifically, performing data cleaning and normalization on the historical safety accident data, and label encoding the type data in the historical safety accident data to obtain historical safety accident optimized data;

[0016] Step S22: Environmental data preprocessing, specifically, obtaining environmental optimization data by performing data cleaning, time alignment, and normalization on the environmental data;

[0017] Step S23: Video surveillance data preprocessing, specifically extracting image frames from the video surveillance data at a rate of one frame per second, denoising and standardizing each frame of image to obtain surveillance standard image data, and dividing all images in the surveillance standard image data into G fixed-size image blocks to obtain surveillance image block data.

[0018] Furthermore, in step S3, the environmental risk prediction is used to predict environmental risks, specifically by using a self-attention multi-scale convolutional long short-term memory model to perform environmental risk prediction based on environmental optimization data to obtain environmental risk prediction information;

[0019] The self-attention multi-scale convolutional long short-term memory model includes an input layer, a multi-scale convolutional subnet, a random inactivation layer, a long short-term memory layer, a self-attention layer and an output layer;

[0020] The multi-scale convolutional subnet is used to capture multi-scale features from the data;

[0021] The long short-term memory layer is used to capture long-term and short-term dependencies from data;

[0022] The self-attention layer is used to focus on key time steps;

[0023] The environmental risk prediction includes the following steps:

[0024] Step S31: Constructing a self-attention multi-scale convolutional long short-term memory model, including the following steps:

[0025] Step S311: constructing an input layer for receiving input data, specifically receiving environmental optimization data as input data of the model through the input layer;

[0026] Step S312: Construct a multi-scale convolutional subnetwork, specifically, setting a parallel first convolutional layer, a second convolutional layer, and a third convolutional layer in the multi-scale convolutional subnetwork, and setting an average pooling layer after each convolutional layer. The output results of each average pooling layer are spliced ​​through a feature splicing layer to obtain multi-scale features of the environment. The first convolutional layer contains 16 4×1 convolution kernels, the second convolutional layer contains 32 4×5 convolution kernels, and the third convolutional layer contains 64 4×5 convolution kernels.

[0027] Step S313: construct a random dropout layer to avoid model overfitting. Specifically, the random dropout layer is used to process the multi-scale features of the environment to obtain the random dropout features of the environment.

[0028] Step S314: constructing a long-short-term memory layer, specifically setting 295 long-short-term memory units in the long-short-term memory layer, processing the random inactivation characteristics of the environment through the long-short-term memory layer to obtain the long-short-term time series characteristics of the environment;

[0029] The long short-term memory unit includes a forget gate, an input gate, an output gate and a memory unit;

[0030] Step S315: Construct a self-attention layer. Specifically, in the self-attention layer, the self-attention mechanism is used to process the long-term and short-term temporal features of the environment to obtain the self-attention features of the environment. The calculation formula is:

[0031] ;

[0032] Where Q is the query matrix, h is the long-term and short-term time series features of the environment, the long-term and short-term time series features are a matrix of u×d, u is the total number of time steps of the long-term and short-term time series features of the environment, d is the dimension of the long-term and short-term time series features of the environment, and W Q is the d×d query weight matrix, K is the key matrix, W K is a d×d key weight matrix, V is a value matrix, W V is a d×d value weight matrix, Att is the self-attention score, T is the transpose operation, and h att is the environmental self-attention feature;

[0033] Step S316: construct an output layer for outputting the model prediction results, specifically processing the environment self-attention features through a fully connected layer and a linear activation function to obtain the model prediction results;

[0034] Step S32: constructing an environmental risk prediction model, specifically, training a self-attention multi-scale convolutional long short-term memory model based on the environmental optimization data, and using a particle swarm optimization algorithm to tune the model parameters to obtain an environmental risk prediction model;

[0035] Step S33: Generate environmental risk prediction information, specifically, perform environmental risk prediction on each grid area through the environmental risk prediction model to obtain environmental risk prediction information, wherein the environmental risk prediction information includes environmental risk values ​​and main environmental risk factors.

[0036] Furthermore, in step S4, the accident risk prediction is used to predict the accident risk, specifically by using a dual-path multimodal deep neural network model to perform accident risk prediction based on environmental risk prediction information, historical safety accident optimization data, and monitoring image block data to obtain accident risk prediction information;

[0037] The dual-path multimodal deep neural network model includes a dual-path feature processing subnet, a multimodal feature fusion layer and a prediction output layer;

[0038] The dual-path feature processing subnet is used to process feature information of multimodal data in parallel;

[0039] The multimodal feature fusion layer is used to fuse multimodal feature information;

[0040] The accident risk prediction includes the following steps:

[0041] Step S41: constructing a dual-path feature processing subnet, specifically setting a parallel first feature processing path and a second feature processing path in the dual-path feature processing subnet, including the following steps:

[0042] Step S411: constructing a first feature processing path, specifically setting a random forest model in the first feature processing path, processing environmental risk prediction information and historical safety accident optimization data through the first feature processing path to obtain environmental accident characteristics;

[0043] Step S412: Constructing a second feature processing path, specifically setting a flattened embedding layer, an image visual transformer model, a long short-term memory layer, and an average pooling layer in the second feature processing path, processing the monitoring image block data through the second feature processing path to obtain monitoring image features, including the following steps:

[0044] Step S4121: construct a flattened embedding layer, specifically flattening each monitoring image block in the monitoring image block data into a vector, and performing linear mapping on the vector to obtain the monitoring image block mapping feature. The calculation formula is:

[0045] ;

[0046] Where z i is the mapping feature of the i-th monitoring image block, i is the monitoring image block index, W p is the linear mapping weight, x i is the flattened vector of the i-th monitoring image block, b p is the linear mapping bias term;

[0047] Step S4122: Constructing a visual transformer model. Specifically, setting a 24-layer encoder in the visual transformer model and setting 16 self-attention heads in each layer. The visual transformer model is used to process the monitoring image block mapping features to obtain the monitoring image global features.

[0048] Step S4123: constructing a long short-term memory layer, setting 295 long short-term memory units in the long short-term memory layer, and processing the global features of the monitoring image through the long short-term memory layer to obtain the monitoring image time series features;

[0049] Step S4124: construct an average pooling layer for aggregating features in the time dimension. Specifically, the monitoring image temporal features are processed by the average pooling layer to obtain the monitoring average pooling features. The calculation formula is:

[0050] ;

[0051] Where, f G is the monitoring average pooling feature, t max is the maximum time step, t is the time step index, c t is the temporal feature of the monitoring image at the tth time step;

[0052] Step S42: Construct a multimodal feature fusion layer, specifically by weighted fusion of environmental accident features and monitoring average pooling features to obtain accident risk features. The calculation formula is:

[0053] ;

[0054] Where, f total is the accident risk characteristic, is the first weighting coefficient, which is used to represent the weight of environmental accident characteristics, f env It is a characteristic of environmental accidents. is the second weighting coefficient, which is used to represent the monitoring average pooling feature weight;

[0055] Step S43: Constructing a prediction output layer, specifically processing the accident risk features through a fully connected layer and a softmax activation function to obtain the final prediction result;

[0056] Step S44: constructing a dual-path multimodal deep neural network model, specifically by constructing a dual-path multimodal deep neural network model through the construction of the dual-path feature processing subnet, the construction of the multimodal feature fusion layer, and the construction of the prediction output layer;

[0057] Step S45: Constructing an accident risk prediction model, specifically, training a dual-path multimodal deep neural network model based on environmental risk prediction information, historical safety accident optimization data, and monitoring image block data, and using a particle swarm optimization algorithm to tune model parameters to obtain an accident risk prediction model;

[0058] Step S46: Generate accident risk prediction information, specifically, perform accident risk prediction through an accident risk prediction model to obtain accident risk prediction information, wherein the accident risk prediction information includes accident risk level and main accident risk factors.

[0059] Furthermore, in step S5, the construction safety management optimization is used to optimize the grid management of construction safety, specifically to optimize the comprehensive management of construction safety based on the accident risk prediction information and the environmental risk prediction information, obtain construction safety management prediction auxiliary information, and conduct grid construction site safety inspections, data aggregation and visual comprehensive analysis based on the construction safety management prediction auxiliary information to generate a construction safety management optimization report to help construction safety managers conduct safety management and decision-making within the grid area.

[0060] The artificial intelligence-based construction safety grid management optimization system provided by the present invention includes: a grid area division module, a data preprocessing module, an environmental risk prediction module, an accident risk prediction module and a construction safety management optimization module;

[0061] The grid area division module is used for grid area division, obtaining historical safety accident data, environmental data and video surveillance data through grid area division, and sending the historical safety accident data, the environmental data and the video surveillance data to the data preprocessing module;

[0062] The data preprocessing module is used for data preprocessing, and obtains historical safety accident optimization data, environmental optimization data and monitoring image block data through data preprocessing, and sends the environmental optimization data to the environmental risk prediction module, and sends the historical safety accident optimization data and the monitoring image block data to the accident risk prediction module;

[0063] The environmental risk prediction module is used for environmental risk prediction, obtains environmental risk prediction information through environmental risk prediction, and sends the environmental risk prediction information to the accident risk prediction module and the construction safety management optimization module;

[0064] The accident risk prediction module is used for accident risk prediction, obtains accident risk prediction information through accident risk prediction, and sends the accident risk prediction information to the construction safety management optimization module;

[0065] The construction safety management optimization module is used for construction safety management optimization, and obtains construction safety management prediction auxiliary information and a construction safety management optimization report through construction safety management optimization.

[0066] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0067] (1) In the existing environmental risk prediction process, the changes in data such as temperature, humidity, and dust concentration in the environment often have trends at different time scales. Traditional environmental prediction models only consider single-scale features, ignore the information complementarity between different scales, and cannot effectively capture multi-scale environmental changes, resulting in poor prediction accuracy. This solution creatively uses a self-attention multi-scale convolutional long short-term memory model for environmental risk prediction, comprehensively capturing information of different scales in environmental-related data, effectively processing long-term dependencies, and enhancing attention to critical moments, thereby accurately capturing important risk signals, improving prediction accuracy and sensitivity, and helping to improve the overall effectiveness and reliability of the method.

[0068] (2) In the existing accident risk prediction process, the accident risk at the construction site needs to be analyzed by comprehensively considering environmental changes, historical safety accidents and real-time monitoring. However, traditional accident risk prediction usually relies on a single data source and cannot fully consider information from multiple dimensions, resulting in a technical problem of reduced prediction accuracy. This solution creatively adopts a dual-path multimodal deep neural network model for accident risk prediction, effectively processes multimodal data related to accident risk, comprehensively captures various factors affecting accident risk, enhances the accuracy of the prediction results, and processes different types of data through dual paths, thereby improving the model calculation efficiency and helping to timely discover potential safety hazards at the construction site.

[0069] (3) In view of the fact that in the process of optimizing the existing grid-based construction safety management, it is difficult to accurately predict and intervene in the potential risks in different areas of the construction site during the existing construction site management process. The main safety hazards at the construction site come from environmental changes and construction site accidents, which leads to the technical problem that traditional safety management lacks regionalization, precise prediction and real-time adjustment. This solution creatively adopts two steps, environmental risk prediction and accident risk prediction, to assist in the optimization of grid-based construction safety management. Through the prediction of environmental risk and accident risk at two levels, it can realize dynamic safety assessment and targeted inspection of each area under the grid management framework, thereby optimizing resource allocation and inspection efficiency, reducing safety hazards, improving the accuracy, response speed and visualization effect of construction safety management, and ensuring that the safety management of each grid area is more scientific and efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 A flow chart of the artificial intelligence-based construction safety grid management optimization method provided by the present invention;

[0071] Figure 2 A schematic diagram of the artificial intelligence-based construction safety grid management optimization system provided by the present invention;

[0072] Figure 3 Schematic diagram of the process of step S3;

[0073] Figure 4 Schematic diagram of the process of step S4.

[0074] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0075] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0076] In the description of the present invention, it should be understood that terms such as "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0077] Example 1, see Figure 1 The present invention provides an artificial intelligence-based construction safety grid management optimization method, which includes the following steps:

[0078] Step S1: grid area division;

[0079] Step S2: data preprocessing;

[0080] Step S3: Environmental risk prediction;

[0081] Step S4: accident risk prediction;

[0082] Step S5: Optimize construction safety management.

[0083] Example 2, see Figure 1 ,This embodiment is based on the above embodiment.,In step S1, the grid area division is used to divide the grid area and perform data collection, specifically, dividing the construction site into N grid areas according to the distribution of the work area, and collecting historical safety accident data of the construction site, and collecting environmental data and video monitoring data in the grid area through sensors;

[0084] The historical safety accident data, including accident date, location, category and level;

[0085] The environmental data, including temperature, humidity, light, noise and dust concentration

[0086] The grid area includes construction work area, storage area, passage area, personnel concentrated activity area and fire safety area.

[0087] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, the data preprocessing is used to optimize the data quality, including the following steps:

[0088] Step S21: pre-processing the historical safety accident data, specifically, performing data cleaning and normalization on the historical safety accident data, and label encoding the type data in the historical safety accident data to obtain historical safety accident optimized data;

[0089] Step S22: Environmental data preprocessing, specifically, obtaining environmental optimization data by performing data cleaning, time alignment, and normalization on the environmental data;

[0090] Step S23: Video surveillance data preprocessing, specifically extracting image frames from the video surveillance data at a rate of one frame per second, denoising and standardizing each frame of image to obtain surveillance standard image data, and dividing all images in the surveillance standard image data into G fixed-size image blocks to obtain surveillance image block data.

[0091] Example 4, see Figure 1 and Figure 3 , this embodiment is based on the above embodiment. In step S3, the environmental risk prediction is used to predict environmental risks. Specifically, based on the environmental optimization data, a self-attention multi-scale convolutional long short-term memory model is used to perform environmental risk prediction to obtain environmental risk prediction information;

[0092] The self-attention multi-scale convolutional long short-term memory model includes an input layer, a multi-scale convolutional subnet, a random inactivation layer, a long short-term memory layer, a self-attention layer and an output layer;

[0093] The multi-scale convolutional subnet is used to capture multi-scale features from the data;

[0094] The long short-term memory layer is used to capture long-term and short-term dependencies from data;

[0095] The self-attention layer is used to focus on key time steps;

[0096] The environmental risk prediction includes the following steps:

[0097] Step S31: Constructing a self-attention multi-scale convolutional long short-term memory model, including the following steps:

[0098] Step S311: constructing an input layer for receiving input data, specifically receiving environmental optimization data as input data of the model through the input layer;

[0099] Step S312: Construct a multi-scale convolutional subnetwork, specifically, setting a parallel first convolutional layer, a second convolutional layer, and a third convolutional layer in the multi-scale convolutional subnetwork, and setting an average pooling layer after each convolutional layer. The output results of each average pooling layer are spliced ​​through a feature splicing layer to obtain multi-scale features of the environment. The first convolutional layer contains 16 4×1 convolution kernels, the second convolutional layer contains 32 4×5 convolution kernels, and the third convolutional layer contains 64 4×5 convolution kernels.

[0100] Step S313: construct a random dropout layer to avoid model overfitting. Specifically, the random dropout layer is used to process the multi-scale features of the environment to obtain the random dropout features of the environment.

[0101] Step S314: constructing a long-short-term memory layer, specifically setting 295 long-short-term memory units in the long-short-term memory layer, processing the random inactivation characteristics of the environment through the long-short-term memory layer to obtain the long-short-term time series characteristics of the environment;

[0102] The long short-term memory unit includes a forget gate, an input gate, an output gate and a memory unit;

[0103] Step S315: Construct a self-attention layer. Specifically, in the self-attention layer, the self-attention mechanism is used to process the long-term and short-term temporal features of the environment to obtain the self-attention features of the environment. The calculation formula is:

[0104] ;

[0105] Where Q is the query matrix, h is the long-term and short-term time series features of the environment, the long-term and short-term time series features are a matrix of u×d, u is the total number of time steps of the long-term and short-term time series features of the environment, d is the dimension of the long-term and short-term time series features of the environment, and W Q is the d×d query weight matrix, K is the key matrix, W K is a d×d key weight matrix, V is a value matrix, W V is a d×d value weight matrix, Att is the self-attention score, T is the transpose operation, and h att is the environmental self-attention feature;

[0106] Step S316: construct an output layer for outputting the model prediction results, specifically processing the environment self-attention features through a fully connected layer and a linear activation function to obtain the model prediction results;

[0107] Step S32: constructing an environmental risk prediction model, specifically, training a self-attention multi-scale convolutional long short-term memory model based on the environmental optimization data, and using a particle swarm optimization algorithm to tune the model parameters to obtain an environmental risk prediction model;

[0108] Step S33: Generate environmental risk prediction information, specifically, perform environmental risk prediction on each grid area using the environmental risk prediction model to obtain environmental risk prediction information, wherein the environmental risk prediction information includes environmental risk values ​​and main environmental risk factors;

[0109] By performing the above operations, in the existing environmental risk prediction process, there are data changes such as temperature, humidity, dust concentration in the environment often have trends on different time scales. The traditional environmental prediction model only considers single-scale features, ignores the information complementarity between different scales, and cannot effectively capture multi-scale environmental changes, resulting in poor prediction accuracy. Technical problems, this solution creatively uses the self-attention multi-scale convolutional long short-term memory model for environmental risk prediction, comprehensively captures different scale information in environmental-related data, effectively handles long-term dependencies, and enhances attention to critical moments, thereby accurately capturing important risk signals, improving prediction accuracy and sensitivity, and helping to improve the overall effectiveness and reliability of the method.

[0110] Example 5, see Figure 1 and Figure 4This embodiment is based on the above embodiment. In step S4, the accident risk prediction is used to predict the accident risk. Specifically, based on the environmental risk prediction information, the historical safety accident optimization data and the monitoring image block data, a dual-path multimodal deep neural network model is used to perform accident risk prediction to obtain accident risk prediction information.

[0111] The dual-path multimodal deep neural network model includes a dual-path feature processing subnet, a multimodal feature fusion layer and a prediction output layer;

[0112] The dual-path feature processing subnet is used to process feature information of multimodal data in parallel;

[0113] The multimodal feature fusion layer is used to fuse multimodal feature information;

[0114] The accident risk prediction includes the following steps:

[0115] Step S41: constructing a dual-path feature processing subnet, specifically setting a parallel first feature processing path and a second feature processing path in the dual-path feature processing subnet, including the following steps:

[0116] Step S411: constructing a first feature processing path, specifically setting a random forest model in the first feature processing path, processing environmental risk prediction information and historical safety accident optimization data through the first feature processing path to obtain environmental accident characteristics;

[0117] Step S412: Constructing a second feature processing path, specifically setting a flattened embedding layer, an image visual transformer model, a long short-term memory layer, and an average pooling layer in the second feature processing path, processing the monitoring image block data through the second feature processing path to obtain monitoring image features, including the following steps:

[0118] Step S4121: construct a flattened embedding layer, specifically flattening each monitoring image block in the monitoring image block data into a vector, and performing linear mapping on the vector to obtain the monitoring image block mapping feature. The calculation formula is:

[0119] ;

[0120] Where z i is the mapping feature of the i-th monitoring image block, i is the monitoring image block index, W p is the linear mapping weight, x i is the flattened vector of the i-th monitoring image block, b p is the linear mapping bias term;

[0121] Step S4122: Constructing a visual transformer model. Specifically, setting a 24-layer encoder in the visual transformer model and setting 16 self-attention heads in each layer. The visual transformer model is used to process the monitoring image block mapping features to obtain the monitoring image global features.

[0122] Step S4123: constructing a long short-term memory layer, setting 295 long short-term memory units in the long short-term memory layer, and processing the global features of the monitoring image through the long short-term memory layer to obtain the monitoring image time series features;

[0123] Step S4124: construct an average pooling layer for aggregating features in the time dimension. Specifically, the monitoring image temporal features are processed by the average pooling layer to obtain the monitoring average pooling features. The calculation formula is:

[0124] ;

[0125] Where, f G is the monitoring average pooling feature, t max is the maximum time step, t is the time step index, c t is the temporal feature of the monitoring image at the tth time step;

[0126] Step S42: Construct a multimodal feature fusion layer, specifically by weighted fusion of environmental accident features and monitoring average pooling features to obtain accident risk features. The calculation formula is:

[0127] ;

[0128] Where, f total is the accident risk characteristic, is the first weighting coefficient, which is used to represent the weight of environmental accident characteristics, f env It is a characteristic of environmental accidents. is the second weighting coefficient, which is used to represent the monitoring average pooling feature weight;

[0129] Step S43: Constructing a prediction output layer, specifically processing the accident risk features through a fully connected layer and a softmax activation function to obtain the final prediction result;

[0130] Step S44: constructing a dual-path multimodal deep neural network model, specifically by constructing a dual-path multimodal deep neural network model through the construction of the dual-path feature processing subnet, the construction of the multimodal feature fusion layer, and the construction of the prediction output layer;

[0131] Step S45: Constructing an accident risk prediction model, specifically, training a dual-path multimodal deep neural network model based on environmental risk prediction information, historical safety accident optimization data, and monitoring image block data, and using a particle swarm optimization algorithm to tune model parameters to obtain an accident risk prediction model;

[0132] Step S46: Generate accident risk prediction information, specifically, perform accident risk prediction using an accident risk prediction model to obtain accident risk prediction information, wherein the accident risk prediction information includes an accident risk level and major accident risk factors;

[0133] By performing the above operations, in the existing accident risk prediction process, accident risks at construction sites need to be analyzed by comprehensively considering environmental changes, historical safety accidents and real-time monitoring. Traditional accident risk prediction usually relies on a single data source and cannot fully consider information from multiple dimensions, resulting in a technical problem of reduced prediction accuracy. This solution creatively adopts a dual-path multimodal deep neural network model for accident risk prediction, effectively processes multimodal data related to accident risks, comprehensively captures various factors affecting accident risks, enhances the accuracy of prediction results, and processes different types of data through dual paths, thereby improving the model calculation efficiency and helping to timely discover potential safety hazards at construction sites.

[0134] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the construction safety management optimization is used to optimize the grid management of construction safety. Specifically, based on the accident risk prediction information and the environmental risk prediction information, the construction safety comprehensive management optimization is performed to obtain construction safety management prediction auxiliary information. Furthermore, based on the construction safety management prediction auxiliary information, grid-based construction site safety inspections, data aggregation, and visual comprehensive analysis are performed to generate a construction safety management optimization report to help construction safety managers perform safety management and decision-making within the grid area.

[0135] By performing the above operations, in the process of optimizing the existing grid management of construction safety, it is difficult to accurately predict and intervene in the potential risks in different areas of the construction site during the existing construction site management process. The main safety hazards at the construction site come from environmental changes and construction site accidents, which leads to the technical problem that traditional safety management lacks regionalization, precise prediction and real-time adjustment. This solution creatively adopts two steps of environmental risk prediction and accident risk prediction to assist in the optimization of grid construction safety management. Through the prediction of environmental risk and accident risk at two levels, it can realize dynamic safety assessment and targeted inspection of each area under the grid management framework, thereby optimizing resource allocation and inspection efficiency, reducing safety hazards, improving the accuracy, response speed and visualization of construction safety management, and ensuring that the safety management of each grid area is more scientific and efficient.

[0136] Example 7, see Figure 2 This embodiment is based on the above embodiment. The artificial intelligence-based construction safety grid management optimization system provided by the present invention includes: a grid area division module, a data preprocessing module, an environmental risk prediction module, an accident risk prediction module and a construction safety management optimization module;

[0137] The grid area division module is used for grid area division, obtaining historical safety accident data, environmental data and video surveillance data through grid area division, and sending the historical safety accident data, the environmental data and the video surveillance data to the data preprocessing module;

[0138] The data preprocessing module is used for data preprocessing, and obtains historical safety accident optimization data, environmental optimization data and monitoring image block data through data preprocessing, and sends the environmental optimization data to the environmental risk prediction module, and sends the historical safety accident optimization data and the monitoring image block data to the accident risk prediction module;

[0139] The environmental risk prediction module is used for environmental risk prediction, obtains environmental risk prediction information through environmental risk prediction, and sends the environmental risk prediction information to the accident risk prediction module and the construction safety management optimization module;

[0140] The accident risk prediction module is used for accident risk prediction, obtains accident risk prediction information through accident risk prediction, and sends the accident risk prediction information to the construction safety management optimization module;

[0141] The construction safety management optimization module is used for construction safety management optimization, and obtains construction safety management prediction auxiliary information and a construction safety management optimization report through construction safety management optimization.

[0142] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0143] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0144] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based construction safety grid management optimization method, characterized by: The method comprises the following steps: Step S1: Grid area division, used to divide the grid area and collect data. Specifically, the construction site is divided into N grid areas according to the distribution of work areas, and historical safety accident data of the construction site is collected. Environmental data and video surveillance data in the grid areas are collected through sensors; Step S2: Data preprocessing, including preprocessing historical safety accident data, environmental data, and video surveillance data respectively to obtain historical safety accident optimization data, environmental optimization data, and monitoring image block data; Step S3: Environmental risk prediction, which is used to predict environmental risks. Specifically, based on the environmental optimization data, a self-attention multi-scale convolutional long short-term memory model is used to predict environmental risks and obtain environmental risk prediction information; The self-attention multi-scale convolutional long short-term memory model includes an input layer, a multi-scale convolutional subnet, a random inactivation layer, a long short-term memory layer, a self-attention layer and an output layer; The multi-scale convolutional subnet is used to capture multi-scale features from the data; The long short-term memory layer is used to capture long-term and short-term dependencies from data; The self-attention layer is used to focus on key time steps; Step S4: Accident risk prediction, which is used to predict accident risks. Specifically, based on environmental risk prediction information, historical safety accident optimization data, and monitoring image block data, a dual-path multimodal deep neural network model is used to perform accident risk prediction and obtain accident risk prediction information. The dual-path multimodal deep neural network model includes a dual-path feature processing subnet, a multimodal feature fusion layer and a prediction output layer; The dual-path feature processing subnet is used to process feature information of multimodal data in parallel; The multimodal feature fusion layer is used to fuse multimodal feature information; In step S4, the accident risk prediction includes the following steps: Step S41: constructing a dual-path feature processing subnet, specifically setting a parallel first feature processing path and a second feature processing path in the dual-path feature processing subnet, including the following steps: Step S411: constructing a first feature processing path, specifically setting a random forest model in the first feature processing path, processing environmental risk prediction information and historical safety accident optimization data through the first feature processing path to obtain environmental accident characteristics; Step S412: Constructing a second feature processing path, specifically setting a flattened embedding layer, an image visual transformer model, a long short-term memory layer, and an average pooling layer in the second feature processing path, processing the monitoring image block data through the second feature processing path to obtain monitoring average pooling features, including the following steps: Step S4121: Construct a flattened embedding layer, specifically flattening each monitoring image block in the monitoring image block data into a vector, and performing linear mapping on the vector to obtain the monitoring image block mapping feature. The calculation formula is: ； Where z i is the mapping feature of the i-th monitoring image block, i is the monitoring image block index, W p is the linear mapping weight, x i is the flattened vector of the i-th monitoring image block, b p is the linear mapping bias term; Step S4122: Constructing a visual transformer model. Specifically, setting a 24-layer encoder in the visual transformer model and setting 16 self-attention heads in each layer. The visual transformer model is used to process the monitoring image block mapping features to obtain the monitoring image global features. Step S4123: constructing a long short-term memory layer, setting 295 long short-term memory units in the long short-term memory layer, and processing the global features of the monitoring image through the long short-term memory layer to obtain the monitoring image time series features; Step S4124: construct an average pooling layer for aggregating features in the time dimension. Specifically, the monitoring image temporal features are processed by the average pooling layer to obtain the monitoring average pooling features. The calculation formula is: ； Where, f G is the monitoring average pooling feature, t max is the maximum time step, t is the time step index, c t is the temporal feature of the monitoring image at the tth time step; Step S42: Construct a multimodal feature fusion layer, specifically by weighted fusion of environmental accident features and monitoring average pooling features to obtain accident risk features. The calculation formula is: ； Where, f total is the accident risk characteristic, is the first weighting coefficient, which is used to represent the weight of environmental accident characteristics, f env It is a characteristic of environmental accidents. is the second weighting coefficient, which is used to represent the monitoring average pooling feature weight; Step S43: Constructing a prediction output layer, specifically processing the accident risk features through a fully connected layer and a softmax activation function to obtain the final prediction result; Step S44: constructing a dual-path multimodal deep neural network model, specifically by constructing a dual-path multimodal deep neural network model through the construction of the dual-path feature processing subnet, the construction of the multimodal feature fusion layer, and the construction of the prediction output layer; Step S45: Constructing an accident risk prediction model, specifically, training a dual-path multimodal deep neural network model based on environmental risk prediction information, historical safety accident optimization data, and monitoring image block data, and using a particle swarm optimization algorithm to tune model parameters to obtain an accident risk prediction model; Step S46: generating accident risk prediction information, specifically performing accident risk prediction using an accident risk prediction model to obtain accident risk prediction information; Step S5: Construction safety management optimization, which is used to optimize the grid management of construction safety, specifically to optimize the comprehensive management of construction safety based on the accident risk prediction information and the environmental risk prediction information, and obtain construction safety management prediction auxiliary information.

2. The artificial intelligence-based construction safety grid management optimization method according to claim 1 is characterized by: In step S3, the environmental risk prediction includes the following steps: Step S31: Constructing a self-attention multi-scale convolutional long short-term memory model, including the following steps: Step S311: constructing an input layer for receiving input data, specifically receiving environmental optimization data as input data of the model through the input layer; Step S312: Construct a multi-scale convolutional subnetwork, specifically, setting a parallel first convolutional layer, a second convolutional layer, and a third convolutional layer in the multi-scale convolutional subnetwork, and setting an average pooling layer after each convolutional layer. The output results of each average pooling layer are spliced ​​through a feature splicing layer to obtain multi-scale features of the environment. The first convolutional layer contains 16 4×1 convolution kernels, the second convolutional layer contains 32 4×5 convolution kernels, and the third convolutional layer contains 64 4×5 convolution kernels. Step S313: construct a random dropout layer to avoid model overfitting. Specifically, the random dropout layer is used to process the multi-scale features of the environment to obtain the random dropout features of the environment. Step S314: constructing a long-short-term memory layer, specifically setting 295 long-short-term memory units in the long-short-term memory layer, processing the random inactivation characteristics of the environment through the long-short-term memory layer to obtain the long-short-term time series characteristics of the environment; The long short-term memory unit includes a forget gate, an input gate, an output gate and a memory unit; Step S315: Construct a self-attention layer. Specifically, in the self-attention layer, the self-attention mechanism is used to process the long-term and short-term temporal features of the environment to obtain the self-attention features of the environment. The calculation formula is: ； Where Q is the query matrix, h is the long-term and short-term time series features of the environment, the long-term and short-term time series features are a matrix of u×d, u is the total number of time steps of the long-term and short-term time series features of the environment, d is the dimension of the long-term and short-term time series features of the environment, and W Q is the d×d query weight matrix, K is the key matrix, W K is a d×d key weight matrix, V is a value matrix, W V is a d×d value weight matrix, Att is the self-attention score, T is the transpose operation, and h att is the environmental self-attention feature; Step S316: construct an output layer for outputting the model prediction results, specifically processing the environment self-attention features through a fully connected layer and a linear activation function to obtain the model prediction results; Step S32: constructing an environmental risk prediction model, specifically, training a self-attention multi-scale convolutional long short-term memory model based on the environmental optimization data, and using a particle swarm optimization algorithm to tune the model parameters to obtain an environmental risk prediction model; Step S33: Generate environmental risk prediction information, specifically, perform environmental risk prediction on each grid area through the environmental risk prediction model to obtain environmental risk prediction information.

3. The artificial intelligence-based construction safety grid management optimization method according to claim 2 is characterized by: In step S2, the data preprocessing is used to optimize the data quality, including the following steps: Step S21: pre-processing the historical safety accident data, specifically, performing data cleaning and normalization on the historical safety accident data, and label encoding the type data in the historical safety accident data to obtain historical safety accident optimized data; Step S22: Environmental data preprocessing, specifically, obtaining environmental optimization data by performing data cleaning, time alignment, and normalization on the environmental data; Step S23: Video surveillance data preprocessing, specifically extracting image frames from the video surveillance data at a rate of one frame per second, denoising and standardizing each frame of image to obtain surveillance standard image data, and dividing all images in the surveillance standard image data into G fixed-size image blocks to obtain surveillance image block data.

4. The artificial intelligence-based construction safety grid management optimization method according to claim 3 is characterized by: In step S5, the construction safety management optimization also includes conducting grid-based construction site safety inspections, data aggregation and visual comprehensive analysis based on the construction safety management prediction auxiliary information, and generating a construction safety management optimization report to help construction safety managers conduct safety management and decision-making within the grid area.

5. An artificial intelligence-based construction safety grid management optimization system, used to implement the artificial intelligence-based construction safety grid management optimization method according to any one of claims 1 to 4, characterized in that: It includes grid area division module, data preprocessing module, environmental risk prediction module, accident risk prediction module and construction safety management optimization module; The grid area division module is used for grid area division, obtaining historical safety accident data, environmental data and video surveillance data through grid area division, and sending the historical safety accident data, the environmental data and the video surveillance data to the data preprocessing module; The data preprocessing module is used for data preprocessing, and obtains historical safety accident optimization data, environmental optimization data and monitoring image block data through data preprocessing, and sends the environmental optimization data to the environmental risk prediction module, and sends the historical safety accident optimization data and the monitoring image block data to the accident risk prediction module; The environmental risk prediction module is used for environmental risk prediction, obtains environmental risk prediction information through environmental risk prediction, and sends the environmental risk prediction information to the accident risk prediction module and the construction safety management optimization module; The accident risk prediction module is used for accident risk prediction, obtains accident risk prediction information through accident risk prediction, and sends the accident risk prediction information to the construction safety management optimization module; The construction safety management optimization module is used for construction safety management optimization, and obtains construction safety management prediction auxiliary information and a construction safety management optimization report through construction safety management optimization.

Images