Intelligent construction site intelligent inspection method, device and system based on space-time diagram convolutional network, and storage medium

Abstract

The invention relates to the technical field of intelligent construction sites, in particular to an intelligent inspection method, device and system for an intelligent construction site based on a space-time diagram convolutional network, and a storage medium, and the method comprises the steps: installing video collection equipment at a construction site, and carrying out the image data collection of construction activities, equipment and stockpile of each region of the construction site; modeling the constructor behavior prediction model, and predicting the flow trend and density of constructors in each region by integrating the construction site construction layout and construction information; processing the acquired image information of each construction area by using an image recognition technology, and recognizing the type, density and related distribution position of equipment; the construction site environment information is fused, the estimated inspection time of each construction area is evaluated, and the relation between the construction equipment density and the inspection time is analyzed; and determining the guarantee range of the spare part maintenance points according to the analysis result of the construction equipment, and adjusting and optimizing the number distribution of the spare part maintenance points according to the utilization efficiency of the existing spare part maintenance points, the flow of construction personnel and the area of the construction region. A reinforcement learning technology is used to optimize inspection roads of inspection personnel, and based on construction personnel flow trend prediction and construction equipment quantity analysis, routing inspection is arranged preferentially in a route of high-utilization-rate equipment and a time period with less construction operation, so that fault hidden dangers are found in time, and the influence on normal construction operation is reduced; and the execution condition of the inspection task is monitored in real time, and the inspection efficiency and the fault risk early warning level are evaluated in combination with the condition fed back by the inspection personnel.

Description

Technical Field

[0001] The present invention relates to the technical field of smart construction sites, and in particular, to a smart construction site intelligent patrol method, device, system and storage medium based on a spatio-temporal graph convolutional network. Background Art

[0002] With the technological development of engineering construction projects, precise design and construction simulation of engineering projects are carried out through a three-dimensional design platform. Around the construction process management, a digital system for construction projects with interconnected collaboration, intelligent production, and scientific management is established. And data mining and analysis are performed on the environmental and equipment engineering information data collected in real time by the three-dimensional visual construction site modeling BIM model and the Internet of Things (IoT) sensor network. Combined with computer vision processing such as safety helmet / reflection vest recognition and behavior analysis, process trend prediction and expert preplans are provided to achieve visual intelligent management of engineering construction, so as to improve the informatization level of engineering management and meet the requirements of higher safety and construction efficiency of smart construction sites.

[0003] However, in the management of large-scale construction sites, the construction area is wide, construction operations are carried out in parallel in each construction area, construction personnel and construction machinery are scheduled in each construction area, the layout scenario of the construction site will change, and the positions of construction machinery and equipment will also be dynamically adjusted. The existing smart construction site patrol methods mainly rely on video monitoring and manual patrols, or use traditional image processing technologies. Manual recording has low efficiency and strong experience dependence, and cannot effectively process spatio-temporal data, has poor adaptability to dynamic scenarios, and insufficient multi-source data fusion. Video analysis and sensor data are processed in isolation, lacking spatio-temporal correlation modeling. The existing graph convolutional networks use fixed adjacency matrices and are difficult to adapt to the dynamic topological changes of smart construction site scenarios. There is a lack of in-depth understanding of complex construction site environments and intelligent decision-making support, and the generated patrol plans are difficult to effectively adapt to the actual situations of smart construction sites in different scenarios.

[0004] In order to solve the above problems, the present invention proposes a smart construction site intelligent patrol method, device, system and storage medium based on a spatio-temporal graph convolutional network. It realizes that through the optimization of the intelligent patrol route, patrol personnel can complete patrol tasks more efficiently; by dynamically adapting to the changes in the layout scenarios of construction site equipment, the effectiveness of the patrol plan is improved. Summary of the Invention

[0005] The present invention provides a smart construction site intelligent patrol method, device, system and storage medium based on a spatio-temporal graph convolutional network.

[0006] A smart construction site intelligent patrol method, device, system and storage medium based on a spatio-temporal graph convolutional network. The smart construction site intelligent patrol method based on a spatio-temporal graph convolutional network includes the following steps:

[0007] S01. Install video acquisition devices at the construction site to collect image data of construction activities, equipment, and stockpiles in each area of the construction site.

[0008] S03. Process the collected image information of each construction area using image recognition technology to identify equipment types, densities, and relevant distribution locations.

[0009] S03. Process the collected image information of each construction area using image recognition technology to identify equipment types, densities, and relevant distribution locations.

[0010] S04. Integrate the environmental information of the construction site, evaluate the estimated inspection time for each construction area, and analyze the relationship between construction equipment density and inspection time.

[0011] S05. Determine the guarantee scope of spare parts maintenance points based on the analysis results of construction equipment. Adjust and optimize the quantity distribution of spare parts maintenance points according to the utilization efficiency of existing spare parts maintenance points, the flow of construction personnel, and the area of construction areas.

[0012] S06. Use reinforcement learning technology to optimize the inspection routes of inspection personnel. Based on the prediction of the flow trend of construction personnel and the analysis of the quantity of construction equipment, arrange inspections preferentially on the routes of highly utilized equipment and during periods with less construction operations, so as to detect potential faults in a timely manner and reduce the impact on normal construction operations.

[0013] S07. Monitor the execution situation of inspection tasks in real time, and evaluate the inspection efficiency and the early warning level of fault risks in combination with the feedback from inspection personnel.

[0014] Optionally, in step S01, use video acquisition devices to obtain video data of each construction area and each construction road in the construction site. Adopt video analysis technology to extract key frames from the collected continuous video stream. The key frames are the frames of the activity states of construction personnel in the construction areas and on the construction roads. Determine the extraction frequency of key frames in each video according to the activity changes of construction personnel, so that the extracted key frames can reflect the construction states at different time periods. Screen the extracted key frames to eliminate the influence of environmental noise in the key frame image data, and perform normalization processing on the key frame images to reduce unnecessary variable interference and improve the accuracy of feature extraction.

[0015] Optionally, in step S02, based on the overall layout of the construction site, describe the smart construction site with a graph structure: G=(V, E); A typical graph structure has vertex attributes and edge attributes. Among them, the vertex V represents each construction area in the construction site, and the edge E represents each construction road. Further, for the constructed smart construction site graph structure G=(V, E), its state description at different times t is: G t = (V t , E t ); Among them, V t represents all vertices at time t, that is, the set of construction areas; E t represents all edges at time t, that is, the set of construction roads. The changes in the attributes of each vertex and edge within the time series can form a spatio-temporal graph sequence input, forming a spatio-temporal graph structure. Vertex v i ∈V represents a construction area, and the vertex attributes include the number of construction workers, construction area characteristics, and environmental information. In the spatio-temporal graph structure, the attributes of the vertex are represented as a feature vector Among them, the number of construction workers is represented by the feature vector to describe the number of construction workers within a certain period; the construction area characteristics are represented by the feature vector including the area A i , equipment type, and equipment quantity; the environmental information is the weather condition of the construction area, represented by an environmental feature vector . Edge e ij ∈E represents the construction road between construction areas v i and v j , and the edge attributes include road length and busyness. In the spatio-temporal graph structure, the road length is the passing distance between construction areas; the busyness describes the flow of construction workers on the construction road. The Spatio-Temporal Graph Convolutional Network ST-GCN model is used to predict the construction worker flow trend in each construction area, the number of construction workers in each construction area, and the busyness of the construction road in the future period of time on the construction site. The time convolutional layer uses one-dimensional convolutional operations to convolve the time series of the vertex features of the graph structure to capture the time dynamic changes. The spatial convolutional layer extracts spatial features from the vertices and adjacent vertices of the graph structure through graph convolutional operations. Through the stacking of multiple spatio-temporal convolutional layers, the spatio-temporal graph structure convolutional network model of the intelligent construction site can capture the complex spatio-temporal dependence relationships between vertices and edges. The last layer outputs the predicted values of the vertices and edges of each graph structure. The loss function of the model is optimized through backpropagation, and the model parameters are gradually adjusted.

[0016] Optionally, in step S03, the monitoring videos of each construction area on the construction site are obtained from the video acquisition device in step S01. For each construction area, key frame images are extracted at regular intervals for subsequent object detection and equipment classification. A semantic segmentation model U-Net based on a convolutional neural network is used to detect and classify the equipment in the images. The semantic segmentation model classifies the pixel points in the input image and segments them into pixel regions of multiple categories, each category represented by a shape with a different identifier, thereby generating a category segmentation map. Each type of equipment is marked with an identifier of a different shape. The semantic segmentation model is trained using a pre-annotated intelligent construction site dataset, using cross-entropy loss as the loss function for semantic segmentation optimization, and fine-tuned for the construction scenario to improve its classification effect on specific equipment types in the intelligent construction site. By calculating the number of each identifier and its shape area, the equipment density in each construction area can be calculated. Based on the results of semantic segmentation, the area of each type of equipment is statistically analyzed, and the equipment distribution density in each area can be calculated and a heat map can be generated to visualize the distribution density of various types of equipment in different construction areas.

[0017] Optionally, in step S04, according to the construction equipment density and construction area attributes, the inspection time for each construction area is estimated, and a linear regression model is used to establish the relationship between the construction equipment density and the inspection time. The part classified as construction equipment is extracted using the semantic segmentation model U-Net based on a convolutional neural network in step S03, and the extracted construction equipment is further analyzed to obtain its density and distribution characteristics. The spatial distribution of the key frame images of each construction area is analyzed, and the density and distribution of construction equipment in different construction areas are calculated. According to the distribution of construction equipment at different distances from the spare parts maintenance points, the change of the construction equipment density with the distance from the spare parts maintenance points is calculated. According to the used inventory and remaining inventory of the spare parts maintenance points, the impact of the usage of the spare parts maintenance points on the construction equipment density is analyzed. The relationship between the number of construction workers in the construction area and the construction equipment density is analyzed.

[0018] Optionally, in step S05, using the above analysis results, the protection range δ of each spare parts maintenance point is calculated j . Calculate the utilization efficiency of the existing spare parts maintenance points, and calculate the number distribution of spare parts maintenance points with respect to the number of construction workers and the construction area. For areas with high utilization efficiency and insufficient protection range, new spare parts maintenance points are added or the inventory of existing spare parts maintenance points is expanded according to the calculated demand; for spare parts maintenance points with low utilization efficiency, consider relocating or removing them and redistributing them to other more efficient construction areas.

[0019] Optionally, in step S06, a graph structure combined with reinforcement learning is used to optimize the inspection route, based on the graph structure G=(V,E) constructed in step S02. In the reinforcement learning state rt The following intelligent construction site graph structure, whose vertex attributes include the construction equipment density M(v i ) of the construction area, the estimated inspection time T(v i ) required to inspect the construction area represented by this vertex, and the number of construction workers D(v i ,t) during a certain future time period t. The edge attributes include the construction road passing distance L(e ij ) between two construction areas, and the construction road busyness degree C(e ij ,t) at a certain future time point t. The displacement a t is to move to the next vertex in the graph by selecting a construction road based on the current vertex. Specifically, the inspector is currently at vertex v i , and can choose to go to any adjacent vertex v j . The goal of the reward function is to improve the inspection efficiency and reduce the impact on construction operations. Use a reinforcement learning method based on a graph convolutional network to generate an optimal road planning strategy, with the goal of maximizing the cumulative reward. For areas with a predicted high density of construction workers or busy roads passed through, to reduce the impact on construction operations, give negative rewards; for reducing the inspection length of the inspector, that is, minimizing the inspection length, give positive rewards; for the inspector passing through areas with a high density of construction equipment and a high utilization rate of spare parts maintenance points, to maximize the inspection efficiency, give positive rewards; for the shortest time according to the estimated inspection time of the vertex and the estimated passing time of the edge, give negative rewards.

[0020] Optionally, in step S07, obtain the optimal road planning strategy generated by the reinforcement learning model from step S06. The model is given in the form of a graph structure G=(V,E), where vertices represent construction areas to be inspected and edges represent each construction road. The road planning strategy not only includes the inspection order of each vertex, but also includes information such as the inspection roads between vertices, the estimated inspection time, and the estimated passing time. By parsing the road planning strategy, detailed operation instructions are generated for each inspector, including the inspection order, inspection area, inspection road, and estimated inspection time. Use a mobile terminal device based on Beidou satellite or other positioning technologies to provide real-time navigation for the inspector, guiding them to move in the construction site according to the optimal road. The system can continuously monitor the actual inspection time of the inspector, equipment maintenance conditions and other progress and environmental changes, and dynamically adjust the inspection road and inspection plan according to the actual inspection progress and monitoring data, update the operation instructions of the inspector, optimize the selection of inspection roads and improve the inspection efficiency.

[0021] The described intelligent inspection device for an intelligent construction site based on a spatio-temporal graph convolutional network includes:

[0022] The data acquisition module 101 includes video acquisition devices to collect data on the activities of construction workers in each construction area within the construction site, and obtain real-time behavior of construction workers, construction conditions of the construction site, and equipment status.

[0023] The construction worker behavior prediction module 102 combines the construction layout and information of the construction site to build a construction worker behavior prediction model, and predicts the flow trend and density of construction workers in each construction area.

[0024] The equipment identification module 103 processes the monitored images of each construction area collected through image processing and feature recognition technologies, and identifies and classifies the equipment type, density, and distribution location.

[0025] The equipment analysis module 104 integrates the environmental information of the construction site, evaluates the estimated inspection time for each area, and analyzes the relationship between construction equipment density and inspection time.

[0026] The spare parts maintenance point optimization and adjustment module 105 determines the guarantee scope of the spare parts maintenance points according to the analysis results of construction equipment; adjusts and optimizes the quantity distribution of the spare parts maintenance points according to the utilization efficiency of the existing spare parts maintenance points, the number of construction workers, and the area of the construction area.

[0027] The inspection route optimization module 106 optimizes the inspection routes of the inspection personnel using the reinforcement learning algorithm according to the construction worker behavior prediction and construction equipment density evaluation results, and preferentially arranges inspections on the routes of high-utilization equipment and during periods with less construction operations, to timely discover potential faults and reduce the impact on normal construction operations.

[0028] The inspection efficiency evaluation module 107 monitors the execution of inspection tasks in real time, collects the feedback from the inspection personnel, and comprehensively evaluates the inspection efficiency and the early warning level of fault risks.

[0029] The described intelligent inspection system for a smart construction site based on a spatio-temporal graph convolutional network includes the following hardware and software components:

[0030] The hardware components include but are not limited to a server for storing and processing data, running the backend service of the intelligent inspection system for a smart construction site based on a spatio-temporal graph convolutional network; data acquisition devices for real-time monitoring of the activities of construction workers and equipment status within the construction site; inspection devices including various inspection tools and inspection vehicles for performing inspection tasks; mobile terminals including laptops and smartphones for use by management personnel and inspection personnel to receive tasks and feedback information in real time.

[0031] The software components include, but are not limited to, software for data processing and analysis of the collected data, which performs tasks such as predicting the behavior of construction workers, identifying and analyzing equipment; the client software provided to management personnel and patrol personnel for displaying the progress of patrol tasks, the density of construction workers, and equipment distribution information; and the communication software responsible for data transmission and communication between the components of the system.

[0032] The present invention provides a computer-readable storage medium, which includes a stored computer program. When the computer program runs, it controls the device where the computer-readable storage medium is located to execute the above-mentioned intelligent patrol method for a smart construction site based on a spatio-temporal graph convolutional network.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. Through intelligent road planning and optimization of spare parts maintenance points, the patrol personnel in large construction sites can complete patrol tasks more efficiently, access spare parts and maintenance tools in a timely manner, thereby improving the overall patrol efficiency, promptly repairing equipment failures, effectively preventing construction risks, and ensuring the progress of engineering projects.

[0035] 2. By controlling the patrol work in densely populated areas of construction workers, the interference to construction operations is reduced, and the construction operation efficiency is improved.

[0036] 3. The system can dynamically adjust the patrol plan according to environmental changes and the actual patrol progress to ensure that the patrol work is always in the optimal state. At the same time, the real-time feedback data helps to continuously optimize the system performance.

[0037] 4. Through the prediction of construction equipment density and the behavior of construction workers, the system can reasonably allocate patrol resources to ensure the reasonable layout of spare parts maintenance points in large construction sites and the effective scheduling of patrol personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flow chart of an intelligent patrol method for a smart construction site based on a spatio-temporal graph convolutional network.

[0039] Figure 2 It is a structural diagram of an intelligent patrol device for a smart construction site based on a spatio-temporal graph convolutional network. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] To make the objectives, technical solutions, and advantages of the present application clearer, the following will further describe the embodiments of the present application in detail with reference to the accompanying drawings.

[0041] Please refer to Figure 1 , the present application provides an intelligent patrol method for a smart construction site based on a spatio-temporal graph convolutional network, including the following steps.

[0042] S01. Install video acquisition devices at the construction site to collect image data of construction activities, equipment, and stockpiles in each area of the construction site.

[0043] S03. Model the behavior prediction model of construction workers, and comprehensively consider the construction layout and construction information of the construction site to predict the flow trend and density of construction workers in each area.

[0044] S06. Process the image information of each construction area collected by using image recognition technology to identify the equipment type, density, and relevant distribution locations.

[0045] S04. Integrate the environmental information of the construction site, evaluate the estimated inspection time of each construction area, and analyze the relationship between the construction equipment density and the inspection time.

[0046] S05. Determine the guarantee scope of the spare parts maintenance points according to the analysis results of the construction equipment, and adjust and optimize the quantity distribution of the spare parts maintenance points according to the utilization efficiency of the existing spare parts maintenance points, the number of construction workers, and the area of the construction area.

[0047] S07. Use reinforcement learning technology to optimize the inspection routes of the inspection personnel. Based on the prediction of the flow trend of construction workers and the analysis of the number of construction equipment, arrange inspections preferentially on the routes of high-utilization equipment and during periods with less construction operations, so as to detect potential faults in a timely manner and reduce the impact on normal construction operations.

[0048] S08. Monitor the execution situation of the inspection tasks in real time, and evaluate the inspection efficiency and the early warning level of fault risks in combination with the feedback from the inspection personnel.

[0049] Further, in step S01, use video acquisition devices to obtain video data of each construction area and each construction road in the construction site, and use video analysis technology to extract key frames from the collected continuous video stream. The key frames are the frames of the activities of construction workers in a stationary, moving, and gathering state in the construction area and on the construction road. Determine the extraction frequency of key frames in each video segment according to the activity changes of construction workers such as moving speed, direction, gathering, and dispersing, so that the extracted key frames can reflect the construction status in different time periods. Screen the extracted key frames, filter out incomplete or unclear frames, eliminate the influence caused by environmental noises such as light and atmospheric transparency in the key frame image data, and perform standardization processing on the key frame images to reduce unnecessary variable interference and improve the accuracy of feature extraction.

[0050] Further, in step S02, based on the overall layout of the construction site, abstract the construction site as a graph structure G=(V, E). Among them, the vertex V represents each construction area in the construction site, and the edge E represents each construction road. The intelligent construction site graph structure has vertex attributes and edge attributes. Each vertex v of the graph structure i∈V represents a construction area, whose attributes include the number of construction workers, the characteristics of the construction area, and environmental information. Among them, the number of construction workers indicates the number of construction workers within a certain period. The target detection model YOLO is used to detect construction workers from the key frames of the monitored video extracted in step S01. The target detection model will draw blue dots in the extracted images to identify and count the construction workers. For a certain construction area, that is, vertex v i , the number of construction workers λ i in its key frame is counted by the number of blue dots detected; the construction area characteristic vector ρ i includes the area A i , equipment type and equipment quantity. The equipment types include excavators, loaders, cranes, concrete machines, pile drivers, earthmoving machinery, steel bar processing machines, formwork construction machines, scaffolding and construction platform equipment, and detection and monitoring equipment, etc., which is expressed as ρ i = [A i , equipment type code, equipment quantity]; the environmental information feature vector μ i describes the weather conditions of the construction area, including temperature, humidity, and rainfall, etc., which is expressed as μ i = [temperature, humidity, wind speed, weather type code]. Edge attributes, each edge e ij ∈ E represents the construction road between areas v i and v j . Its attributes include road length and busyness degree. The road length represents the passing distance between construction areas, which can be obtained from geographic information system (GIS) data or directly measured, denoted as D ij ; the busyness degree is calculated based on the number of construction workers on the construction road and combined with the marked scope of the construction road, denoted as C ij .

[0051] The spatio-temporal graph convolutional network (ST-GCN) model is adopted to predict the construction worker flow trend in each construction area, the number of construction workers in each area, and the busyness degree of the construction roads in the construction site in the future for a period of time. The main structure of the ST-GCN model includes: performing convolutional operations on the spatial convolutional layer for the graph structure data to extract the spatial features of vertices and their adjacent vertices; performing convolutional operations on the time series for the time convolutional layer to capture the changes of vertex attributes over time. For the constructed intelligent construction site graph: G = (V, E), its state at different times t is represented by G t = (V t , E t ). Among them, is the set of all vertices at time t, and n is the number of construction areas; refers to a specific vertex at time t, that is, a specific construction area in the construction site. is the set of all edges at time t. The vertex attributes are represented as a feature vector where is the number of construction workers obtained by the object detection model, is the construction area characteristics, including area, equipment type and equipment quantity, is the environmental information of the construction area including weather conditions. Each edge attributes are represented as, where D ij is the length of the construction road, that is, the passing distance. is the degree of busyness, expressed by the following formula: where P ij represents the set of all detection areas on the construction road e ij , A ij is the marked range of the construction road e ij , is the number of construction workers on the construction road e ij at time t, that is, the number of blue dots extracted by the object detection model. The changes of the attributes of each vertex and edge within the time series can form a spatio-temporal graph sequence input: X = {G t , G t-1 , …, G t-T+1}; where T is the set size of the observation time window.

[0052] The spatial convolutional layer extracts spatial features from the vertices and adjacent vertices of the intelligent construction map structure through graph convolution operations. For a vertex v i , its update rule is: where the σ function is an activation function, such as ReLU. is the hidden feature of the vertex v i at the l-th layer, Δ(i) is the set of adjacent vertices of the vertex v i , d i is the degree of the vertex v i , W (l) and π (l) are the trainable parameters of the l-th layer, capturing the relationship between vertices and adjacent vertices in the intelligent construction map structure.

[0053] The temporal convolutional layer performs one-dimensional convolution operations on the time series of the vertex features of the intelligent construction map structure to capture temporal dynamic changes. For the vertex v i , the temporal convolution operation is: Among them, Conv1D is a one-dimensional convolution operation, is the parameter of the temporal convolution kernel, represents vertex v i in the feature sequence from time t to t - K + 1, where K is the size of the temporal convolution kernel.

[0054] By stacking multiple spatio-temporal convolutional layers, the model can capture the complex spatio-temporal dependencies between the vertices and edges of the intelligent construction site map structure. The last layer outputs the predicted values for each vertex and edge. For each vertex v i , the model outputs the characteristics of the number of construction workers in the future time periods t + 1, t + 2, …, t + H: Among them, ψ is the function of the spatio-temporal graph convolutional network ST-GCN model, which takes the input and outputs the characteristics of the number of construction workers at vertex v i at the future time step. G t represents the state of the construction site map at the current time step t, including the vertices (construction areas) and their connection relationships. G t-1 , G t-2 ,..., G t-T+1 represents the historical states of the intelligent construction site map from the current time step t back T - 1 time steps. T is the size of the historical time window, which is used to capture the changes in the intelligent construction site map structure in the past time period. represents the characteristics of the number of construction workers at vertex v i at the future time step t + h. h is the index of the future time step, ranging from 1 to H, and H is the number of predicted time steps. The loss function is optimized through backpropagation to gradually adjust the model parameters.

[0055] Similarly, for each edge e ij , the model outputs the degree of busyness in the future time period: Among them, φ is the function of the spatio-temporal graph convolutional network ST-GCN model, which takes the input and outputs the predicted degree of busyness. represents the degree of busyness of the edge from vertex v i to vertex v j at the future time t + h, where h is the future time step. G t represents the state of the construction site map at the current time step t, including the vertices (construction areas) and their connection relationships. G t-1 , G t-2 ,..., G t-T+1Denotes the historical state of the intelligent construction site map from the current time t back T - 1 time steps. T is the size of the historical time window, used to capture the structural changes of the intelligent construction site map within the past time period.

[0056] The model is trained using the historical data of the intelligent construction site, and the supervised learning method is adopted during the training process. The loss function can take the following form: where L is the loss function used during the model training process, used to measure the difference between the model prediction value and the true value. Denotes vertex v i The actual number of construction workers at future time t + h. Denotes the construction worker number of vertex v i predicted by the model at future time t + h. Denotes from vertex v i to vertex v j The actual busyness degree of the edge at future time t + h. Denotes the busyness degree of the edge predicted by the model from vertex v i to vertex v j at future time t + h. By optimizing this loss function, the optimal model parameters can be learned.

[0057] Furthermore, in step S03, the monitoring videos of each construction area on the construction site are obtained from the video acquisition device in step S01. For each construction area, key frame images are extracted at regular intervals for subsequent object detection and equipment classification. The image quality is enhanced by adjusting the saturation and sharpness, and according to the input size requirements of the convolutional neural network, the images are scaled to the same size and the effective areas with equipment are cropped out. The semantic segmentation model U - Net based on the convolutional neural network is used to detect and classify the equipment in the images. The semantic segmentation model classifies the pixel points in the input image, divides them into pixel regions of multiple categories, and each category is represented by a shape with a different identifier, thus generating a category segmentation map. Each type of equipment is marked with an identifier of a different shape. By calculating the number of each identifier and the pixel area of the shape, the number and shape area of different types of equipment are obtained. For example: construction equipment, such as excavators, loaders, cranes, etc., are represented by one identifier; construction materials piles, such as sand and gravel, steel bars, cement, wood, plaster ponds, etc., which are the materials required for construction equipment, are represented by another identifier. The semantic segmentation model is trained using a pre - annotated intelligent construction site dataset, uses the cross - entropy loss as the loss function for semantic segmentation optimization, and is fine - tuned for the construction scenario to improve its classification effect on specific equipment types in the intelligent construction site. The calculation formula for the equipment distribution density is as follows, where, Denote the distribution density of devices of category x in area i, is the total area of devices of category x in area i, A i is the area of area i. Based on the results of semantic segmentation, the area of each type of device is counted, and the device distribution density of each area can be calculated and a heat map can be generated to visualize the distribution density of various devices in different construction areas.

[0058] Furthermore, in step S04, according to the construction equipment density and the attributes of the construction area, estimate the inspection time of each construction area, and use a linear regression model to establish the relationship between the construction equipment density and the inspection time: where, T i is the estimated inspection time of area i, D i is the construction equipment density of construction area i, represents other attributes of area i and the vertex attributes described in step S02, and α, β, γ are the parameters of the regression model. Use the semantic segmentation model based on convolutional neural network in step S03 to extract the part classified as construction equipment, and further analyze the extracted construction equipment to obtain its density and distribution characteristics. Conduct a spatial distribution analysis on the key frame images of each area, and calculate the density D i and distribution P i of construction equipment in different areas: where, D i represents the density of construction equipment in construction area i; represents the number of construction equipment in construction area i; A i is the area of construction area i. P i represents the distribution characteristics of construction equipment in construction area i. According to the used inventory K j and the remaining inventory of the spare parts maintenance point, analyze the influence of the usage situation of the spare parts maintenance point on the construction equipment density D i , The function f describes the relationship between the usage situation of the spare parts maintenance point and the construction equipment density. Calculate the change situation Z i,j of the construction equipment density with the distance d ij from the spare parts maintenance point j, The function can be obtained by fitting the actual data of the intelligent construction site to analyze the distribution of construction equipment density at different distances from the spare parts maintenance point. Analyze the influence of the number of construction workers on the construction equipment density and the inspection time, and analyze the relationship between the number of construction workers L i in construction area i and the construction equipment density D i , Function describes the influence of the number of construction workers on the density of construction equipment.

[0059] Furthermore, in step S05, calculate the support range of the spare parts maintenance point, and use the analysis result of step S04 to calculate the support range δ of each spare parts maintenance point j j : δ j = Γ{d i,j |D ij ≤ τ}, where τ is the set threshold of construction equipment density, used to determine the effective support range of the spare parts maintenance point. Calculate the utilization efficiency E of the existing spare parts maintenance point j = U j / S j , where U j is the average used inventory of the spare parts maintenance point j over a period of time, and S j is the total inventory of the spare parts maintenance point. The range of E j is 0 ≤ E j ≤ 1. For construction area i, calculate the number distribution of the number of construction workers and the construction area on the spare parts maintenance point. The number distribution M i of the spare parts maintenance point can be calculated based on the following formula: where, is the average number of construction workers in construction area i; A i is the area of construction area i; is the average inventory of the spare parts maintenance point; is the average support range of the spare parts maintenance point. For areas with high utilization efficiency, such as E j > 0.9 and insufficient support range, new spare parts maintenance points are added according to the calculated demand M i , or the inventory of the existing spare parts maintenance point is expanded; for spare parts maintenance points with low utilization efficiency, such as E j < 0.2, consider relocating or removing and reallocating to other more efficient areas.

[0060] Furthermore, in step S06, use the graph structure combined with reinforcement learning to optimize the inspection path. Based on the intelligent construction site graph structure G = (V, E) constructed in step S02, change the vertex attributes and edge attributes of the graph structure as the state r of reinforcement learning t , the vertex attribute is the estimated inspection time T(v i ), representing the time required to inspect the construction area of this vertex, which can be obtained from the linear regression model in step S04; the construction equipment density M(v i), representing the construction equipment density in the vertex construction area, which can be calculated from the densities of different types of equipment detected in step S03; the number of construction workers D(v i ,t), representing the predicted number of construction workers in a future time period t, which can be predicted by the ST-GCN model in step S02. The edge attribute is the construction road length L(e ij ), indicating the passing distance between two construction areas; the construction road busyness C(e ij ,t), representing the predicted busyness of the construction road at a future time point t, which can be predicted by the ST-GCN model in step S02. The displacement a t is to move to the next vertex in the graph along a road based on the current vertex. Specifically, the inspection personnel are currently at vertex v i , and can choose to go to any adjacent vertex v j , corresponding to the edge e ij ∈ E, a t = {v i → v j}, and the objective of the designed reward function is to improve the inspection efficiency and reduce the impact on construction workers. For areas with a predicted high density of construction workers or relatively busy roads passed through, to reduce the impact on construction operations, a negative reward is given; for reducing the inspection length of the inspection personnel, that is, minimizing the inspection length, a positive reward is given; for areas where the inspection personnel pass through areas with a high density of construction equipment and a high utilization rate of spare parts maintenance points, to maximize the inspection efficiency, a positive reward is given; for the predicted inspection time of the vertex and the predicted passing time of the edge, to strive for the shortest time, a negative reward is given. The reward function can be expressed as: Among them, w1, w2, w3, w4 are coefficients for adjusting the weights of different reward items, used to balance the importance of different factors. C(e ij ,t) is the busyness of the edge e ij during the time period t; D(v j ,t) is the construction worker density of the vertex v j during the time period t, M(v j ) is the construction equipment density of the vertex v j , E(v j ) is the utilization rate of the spare parts maintenance point, that is, the ratio of the used inventory to the total inventory; L(e ij ) is the construction road length, T(v j ) is the predicted inspection time of the vertex v j , estimated according to factors such as the construction equipment density and the construction area, P(e ij ) is the edge e ijThe estimated travel time can be estimated based on the length of the construction road and the current level of busyness. Use a reinforcement learning method based on graph convolutional networks to learn the optimal road planning strategy, with the goal of maximizing the cumulative reward: where Γ θ is the indicator function of the maximum value of the parameter θ, representing the policy θ that maximizes the expected cumulative reward. E is used to calculate the expected value of the cumulative reward under the policy θ. is the discount factor, where The discount factor is used to balance the importance of the current reward and future rewards. The closer it is to 1, the greater the impact of future rewards; The closer it is to 0, the greater the impact of the current reward. is the immediate reward at the state r t and displacement a t at time t. θ * (s) is the optimal policy at the state s, which maps the state to the displacement to maximize the cumulative reward. The training process includes initializing the model parameters and using the graph convolutional network GCN to initialize the weights of the neural network; encoding the vertex and edge features in the intelligent construction site map structure into low-dimensional embedding vectors through GCN for graph convolutional operations to capture the dependencies and spatio-temporal features between regions in the intelligent construction site map structure. The inspection personnel execute the displacement according to the current policy and collect data, recording the state transition Sampling state transition samples from the intelligent construction site training dataset for training feedback, which is used to train the neural network to generate policy parameters; based on the policy gradient or value iteration method, updating the parameters of the policy network to optimize the road planning strategy. Repeat the above training process iteratively until the model converges to a stable optimal policy.

[0061] Further, in step S07, the optimal road planning strategy generated by the reinforcement learning model is obtained from step S06. The roads are given in the form of a graph structure, where the vertices represent the areas to be inspected, and the edges represent the construction roads. The road planning strategy not only includes the inspection order of each vertex but also the inspection roads between vertices, the estimated inspection time, and the estimated passing time information. By parsing the road planning strategy, detailed operation instructions are generated for each inspector, including the inspection order, inspection area, inspection road, and estimated inspection time. A mobile terminal device based on Beidou satellite positioning or other positioning technologies is used to provide real-time navigation for the inspectors, guiding them to move in the construction site along the optimal road. The system can continuously monitor the actual inspection time of the inspectors, the equipment maintenance status, and other progress and environmental changes, and dynamically adjust the inspection roads and inspection plans according to the actual inspection progress and monitoring data, update the operation instructions of the inspectors, optimize the selection of inspection roads, and improve the inspection efficiency. For example, if the inspection time of a certain construction area exceeds the estimated time, the system will re-analyze the remaining time and roads and generate new operation instructions. The actual inspection data such as the inspection time of the inspectors and the equipment inspection status will be used as feedback input to update the reinforcement learning model and optimize future road planning and inspection plans.

[0062] Please refer to Figure 2 , this application provides an intelligent inspection device for a smart construction site based on a spatio-temporal graph convolutional network, including:

[0063] The data acquisition module 101, including video acquisition devices, collects data on the construction activities in each area of the construction site to obtain real-time construction personnel behavior, construction site conditions, and equipment status.

[0064] The construction personnel behavior prediction module 102 constructs a construction personnel behavior prediction model by integrating the construction layout and construction information of the construction site to predict the flow trend and density of construction personnel in each area.

[0065] The equipment identification module 103 processes the monitored images of each area collected by using image processing and feature recognition technologies to identify and classify the equipment types, densities, and distribution positions.

[0066] The equipment analysis module 104 integrates the construction site environment information, evaluates the estimated inspection time of each area, and analyzes the relationship between the construction equipment density and the inspection time.

[0067] The spare parts maintenance point optimization and adjustment module 105 determines the guarantee scope of the spare parts maintenance points according to the analysis results of the construction equipment, and adjusts and optimizes the quantity distribution of the spare parts maintenance points according to the utilization efficiency of the existing spare parts maintenance points, the construction personnel flow, and the construction area.

[0068] The patrol route optimization module 106 optimizes the patrol routes of the patrol personnel by using the reinforcement learning algorithm according to the prediction of the construction personnel's behavior and the evaluation result of the construction equipment density. It preferentially arranges patrols on the routes of high-utilization equipment and during periods with less construction operations, discovers potential faults in a timely manner, and reduces the impact on normal construction operations.

[0069] The patrol efficiency evaluation module 107 monitors the execution of the patrol tasks in real time, collects the feedback from the patrol personnel, and comprehensively evaluates the patrol efficiency and the early warning level of fault risks.

[0070] This application provides a smart construction site intelligent patrol system based on a spatio-temporal graph convolutional network. The system includes the following hardware and software components:

[0071] The hardware components include, but are not limited to, a server for storing and processing data, running the backend service of the smart construction site intelligent patrol system based on the spatio-temporal graph convolutional network; data acquisition devices for real-time monitoring of the activities of construction personnel and the status of equipment on the construction site; patrol devices including various patrol tools and patrol vehicles for performing patrol tasks; and mobile terminals including portable computers and smartphones for use by management personnel and patrol personnel to receive tasks and feedback information in real time.

[0072] The software components include, but are not limited to, software for processing and analyzing the collected data, performing tasks such as predicting the behavior of construction personnel, identifying and analyzing equipment; user-end software provided to management personnel and patrol personnel for displaying the progress of patrol tasks, the density of construction personnel, and equipment distribution information; and communication software responsible for data transmission and communication between the components of the system.

[0073] This application provides a computer-readable storage medium. The computer-readable storage medium includes a stored computer program. When the computer program runs, it controls the device where the computer-readable storage medium is located to execute the above-mentioned smart construction site intelligent patrol method based on the spatio-temporal graph convolutional network.

[0074] As mentioned above, 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 foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent inspection method for smart construction sites based on spatio-temporal graph convolutional network, characterized in that, It includes the following steps: S01. Install video acquisition devices at the construction site to collect image data of the construction activities, equipment, and stockpiles in each area of the construction site. S02. Build a prediction model for construction workers' behavior. Integrate the construction layout and construction information of the construction site to predict the flow trend and density of construction workers in each area. S03. Process the collected image information of each construction area using image recognition technology to identify the equipment type, density, and relevant distribution locations. S04. Integrate the environmental information of the construction site, evaluate the estimated patrol time of each construction area, and analyze the relationship between the construction equipment density and the patrol time. S05. Determine the guarantee scope of the spare parts maintenance points according to the analysis results of the construction equipment. Adjust and optimize the quantity distribution of the spare parts maintenance points according to the utilization efficiency of the existing spare parts maintenance points, the flow of construction workers, and the area of the construction area. S06. Use reinforcement learning technology to optimize the patrol routes of the patrol personnel. Based on the prediction of the flow trend of construction workers and the analysis of the quantity of construction equipment, arrange the patrol during the period when the routes of high-utilization equipment and the construction operations are less, so as to detect potential faults in time and reduce the impact on normal construction operations. S07. Monitor the execution situation of the patrol tasks in real time. Combine the situation feedback by the patrol personnel to evaluate the patrol efficiency and the early warning level of the fault risk.

2. The intelligent inspection method for a smart construction site based on a spatio-temporal graph convolutional network according to claim 1, wherein, It includes: In step S02, based on the overall layout of the construction site, a graph structure is used to describe the intelligent construction site G = (V, E); where the vertices V represent the various construction areas in the construction site, and the edges E represent the various construction roads. For the constructed intelligent construction site graph structure G = (V, E), its state at different times t is described as G t =(V t , E t ); where V t represents all vertices at time t, that is, the set of construction areas; E t represents all edges at time t, that is, the set of construction roads. The changes in the attributes of each vertex and edge within the time series can form a spatio-temporal graph sequence input, forming a spatio-temporal graph structure. The vertex v i ∈V represents a construction area, and the vertex attributes include the number of construction workers, the characteristics of the construction area, and environmental information. In the spatio-temporal graph structure, the attribute of the vertex is represented as a feature vector Among them, the number of construction workers is represented by the feature vector , describing the number of construction workers within a certain time period; the characteristics of the construction area are represented by the feature vector , including the area A i , equipment type, and equipment quantity; the environmental information is the weather condition of the construction area, represented by an environmental feature vector . The edge e ij ∈E represents the construction road between the construction areas v i and v j , and the edge attributes include road length and busyness. In the spatio-temporal graph structure, is the set of all edges at time t. The attribute of each edge is represented as Among them, D ij is the construction road length, busyness. The time convolution layer uses one-dimensional convolution operations to convolve the time series of the graph structure vertex features to capture time dynamic changes. The spatial convolution layer extracts spatial features from the vertices and adjacent vertices of the graph structure through graph convolution operations. Through the stacking of multiple spatio-temporal convolution layers, the intelligent construction site spatio-temporal graph structure convolution network model can capture the complex spatio-temporal dependence relationships between vertices and edges. The last layer outputs the predicted values of the vertices and edges of each graph structure. The loss function of the model is optimized through backpropagation, and the model parameters are gradually adjusted.

3. The intelligent inspection method for a smart construction site based on a spatio-temporal graph convolutional network according to claim 1, wherein, It includes: Use the target detection model YOLO to detect construction workers from the key frames of the surveillance video extracted in step S01. The target detection model will draw blue dots in the extracted images to identify and count the construction workers, vertices The number of construction workers is counted by the number of blue dots obtained through detection. The busyness level is calculated based on the number of construction workers on the construction road and in combination with the marked scope of the construction road. For the side busyness level is expressed by the following formula where P ij represents the set of all detection areas on the construction road e ij , A ij is the marked scope of the construction road e ij , is the number of construction workers on the construction road e ij at time t, that is, the number of blue dots extracted by the target detection model.

4. The intelligent inspection method for a smart construction site based on a spatio-temporal graph convolutional network according to claim 1, wherein, It includes: In step S03, obtain the surveillance videos of each construction area at the construction site from the video acquisition devices in step S01. For each construction area, extract key-frame images regularly for subsequent object detection and equipment classification. Enhance the image quality by adjusting the saturation and sharpness, and scale the images to the same size according to the input size requirements of the convolutional neural network, and crop out the effective areas where there are equipment. Adopt the semantic segmentation model U-Net based on the convolutional neural network to detect and classify the equipment in the images. The semantic segmentation model classifies the pixel points in the input image and divides them into pixel regions of multiple categories, and each category is represented by a shape with a different identifier, so as to generate a category segmentation map. Each type of equipment is marked with a different-shaped identifier. The semantic segmentation model is trained using a pre-annotated intelligent construction site dataset, uses the cross-entropy loss as the loss function for semantic segmentation optimization, and is fine-tuned for the construction scenario to improve its classification effect on the specific equipment types in the intelligent construction site. By calculating the quantity of each identifier and its shape area, the equipment density of each construction area can be calculated. Based on the results of semantic segmentation, the area of each type of equipment is statistically analyzed, and the equipment distribution density of each area can be calculated and a heat map is generated to visualize the distribution density of various types of equipment in different construction areas.

5. The intelligent inspection method for a smart construction site based on a spatio-temporal graph convolutional network according to claim 1, characterized in that, It includes: In step S04, according to the construction equipment density and the construction area attributes, estimate the inspection time for each construction area, and use a linear regression model to establish the relationship T between the construction equipment density and the inspection time i = αD i + where T i is the estimated inspection time for area i, D i is the construction equipment density of construction area i, represents the other attributes of area i and the vertex attributes described in step S02, and α, β, γ are the parameters of the regression model. Use the semantic segmentation model based on a convolutional neural network in step S03 to extract the part classified as construction equipment, and further analyze the extracted construction equipment to obtain its density and distribution characteristics. Conduct a spatial distribution analysis on the key frame images of each area, and calculate the density D i and distribution P i of the construction equipment in different areas, where D i represents the density of construction equipment in construction area i; represents the number of construction equipment in construction area i; A i is the area of construction area i. P i represents the construction equipment distribution characteristics in construction area i. Analyze the impact of the used inventory K j and the remaining inventory of the spare parts maintenance point on the construction equipment density D i , The function f describes the relationship between the usage situation of the spare parts maintenance point and the construction equipment density. Calculate the change Z i,j of the construction equipment density with the distance d ij from the spare parts maintenance point j, The function can be obtained by fitting the actual data of the intelligent construction site to analyze the distribution of the construction equipment density at different distances from the spare parts maintenance point. Analyze the influence of the number of construction workers on the construction equipment density and the inspection time, and analyze the relationship between the number of construction workers L i in construction area i and the construction equipment density D i , The function describes the influence of the number of construction workers on the construction equipment density.

6. The intelligent inspection method for a smart construction site based on a spatio-temporal graph convolutional network according to claim 1, wherein, It includes: In step S05, calculate the guarantee scope of the spare parts maintenance points. Using the analysis results of step S04, calculate the guarantee scope δ of each spare parts maintenance point j j , δ j = Γ{d i,j |D ij ≤ τ}. Where τ is the set density threshold of construction equipment, used to determine the effective guarantee scope of the spare parts maintenance points. Calculate the utilization efficiency E j = U j / S j , where U j is the average used inventory of the spare parts maintenance point j over a period of time, and S j is the total inventory of the spare parts maintenance point. The range of E j is 0 ≤ E j ≤ 1. For the construction area i, calculate the number distribution of construction personnel and the area of the construction area for the spare parts maintenance points. The number distribution M i of the spare parts maintenance points can be calculated based on the following formula where is the average number of construction personnel in the construction area i; A i is the area of the construction area i; is the average inventory of the spare parts maintenance points; is the average guarantee scope of the spare parts maintenance points. For areas with high utilization efficiency but insufficient guarantee scope, add new spare parts maintenance points according to the calculated demand M i , or expand the inventory of existing spare parts maintenance points; for spare parts maintenance points with low utilization efficiency, consider relocating or removing them and reallocating them to other more efficient areas.

7. The intelligent inspection method for a smart construction site based on a spatio-temporal graph convolutional network according to claim 1, wherein It includes: In step S06, a graph structure is used in combination with reinforcement learning to optimize the inspection route. Based on the intelligent construction site graph structure G = (V, E) constructed in step S02, the vertex attributes and edge attributes of the graph structure are changed, serving as the state r of the reinforcement learning t . The vertex attributes are the predicted inspection time T(v i ), the construction equipment density M(v i ), and the number of construction workers D(v i , t). The edge attributes are the construction road length L(e ij ), and the construction road busyness level C(e ij , t). The displacement a t is to move to the next vertex in the graph along a road based on the current vertex. The objective of designing the reward function is to improve the inspection efficiency and reduce the impact on construction workers. For areas with a predicted high density of construction workers or relatively busy roads passed through, to reduce the impact on construction operations, a negative reward is given; for reducing the inspection length of inspection personnel, that is, minimizing the inspection length, a positive reward is given; for inspection personnel passing through areas with a high density of construction equipment and a high utilization rate of spare parts maintenance points, to maximize the inspection efficiency, a positive reward is given; for striving for the shortest time according to the predicted inspection time of the vertex and the predicted passing time of the edge, a negative reward is given. The reward function can be expressed as where w1, w2, w3, w4 are coefficients for adjusting the weights of different reward terms, used to balance the importance of different factors. C(e ij , t) is the busyness level of edge e ij during time period t; D(v j , t) is the density of construction workers at vertex v j during time period t, M(v j ) is the construction equipment density at vertex v j , E(v j ) is the utilization rate of the spare parts maintenance point, that is, the ratio of the used inventory to the total inventory; L(e ij ) is the construction road length, T(v j ) is the predicted inspection time of vertex v j , estimated according to factors such as construction equipment density and construction area, P(e ij ) is the predicted passing time of edge e ij , which can be estimated based on the construction road length and the current busyness level. Use a reinforcement learning method based on a graph convolutional network to learn the optimal road planning strategy, with the goal of maximizing the cumulative reward where Γ θ is an indicator function of the maximum value of parameter θ, representing the strategy θ that maximizes the expected cumulative reward. E is used to calculate the expected value of the cumulative reward under strategy θ is the discount factor, where the discount factor is used to balance the importance of current rewards and future rewards,[[]] the closer it is to 1, the greater the impact of future rewards; the closer it is to 0, the greater the impact of current rewards. is the state r at time t t and displacement a t under the immediate reward. θ * (s) is the optimal policy in state s, which maps the state to displacement to maximize the cumulative reward. The training process includes initializing the model parameters and initializing the weights of the neural network using the graph convolutional network GCN; encoding the vertex and edge features in the intelligent construction site graph structure into low-dimensional embedding vectors through GCN for graph convolutional operations to capture the dependencies and spatio-temporal features between regions in the intelligent construction site graph structure. The inspection personnel execute displacements and collect data according to the current policy, and record the state transitions sampling state transition samples from the intelligent construction site training dataset for training feedback, which is used to train the neural network to generate policy parameters; based on the policy gradient or value iteration method, updating the parameters of the policy network to optimize the road planning policy. Repeat the above training process iteratively until the model converges to a stable optimal policy.

8. An intelligent inspection device for a smart construction site based on a spatio-temporal graph convolutional network, which adopts the intelligent inspection method for a smart construction site based on a spatio-temporal graph convolutional network according to any one of claims 1-7, characterized in that, It includes: The data acquisition module 101 includes a video acquisition device to collect data on the activities of construction workers in each construction area within the construction site, and obtain real-time construction worker behaviors, construction site conditions, and equipment status. The construction worker behavior prediction module 102 constructs a construction worker behavior prediction model in combination with the construction layout and information of the construction site to predict the flow trend and density of construction workers in each construction area. The equipment identification module 103 processes the monitored images of each construction area collected through image processing and feature recognition technologies to identify and classify the equipment types, density, and distribution locations. The equipment analysis module 104 integrates the construction site environment information to evaluate the estimated inspection time for each area and analyze the relationship between the construction equipment density and the inspection time. The spare parts maintenance point optimization and adjustment module 105 determines the guarantee scope of the spare parts maintenance points according to the analysis results of the construction equipment; adjusts and optimizes the quantity distribution of the spare parts maintenance points according to the utilization efficiency of the existing spare parts maintenance points, the number of construction workers, and the construction area. The inspection route optimization module 106 optimizes the inspection routes of the inspection personnel using the reinforcement learning algorithm based on the construction worker behavior prediction and construction equipment density evaluation results, and preferentially arranges inspections on the routes of high-utilization equipment and during periods with less construction operations to promptly discover potential faults and reduce the impact on normal construction operations. The inspection efficiency evaluation module 107 monitors the execution of the inspection tasks in real time, collects the feedback from the inspection personnel, and comprehensively evaluates the inspection efficiency and the early warning level of fault risks.

9. A smart construction site intelligent patrol system based on a spatio-temporal graph convolutional network using the smart construction site intelligent patrol method based on the spatio-temporal graph convolutional network according to any one of claims 1-7, characterized in that, It includes the following hardware and software components: The hardware components include, but are not limited to, a server for storing and processing data, running the backend service of the intelligent inspection system for the smart construction site based on the spatio-temporal graph convolutional network; Data acquisition devices for real-time monitoring of the activities of construction workers and equipment status within the construction site; Inspection devices, including various inspection tools and inspection vehicles, for performing inspection tasks; Mobile terminals, including laptops and smartphones, for use by management personnel and inspection personnel to receive tasks and feedback information in real time. The software components include, but are not limited to, software for processing and analyzing the collected data to perform construction worker behavior prediction, equipment identification, and analysis tasks; user-end software provided to management personnel and inspection personnel for displaying the progress of inspection tasks, construction worker density, and equipment distribution information; communication software responsible for data transmission and communication between the components of the system.

10. A computer-readable storage medium, characterized in that: It stores a computer program that can be loaded and executed by a processor, such as the intelligent inspection method for the smart construction site based on the spatio-temporal graph convolutional network described in any one of claims 1-7.

Images