Large hydro-junction sluice deep foundation pit excavation construction and monitoring method

Abstract

The invention discloses a large hydro-junction sluice deep foundation pit excavation construction and monitoring method. The method comprises the steps that historical image data sets of all areas are obtained through a cloud platform system; performing classification processing on the historical image data set of each region through a K-prototype clustering algorithm, only retaining key frame data, and training the image data set of each region through a CNN-gMLP neural network model to obtain a CNN-gMLP neural network model corresponding to each region; whether risks exist in all areas of the divided water drainage gate deep foundation pit excavation construction or not is judged through a CNN-gMLP neural network model; the key frame data reserved by the method can more accurately reflect the actual data monitoring condition of the monitored part of the excavation construction of the deep foundation pit of the sluice; cNN-gMLP neural network model training is carried out on the key frame data, training sample diversity expansion is carried out on the key frame data by using an adversarial network, and the obtained CNN-gMLP neural network model is more accurate in risk prediction of a monitored area of water release gate deep foundation pit excavation construction, so that the prediction timeliness is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence monitoring technology, and in particular to a deep foundation pit excavation construction and monitoring method for a sluice gate of a large-scale water conservancy hub. Background Art

[0002] As a major component of the infrastructure of water conservancy and hydropower projects, the sluice gates of large-scale water conservancy hubs play an important role in flood control and drainage, water supply and diversion, hydropower generation, etc. As a key hydraulic structure, the construction technology of the sluice gates of large-scale water conservancy hubs is particularly important. The sluice gates are usually built at key locations of reservoirs or rivers. Their construction needs to cope with complex hydrological conditions and geological environments. Deep foundation pit excavation construction and monitoring work are particularly important. Due to the large amount of engineering, tight construction period, complex construction environment, many uncertain factors, high safety risks, and high frequency of accidents in deep foundation pit projects, once a foundation pit safety accident occurs, it will cause huge economic losses and casualties, which will not only delay the construction period, but also cause adverse social impacts.

[0003] Chinese patent application number CN202210835431.4 discloses a BIM-based deep foundation pit construction safety intelligent management and control method and system. The present invention provides a BIM-based deep foundation pit construction safety intelligent management and control method and system, the system comprising: a geological model construction module for constructing a three-dimensional geological model of a deep foundation pit project according to design documents and on-site survey documents; a safety model construction module for constructing a BIM safety management and control model of a deep foundation pit project according to hazard source information combined with a three-dimensional geological model; a deep foundation pit monitoring module for monitoring the structure of a deep foundation pit project, obtaining monitoring information of monitoring points of a deep foundation pit project, and updating the deep foundation pit project BIM safety management and control model according to the acquired monitoring information; a control target monitoring module for dynamically acquiring the location information of the control target, and feeding back the acquired location information to the deep foundation pit project BIM safety management and control model; a safety management and control module for performing safety analysis according to the monitoring information of the monitoring points of the deep foundation pit project and the location information of the control target, and obtaining a safety analysis result. The present invention is helpful to improve the intelligent level of safety management and control of deep foundation pit construction.

[0004] The above scheme still has several limitations, specifically: the analysis of construction risks is not timely and inaccurate; the cost of construction monitoring is high; the monitoring data is not accurately analyzed through artificial intelligence algorithms; and construction risks cannot be predicted in advance. Summary of the invention

[0005] The purpose of the present invention is to provide a deep foundation pit excavation construction and monitoring method for a large-scale water conservancy project sluice gate, so as to solve the above-mentioned problems existing in the prior art.

[0006] The specific application is as follows:

[0007] The deep foundation pit excavation construction and monitoring method of the sluice gate of a large-scale water conservancy hub includes the following steps:

[0008] S1. Deploy several cameras to monitor the construction site of the deep foundation pit of the sluice gate, collect the entire regional image of the construction site through the cameras, and divide N regional images through image analysis algorithms, wherein the regional images at least include construction personnel, construction equipment, construction materials and construction environment;

[0009] S2, connect the cameras installed in all areas to the cloud platform system, and obtain the historical image data set of each area through the cloud platform system;

[0010] S3. Classify the historical image data sets of each region using the K-prototype clustering algorithm, and divide the image data into two categories: key frame data and other frame data; retain the key frame data and remove other frame data; after classifying the historical image data set using the clustering algorithm, take the most densely distributed and top three highest-scoring categories as key frame data, and take all other categories as other frame data;

[0011] S4, representing each of the retained key frame data with a data set {Bi}, respectively, training the data set {Bi} of each region respectively through the CNN-gMLP neural network model, and obtaining the CNN-gMLP neural network model corresponding to each region;

[0012] S5. Requests for obtaining data of corresponding areas are continuously sent to cameras connected to N areas through the cloud platform system. After receiving the request, the cameras in each area feed back the collected image data of the corresponding area to the cloud platform system. The cloud platform system inputs the received image data into the corresponding CNN-gMLP neural network model to determine whether there are risks in each area of ​​the deep foundation pit excavation construction of the divided sluice gate;

[0013] S6. If it is determined that there is a risk in a certain area of ​​the deep foundation pit excavation for the sluice gate, a warning must be issued in a timely manner to notify the construction personnel to stop the work and notify the technical personnel to inspect and repair the area.

[0014] Further, the monitoring by deploying a number of cameras at the construction site of the deep foundation pit excavation of the sluice gate in step S1 includes: ensuring that after the cameras are installed, the monitoring data of the deep foundation pit excavation of the sluice gate in the entire area can be collected; the monitoring data of the construction data includes the construction operation data of the construction personnel, the construction operation data of the construction equipment, the construction material data and the construction environment data, the construction material data includes the location of the construction materials and whether the construction materials are used up, and the construction environment data includes whether the construction structure is deformed;

[0015] The dividing of the N regional images by an image analysis algorithm comprises:

[0016] The area is divided according to the number of construction workers. If there are construction workers in the area, it is judged as an area that needs to be monitored;

[0017] If there are no construction workers in the area and no construction equipment in the area, the area is judged as an area that does not need to be monitored.

[0018] Further, the step S2 of obtaining the historical image data set of each area through the cloud platform system includes: obtaining the historical image data set of each area by consulting the historical data of the deep foundation pit excavation construction of the sluice gate, wherein the historical image data set represents the historical data collected from each area, including risk data and non-risk data, and whether it is risk data is retrieved and confirmed by the deep foundation pit knowledge base through the improved whale algorithm;

[0019] The deep foundation pit knowledge base includes technical specifications for construction safety of deep foundation pits, technical specifications for monitoring of deep foundation pits, basic parameter database of deep foundation pits, engineering geological database of deep foundation pits, hydrogeological database of deep foundation pits, safety technical specifications for deep foundation pits of sluice gates of water conservancy hubs, and monitoring technical specifications for deep foundation pits of sluice gates of water conservancy hubs.

[0020] Furthermore, the step S2 of obtaining the historical image data set of each area through the cloud platform system also includes:

[0021] S21. If the historical image data set of a monitored area only contains non-risk data, it is necessary to conduct security algorithm analysis and risk simulation experiments on the area to obtain a series of risk data, and add the obtained risk data to the non-risk data set to form a new data set;

[0022] S22. The risk simulation experiment means assuming that a certain area of ​​the deep foundation pit excavation construction of the sluice gate is in a certain state, and through the safety algorithm analysis of the state, it is found that the state is risky, then the video of the state is collected by video recording, and the image frame is extracted from the video recording to generate image data, and the image data is risky;

[0023] S23. Obtain a series of risk data by continuously conducting risk simulation experiments.

[0024] Furthermore, in step S22, extracting image frames from the video recording and generating image data also includes:

[0025] The specific process of expanding the image samples of the generated image data by using an adversarial network is as follows: according to the original image of the generated image data, adversarial image samples are generated, and the specifications of the adversarial image samples are: risk data diversification features, risk data position non-fixed features and risk data image brightness diversification features. One given original image is input, and a perturbation factor is added through the adversarial network to generate 1,000 image sample images.

[0026] Furthermore, the classification processing of the historical image data sets of each region by the K-prototype clustering algorithm in step S3 includes: if it is found that the historical image data set only contains non-risk data, after obtaining a series of risk data through a risk simulation experiment, the risk data is added to the historical image data set, and the K-prototype clustering algorithm analysis is performed on the newly generated historical image data set;

[0027] The K-prototype clustering algorithm performs clustering on the sample set D = {y 1 ,y 2 ,…y n The classification method is: by continuously updating the central object, the cluster division C = {C 1 , C 2 , …C k}, C 1 , C 2 , C k Each represents a single cluster, where each cluster includes multiple single samples in the same cluster; the clustering rule is to minimize the mean square error, and the mean square error formula is: Among them, y represents being classified into cluster C i The sample data in , δ represents the Hamming distance, w i It is cluster C i The mean vector, P d The smaller the value of is, the higher the similarity of samples y within the cluster.

[0028] Furthermore, the specific structure of the CNN-gMLP neural network model includes an image feature extraction 3DResNet network and a gMLP perceptron network, the 3DResNet network is connected in series with the gMLP perceptron network, and the output of the 3DResNet network is used as the input of the gMLP perceptron network;

[0029] The 3DResNet network is used to extract the spatiotemporal features of four submodules from the video frame image sequence of the acquisition camera; the four submodules are: construction operation data of construction personnel, construction operation data of construction equipment, construction material data and construction environment data;

[0030] The gMLP sensor network includes a one-dimensional channel mapping unit and a three-dimensional space-time mapping unit;

[0031] The one-dimensional channel mapping unit is used to generate a channel fine-grained feature map; the three-dimensional spatiotemporal mapping unit is used to generate a spatiotemporal fine-grained feature map; the real-time situation of each area of ​​the deep foundation pit excavation construction site is obtained according to the generated channel fine-grained feature map and the spatiotemporal fine-grained feature map, and whether there is a risk warning in each area is output; the real-time situation of the deep foundation pit excavation construction site includes: construction operation data of construction personnel during the construction process, construction operation data of construction equipment during the construction process, construction material data during the construction process, and construction environment data during the construction process;

[0032] The one-dimensional channel mapping unit includes: the step of generating a channel fine-grained feature map through the one-dimensional channel mapping unit is:

[0033] G1(g)=σ[a*gMLP(Αvg(g))+b*gMLP(Μax(g))],

[0034]

[0035] Where σ is the sigmoid activation function, G1 represents the one-dimensional channel mapping, gMLP represents the perceptron, Avg and Max represent average pooling and maximum pooling respectively, g and g′ correspond to the first input feature map and channel fine-grained feature map respectively, a represents the weight coefficient of Avg, and b represents the weight coefficient of Max;

[0036] The three-dimensional space-time mapping unit includes: the steps of generating a space-time refinement feature map through the three-dimensional space-time mapping unit are:

[0037]

[0038] Among them, σ is the sigmoid activation function, G2 represents the three-dimensional spatiotemporal mapping, Avg and Max represent average pooling and maximum pooling respectively, Conv represents convolution, g′ and g″ represent channel fine-grained feature mapping and spatiotemporal fine-grained feature mapping respectively, c represents the weight coefficient of Avg, and d represents the weight coefficient of Max.

[0039] Furthermore, in step S5, the cloud platform system inputs the received image data into corresponding CNN-gMLP neural network models respectively, including: performing CNN-gMLP neural network model training on the image data of each area respectively to obtain a CNN-gMLP neural network model for each area, and the obtained CNN-gMLP neural network model is one-to-one corresponding to each area; therefore, when the cloud platform system receives the image data fed back by each camera, the image data is input into the corresponding CNN-gMLP neural network model according to the correspondence between the CNN-gMLP neural network model and each area, and determines whether there are risks in each area of ​​the deep foundation pit excavation construction divided for the sluice gate through the result of the CNN-gMLP neural network model.

[0040] Furthermore, the determination of whether there are risks in each area of ​​deep foundation pit excavation construction divided by the result of the CNN-gMLP neural network model includes: if the image data of a certain area is input into the corresponding CNN-gMLP neural network model, and the predicted result is that there is a risk, then it is determined that the certain area is at risk; if the image data of a certain area is input into the corresponding CNN-gMLP neural network model, and the predicted result is that there is no risk, then it is determined that there is no risk in the certain area.

[0041] Furthermore, in step S6, if it is determined through judgment that a certain area of ​​the deep foundation pit excavation construction of the sluice gate is at risk, the image data is input into the corresponding CNN-gMLP neural network model. If it is determined that a certain area is at risk, an early warning is immediately issued to the construction site; construction personnel are required to stop operations in time; and relevant technical personnel are notified to check and repair the said area at risk in time.

[0042] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0043] The embodiment of the present invention provides a method for obtaining historical image data sets of various regions through a cloud platform system; classifying and processing the historical image data sets of various regions respectively through a K-prototype clustering algorithm, retaining only key frame data, and training the image data sets of various regions respectively through a CNN-gMLP neural network model to obtain a CNN-gMLP neural network model corresponding to each region; and judging whether there are risks in each divided area for deep foundation pit excavation construction of a sluice gate through the CNN-gMLP neural network model; the key frame data retained by the method will more accurately reflect the actual monitoring situation of the data of the monitored part of the deep foundation pit excavation construction of the sluice gate; the key frame data is used to train the CNN-gMLP neural network model, and the key frame data is used to expand the diversity of training samples using an adversarial network, and the obtained CNN-gMLP neural network model will make the risk prediction of the monitored area of ​​the deep foundation pit excavation construction of the sluice gate more accurate, improve the prediction timeliness, reduce the construction monitoring cost, and predict the construction risk in advance. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flow chart of a deep foundation pit excavation construction and monitoring method for a sluice gate of a large-scale water conservancy hub provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0046] Example 1

[0047] First, the technical terms involved in the embodiments of the present application are explained.

[0048] (1) gMLP neural network

[0049] gMLP (Gated Multi-Layer Perceptron) is a network architecture based on MLP. It does not contain the traditional self-attention structure. gMLP is mainly composed of static parameterized channel projections and spatial projections. Experiments have found that when the spatial mapping is linear and paired with multiplication gating, the effect is best. Through this structure, gMLP can achieve performance comparable to Transformer in key language and vision applications. In some cases, gMLP can even achieve the same level as Transformer in pre-trained perplexity indicators without self-attention mechanism, and perform better on some downstream tasks. This shows that self-attention may not be necessary for Vision Transformer, because gMLP can achieve the same accuracy. In general, gMLP shows that the MLP structure can achieve good results in complex tasks without relying on the self-attention mechanism;

[0050] (2) 3DResNet network

[0051] 3DResNet (3D Residual Network) is a deep learning model for video understanding and action recognition. It combines the advantages of 3D convolutional neural networks (3D CNNs) and residual networks (ResNets). 3DResNet effectively improves the performance of video understanding and action recognition by combining 3D convolution and residual learning, while solving some challenges in deep network training.

[0052] The embodiment of the present invention provides a deep foundation pit excavation construction and monitoring method for a large-scale water conservancy project sluice gate, such as Figure 1 , including the following steps:

[0053] A deep foundation pit excavation construction and monitoring method for a large-scale water conservancy project sluice gate is characterized by comprising the following steps:

[0054] S1. Deploy several cameras to monitor the construction site of the deep foundation pit of the sluice gate, collect the entire regional image of the construction site through the cameras, and divide N regional images through image analysis algorithms, wherein the regional images at least include construction personnel, construction equipment, construction materials and construction environment;

[0055] S2, connect the cameras installed in all areas to the cloud platform system, and obtain the historical image data set of each area through the cloud platform system;

[0056] S3. Classify the historical image data sets of each region using the K-prototype clustering algorithm, and divide the image data into two categories: key frame data and other frame data; retain the key frame data and remove other frame data; after classifying the historical image data set using the clustering algorithm, take the most densely distributed and top three highest-scoring categories as key frame data, and take all other categories as other frame data;

[0057] S4, representing each of the retained key frame data with a data set {Bi}, respectively, training the data set {Bi} of each region respectively through the CNN-gMLP neural network model, and obtaining the CNN-gMLP neural network model corresponding to each region;

[0058] S5. Requests for obtaining data of corresponding areas are continuously sent to cameras connected to N areas through the cloud platform system. After receiving the request, the cameras in each area feed back the collected image data of the corresponding area to the cloud platform system. The cloud platform system inputs the received image data into the corresponding CNN-gMLP neural network model to determine whether there are risks in each area of ​​the deep foundation pit excavation construction of the divided sluice gate;

[0059] S6. If it is determined that there is a risk in a certain area of ​​the deep foundation pit excavation for the sluice gate, a warning must be issued in a timely manner to notify the construction personnel to stop the work and notify the technical personnel to inspect and repair the area.

[0060] Specifically, the present invention includes firstly dividing the construction site of the deep foundation pit excavation of the sluice gate into different areas through an image analysis algorithm; the image segmentation algorithm includes but is not limited to a neural network segmentation algorithm, a Canny edge detector, a Sobel operator, a Prewitt operator, a Roberts operator, etc.; installing cameras in each area for monitoring; connecting the cameras installed in all areas to a cloud platform system, and obtaining a historical data information set {Xi} of each area through the cloud platform system; then, classifying and processing the historical image data set {Xi} of each area respectively through a K-prototype clustering algorithm, and dividing the data information into two categories: key frame data and data of other frames; only retaining the key frame data; representing the retained key frame data of each area with a data set {Bi}, and performing clustering on the data set {Bi} of each area respectively through a CNN-gMLP neural network model. Training is performed to obtain the CNN-gMLP neural network model corresponding to each area; requests for obtaining image data of the corresponding area are continuously sent to the cameras connected to each area through the cloud platform system. After receiving the request, the cameras in each area feed back the collected image data of the corresponding area to the cloud platform system. The cloud platform system inputs the received image data into the corresponding CNN-gMLP neural network model to determine whether there are risks in each area of ​​the divided deep foundation pit excavation construction site of the sluice gate; if there is a risk in a certain area of ​​the divided deep foundation pit excavation construction site of the sluice gate, it is necessary to promptly warn of the risk, notify the construction personnel to stop working, and notify relevant technical personnel to repair and reinforce the area. The cameras include but are not limited to high-definition cameras and infrared high-definition cameras. Compared with the existing methods, only image key frame data is processed through the K-prototype clustering algorithm to accelerate the risk identification speed and recognition accuracy.

[0061] In the above embodiment, specifically, it is ensured that after the camera is installed, the monitoring data of the deep foundation pit excavation construction of the sluice gate in the entire area can be collected; the monitoring data of the construction data includes the construction operation data of the construction personnel, the construction operation data of the construction equipment, the construction material data and the construction environment data, the construction material data includes the location of the construction material and whether the construction material is used up, and the construction environment data includes whether the construction structure is deformed;

[0062] The dividing of the N regional images by an image analysis algorithm comprises:

[0063] The area is divided according to the number of construction workers. If there are construction workers in the area, it is judged as an area that needs to be monitored;

[0064] If there are no construction workers in the area and no construction equipment in the area, the area is judged as an area that does not need to be monitored.

[0065] In the above embodiment, specifically, by consulting the historical data of the deep foundation pit excavation construction of the sluice gate, the historical image data set of each area is obtained, and the historical image data set represents the historical data collected in each area, including risk data and non-risk data. Whether it is risk data is retrieved and confirmed by the deep foundation pit knowledge base through the improved whale algorithm;

[0066] The deep foundation pit knowledge base includes technical specifications for construction safety of deep foundation pits, technical specifications for monitoring of deep foundation pits, basic parameter database of deep foundation pits, engineering geological database of deep foundation pits, hydrogeological database of deep foundation pits, safety technical specifications for deep foundation pits of sluice gates of water conservancy hubs, and monitoring technical specifications for deep foundation pits of sluice gates of water conservancy hubs.

[0067] Specifically, for the expert system of deep foundation pits, a knowledge base and a reasoning model are constructed. In this embodiment, the knowledge base includes relevant specifications for deep foundation pits, historical monitoring data and expert experience, wherein the relevant specifications for deep foundation pits include technical specifications for construction safety of deep foundation pits, technical specifications for monitoring of deep foundation pits, basic parameter database of deep foundation pits, engineering geological database of deep foundation pits, hydrogeological database of deep foundation pits, technical specifications for safety of deep foundation pits of water conservancy hub sluice gates and technical specifications for monitoring of deep foundation pits of water conservancy hub sluice gates.

[0068] Improved whale algorithm implementation process:

[0069] D=|CX * (t)-X(t)|,

[0070] X(t+1)=X * (t)-AD,

[0071] t is the current iteration number, A and C are coefficients, X * (t) represents the best position so far, i.e. the optimal solution, and X(t) represents the current search position;

[0072] A=2ar 1 -a,

[0073] C=2r 2 ,

[0074] a=2-2t / Tmax,where r 1 and r 2 is a random number in (0,1), the value of a decreases linearly from 2 to 0, t represents the current number of iterations, Tmax represents the maximum number of iterations,

[0075] X(t+1)=X * (t)+D * *e b.d *cos(2πd),

[0076] Among them, D * =|X * (t)-X(t)| represents the distance between the current historical image data location to be retrieved and the historical image data location of the retrieved knowledge base, X * (t) is the best position vector so far, b is a constant used to define the shape of the spiral, and d is a random number in (-1,1).

[0077] D=|C*X rand -X(t)|,

[0078] X(t+1)=X rand -A*D,

[0079] Among them, X rand is the randomly selected position vector of the historical image data to be retrieved. The algorithm sets that when A≥1, a search agent is randomly selected, the number of historical image data of the retrieved knowledge base is set to N, the historical image data boundary of the retrieved knowledge base is set to the time boundary of N months of the retrieved knowledge base, the number of iterations is set to T, and the optimal result is finally retrieved. The result is used to determine whether the current historical image data to be retrieved is risk data;

[0080] Compared with the existing technology, the use of the improved whale algorithm can quickly identify whether the image data is risky data in massive knowledge base data.

[0081] In the above embodiment, specifically, obtaining the historical image data set of each area through the cloud platform system includes:

[0082] S21. If the historical image data set of a monitored area only contains non-risk data, it is necessary to conduct security algorithm analysis and risk simulation experiments on the area to obtain a series of risk data, and add the obtained risk data to the non-risk data set to form a new data set;

[0083] S22. The risk simulation experiment means assuming that a certain area of ​​the deep foundation pit excavation construction of the sluice gate is in a certain state, and through the safety algorithm analysis of the state, it is found that the state is risky, then the video of the state is collected by video recording, and the image frame is extracted from the video recording to generate image data, and the image data is risky;

[0084] S23. Obtain a series of risk data by continuously conducting risk simulation experiments.

[0085] In the above embodiment, specifically, the specific process of expanding the image samples of the generated image data by using an adversarial network is: according to the original image of the generated image data, adversarial image samples are generated, and the specifications of the adversarial image samples are: risk data diversification features, risk data position non-fixed features and risk data image brightness diversification features, input 1 given original image, add perturbation factors through the adversarial network, and generate 1000 image sample images.

[0086] It should be noted that, in this embodiment, the adversarial network adopts the BigGAN adversarial generative network.

[0087] In the above embodiment, specifically, the K-prototype clustering algorithm performs clustering on the sample set D = {y 1 ,y 2 ,…y n The classification method is: by continuously updating the central object, the cluster division C = {C 1 , C 2 , …C k}, C 1 , C 2 , C k Each represents a single cluster, where each cluster includes multiple single samples in the same cluster; the clustering rule is to minimize the mean square error, and the mean square error formula is: Among them, y represents being classified into cluster C i The sample data in , δ represents the Hamming distance, w i It is cluster C i The mean vector, P d The smaller the value, the higher the similarity of the sample y in the cluster; after the historical image data set is classified by the clustering algorithm, the top three classifications with the most densely distributed sample data sets are used as key frame data, and the remaining all other classified sample data sets are used as data for other frames;

[0088] In the above embodiment, specifically, the historical image data sets of each region are classified and processed respectively by the K-prototype clustering algorithm, including: if it is found that the historical image data set only contains non-risk data, then after obtaining a series of risk data through a risk simulation experiment, the risk data is added to the historical image data set, and the newly generated historical image data set is analyzed by the K-prototype clustering algorithm.

[0089] In the above embodiment, specifically, the specific structure of the CNN-gMLP neural network model includes an image feature extraction 3DResNet network and a gMLP perceptron network, the 3DResNet network is connected in series with the gMLP perceptron network, and the output of the 3DResNet network is used as the input of the gMLP perceptron network;

[0090] The 3DResNet network is used to extract the spatiotemporal features of four submodules from the video frame image sequence of the acquisition camera; the four submodules are: construction operation data of construction personnel, construction operation data of construction equipment, construction material data and construction environment data;

[0091] The gMLP sensor network includes a one-dimensional channel mapping unit and a three-dimensional space-time mapping unit;

[0092] The one-dimensional channel mapping unit is used to generate a channel fine-grained feature map; the three-dimensional spatiotemporal mapping unit is used to generate a spatiotemporal fine-grained feature map; the real-time situation of each area of ​​the deep foundation pit excavation construction site is obtained according to the generated channel fine-grained feature map and the spatiotemporal fine-grained feature map, and whether there is a risk warning in each area is output; the real-time situation of the deep foundation pit excavation construction site includes: construction operation data of construction personnel during the construction process, construction operation data of construction equipment during the construction process, construction material data during the construction process, and construction environment data during the construction process;

[0093] The one-dimensional channel mapping unit includes: the step of generating a channel fine-grained feature map through the one-dimensional channel mapping unit is:

[0094] G1(g)=σ[a*gMLP(Αvg(g))+b*gMLP(Μax(g))],

[0095]

[0096] Where σ is the sigmoid activation function, G1 represents the one-dimensional channel mapping, gMLP represents the perceptron, Avg and Max represent average pooling and maximum pooling respectively, g and g′ correspond to the first input feature map and channel fine-grained feature map respectively, a represents the weight coefficient of Avg, and b represents the weight coefficient of Max;

[0097] The three-dimensional space-time mapping unit includes: the steps of generating a space-time refinement feature map through the three-dimensional space-time mapping unit are:

[0098]

[0099] Among them, σ is the sigmoid activation function, G2 represents the three-dimensional spatiotemporal mapping, Avg and Max represent average pooling and maximum pooling respectively, Conv represents convolution, g′ and g″ represent channel fine-grained feature mapping and spatiotemporal fine-grained feature mapping respectively, c represents the weight coefficient of Avg, and d represents the weight coefficient of Max.

[0100] It should be noted that compared with the existing technology, the model that combines the 3DResNet network with the gMLP perceptron network has a high accuracy in identifying risks in images. In the complex environment of deep foundation pit excavation construction of large-scale water conservancy hub sluice gates, one-dimensional channels or three-dimensional space-time channels can be used to identify risks in different fine-grained feature directions, thereby improving the robustness of risk event identification.

[0101] In the above embodiment, specifically, the cloud platform system inputs the received image data into the corresponding CNN-gMLP neural network model respectively, including: performing CNN-gMLP neural network model training on the image data of each area respectively to obtain a CNN-gMLP neural network model for each area, and the obtained CNN-gMLP neural network model is one-to-one corresponding to each area; therefore, when the cloud platform system receives the image data fed back by each camera, the image data is input into the corresponding CNN-gMLP neural network model according to the correspondence between the CNN-gMLP neural network model and each area, and determines whether there are risks in each area of ​​the deep foundation pit excavation construction divided for the sluice gate through the results of the CNN-gMLP neural network model.

[0102] In the above embodiment, specifically, the determination of whether there are risks in each area of ​​deep foundation pit excavation construction divided by the result of the CNN-gMLP neural network model includes: if the image data of a certain area is input into the corresponding CNN-gMLP neural network model, and the predicted result is that there is a risk, then it is determined that the certain area is at risk; if the image data of a certain area is input into the corresponding CNN-gMLP neural network model, and the predicted result is that there is no risk, then it is determined that there is no risk in the certain area.

[0103] In the above embodiment, specifically, if it is determined through judgment that a certain area of ​​the deep foundation pit excavation construction of the sluice gate is at risk, the method includes: inputting the image data into the corresponding CNN-gMLP neural network model; if it is determined that a certain area is at risk, an early warning is immediately issued to the construction site; construction personnel are required to stop operations in time; and relevant technical personnel are notified to promptly check and repair the area at risk.

[0104] Specifically, compared with the prior art, the entire method process of this embodiment can be automatically completed between the cloud platform system and the camera, without the need for manpower, simple and convenient, low cost, easy to deploy, and quickly identify risks. The cloud computing of the cloud platform system can dynamically adjust computing resources according to demand and provide a flexible computing environment for intelligent algorithms. Cloud computing provides massive data storage and processing capabilities, and provides data support for neural network algorithms. Cloud computing provides powerful computing resources and supports neural network training on large-scale data sets. Cloud computing quickly obtains a large amount of computing resources, which reduces hardware costs and improves computing efficiency in the complex deep foundation pit construction environment of large-scale water conservancy hub sluice gates.

[0105] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing such systems. In addition, the present invention is not directed to any specific programming language either. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the description of the above specific languages ​​is for disclosing the best mode of the present invention.

[0106] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0107] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting the intention that the claimed invention requires more features than those explicitly recited in each claim. More specifically, as reflected in the claims below, inventive aspects lie in less than all the features of the individual embodiments disclosed above. Therefore, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, with each claim itself serving as a separate embodiment of the present invention.

[0108] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition they may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0109] In addition, those skilled in the art will appreciate that, although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present invention and form different embodiments. For example, in the claims below, any one of the claimed embodiments may be used in any combination.

[0110] The various component embodiments of the present invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) may be used in practice to implement some or all of the functions of some or all of the components in the apparatus according to an embodiment of the present invention. The present invention may also be implemented as a device or device program (e.g., a computer program and a computer program product) for executing part or all of the methods described herein. Such a program implementing the present invention may be stored on a computer-readable medium, or may have the form of one or more signals. Such a signal may be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

Claims

1. A deep foundation pit excavation construction and monitoring method for a large-scale water conservancy project sluice gate, characterized in that: The following steps are involved: S1. Deploy several cameras to monitor the construction site of the deep foundation pit of the sluice gate, collect the entire regional image of the construction site through the cameras, and divide it into N regional images through an image analysis algorithm, each of the N regional images at least including construction personnel, construction equipment, construction materials and construction environment; S2, connect the cameras installed in all areas to the cloud platform system, and obtain the historical image data set of each area through the cloud platform system; S3. Classify the historical image data sets of each region using the K-prototype clustering algorithm, and divide the image data into two categories: key frame data and other frame data; retain the key frame data and remove other frame data; after classifying the historical image data set using the clustering algorithm, take the most densely distributed and top three highest-scoring categories as key frame data, and take all other categories as other frame data; S4, representing each of the retained key frame data with a data set {Bi}, respectively, training the data set {Bi} of each region respectively through the CNN-gMLP neural network model, and obtaining the CNN-gMLP neural network model corresponding to each region; S5. Requests for obtaining data of corresponding areas are continuously sent to cameras connected to N areas through the cloud platform system. After receiving the request, the cameras in each area feed back the collected image data of the corresponding area to the cloud platform system. The cloud platform system inputs the received image data into the corresponding CNN-gMLP neural network model to determine whether there are risks in each area of ​​the deep foundation pit excavation construction of the divided sluice gate; S6. If it is determined that there is a risk in a certain area of ​​the deep foundation pit excavation for the sluice gate, a warning must be issued in a timely manner to notify the construction personnel to stop the work and notify the technical personnel to inspect and repair the area.

2. The deep foundation pit excavation construction and monitoring method for a large-scale water conservancy project sluice gate according to claim 1 is characterized in that: The monitoring by deploying a number of cameras at the construction site of the deep foundation pit excavation of the sluice gate in step S1 includes: ensuring that after the cameras are installed, the monitoring data of the deep foundation pit excavation of the sluice gate in the entire area can be collected; the monitoring data of the construction data includes the construction operation data of the construction personnel, the construction operation data of the construction equipment, the construction material data and the construction environment data, the construction material data includes the location of the construction materials and whether the construction materials are used up, and the construction environment data includes whether the construction structure is deformed; The dividing of the N regional images by an image analysis algorithm comprises: The area is divided according to the number of construction workers. If there are construction workers in the area, it is judged as an area that needs to be monitored; If there are no construction workers in the area and no construction equipment in the area, the area is judged as an area that does not need to be monitored.

3. The deep foundation pit excavation construction and monitoring method for a large-scale water conservancy project sluice gate according to claim 1 is characterized in that: The step S2 of obtaining the historical image data set of each area through the cloud platform system includes: obtaining the historical image data set of each area by consulting the historical data of the deep foundation pit excavation construction of the sluice gate, wherein the historical image data set represents the historical data collected from each area, including risk data and non-risk data, and whether the risk data is risk data is retrieved and confirmed by the deep foundation pit knowledge base through the improved whale algorithm; The deep foundation pit knowledge base includes technical specifications for construction safety of deep foundation pits, technical specifications for monitoring of deep foundation pits, basic parameter database of deep foundation pits, engineering geological database of deep foundation pits, hydrogeological database of deep foundation pits, safety technical specifications for deep foundation pits of sluice gates of water conservancy hubs, and monitoring technical specifications for deep foundation pits of sluice gates of water conservancy hubs.

4. The deep foundation pit excavation construction and monitoring method for a large-scale water conservancy project sluice gate according to claim 1 is characterized in that: The step S2 of obtaining the historical image data set of each region through the cloud platform system also includes: S21: If the historical image data set of a monitored area only contains non-risk data, it is necessary to conduct security algorithm analysis and risk simulation experiments on the area to obtain a series of risk data, and add the obtained risk data to the set of non-risk data to form a new data set; S22. The risk simulation experiment means assuming that a certain area of ​​the deep foundation pit excavation construction of the sluice gate is in a certain state, and through the safety algorithm analysis of the state, it is found that the state is risky, then the video of the state is collected by video recording, and the image frame is extracted from the video recording to generate image data, and the image data is risky; S23. Obtain a series of risk data by continuously conducting risk simulation experiments.

5. The deep foundation pit excavation construction and monitoring method for a large-scale water conservancy project sluice gate according to claim 4 is characterized in that: In step S22, image frames are extracted from the video recording to generate image data, which further includes: The specific process of expanding the image samples of the generated image data by using an adversarial network is as follows: according to the original image of the generated image data, adversarial image samples are generated, and the specifications of the adversarial image samples are: risk data diversification features, risk data position non-fixed features and risk data image brightness diversification features. One given original image is input, and a perturbation factor is added through the adversarial network to generate 1,000 image sample images.

6. The deep foundation pit excavation construction and monitoring method for a large-scale water conservancy project sluice gate according to claim 1 is characterized in that: The step S3 of classifying the historical image data sets of each region by using the K-prototype clustering algorithm includes: if it is found that the historical image data set only contains non-risk data, then after obtaining a series of risk data through a risk simulation experiment, the risk data is added to the historical image data set, and the newly generated historical image data set is analyzed by the K-prototype clustering algorithm; The K-prototype clustering algorithm performs clustering on the sample set D = {y1, y2, ... n The classification method is: by continuously updating the central object, the cluster division C = {C1, C2, ...C k }, C1, C2, C k Each represents a single cluster, where each cluster includes multiple single samples in the same cluster; the clustering rule is to minimize the mean square error, and the mean square error formula is: Among them, y represents being classified into cluster C i The sample data in , δ represents the Hamming distance, w i It is cluster C i The mean vector, P d The smaller the value of is, the higher the similarity of samples y within the cluster.

7. The deep foundation pit excavation construction and monitoring method for a large-scale water conservancy project sluice gate according to claim 1 is characterized in that: The specific structure of the CNN-gMLP neural network model includes an image feature extraction 3DResNet network and a gMLP perceptron network, wherein the 3DResNet network is connected in series with the gMLP perceptron network, and the output of the 3DResNet network is used as the input of the gMLP perceptron network; The 3DResNet network is used to extract the spatiotemporal features of four submodules from the video frame image sequence of the acquisition camera; the four submodules are: construction operation data of construction personnel, construction operation data of construction equipment, construction material data and construction environment data; The gMLP sensor network includes a one-dimensional channel mapping unit and a three-dimensional space-time mapping unit; The one-dimensional channel mapping unit is used to generate a channel fine-grained feature map; the three-dimensional spatiotemporal mapping unit is used to generate a spatiotemporal fine-grained feature map; the real-time situation of each area of ​​the deep foundation pit excavation construction site is obtained according to the generated channel fine-grained feature map and the spatiotemporal fine-grained feature map, and whether there is a risk warning in each area is output; the real-time situation of the deep foundation pit excavation construction site includes: construction operation data of construction personnel during the construction process, construction operation data of construction equipment during the construction process, construction material data during the construction process, and construction environment data during the construction process; The one-dimensional channel mapping unit comprises: the step of generating a channel fine-grained feature map through the one-dimensional channel mapping unit is: G1(g)=σ[a*gMLP(Αvg(g))+b*gMLP(Μax(g))], Where σ is the sigmoid activation function, G1 represents the one-dimensional channel mapping, gMLP represents the perceptron, Avg and Max represent average pooling and maximum pooling respectively, g and g′ correspond to the first input feature map and channel fine-grained feature map respectively, a represents the weight coefficient of Avg, and b represents the weight coefficient of Max; The three-dimensional space-time mapping unit includes: the steps of generating a space-time refinement feature map by the three-dimensional space-time mapping unit are: Among them, σ is the sigmoid activation function, G2 represents the three-dimensional spatiotemporal mapping, Avg and Max represent average pooling and maximum pooling respectively, Conv represents convolution, g′ and g″ represent channel fine-grained feature mapping and spatiotemporal fine-grained feature mapping respectively, c represents the weight coefficient of Avg, and d represents the weight coefficient of Max.

8. The deep foundation pit excavation construction and monitoring method for a large-scale water conservancy project sluice gate according to claim 1 is characterized in that: In step S5, the cloud platform system inputs the received image data into the corresponding CNN-gMLP neural network model respectively, including: performing CNN-gMLP neural network model training on the image data of each area respectively to obtain a CNN-gMLP neural network model for each area, and the obtained CNN-gMLP neural network model is one-to-one corresponding to each area; therefore, when the cloud platform system receives the image data fed back by each camera, the image data is input into the corresponding CNN-gMLP neural network model according to the correspondence between the CNN-gMLP neural network model and each area, and determines whether there are risks in each area of ​​the deep foundation pit excavation construction divided for the sluice gate through the result of the CNN-gMLP neural network model.

9. The deep foundation pit excavation construction and monitoring method for a large-scale water conservancy project sluice gate according to claim 8 is characterized in that: The method of determining whether there are risks in each area of ​​deep foundation pit excavation construction divided by the results of the CNN-gMLP neural network model includes: if the image data of a certain area is input into the corresponding CNN-gMLP neural network model, and the predicted result is that there is a risk, then it is determined that the certain area is at risk; if the image data of a certain area is input into the corresponding CNN-gMLP neural network model, and the predicted result is that there is no risk, then it is determined that there is no risk in the certain area.

10. The deep foundation pit excavation construction and monitoring method for a large-scale water conservancy project sluice gate according to claim 1 is characterized in that: If it is determined in step S6 that a certain area of ​​the deep foundation pit excavation construction of the sluice gate is at risk, the method includes: inputting the image data into the corresponding CNN-gMLP neural network model; if it is determined that a certain area is at risk, an early warning is immediately issued to the construction site; construction personnel are required to stop work in time; and relevant technical personnel are notified to check and repair the certain area at risk in time.

Images