A method and system for monitoring the urban underground space environment based on the Internet of Things

Abstract

The present invention discloses a method and system for monitoring the urban underground space environment based on the Internet of Things, including: generating a point cloud map according to a binocular image, calculating the minimum distance of the point cloud map, obtaining a change cloud map, performing three-dimensional modeling based on the change cloud map, calculating the adjacent correlation degree according to the environmental data of the underground space, performing a first cluster analysis based on the adjacent correlation degree, obtaining a number of spatial regions with similar characteristics, further calculating the degree of environmental anomaly, constructing a prediction model to obtain a trend curve of the degree of anomaly, calculating the probability of intervention and performing a second cluster analysis, obtaining an intervention hotspot area, color-coding the three-dimensional model according to the intervention hotspot area, and sending the color-coded three-dimensional model and the trend curve of the degree of anomaly to the monitoring platform as an early warning. This method can not only improve the efficiency and real-time performance of underground space environmental monitoring, but also has good interpretability, and can be directly applied to the urban underground space environmental monitoring system.

Description

Technical Field

[0001] The present invention relates to the field of environmental monitoring, and in particular to an urban underground space environmental monitoring method and system based on the Internet of Things. Background Art

[0002] With the rapid development of the Internet of Things, artificial intelligence and big data, the intelligence and automation levels of environmental monitoring methods have been continuously improved, and the accuracy and timeliness of monitoring have become key factors in ensuring environmental safety.

[0003] Traditional underground space environmental monitoring methods mainly rely on geological survey technology and physical detection technology, but this method has problems such as high data collection cost and low efficiency. In recent years, with the rapid development of technology, more and more environmental monitoring methods have begun to use intelligent algorithms or big data for monitoring to reduce costs and improve monitoring efficiency and accuracy.

[0004] At present, although there are some urban underground space environment monitoring methods or systems based on intelligent algorithms or big data, these monitoring methods have limitations in real-time and accuracy. They usually focus on detecting anomalies in data, but ignore the comprehensive consideration of space and time. They also lack predictions of future environmental trends and cannot effectively provide early warnings, which affects the monitoring effect. Summary of the invention

[0005] The purpose of the present invention is to provide an urban underground space environment monitoring method and system based on the Internet of Things.

[0006] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0007] The first aspect of the present invention provides an urban underground space environment monitoring method based on the Internet of Things, comprising the following steps:

[0008] Step S1, obtaining a binocular image of the underground space, generating point cloud data based on the binocular image, calculating a minimum distance of the point cloud data, obtaining a change point cloud, and performing three-dimensional modeling based on the change point cloud to obtain a three-dimensional model;

[0009] Step S2, collecting environmental data of the underground space, calculating the adjacent correlation degree between the environmental data, and performing a first cluster analysis in the three-dimensional model based on the adjacent correlation degree to obtain a plurality of spatial regions with similar characteristics;

[0010] Step S3, obtaining the gas concentration in the spatial region with similar characteristics, calculating the degree of environmental abnormality according to the gas concentration distribution and the gas concentration exceeding standard ratio, constructing a prediction model to predict the spatial region with similar characteristics, and obtaining a trend curve of abnormality degree;

[0011] Step S4, calculating the intervention probability according to the environmental abnormality degree and the abnormality degree trend curve, performing a second cluster analysis based on the intervention probability to obtain the intervention hotspot area, color-coding the three-dimensional model according to the intervention hotspot area, and sending the color-coded three-dimensional model and the abnormality degree trend curve as an early warning to the monitoring platform.

[0012] Furthermore, the method for generating point cloud data based on the binocular image includes:

[0013] Calculate the binocular camera parallax, the calculation formula is:

[0014]

[0015] in, is the binocular camera parallax, is the horizontal position of the light spot on the imaging plane, is the focal length of the binocular camera, is the distance from the binocular camera to the imaging plane, is the baseline length of the binocular camera, is the viewing angle of the binocular camera. The depth of the image is calculated based on the parallax. The calculation formula is:

[0016]

[0017] in, is the focal length of the binocular camera, is the baseline length of the binocular camera, is the binocular camera parallax;

[0018] The points in the binocular camera image are converted into coordinates in three-dimensional space. The expression is:

[0019]

[0020] in, , and are the coordinates of the points in three-dimensional space, and are the horizontal and vertical coordinates of the midpoints of the left and right camera images, and are the camera principal point coordinates, is the focal length of the binocular camera, is the depth of the image;

[0021] The obtained coordinates in the three-dimensional space are collected to obtain point cloud data.

[0022] Furthermore, the method of calculating the minimum distance of the point cloud data to obtain the change point cloud includes:

[0023] Preprocess the point cloud data and align the points in the overlapping point clouds. The expression is:

[0024]

[0025] in, is the number of points in the point cloud data, is the three-dimensional transformation matrix of the point cloud, and are represented as the source point cloud and the target point cloud respectively. Points, Expressed as The normal vector of

[0026] Perform spatial search for each point cloud in the point cloud data to find the nearest neighbor point, calculate the Euclidean distance between each point cloud and its nearest neighbor point, take the minimum value as the densest area in the point cloud, and generate a distance field based on the distance from each point cloud to its nearest neighbor point;

[0027] Cluster analysis is performed on the distance field, and the point cloud data is grouped into clusters according to the local density of the point cloud to obtain the change point cloud.

[0028] Furthermore, the method for performing three-dimensional modeling based on the change point cloud includes:

[0029] The surface of the change point cloud is reconstructed by solving the Poisson equation, which is expressed as:

[0030]

[0031] in, Represented as a point in the change point cloud, is the set of points in the change point cloud, Expressed as the Laplace operator, Represented as a distance field The divergence of It is expressed as a spatial function, i.e. the function to be solved;

[0032] The scale space is generated according to the change point cloud after surface reconstruction. The expression is:

[0033]

[0034] in, is the Gaussian kernel width, is the spatial scale parameter, , and They are respectively represented as the location information of the change point cloud, Represented as a convolution operation, It is represented as the target model after downsampling of the change point cloud;

[0035] In the scale space, edge, corner and ridge features are extracted as key points of the control mesh. The control mesh is constructed by connecting the key points, and the control mesh is converted into a polygonal mesh to obtain a three-dimensional model.

[0036] Furthermore, the method of calculating the adjacent correlation degree between the environmental data and performing a first cluster analysis in the three-dimensional model based on the adjacent correlation degree to obtain a plurality of spatial regions with similar characteristics includes:

[0037] Calculate the adjacent association degree, the calculation formula is:

[0038]

[0039] in, Indicates the normalization operation of [0, 1]. Represented as data points in the environmental data The neighborhood size of Any data point in the environmental data and data points The Euclidean distance of is the number of data points in the environmental data;

[0040] Constructing a similarity matrix , the elements in the matrix Represents data points and data points Adjacent correlation degree; construct degree matrix , where the diagonal elements Represents data points The degree of all data points The sum of the weights of the connected edges;

[0041] Usage Matrix and the similarity matrix Constructing the Laplacian Matrix ,in , perform eigenvalue decomposition on the Laplace matrix and obtain non-zero eigenvectors and their corresponding features are adjacent;

[0042] Will The eigenvectors are used as column vectors to form an eigenvector matrix , the eigenvector matrix As data points in the new feature space, the k-means clustering algorithm is used to cluster them along the x, y and z axes to obtain spatial regions with similar features.

[0043] Furthermore, the method for calculating the degree of environmental abnormality according to the gas concentration distribution and the gas concentration exceeding the standard ratio includes:

[0044] Calculate the gas concentration distribution in spatial regions with similar characteristics. The calculation formula is:

[0045]

[0046] in, is the amount of polluted gas added per unit time of 1 hour, , and They represent the three-dimensional coordinates of the monitoring position, , and are respectively represented as the three-dimensional coordinates of the pollution source, , and The pollutant , and The diffusion coefficient of the axis, is the average wind speed;

[0047] The degree of abnormality in spatial areas with similar characteristics is calculated based on the gas concentration exceeding the standard ratio and the gas concentration exceeding the standard ratio. The calculation formula is:

[0048]

[0049] in, is the rate at which gas concentration exceeds the standard, To monitor the quantity of gas species, is the number of times the gas concentration exceeds the standard within 1 hour, is the length of time that the gas concentration exceeds the standard continuously, is the gas concentration distribution, is the average value of the gas concentration exceeding the standard ratio, is the area ratio of the area exceeding the standard to the total monitoring area, is the gas concentration attenuation coefficient.

[0050] Furthermore, the method of constructing a prediction model to predict the spatial area with similar characteristics and obtaining the abnormality trend curve of the environmental data includes:

[0051] The prediction model is built based on the neural network. The prediction model structure includes: input layer, convolution layer, LSTM layer, fully connected layer and output layer, which are specifically:

[0052] The input layer is used to receive the time series data obtained by sliding the window of the environmental data and the corresponding degree of abnormality;

[0053] The convolution layer is composed of a convolutional neural network, which performs convolution operations on the input data. A maximum pooling layer is added after the convolution layer to downsample the feature map output by the convolution layer.

[0054] The LSTM layer is composed of a long short-term memory neural network, and uses LSTM units to learn the feature vectors extracted from the convolutional layer;

[0055] The fully connected layer introduces a self-attention mechanism to convert the hidden information of all neural units in the LSTM layer into three sets of feature vector sequences: query vector, key vector and value vector. The query vector and key vector are used to calculate the similarity to obtain a similarity matrix. The similarity matrix is ​​normalized to obtain a weight matrix. The normalized weight matrix is ​​used to perform weighted summation on the value vector to obtain a new feature representation.

[0056] The output layer maps the new feature representation into output data to obtain the predicted value, and plots the predicted value after time sorting into a chart to obtain the abnormality trend curve.

[0057] Furthermore, the method of calculating the intervention probability according to the environmental abnormality degree and the abnormality degree trend curve, and performing a second cluster analysis based on the intervention probability to obtain the intervention hotspot area includes:

[0058] Calculate the intervention probability using the formula:

[0059]

[0060] in, Indicates the normalization operation of [0, 1]. is the integral area of ​​the abnormality trend curve, For abnormal degree, is the rate at which the gas concentration decays under intervention, is the smooth slope of the abnormality trend curve, is the peak value of the abnormality trend curve, is the valley value of the abnormality trend curve, is the sliding window time size in the abnormality degree trend curve;

[0061] Cluster analysis was performed based on intervention probability, and the intervention probability was located in The spatial area with high intervention probability is regarded as a high intervention hotspot area, indicating that it is necessary to take immediate measures on the underground space environment; The spatial area with a probability of intervention is regarded as a medium intervention hotspot area, indicating that the spatial area is at risk and needs to be continuously monitored. The spatial area is regarded as a low-intervention hotspot area, indicating that the environmental conditions in the spatial area are relatively stable and no intervention is required.

[0062] Furthermore, the method of color-coding the three-dimensional model according to the hotspot areas includes: mapping the colors of three intervention hotspot areas in the three-dimensional model, wherein the color corresponding to the high intervention hotspot area is red, the color corresponding to the medium intervention hotspot area is yellow, and the color corresponding to the low intervention hotspot area is green, so as to obtain a three-dimensional model of the spatial distribution of the intervention hotspot areas.

[0063] The second aspect of the present invention also provides an urban underground space environment monitoring system based on the Internet of Things, comprising:

[0064] The three-dimensional modeling module obtains binocular images of the underground space, generates a point cloud map based on the binocular images, calculates the minimum distance of the point cloud map, obtains a change cloud map, and performs three-dimensional modeling based on the change cloud map;

[0065] The module for dividing regions with similar characteristics collects underground space environmental data, calculates the adjacent correlation between environmental data, and performs the first cluster analysis in the three-dimensional model based on the adjacent correlation to obtain several spatial regions with similar characteristics;

[0066] The abnormality assessment module calculates the degree of environmental abnormality based on the gas concentration distribution and the proportion of gas concentration exceeding the standard, and constructs a prediction model to obtain the trend curve of the degree of abnormality in spatial areas with similar characteristics;

[0067] The monitoring and visualization module calculates the intervention probability, performs a second cluster analysis based on the intervention probability, obtains the intervention hotspot area, builds a three-dimensional visualization anomaly model based on the intervention hotspot area, and sends the three-dimensional visualization anomaly model and the anomaly degree trend curve as an early warning to the monitoring platform.

[0068] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0069] (1) The present invention provides an automatic constant temperature control method and system based on the Internet of Things and fuzzy judgment, which performs three-dimensional modeling based on binocular images, calculates adjacent correlation based on environmental data of underground space, performs a first cluster analysis based on adjacent correlation, obtains a number of spatial regions with similar characteristics, further calculates the degree of environmental anomaly, constructs a prediction model to obtain a trend curve of the degree of anomaly, calculates the intervention probability and performs a second cluster analysis to obtain an intervention hotspot area, performs color coding on the three-dimensional model according to the intervention hotspot area, and sends the color-coded three-dimensional model and the trend curve of the degree of anomaly as an early warning to the monitoring platform. This method uses three-dimensional modeling and visualization of anomalies to more clearly reflect the distribution of environmental data anomalies, which helps to quickly identify abnormal areas;

[0070] (2) The present invention calculates the adjacent correlation between different environmental data points and performs the first clustering based on the adjacent correlation, which helps to identify the correlation between data points. The second clustering analysis is further performed based on the intervention probability to identify the hot spots that need intervention, which helps to formulate targeted governance plans for areas with the same environmental change characteristics and optimize resource allocation in environmental governance.

[0071] (3) The present invention constructs a prediction model to obtain a trend curve of the degree of abnormality, and feeds the trend curve back to the monitoring platform, thereby considering the degree of abnormality in the time dimension, which can help the monitoring platform to continuously track environmental conditions and help take preventive measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 The present invention is a flowchart of the steps of an urban underground space environment monitoring method and system based on the Internet of Things. DETAILED DESCRIPTION

[0073] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0074] Reference Figure 1 As shown, the present invention provides an urban underground space environment monitoring method and system based on the Internet of Things, including:

[0075] Step S1, obtaining a binocular image of the underground space, generating point cloud data based on the binocular image, calculating a minimum distance of the point cloud data, obtaining a change point cloud, and performing three-dimensional modeling based on the change point cloud;

[0076] In the actual evaluation, the binocular camera parameter data is obtained, stereo matching is performed using the stereo vision library OpenCV, and the disparity map is calculated. Based on the disparity map and camera parameters, the depth of the image is calculated. , convert the points in the image into three-dimensional coordinates, collect all the converted three-dimensional coordinate points to form point cloud data, pre-process the point cloud data, and align the points between overlapping point clouds;

[0077] In the actual evaluation, the KD tree algorithm is used to perform spatial search to find the nearest neighbor points of the point cloud. The DBSCAN algorithm is further used to perform cluster analysis based on the local density of the point cloud to obtain the changed point cloud. The surface of the changed point cloud is reconstructed by solving the Poisson equation. The scale space theory is used to extract surface features of different scales through Gaussian filtering. According to the feature extraction results of the scale space, a control grid is constructed, which is converted into a polygonal mesh. The subdivision surface algorithm is used to generate a smooth three-dimensional model.

[0078] Step S2, collecting environmental data of the underground space, calculating the adjacent correlation degree between the environmental data, and performing a first cluster analysis in the three-dimensional model based on the adjacent correlation degree to obtain a plurality of spatial regions with similar characteristics;

[0079] In the actual assessment, the environmental data of the underground space is obtained, which consists of 100 points, each with Coordinates, for each point, calculate its Euclidean distance to all other points, and calculate the adjacent correlation between data points to construct The similarity matrix and degree matrix of The Laplace matrix of the vector is decomposed and eigenvalues ​​are decomposed to obtain 5 non-zero eigenvalues ​​and their corresponding eigenvectors. The 5 eigenvectors are used as data points in the new feature space. The k-means clustering algorithm is used to perform the first clustering analysis in the x-, y-, and z-axis directions to obtain 5 spatial regions with similar characteristics.

[0080] Step S3, obtaining the gas concentration in the spatial region with similar characteristics, calculating the degree of environmental abnormality according to the gas concentration distribution and the gas concentration exceeding standard ratio, constructing a prediction model to predict the spatial region with similar characteristics, and obtaining a trend curve of abnormality degree;

[0081] In the actual evaluation, O2, CO2, CO, and NH3 are monitored, the gas concentration distribution and the gas concentration exceeding the standard ratio are calculated, and the degree of environmental abnormality is further calculated. The environmental data is operated by sliding window, and the window size is set to 1 point. The time series matrix data is obtained, and the prediction model is constructed based on the neural network. The input data is The matrix represents the gas concentration values ​​of the four gases at a time point of 30 minutes. The local features of the time series matrix data are extracted through the convolution layer of the prediction model, and the long-term dependencies are further learned through the LSTM layer to obtain the degree of abnormality within the next hour. The predicted values ​​are sorted by time and plotted into a curve to obtain the abnormality trend curve of the environmental data.

[0082] Step S4, calculating the intervention probability according to the environmental abnormality degree and the abnormality degree trend curve, performing a second cluster analysis based on the intervention probability to obtain the intervention hotspot area, color-coding the three-dimensional model according to the intervention hotspot area, and sending the color-coded three-dimensional model and the abnormality degree trend curve as an early warning to the monitoring platform.

[0083] In the actual assessment, the intervention probability is calculated based on the abnormality degree and the abnormality trend curve. Based on the obtained interference probability, a hierarchical cluster analysis is performed to divide the underground space into hot spots with high, medium and low intervention levels, and a total of 10 intervention probability areas are obtained. In the three-dimensional model, the high intervention area is coded as red, the medium intervention area is coded as yellow, and the low intervention area is coded as green. The color-coded three-dimensional model is loaded through Three.js, and the abnormality trend curve and the color-coded three-dimensional model are integrated into a visual early warning report. The early warning report is sent to the monitoring platform through the API interface.

[0084] In this embodiment, the method for generating point cloud data based on the binocular image includes:

[0085] Calculate the binocular camera parallax, the calculation formula is:

[0086]

[0087] in, is the binocular camera parallax, is the horizontal position of the light spot on the imaging plane, is the focal length of the binocular camera, is the distance from the binocular camera to the imaging plane, is the baseline length of the binocular camera, is the viewing angle of the binocular camera. The depth of the image is calculated based on the parallax. The calculation formula is:

[0088]

[0089] in, is the focal length of the binocular camera, is the baseline length of the binocular camera, is the binocular camera parallax;

[0090] The points in the binocular camera image are converted into coordinates in three-dimensional space. The expression is:

[0091]

[0092] in, , and are the coordinates of the points in three-dimensional space, and are the horizontal and vertical coordinates of the midpoints of the left and right camera images, and are the camera principal point coordinates, is the focal length of the binocular camera, is the depth of the image;

[0093] The obtained coordinates in the three-dimensional space are collected to obtain point cloud data.

[0094] In this embodiment, the method of calculating the minimum distance of the point cloud data to obtain the change point cloud includes:

[0095] Preprocess the point cloud data and align the points in the overlapping point clouds. The expression is:

[0096]

[0097] in, is the number of points in the point cloud data, is the three-dimensional transformation matrix of the point cloud, and are represented as the source point cloud and the target point cloud respectively. Points, Expressed as The normal vector of

[0098] Perform spatial search for each point cloud in the point cloud data to find the nearest neighbor point, calculate the Euclidean distance between each point cloud and its nearest neighbor point, take the minimum value as the densest area in the point cloud, and generate a distance field based on the distance from each point cloud to its nearest neighbor point;

[0099] Cluster analysis is performed on the distance field, and the point cloud data is grouped into clusters according to the local density of the point cloud to obtain the change point cloud.

[0100] In this embodiment, the method for performing three-dimensional modeling based on the change point cloud includes:

[0101] The surface of the change point cloud is reconstructed by solving the Poisson equation, which is expressed as:

[0102]

[0103] in, Represented as a point in the change point cloud, is the set of points in the change point cloud, Expressed as the Laplace operator, Represented as a distance field The divergence of It is expressed as a spatial function, i.e. the function to be solved;

[0104] The scale space is generated according to the change point cloud after surface reconstruction. The expression is:

[0105]

[0106] in, is the Gaussian kernel width, is the spatial scale parameter, , and They are respectively represented as the location information of the change point cloud, Represented as a convolution operation, It is represented as the target model after downsampling of the change point cloud;

[0107] In the scale space, edge, corner and ridge features are extracted as key points of the control mesh. The control mesh is constructed by connecting the key points, and the control mesh is converted into a polygonal mesh to obtain a three-dimensional model.

[0108] In this embodiment, the method of calculating the adjacent correlation degree between the environmental data, performing a first cluster analysis in the three-dimensional model based on the adjacent correlation degree, and obtaining a plurality of spatial regions with similar characteristics includes:

[0109] Calculate the adjacent association degree, the calculation formula is:

[0110]

[0111] in, Indicates the normalization operation of [0, 1]. Represented as data points in the environmental data The neighborhood size of Any data point in the environmental data and data points The Euclidean distance of is the number of data points in the environmental data;

[0112] Constructing a similarity matrix , the elements in the matrix Represents data points and data points Adjacent correlation degree; construct degree matrix , where the diagonal elements Represents data points The degree of all data points The sum of the weights of the connected edges;

[0113] Usage Matrix and the similarity matrix Constructing the Laplacian Matrix ,in , perform eigenvalue decomposition on the Laplace matrix and obtain non-zero eigenvectors and their corresponding features are adjacent;

[0114] Will The eigenvectors are used as column vectors to form an eigenvector matrix , the eigenvector matrix As data points in the new feature space, the k-means clustering algorithm is used to cluster them along the x, y and z axes to obtain spatial regions with similar features.

[0115] In this embodiment, the method for calculating the degree of environmental abnormality based on the gas concentration distribution and the gas concentration exceeding the standard ratio includes:

[0116] Calculate the gas concentration distribution in spatial regions with similar characteristics. The calculation formula is:

[0117]

[0118] in, is the amount of polluted gas added per unit time of 1 hour, , and They represent the three-dimensional coordinates of the monitoring position, , and are respectively represented as the three-dimensional coordinates of the pollution source, , and The pollutant , and The diffusion coefficient of the axis, is the average wind speed;

[0119] The degree of abnormality in spatial areas with similar characteristics is calculated based on the gas concentration exceeding the standard ratio and the gas concentration exceeding the standard ratio. The calculation formula is:

[0120]

[0121] in, is the rate at which gas concentration exceeds the standard, To monitor the quantity of gas species, is the number of times the gas concentration exceeds the standard within 1 hour, is the length of time that the gas concentration exceeds the standard continuously, is the gas concentration distribution, is the average value of the gas concentration exceeding the standard ratio, is the area ratio of the area exceeding the standard to the total monitoring area, is the gas concentration attenuation coefficient.

[0122] In this embodiment, the method for calculating the degree of environmental abnormality based on the gas concentration distribution and the gas concentration exceeding the standard ratio includes:

[0123] Calculate the gas concentration distribution in spatial regions with similar characteristics. The calculation formula is:

[0124]

[0125] in, is the intensity of the pollution source, i.e. the amount of pollutants emitted per hour. , and They represent the three-dimensional coordinates of the monitoring position, , and are respectively represented as the three-dimensional coordinates of the pollution source, , and The pollutant , and The diffusion coefficient of the axis, is the average wind speed;

[0126] The degree of abnormality in spatial areas with similar characteristics is calculated based on the gas concentration exceeding the standard ratio and the gas concentration exceeding the standard ratio. The calculation formula is:

[0127]

[0128] in, is the rate at which gas concentration exceeds the standard, To monitor the quantity of gas species, The number of times the gas concentration exceeds the standard within 1 hour. is the length of time that the gas concentration exceeds the standard continuously, is the concentration distribution, is the average value of the gas concentration exceeding the standard ratio, is the area ratio of the area exceeding the standard to the total observation area, is the gas concentration attenuation coefficient.

[0129] In this embodiment, the method of constructing a prediction model to predict the spatial region with similar characteristics and obtaining the abnormality trend curve of the environmental data includes:

[0130] The prediction model is built based on the neural network. The prediction model structure includes: input layer, convolution layer, LSTM layer, fully connected layer and output layer, which are specifically:

[0131] The input layer is used to receive the time series data obtained by sliding the window of the environmental data and the corresponding degree of abnormality;

[0132] The convolution layer is composed of a convolutional neural network, which performs convolution operations on the input data. A maximum pooling layer is added after the convolution layer to downsample the feature map output by the convolution layer.

[0133] The LSTM layer is composed of a long short-term memory neural network, and uses LSTM units to learn the feature vectors extracted from the convolutional layer;

[0134] The fully connected layer introduces a self-attention mechanism to convert the hidden information of all neural units in the LSTM layer into three sets of feature vector sequences: query vector, key vector and value vector. The query vector and key vector are used to calculate the similarity to obtain a similarity matrix. The similarity matrix is ​​normalized to obtain a weight matrix. The normalized weight matrix is ​​used to perform weighted summation on the value vector to obtain a new feature representation.

[0135] The output layer maps the new feature representation into output data to obtain the predicted value, and plots the predicted value after time sorting into a chart to obtain the abnormality trend curve.

[0136] In this embodiment, the method of calculating the intervention probability according to the environmental abnormality degree and the abnormality degree trend curve, and performing a second cluster analysis based on the intervention probability to obtain the intervention hotspot area includes:

[0137] Calculate the intervention probability using the formula:

[0138]

[0139] in, Indicates the normalization operation of [0, 1]. is the integral area of ​​the abnormality trend curve, For abnormal degree, is the rate at which the gas concentration decays under intervention, is the smooth slope of the abnormality trend curve, is the peak value of the abnormality trend curve, is the valley value of the abnormality trend curve, is the sliding window time size in the abnormality degree trend curve;

[0140] Cluster analysis was performed based on intervention probability, and the intervention probability was located in The spatial area with high intervention probability is regarded as a high intervention hotspot area, indicating that it is necessary to take immediate measures on the underground space environment; The spatial area with a probability of intervention is regarded as a medium intervention hotspot area, indicating that the spatial area is at risk and needs to be continuously monitored. The spatial area is regarded as a low-intervention hotspot area, indicating that the environmental conditions in the spatial area are relatively stable and no intervention is required.

[0141] In this embodiment, the method of color coding the three-dimensional model according to the hot spot areas includes: mapping the colors of three intervention hot spot areas in the three-dimensional model, wherein the color corresponding to the high intervention hot spot area is red, the color corresponding to the medium intervention hot spot area is yellow, and the color corresponding to the low intervention hot spot area is green, so as to obtain a three-dimensional model of the spatial distribution of the intervention hot spot areas.

[0142] The second aspect of the present invention also provides an urban underground space environment monitoring system based on the Internet of Things, comprising:

[0143] The three-dimensional modeling module obtains binocular images of the underground space, generates a point cloud map based on the binocular images, calculates the minimum distance of the point cloud map, obtains a change cloud map, and performs three-dimensional modeling based on the change cloud map;

[0144] The module for dividing regions with similar characteristics collects underground space environmental data, calculates the adjacent correlation between environmental data, and performs the first cluster analysis in the three-dimensional model based on the adjacent correlation to obtain several spatial regions with similar characteristics;

[0145] The abnormality assessment module calculates the degree of environmental abnormality based on the gas concentration distribution and the proportion of gas concentration exceeding the standard, and constructs a prediction model to obtain the trend curve of the degree of abnormality in spatial areas with similar characteristics;

[0146] The monitoring and visualization module calculates the intervention probability, performs a second cluster analysis based on the intervention probability, obtains the intervention hotspot area, builds a three-dimensional visualization anomaly model based on the intervention hotspot area, and sends the three-dimensional visualization anomaly model and the anomaly degree trend curve as an early warning to the monitoring platform.

[0147] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A method for monitoring the urban underground space environment based on the Internet of Things, characterized in that: The following steps are involved: Step S1, obtaining a binocular image of the underground space, generating point cloud data based on the binocular image, calculating a minimum distance of the point cloud data, obtaining a change point cloud, and performing three-dimensional modeling based on the change point cloud to obtain a three-dimensional model; Step S2, collecting environmental data of the underground space, calculating the adjacent correlation degree between the environmental data, and performing a first cluster analysis in the three-dimensional model based on the adjacent correlation degree to obtain a plurality of spatial regions with similar characteristics; Step S3, obtaining the gas concentration in the spatial region with similar characteristics, calculating the degree of environmental abnormality according to the gas concentration distribution and the gas concentration exceeding standard ratio, constructing a prediction model to predict the spatial region with similar characteristics, and obtaining a trend curve of abnormality degree; Step S4, calculating the intervention probability according to the environmental abnormality degree and the abnormality degree trend curve, performing a second cluster analysis based on the intervention probability to obtain an intervention hotspot area, color-coding the three-dimensional model according to the intervention hotspot area, and sending the color-coded three-dimensional model and the abnormality degree trend curve as an early warning to the monitoring platform; The method of calculating the adjacent correlation degree between the environmental data, and performing a first cluster analysis in the three-dimensional model based on the adjacent correlation degree to obtain a plurality of spatial regions with similar characteristics includes: Calculate the adjacent association degree, the calculation formula is: ； in, Indicates the normalization operation of [0, 1]. Represented as data points in the environmental data The neighborhood size of Any data point in the environmental data and data points The Euclidean distance of is the number of data points in the environmental data; Constructing a similarity matrix , the elements in the matrix Represents data points and data points Adjacent correlation degree; construct degree matrix , where the diagonal elements Represents data points The degree of all data points The sum of the weights of the connected edges; Usage Matrix and the similarity matrix Constructing the Laplacian Matrix ,in , perform eigenvalue decomposition on the Laplace matrix and obtain non-zero eigenvectors and their corresponding features are adjacent; Will The eigenvectors are used as column vectors to form an eigenvector matrix , the eigenvector matrix As data points in the new feature space, the k-means clustering algorithm is used to cluster them along the x, y and z axes to obtain spatial regions with similar features; The method for calculating the degree of environmental abnormality according to the gas concentration distribution and the gas concentration exceeding the standard ratio comprises: Calculate the gas concentration distribution in spatial regions with similar characteristics. The calculation formula is: ； in, is the amount of polluted gas added per unit time of 1 hour, , and They represent the three-dimensional coordinates of the monitoring position, , and are respectively represented as the three-dimensional coordinates of the pollution source, , and The pollutant , and The diffusion coefficient of the axis, is the average wind speed; The degree of abnormality in spatial areas with similar characteristics is calculated based on the gas concentration exceeding the standard ratio and the gas concentration exceeding the standard ratio. The calculation formula is: ； in, is the rate at which gas concentration exceeds the standard, To monitor the quantity of gas species, is the number of times the gas concentration exceeds the standard within 1 hour, is the length of time that the gas concentration exceeds the standard continuously, is the gas concentration distribution, is the average value of the gas concentration exceeding the standard ratio, is the area ratio of the area exceeding the standard to the total monitoring area, is the gas concentration attenuation coefficient; The method of calculating the intervention probability according to the environmental abnormality degree and the abnormality degree trend curve, and performing a second cluster analysis based on the intervention probability to obtain the intervention hotspot area includes: Calculate the intervention probability using the formula: ； in, Indicates the normalization operation of [0, 1]. is the integral area of ​​the abnormality trend curve, For abnormal degree, is the rate at which the gas concentration decays under intervention, is the smooth slope of the abnormality trend curve, is the peak value of the abnormality trend curve, is the valley value of the abnormality trend curve, is the sliding window time size in the abnormality degree trend curve; Cluster analysis was performed based on intervention probability, and the intervention probability was located in The spatial area with high intervention probability is regarded as a high intervention hotspot area, indicating that it is necessary to take immediate measures on the underground space environment; The spatial area with a probability of intervention is regarded as a medium intervention hotspot area, indicating that the spatial area is at risk and needs to be continuously monitored. The spatial area is regarded as a low-intervention hotspot area, indicating that the environmental conditions in the spatial area are relatively stable and no intervention is required.

2. The method for monitoring the urban underground space environment based on the Internet of Things according to claim 1 is characterized in that: The method for generating point cloud data based on the binocular image comprises: Calculate the binocular camera parallax, the calculation formula is: ； in, is the binocular camera parallax, is the horizontal position of the light spot on the imaging plane, is the focal length of the binocular camera, is the distance from the binocular camera to the imaging plane, is the baseline length of the binocular camera, is the viewing angle of the binocular camera. The depth of the image is calculated based on the parallax. The calculation formula is: ； in, is the focal length of the binocular camera, is the baseline length of the binocular camera, is the binocular camera parallax; The points in the binocular camera image are converted into coordinates in three-dimensional space. The expression is: ； in, , and are the coordinates of the points in three-dimensional space, and are the horizontal and vertical coordinates of the midpoints of the left and right camera images, and are the camera principal point coordinates, is the focal length of the binocular camera, is the depth of the image; The obtained coordinates in the three-dimensional space are collected to obtain point cloud data.

3. The method for monitoring the urban underground space environment based on the Internet of Things according to claim 1 is characterized in that: The method of calculating the minimum distance of the point cloud data to obtain the change point cloud includes: Preprocess the point cloud data and align the points in the overlapping point clouds. The expression is: ； in, is the number of points in the point cloud data, is the three-dimensional transformation matrix of the point cloud, and are represented as the source point cloud and the target point cloud respectively. Points, Expressed as The normal vector of Perform spatial search for each point cloud in the point cloud data to find the nearest neighbor point, calculate the Euclidean distance between each point cloud and its nearest neighbor point, take the minimum value as the densest area in the point cloud, and generate a distance field based on the distance from each point cloud to its nearest neighbor point; Cluster analysis is performed on the distance field, and the point cloud data is grouped into clusters according to the local density of the point cloud to obtain the change point cloud.

4. The method for monitoring the urban underground space environment based on the Internet of Things according to claim 1, characterized in that: The method for performing three-dimensional modeling based on the change point cloud comprises: The surface of the change point cloud is reconstructed by solving the Poisson equation, which is expressed as: ； in, Represented as a point in the change point cloud, is the set of points in the change point cloud, Expressed as the Laplace operator, Represented as a distance field The divergence of It is expressed as a spatial function, i.e. the function to be solved; The scale space is generated according to the change point cloud after surface reconstruction. The expression is: ； in, is the Gaussian kernel width, is the spatial scale parameter, , and They are respectively represented as the location information of the change point cloud, Represented as a convolution operation, It is represented as the target model after downsampling of the change point cloud; In the scale space, edge, corner and ridge features are extracted as key points of the control mesh. The control mesh is constructed by connecting the key points, and the control mesh is converted into a polygonal mesh to obtain a three-dimensional model.

5. The method for monitoring the urban underground space environment based on the Internet of Things according to claim 1 is characterized in that: The method of constructing a prediction model to predict the spatial area with similar characteristics and obtaining the abnormality trend curve of the environmental data includes: The prediction model is built based on the neural network. The prediction model structure includes: input layer, convolution layer, LSTM layer, fully connected layer and output layer, which are specifically: The input layer is used to receive the time series data obtained by sliding the window of the environmental data and the corresponding degree of abnormality; The convolution layer is composed of a convolutional neural network, which performs convolution operations on the input data. A maximum pooling layer is added after the convolution layer to downsample the feature map output by the convolution layer. The LSTM layer is composed of a long short-term memory neural network, and uses LSTM units to learn the feature vectors extracted from the convolutional layer; The fully connected layer introduces a self-attention mechanism to convert the hidden information of all neural units in the LSTM layer into three sets of feature vector sequences: query vector, key vector and value vector. The query vector and key vector are used to calculate the similarity to obtain a similarity matrix. The similarity matrix is ​​normalized to obtain a weight matrix. The normalized weight matrix is ​​used to perform weighted summation on the value vector to obtain a new feature representation. The output layer maps the new feature representation into output data to obtain the predicted value, and plots the predicted value after time sorting into a chart to obtain the abnormality trend curve.

6. The method for monitoring the urban underground space environment based on the Internet of Things according to claim 1 is characterized in that: The method for color coding the three-dimensional model according to the hot spot areas includes: mapping the colors of three intervention hot spot areas in the three-dimensional model, wherein the color corresponding to the high intervention hot spot area is red, the color corresponding to the medium intervention hot spot area is yellow, and the color corresponding to the low intervention hot spot area is green, so as to obtain a three-dimensional model of the spatial distribution of the intervention hot spot areas.

7. An urban underground space environment monitoring system based on the Internet of Things, used to execute the method of claims 1-6, characterized in that: include: The three-dimensional modeling module obtains binocular images of the underground space, generates a point cloud map based on the binocular images, calculates the minimum distance of the point cloud map, obtains a change cloud map, and performs three-dimensional modeling based on the change cloud map; The module for dividing regions with similar characteristics collects underground space environmental data, calculates the adjacent correlation between environmental data, and performs the first cluster analysis in the three-dimensional model based on the adjacent correlation to obtain several spatial regions with similar characteristics; The abnormality assessment module calculates the degree of environmental abnormality based on the gas concentration distribution and the proportion of gas concentration exceeding the standard, and constructs a prediction model to obtain the trend curve of the degree of abnormality in spatial areas with similar characteristics; The monitoring and visualization module calculates the intervention probability, performs a second cluster analysis based on the intervention probability, obtains the intervention hotspot area, builds a three-dimensional visualization anomaly model based on the intervention hotspot area, and sends the three-dimensional visualization anomaly model and the anomaly degree trend curve as an early warning to the monitoring platform.

Images