Underground pipeline abnormal data analysis monitoring method and system based on artificial intelligence

Abstract

The embodiment of the invention provides an underground pipeline abnormal data analyzing and monitoring method and system based on artificial intelligence. The method comprises the steps that S11, monitoring devices are divided into different subdivision grids according to a map; s12, acquiring monitoring data of the monitoring device, endowing different monitoring data with corresponding colors or gray levels into the rectangular pattern, and storing the colors or gray levels in the historical matrix pattern; s13, constructing a convolutional neural network model ResNET, and training the convolutional neural network model ResNET according to the historical matrix pattern; and S14, acquiring monitoring data of a real-time monitoring device, inputting the real-time monitoring data into the convolutional neural network model ResNET, and outputting a corresponding color or gray scale into the real-time matrix pattern. According to the analysis monitoring method of the embodiment, the abnormal monitoring data of the monitoring device can be visually displayed, so that a user can conveniently locate the abnormal monitoring device in real time, and the efficiency of processing the fault of the monitoring device is improved.

Description

Technical Field

[0001] The embodiments disclosed herein belong to the field of intelligent monitoring technology, specifically relating to an artificial intelligence-based method and system for analyzing and monitoring abnormal data of underground pipelines. Background Technology

[0002] In existing technologies, underground pipelines typically monitor physical parameters and material leaks at various geographical coordinates using sensors and instruments. However, due to the large amount of data generated by underground pipelines, a significant amount of data needs to be monitored simultaneously. Current technologies often compare the monitored data with normal values ​​one by one, analyzing changes in parameters such as pressure to pinpoint anomalies. However, this monitoring method cannot visually represent the actual geographical location of anomalies, thus hindering the efficiency of troubleshooting. Summary of the Invention

[0003] The embodiments disclosed herein aim to at least solve one of the technical problems existing in the prior art, and provide an artificial intelligence-based method and system for analyzing and monitoring abnormal data of underground pipelines.

[0004] The embodiments of this disclosure provide an artificial intelligence-based method for analyzing and monitoring abnormal data of underground pipelines, the analysis and monitoring method comprising:

[0005] S11: Divide each monitoring device into different subdivided grids according to the map;

[0006] S12: Acquire monitoring data from the monitoring device, assign different colors or grayscale values ​​to the rectangular pattern, and store them in the historical matrix pattern;

[0007] S13: Construct the convolutional neural network model ResNET and train the convolutional neural network model ResNET based on the historical matrix pattern;

[0008] S14: Obtain monitoring data from the real-time monitoring device, input the real-time monitoring data into the convolutional neural network model ResNET, and output the corresponding color or grayscale to the real-time matrix pattern.

[0009] In some embodiments of this disclosure, the construction of the convolutional neural network model ResNET, and the training of the convolutional neural network model ResNET based on historical matrix patterns, specifically includes:

[0010] S131: Construct the ResNET convolutional neural network model;

[0011] S132: Select anomaly monitoring data from the historical matrix pattern, randomly assign pixel values ​​of the anomaly monitoring data to blank pixels in the historical matrix pattern, and divide the historical matrix pattern into a training set and a validation set.

[0012] S133: Optimize the ResNET convolutional neural network model based on the training and validation sets.

[0013] In some embodiments of this disclosure, optimizing the ResNET convolutional neural network model based on the training and validation sets specifically includes:

[0014] S1331: Construct the loss function L, and train the convolutional neural network model ResNET based on the training set and the loss function L;

[0015] S1332: Optimize the ResNET convolutional neural network model based on the validation set;

[0016] S1333: Based on the stable binary classification prediction results and loss function L values ​​on the validation set, the convolutional neural network model ResNET is determined.

[0017] In some embodiments of this disclosure, the random assignment method includes:

[0018] S31: Select different abnormal monitoring data by using the target selection method to randomly assign different pixel values ​​to the different blank pixel numbers in the target historical rectangular pattern;

[0019] S32: Traverse all selection methods to select different anomaly monitoring data, and randomly assign different pixel values ​​to different blank pixel numbers;

[0020] S33: Traverse all historical rectangular patterns of anomaly monitoring data, select different anomaly monitoring data, and randomly assign different pixel values ​​to different blank pixel numbers.

[0021] In some embodiments of this disclosure, the monitoring method further includes a method for encrypting static data, wherein the method for encrypting static data includes:

[0022] S21: Based on the geographic coordinates of the geographic regions that have been divided into grids, establish a composite model for each grid and form a raster data layer. Divide the smooth surface of the origin reference plane that projects over the raster data layer into multiple regions.

[0023] S22: Generate encryption matrices in multiple regions based on a multi-parameter static random number generation algorithm using random geographic location coordinates;

[0024] S23: Encrypt the raster data layer and the projection of random geographic location coordinates onto the reference plane or smooth surface using an encryption matrix, and form first-level encrypted data and an offset matrix;

[0025] S24: Perform secondary encryption on the primary encrypted data using an offset matrix, and generate a secondary encrypted map.

[0026] In some embodiments of this disclosure, the multi-parameter static random number generation algorithm based on random geographic location coordinates generates an encryption matrix in multiple regions, specifically including:

[0027] S221: Set the random number generation algorithm:

[0028] rand(x,y,w)∈(-1,1);

[0029] Where x and y are the random geographic coordinates of the projection of the raster data layer onto the smooth surface in coordinate system E, and w is determined according to the following formula:

[0030]

[0031] Where g = 1, 2, 3, st is a status indicator; st not displayed indicates the parameter exists, st = 0 indicates the parameter does not exist, R is the area of ​​the encrypted geographic data range, W and H are the width and height of the encryption matrix corresponding to randomly selected geographic coordinates other than those mentioned above, ID is the additional code, Sx is the allowable offset of the X-axis, and Sy is the allowable offset of the Y-axis. The average length of the projection of each region onto the reference plane. The average width of each region projected onto the reference plane, where Xo and Yo are preset coordinates in the projection;

[0032] Sx and Sy satisfy the following formula:

[0033]

[0034] Where r is the spatial resolution of the raster data layer, k is negatively correlated with the number of regions, k = NC(num), num is the number of regions, k ∈ [1, K], K > 1, and when the number of regions is 1, k = K. The number of regions is such that the average area projected onto the reference plane by each region is qr. 2 When q∈(1,2], k=1;

[0035] S222: Fix Cg and NC values, and change s x s y , The values ​​of Xo and Yo are used to obtain the value of k, kd, in order to obtain the intermediate value of w. proc ;

[0036] Wherein, kd satisfies the following condition:

[0037]

[0038]

[0039] |s x +k D rand(x,y,w)| (x,y)∈R |∈(2|r|,2k D |r|],|s y +k D rand(x,y,w)| (x,y)∈R |∈(2|r|,2k D |r|];

[0040] According to w proc To obtain the average of the derivatives at p points on each boundary of each region:

[0041]

[0042]

[0043] Where i = 1, 2, 3, 4, p ∈ [1, +∞), b j Let w be a point on boundary b, and j = 1, 2, ..., p. proc =C gproc ;

[0044] S223: Calculate Wd using the following formula and form the encryption matrix JM:

[0045]

[0046] JM = k D rand(x,y,w) D );

[0047] Where g = 1, 2, 3; G = 11, 22, 33; C G Similarly, for linear and / or nonlinear combinations, G = 11, 22, 33 represent linear, nonlinear, and mixed linear and nonlinear combinations, respectively, abbreviated as C. 11 C 22 and C 33 ,and

[0048] In some embodiments of this disclosure, the encryption of the raster data layer and the projection of random geographic location coordinates onto a reference plane or smooth surface using an encryption matrix to form first-level encrypted data and an offset matrix specifically includes:

[0049] S231: Projecting the raster data layer and the user's location coordinates onto a smooth surface yields the following formula:

[0050]

[0051] S232: Superimpose X1 and Xap, and superimpose Y1 and Yap to obtain the first-level encrypted data:

[0052]

[0053] Offset matrix:

[0054]

[0055] The coordinate system changes from E to E(1);

[0056] S233: Translate the raster data layer before encryption along with the coordinate system E so that E and E (1) Data on the first-level encrypted smooth surface is merged and deleted to form a first-level encrypted raster data layer M1.

[0057] In some embodiments of this disclosure, the step of performing secondary encryption on the primary encrypted data using an offset matrix and generating a secondary encrypted map specifically includes:

[0058] S241: Using the offset matrix M s The first-level encrypted data J1 is then subjected to second-level encryption to generate second-level encrypted data:

[0059]

[0060] S242: Connect the first-level encrypted raster data layer M1 with the coordinate system E (1) The coordinate system E is translated together. (1) With E (2) Data on the secondary smooth surface is merged and deleted to form a secondary encrypted raster data layer M2;

[0061] S243: Preset additional key A∈(0,1), calculate E (2) With E (1) The length L of the line segment connecting the origins of coordinates E is used to connect the secondary encrypted raster data layer M2 along the line segment with the coordinate system E. (2) Translate together by a distance AL such that E (2) The origin of the coordinate system is close to E. (1) The origin of the coordinate system is used to obtain the final encrypted raster data layer M. f ;

[0062] S244: Based on M1, M2, M f If the offset of the encrypted matrix area corresponding to the location requested by the user is within a preset range, then at least one of them will be returned to the user for display.

[0063] According to M1, M2, M fIf the offset in the encryption matrix region corresponding to the location requested by the user is outside the preset range, repeat the encryption steps until M1, M2, and M are obtained. f If the offset of at least one of the encrypted matrix regions corresponding to random positions x and y is within a preset range, select at least one of these at least one and return it to the user for decryption and display.

[0064] This disclosure also proposes an artificial intelligence-based underground pipeline anomaly data analysis and monitoring system, which is used to execute the artificial intelligence-based underground pipeline anomaly data analysis and monitoring method according to any of the above embodiments. The artificial intelligence-based underground pipeline anomaly data analysis and monitoring system includes: an infrastructure layer, an operating environment, a data layer, a support layer, a service layer, an application layer, and a user layer.

[0065] The application layer includes an intelligent detection module, which is used to divide each monitoring device into different subdivided grids according to the map; the intelligent module is used to acquire monitoring data from the monitoring devices, assign different colors or grayscale values ​​to rectangular patterns for different monitoring data, and store them in a historical matrix pattern; the intelligent module is used to construct a convolutional neural network model ResNET, and train the convolutional neural network model ResNET according to the historical matrix pattern; the intelligent module is used to acquire monitoring data from real-time monitoring devices, input the real-time monitoring data into the convolutional neural network model ResNET, and output the corresponding colors or grayscale values ​​to the real-time matrix pattern.

[0066] In some embodiments of this disclosure, the application layer further includes a data encryption module. This data encryption module is used to establish a composite model for each grid and form a raster data layer based on the geographic coordinates of the divided geographic regions. It then divides the smooth surface of the origin reference plane projected onto the raster data layer into multiple regions. The data encryption module is used to generate encryption matrices in these multiple regions based on a multi-parameter static random number generation algorithm for random geographic location coordinates. The data encryption module is used to encrypt the raster data layer and the projection of random geographic location coordinates onto the reference plane or smooth surface using the encryption matrix, forming primary encrypted data and an offset matrix. Finally, the data encryption module is used to perform secondary encryption on the primary encrypted data using the offset matrix, generating a secondary encrypted map.

[0067] The artificial intelligence-based method and system for analyzing and monitoring abnormal data of underground pipelines disclosed in this embodiment first divides the monitoring device into different subdivided grids, then acquires the monitoring data from the monitoring device, assigns different colors or grayscale values ​​to different monitoring data, displays them in a rectangular pattern, and stores the colors or grayscale values ​​at different times in a historical matrix pattern. Finally, a convolutional neural network model ResNET is built and optimized using the historical matrix pattern. When real-time monitoring data is input into the ResNET model, the ResNET model outputs the corresponding color or grayscale value to the real-time matrix pattern, so that users can obtain the abnormal status of the monitoring device in real time. The analysis and monitoring method of this embodiment can intuitively display the abnormal monitoring data of the monitoring device, thereby facilitating users to locate the monitoring device with abnormalities in real time, and improving the efficiency of handling monitoring device failures. Attached Figure Description

[0068] Figure 1 This is a logic flowchart of an artificial intelligence-based underground pipeline anomaly data analysis and monitoring method according to an embodiment of this disclosure;

[0069] Figure 2 This is a mesh partitioning diagram of an embodiment of this disclosure;

[0070] Figure 3 for Figure 2 A magnified view of a portion of grid A6 shown;

[0071] Figure 4 This is a rectangular pattern layout diagram of an embodiment of this disclosure;

[0072] Figure 5 This is a flowchart illustrating the training process of the ResNET convolutional neural network model, an embodiment of this disclosure.

[0073] Figure 6 A flowchart illustrating a randomization method according to an embodiment of this disclosure;

[0074] Figure 7 This is a diagram of a data upload interface according to an embodiment of this disclosure;

[0075] Figure 8 A service registration interface diagram for an embodiment of this disclosure;

[0076] Figure 9 This is a diagram showing the oil pipe display interface according to an embodiment of this disclosure;

[0077] Figure 10 This is a diagram of an offline data loading interface according to an embodiment of this disclosure;

[0078] Figure 11 This is a diagram of the database monitoring interface according to an embodiment of this disclosure. Detailed Implementation

[0079] To enable those skilled in the art to better understand the technical solutions of this disclosure, the disclosure will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0080] This disclosure proposes an artificial intelligence-based underground pipeline anomaly data analysis and monitoring system, which includes: an infrastructure layer, an operating environment, a data layer, a support layer, a service layer, an application layer, and a user layer.

[0081] Specifically, the infrastructure layer includes: servers, storage devices, network switches, hardware firewalls, and monitoring facilities—basic hardware infrastructure that provides physical support; the operating environment includes: one Windows host and three Linux hosts built on the infrastructure as the platform and application operating environment; the data layer includes: SAN storage, spatial databases, and non-spatial databases, providing data storage functions for the service layer; the support layer includes: identity authentication, access control, log management, caching, load balancing, etc., providing software support throughout the system's operation; and the service layer includes: intelligent data monitoring services, monitoring data analysis services, data encryption services, data management services, and operation and maintenance management services, providing business support capabilities for the application layer. The service layer employs at least one server, which is used for intelligent monitoring and analysis of underground pipeline data within the grid, receiving and decrypting encrypted data transmitted by a multi-data tile processing encryption system. The multi-data tile processing encryption system uses a static multi-parameter adjustable encryption matrix data encryption method. The application layer includes intelligent monitoring, comprehensive display, data encryption, data management, data analysis, service management, and operation and maintenance management, providing application capabilities to the user layer. The user layer includes professionals in surveying, exploration, geotechnical engineering, and civil engineering who access the system via PC browsers and mobile apps.

[0082] The application layer includes an intelligent detection module, which is used to divide each monitoring device into different subdivided grids according to the map; the intelligent module is used to acquire monitoring data from the monitoring devices, assign different colors or grayscale values ​​to rectangular patterns for different monitoring data, and store them in a historical matrix pattern; the intelligent module is used to construct a convolutional neural network model ResNET, and train the convolutional neural network model ResNET according to the historical matrix pattern; the intelligent module is used to acquire monitoring data from real-time monitoring devices, input the real-time monitoring data into the convolutional neural network model ResNET, and output the corresponding colors or grayscale values ​​to the real-time matrix pattern.

[0083] The application layer also includes a data encryption module. This module is used to establish a composite model for each grid and form a raster data layer based on the geographic coordinates of the divided geographic regions. It then divides the smooth surface of the origin reference plane projected onto the raster data layer into multiple regions. The data encryption module generates encryption matrices in these regions using a multi-parameter static random number generation algorithm based on random geographic coordinates. Finally, it encrypts the raster data layer and the projections of random geographic coordinates onto the reference plane or smooth surface using the encryption matrix, forming primary encrypted data and an offset matrix. The data encryption module further encrypts the primary encrypted data using the offset matrix, generating a secondary encrypted map.

[0084] Another aspect of this disclosure proposes an artificial intelligence-based method for analyzing and monitoring abnormal data from underground pipelines, the method comprising:

[0085] S11: Divide each monitoring device into different subdivided grids according to the map;

[0086] S12: Acquire monitoring data from the monitoring device, assign different colors or grayscale values ​​to the rectangular pattern, and store them in the historical matrix pattern;

[0087] S13: Construct the convolutional neural network model ResNET and train the convolutional neural network model ResNET based on the historical matrix pattern;

[0088] S14: Obtain monitoring data from the real-time monitoring device, input the real-time monitoring data into the convolutional neural network model ResNET, and output the corresponding color or grayscale to the real-time matrix pattern.

[0089] According to the AI-based underground pipeline anomaly data analysis and monitoring method of this disclosure, the monitoring device is first divided into different subdivided grids. Then, the monitoring data of the monitoring device is acquired, and different colors or grayscale values ​​are assigned to different monitoring data and displayed in a rectangular pattern. The colors or grayscale values ​​at different times are stored in a historical matrix pattern. Finally, a convolutional neural network model ResNET is built and optimized using the historical matrix pattern. When real-time monitoring data is input into the ResNET model, the ResNET model outputs the corresponding color or grayscale value to the real-time matrix pattern, so that users can obtain the anomaly status of the monitoring device in real time. The analysis and monitoring method of this embodiment can intuitively display the abnormal monitoring data of the monitoring device, thereby facilitating users to locate the monitoring device with anomalies in real time, and improving the efficiency of handling monitoring device failures.

[0090] S11: Divide each monitoring device into different subdivided grids according to the map;

[0091] Specifically, the area where the underground pipelines are located is first divided into multiple zones, and then each zone is further divided into multiple subdivided grids. The underground pipelines include multiple pipes, such as water pipes, oil pipes, cables, and fiber optic cables. At least one type of monitoring device is installed on each pipe to monitor the data of that pipe. There can be multiple monitoring devices of this type, which are spaced apart on the pipe to monitor data at different locations of the pipe, thereby improving the accuracy of pipeline monitoring.

[0092] For example, each oil pipe can be equipped with multiple electronic noses at intervals to monitor the fuel gas molecule data; each oil pipe can be equipped with multiple second pressure gauges at intervals to monitor the pressure data within the oil pipe; electronic noses and second pressure gauges must be installed at the bends of each oil pipe to monitor the fuel gas molecule data and pressure data at the bends, preventing erosion from the diverting fluid. Similarly, each water pipe can be equipped with multiple fiber optic hydrophones at intervals to monitor the underwater acoustic data; each water pipe can be equipped with multiple first pressure gauges at intervals to monitor the pressure data. Likewise, each cable can be equipped with a surge generator and a reflected wave receiver, and underground optical cables can be equipped with WDM (Wavelength Division Multiplexing) combiners, filters, and optical time-domain reflectometers (OTDRs).

[0093] For the same type of monitoring device, the subdivision grid can be uniform or non-uniform. The purpose of subdivision grid is to determine the geographical location range of monitoring devices or abnormal monitoring data. Therefore, since the spatial distribution of various monitoring devices may be uneven, it is difficult to achieve uniformity in the subdivision grid. That is, most of the subdivision grid is non-uniform, while a small number of subdivision grids can achieve uniformity because some monitoring devices are evenly distributed in a certain area.

[0094] S12: Acquire monitoring data from the monitoring device, assign different colors or grayscale values ​​to the rectangular pattern, and store them in the historical matrix pattern;

[0095] Traverse all subdivided grids in all regions to obtain monitoring data from each monitoring device, that is, obtain monitoring data from each type of monitoring device for each pipeline, so as to monitor each pipeline of underground pipelines in the entire region, such as monitoring oil pipes, water pipes, cables and optical cables in the entire region.

[0096] The monitoring data from the monitoring devices can be arranged in a rectangular pattern according to a certain order, using colors or grayscale values. For example, firstly, monitoring data from the monitoring devices in the first row of subdivided grids within a certain area are acquired sequentially from east to west. Then, monitoring data from the monitoring devices in the second row of subdivided grids within each area are acquired sequentially from north to south, again from east to west. This process is repeated to acquire the monitoring data for each subdivided grid within the area. Following this sequence, monitoring data for each subdivided grid in each area of ​​the underground pipeline is acquired. Different types of monitoring devices are assigned colors or grayscale values ​​to the matrix pattern to display the data from different types of monitoring devices at different locations. The colors or grayscale values ​​corresponding to monitoring devices at different times are stored in the historical matrix pattern to ensure data consistency across different periods.

[0097] S13: Construct the convolutional neural network model ResNET and train the convolutional neural network model ResNET based on the historical matrix pattern.

[0098] S13 specifically includes the following steps:

[0099] S131: Construct the ResNET convolutional neural network model;

[0100] S132: Select anomaly monitoring data from the historical matrix pattern, randomly assign pixel values ​​of the anomaly monitoring data to blank pixels in the historical matrix pattern, and divide the historical matrix pattern into a training set and a validation set.

[0101] S133: Optimize the ResNET convolutional neural network model based on the training and validation sets.

[0102] First, we build a convolutional neural network model called ResNET.

[0103] Then, anomaly monitoring data is selected from the historical matrix pattern, and corresponding pixel values ​​are randomly assigned to the anomaly monitoring data. These pixel values ​​are then assigned to blank pixels. After all the selected anomaly monitoring data have their corresponding pixel values ​​assigned to blank pixels, a new historical matrix pattern is formed. Here, anomaly monitoring data refers to data whose monitoring data exceeds a monitoring threshold. The data in the new historical matrix pattern is divided into a training set and a validation set, with a ratio of 2 to 4:1. In this embodiment, a ratio of 3:1 is selected.

[0104] Specifically, for different monitoring devices, 0 to 11 abnormal monitoring data points are selected. Examples include: 0 to 11 abnormal monitoring data points for the electronic nose of oil pipes; 0 to 11 abnormal monitoring data points for the second pressure gauge of oil pipes; 0 to 11 abnormal monitoring data points for the fiber optic hydrophone of water pipes; 0 to 11 abnormal monitoring data points for the first pressure gauge of water pipes; 0 to 11 abnormal monitoring data points for the VDM multiplexer of underground optical cables; 0 to 11 abnormal monitoring data points for the filter of underground optical cables; 0 to 11 abnormal monitoring data points for the optical time domain reflectometer of underground optical cables; 0 to 11 abnormal monitoring data points for the surge generator of cables; and 0 to 11 abnormal monitoring data points for the reflected wave receiver of cables. The pixel values ​​corresponding to the abnormal monitoring data are randomly assigned to blank pixels in the historical matrix graphic. The 0 to 11 abnormal monitoring data points represent data exceeding thresholds of 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, and 100%, respectively. Here, pixel value refers to the different colors or grayscale values ​​corresponding to different monitoring data points.

[0105] Finally, the parameters of the ResNET convolutional neural network model are optimized based on the training and validation sets.

[0106] Specifically, the training set is input into the ResNET convolutional neural network model to train the ResNET convolutional neural network model, and the validation set is input into the ResNET convolutional neural network model. The ResNET convolutional neural network model is optimized based on the binary classification prediction results of the ResNET convolutional neural network model to optimize the parameters of the ResNET convolutional neural network model.

[0107] S133: Optimize the ResNET convolutional neural network model based on the training and validation sets, specifically including the following steps:

[0108] S1331: Construct the loss function L, and train the convolutional neural network model ResNET based on the training set and the loss function L;

[0109] S1332: Optimize the ResNET convolutional neural network model based on the validation set;

[0110] S1333: Based on the stable binary classification prediction results and loss function L values ​​on the validation set, the convolutional neural network model ResNET is determined.

[0111] Specifically, first, a loss function L is constructed. The training set is input into the ResNET convolutional neural network model. The ResNET model outputs binary classification prediction results through fully connected layers (FC) and a softmax function. The parameters of the ResNET model are optimized based on the binary classification prediction results and the loss function L. Then, the validation set is input into the ResNET model. The ResNET model outputs binary classification prediction results through fully connected layers (FC) and a softmax function. The parameters of the ResNET model are optimized based on the binary classification prediction results and the loss function L. When both the binary classification prediction results and the loss function L tend to stabilize, the ResNET model training is complete.

[0112] In some embodiments of this disclosure, the random assignment method in S132 includes:

[0113] S31: Select different abnormal monitoring data by using the target selection method to randomly assign different pixel values ​​to the different blank pixel numbers in the target historical rectangular pattern;

[0114] A historical matrix pattern of a selected region is used as the target historical matrix pattern. Blank pixels in the target historical matrix pattern are numbered; these blank pixels represent unassigned areas. A random number generator selects a blank pixel as the first number. Then, a random number generator in the target selection method randomly selects an anomaly detection data point as the first anomaly detection data point. The pixel value corresponding to the first anomaly detection data point is assigned to the blank pixel with the first number. The random number generator selects a number from the remaining blank pixel numbers as the second number. Then, a random number generator in the target selection method randomly selects the remaining anomaly detection data point as the second anomaly detection data point. The pixel value corresponding to the second anomaly detection data point is assigned to the blank pixel with the second number. This process is repeated until all anomaly detection data points have been assigned corresponding pixel values.

[0115] Among them, the target selection method is any one of the selection methods, and there is no special characteristic of the target selection method.

[0116] S32: Traverse all selection methods to select different anomaly monitoring data, and randomly assign different pixel values ​​to different blank pixel numbers;

[0117] Change the selection method and repeat step S31 until all selection methods are traversed. That is, assign the corresponding pixel value to all anomaly monitoring data through each selection method. After traversing all selection methods, the anomaly monitoring data in the target historical matrix pattern has been assigned the corresponding pixel value.

[0118] S33: Traverse all historical rectangular patterns of anomaly monitoring data, select different anomaly monitoring data, and randomly assign different pixel values ​​to different blank pixel numbers.

[0119] Select the remaining historical matrix patterns other than the target historical matrix pattern. Repeat steps S31 to S32 for each of the remaining historical matrix patterns to obtain the corresponding pixel value for the abnormal monitoring data in each of the remaining historical matrix patterns. That is, the pixel value corresponding to the abnormal monitoring data in all historical matrix patterns is assigned to the blank pixel number.

[0120] Specifically, the random assignment method steps S31 to S33 are as follows: Figure 5 As shown, firstly, a historical matrix pattern is selected from multiple historical matrix patterns. Then, blank pixels are assigned blank numbers. Using a random selection method, the middle abnormal monitoring data point (three abnormal values) is selected from three abnormal monitoring data points, and its corresponding pixel value is assigned to a blank number. Next, the random selection method selects the left abnormal monitoring data point from the remaining two abnormal monitoring data points, and its corresponding pixel value is assigned to the remaining blank numbers. Finally, the random selection method selects the last abnormal monitoring data point, i.e., the right abnormal monitoring data point, and its corresponding pixel value is assigned to a blank number not assigned by the middle or left abnormal monitoring data points. This process is repeated for all selection methods, assigning the pixel values ​​corresponding to all abnormal monitoring data points in the historical matrix pattern to blank numbers. Finally, this process is repeated for all historical matrix patterns, assigning the pixel values ​​corresponding to all abnormal monitoring data points in the region containing all historical matrix patterns to blank numbers.

[0121] By randomly assigning data, simulation sets of training and validation sets of historical matrix patterns can be constructed, ensuring sufficient data for training and guaranteeing the accuracy of the ResNET model.

[0122] S14: Obtain monitoring data from the real-time monitoring device, input the real-time monitoring data into the convolutional neural network model ResNET, and output the corresponding color or grayscale to the real-time matrix pattern.

[0123] Once the ResNet convolutional neural network model is trained, it can display the specific location of real-time monitoring data. Real-time monitoring data is input into the ResNet model, which outputs the corresponding color or grayscale value to a real-time matrix pattern to display the data at the designated location. This allows users to observe whether there are any anomalies in the monitoring data in real time. Specifically, the monitoring data may be normal or abnormal. When normal monitoring data is input into the ResNet model, the ResNet model outputs the corresponding color or grayscale value; when abnormal monitoring data is input into the ResNet model, the ResNet model outputs the corresponding color or grayscale value.

[0124] In some embodiments of this disclosure, the monitoring method further includes: S20, a method for encrypting static data, wherein S20: encrypting static data specifically includes the following steps:

[0125] S21: Based on the geographic coordinates of the geographic regions that have been divided into grids, establish a composite model for each grid and form a raster data layer. Divide the smooth surface of the origin reference plane that projects over the raster data layer into multiple regions.

[0126] First, the geographic coordinate system E, which is divided into multiple regions, is determined; then, a composite model is built for each region through the modeling subsystem, in which the raster data layer determines the origin reference plane of the multiple coordinate systems and the projection on the reference plane is completely covered.

[0127] S22: Generate encryption matrices in multiple regions based on a multi-parameter static random number generation algorithm using random geographic location coordinates;

[0128] Specifically, S22 includes the following steps:

[0129] S221: Set the random number generation algorithm:

[0130] rand(x,y,w)∈(-1,1);

[0131] Where x and y are the random geographic coordinates of the projection of the raster data layer onto the smooth surface in coordinate system E at the time of request, and w is a preset parameter. For a given value of w, x and y are the geographic coordinates of a randomly selected server (not at least one) on the smooth surface. Given x and y, the generated random values ​​are the same.

[0132] w is determined according to the following formula:

[0133]

[0134] Where w represents linear, nonlinear, and mixed linear-nonlinear combinations, g = 1, 2, 3, st is a status indicator; st not displayed indicates the presence of the parameter, st = 0 indicates its absence, R is the area of ​​the encrypted geographic data range, W and H are the width and height of the encryption matrix corresponding to randomly selected geographic coordinates other than those on the aforementioned servers, ID is the additional code, and S... x S is the allowable offset on the X-axis. y This is the allowable offset for the Y-axis. Projecting each region onto the reference plane

[0135] Average length, The average width of each region projected onto the reference plane, where Xo and Yo are preset coordinates in the projection;

[0136] Understandably, the area R of the encrypted geographic data range is determined by the size of the raster data layer. It can be the entire raster data layer or a portion, such as a pre-defined range around a pipeline segment of concern for disaster relief, maintenance, or quality inspection. The area where the encryption matrix corresponding to x and y is located encompasses at least one of the defined regions, the range of which is user-defined and belongs to the region of interest (ROI) within the encrypted geographic data range. When at least one server makes a request, its actual coordinates may be within the region where the encryption matrix is ​​located (e.g., near x and y) or not. The average area is the projected area of ​​the smooth surface onto the reference plane divided by the number of defined regions.

[0137] S x S y Satisfy the following formula:

[0138]

[0139] Where r is the spatial resolution of the raster data layer, k is negatively correlated with the number of regions, k = NC(num), num is the number of regions, k ∈ [1, K], K > 1, and when the number of regions is 1, k = K. The number of regions is such that the average area projected onto the reference plane by each region is qr. 2 When q∈(1,2], k=1;

[0140] After obtaining w, the average value of the derivatives at p points on each boundary of each region in the multiple regions is defined as follows:

[0141]

[0142]

[0143] Where x j →x and y j→y is defined as the point x on each boundary. j ,y j It moves towards an adjacent pixel in both the x and y directions along the X and Y axes, respectively.

[0144] Specifically, the derivative calculation for points on the grid region boundary corresponding to the projection of multiple regions onto the reference plane is performed using the existing X and Y direction derivatives. For example, if the grid region boundary is a rectangular edge, then for a sub-rectangle located at the corner of the rectangle, there are two boundaries at the corresponding corner vertex, and the derivative at the vertex of that corner is calculated using the existing X and Y direction derivatives respectively. When the corner of the grid region boundary is a non-rectangular corner, it is treated as a two-boundary division with the midpoint of the boundary length, and the existing X and Y direction derivatives are calculated accordingly. In this case, the corner sub-region still has a maximum of eight sub-regions. and parameter.

[0145] S222: Fixed C, fixed w and NC values, i.e., fixed C. g and NC value, and change s x s y , The values ​​of Xo and Yo are used to obtain the value of k. d Thus, we obtain k d Related w proc ;

[0146] Where, k d The following conditions must be met:

[0147]

[0148]

[0149] |s x +k D rand(x, y, w)| (x，y)∈R |∈(2|r|,2k D |r|],|s y +k D rand(x, y, w)| (x，y)∈R |∈(2|r|,2k D |r|];

[0150] via k d The intermediate value w can be obtained. proc .

[0151] According to w proc To obtain the average of the derivatives at p points on each boundary of each region:

[0152]

[0153]

[0154] Where i = 1, 2, 3, 4, p ∈ [1, +∞), b j Let j be a point on boundary b, and j = 1, 2, 3, 4, w proc =C gpro c);

[0155] S223: Calculate Wd using the following formula and form the encryption matrix JM:

[0156]

[0157] JM = k D rand(x, y, w) D );

[0158] Where g = 1, 2, 3; G = 11, 22, 33; C G Similarly, for linear and / or nonlinear combinations, G = 11, 22, 33 represent linear, nonlinear, and mixed linear and nonlinear combinations, respectively, abbreviated as C. 11 C 22 and C 33 ,and

[0159] It should be understood that when calculating the above limits or summation of limits, the coordinate values ​​x and y are the coordinate values ​​of points on the partition boundary b, not the locations of at least one server. However, the coordinate values ​​x and y in the encryption matrix, defined by the multi-parameter random number generation algorithm, represent random locations that are not at least one server. In reality, there are eight times the number of parameters involved in the combination of the division regions. To keep the writing concise, we will not use additional subscripts to represent the average derivatives of different division regions.

[0160] Understandably, on the one hand, a smooth surface, s x s y X o Y o C1, C2, C3 p, Modifications to these parameters will cause w to differ; on the other hand, under offset constraints, boundary constraints, and R-domain constraints, due to As w changes, k also changes negatively. Therefore, these two factors cause the encryption matrix JM to change jointly with w and k, increasing the difficulty of cracking the encryption matrix JM. A change in k also means a change in s. x s y The range of values ​​thus changes. Here, the negative correlation function NC is an independent variable, while X... oY o Let be independent variables, the width Wid and height H of the encryption matrix be requested variables, the rectangular area R of the grid region be a preset value, and p∈[1,+∞) be an independent variable. and As p changes, the selected smooth surface is also an independent variable. The calculation result differs depending on the function form of rand(x,y,w). Therefore, rand(x,y,w) itself is an independent variable, and the combination w = C. g g = 1, 2, 3 and C G G = 11, 22, 33 are each three independent variables.

[0161] Therefore, for a given R, W, H, and ID code, the method uses a set of independent variables. There are actually 16 independent variables, which is the set of variables needed to decrypt the code. There are 20 variables, and considering and The construction of the encryption matrix requires all parameter sets to be known. There are 20+8mun parameters. The generation of the encryption matrix depends on a maximum of 20+8mun parameters. Adjusting any one of these parameters will cause a change in the random encryption value of the encryption matrix, giving the encryption matrix a fairly high resistance to reverse decoding.

[0162] The encryption matrix can be adjusted by parameters to control its detailed variations, ensuring a rich range of nuances. The finer the region is divided, the less important the smoothness of its internal components becomes. This is because, firstly, finer divisions bring the internal dimensions closer to the resolution, making distortion less visually perceptible; secondly, finer divisions create more boundaries, making distortion more dependent on the smoothness of those boundaries. Furthermore, since k is close to 1 and the absolute value of rand(x,y,w) is less than 1, the overall internal offset is guaranteed to remain below the resolution, thus minimizing the impact on smoothness. Even with a smaller number of divisions, larger internal offsets may occur, but due to the larger region area, the offset and distortion may still appear visually acceptable. However, if visually perceived as unreasonable, a suitable s can be selected during the second-level encryption process. x s y To adjust the offset.

[0163] The coarser the division, the smoother the internal region remains, as it is guaranteed by a smooth surface. Therefore, even with the k-fold rand(x,y,w) offset generated after the first-level encryption and the offset generated after the second-level encryption, the smoothness of the internal region remains. The offset within a resolution multiple of s. Due to the negative correlation, although k is large and the absolute value of rand(x,y,w) is less than 1, the encryption matrix may also be large. However, a small offset can still be chosen during secondary encryption (i.e., according to s). x s y (Selection) thus ensures that the overall smoothness is not greatly affected, and at this time, as long as the boundary is smooth, the encryption matrix can also be considered smooth.

[0164] Therefore, in summary, this invention only selects to calculate the directional derivatives on the X and Y sides of important boundaries. If the directional derivatives exist, it is considered smooth, or a suitable s is selected during the second-level encryption process. x s y Adjusting the offset to achieve a satisfactory smoothness greatly reduces the computational load.

[0165] It's important to emphasize that if the number randomly generated by rand(x,y,w) is very small (and its probability is also small), the smoothness within the region is less affected. Although this might result in the first-level encryption offset being smaller than the resolution, or the encryption matrix still being relatively large, a suitable s can still be chosen during the second-level encryption process. x s y Ensure that digital map images can produce appropriate offsets and distortions.

[0166] S23: Encrypt the raster data layer and the projection of random geographic location coordinates onto the reference plane or smooth surface using an encryption matrix, and form first-level encrypted data and an offset matrix;

[0167] Specifically, S23 includes the following steps:

[0168] S231: Project the raster data layer and the user's positioning coordinates onto a smooth surface to obtain...

[0169]

[0170] Where X1 and Y1 are the coordinates of the raster data layer after projection onto the smooth surface, and x and y are the coordinates of the raster data layer before projection onto the smooth surface, respectively. ap and Y ap These are the user's location coordinates projected onto the smooth surface, x and x. ap and y ap These are the user's location coordinates before being projected onto the smooth surface;

[0171] S232: Connect X1 and X ap Superimpose Y1 and Y ap By superimposing the data, we obtain the first-level encrypted data:

[0172]

[0173] Offset matrix:

[0174]

[0175] The coordinate system changes from E to E(1);

[0176] It should be understood that at this point, the smooth surface is also subjected to first-level encryption, forming a first-level encrypted smooth surface.

[0177] S233: Translate the raster data layer before encryption along with the coordinate system E so that E and E (1) Data on the first-level encrypted smooth surface is merged and deleted to form a first-level encrypted raster data layer M1.

[0178] It is evident that the location of at least one server cannot be decrypted due to the superposition of random locations.

[0179] S24: Perform secondary encryption on the primary encrypted data using an offset matrix, and generate a secondary encrypted map.

[0180] Specifically, S24 includes the following steps:

[0181] S241: Using the offset matrix M s The first-level encrypted data J1 is then subjected to second-level encryption to generate second-level encrypted data:

[0182]

[0183] It should be understood that at this point, the smooth surface of the first-level encryption is also subjected to second-level encryption, forming a second-level encrypted smooth surface. When the first-level encryption produces an unsatisfactory offset, it can still be corrected during the second-level encryption process by adjusting s. x s y (Thus JM also changes) to ensure that the digital map can produce appropriate offsets and distortions.

[0184] S242: Connect the first-level encrypted raster data layer M1 with the coordinate system E (1) The coordinate system E is translated together. (1) With E (2) Data on the secondary smooth surface is merged and deleted to form a secondary encrypted raster data layer M2;

[0185] S243: Preset additional key A∈(0,1), calculate E (2) With E (1) The length L of the line segment connecting the origins of coordinates E is used to connect the secondary encrypted raster data layer M2 along the line segment with the coordinate system E. (2) Translate together by a distance AL such that E (2) The origin of the coordinate system is close to E. (1)The origin of the coordinate system is used to obtain the final encrypted raster data layer M. f ;

[0186] S244: Based on M1, M2, M f If the offset of the encrypted matrix area corresponding to the location requested by the user is within a preset range, then at least one of them will be returned to the user for display.

[0187] According to M1, M2, M f If the offset in the encryption matrix region corresponding to the location requested by the user is outside the preset range, repeat the encryption steps until M1, M2, and M are obtained. f If the offset of at least one of the encrypted matrix regions corresponding to random positions x and y is within a preset range, select at least one of these at least one and return it to the user for decryption and display.

[0188] After introducing the additional key A∈(0,1), the set of independent variables in the method becomes There are actually 17 independent variables, so the set of variables needed to decrypt the code becomes... There are 21 variables, and considering and The set of all parameters required to construct the encryption matrix then becomes:

[0189]

[0190] There are 21+8mun parameters.

[0191] It is understandable that the additional code is also another way to adjust the offset.

[0192] Example 1

[0193] An artificial intelligence-based underground pipeline anomaly data analysis and monitoring system includes an infrastructure layer: including servers, storage devices, network switches, hardware firewalls, and basic monitoring hardware facilities, which provide physical support.

[0194] Service Layer: Provides intelligent data monitoring services, monitoring data analysis services, data encryption services, data management services, and operation and maintenance management services to the application layer. The service layer is implemented using at least one server. This server is used for intelligent data monitoring and analysis of underground pipeline data within the grid, receiving and decrypting encrypted data transmitted from the multi-data tile processing encryption system, and transmitting re-encrypted data and information to the user terminal. The multi-data tile processing encryption system uses a static multi-parameter adjustable encryption matrix data encryption method.

[0195] Application layer: Includes intelligent monitoring, comprehensive display, data encryption, data management, data analysis, service management, and operation and maintenance management, providing application capabilities for the user layer.

[0196] The intelligent data monitoring includes fusing data from utility tunnels, digital maps of underground pipeline distribution, and monitoring device data. This process embeds the utility tunnel data and monitoring device data into the digital map of underground pipeline distribution in real time, enabling the acquisition of real-time information on any underground pipeline within each grid, as well as intelligent visualization and identification of monitoring device data.

[0197] like Figures 2 to 6 As shown, the mountainous area is divided into six grids, A1-A6. Within each grid, all underground oil and water pipes exist as single, intact lines; no single pipe spans multiple grids. Grid A6 represents a relatively densely populated geographical area.

[0198] A low-altitude drone (not shown in the figure) equipped with a high-definition camera collects the modeling data required for raster data modeling of mountain grids, and forms a raster data model.

[0199] The at least one server, such as Figure 2 The image shows the main server in grid A3 and the sub-servers in grid A6. The main server receives various types of modeling data transmitted from the sub-servers, and can also independently establish raster data models. Like all sub-servers within the grid, it can also monitor and analyze underground pipeline data within all grids, receive encrypted data transmitted by the multi-data tile processing encryption system and decrypt the data, and transmit re-encrypted data and information communication to the user terminal.

[0200] This embodiment features an underground pipeline digital map data acquisition and modeling system, used to collect the data needed to create a digital map of underground pipeline distribution and for 3D reconstruction. Specifically, it includes remote sensing satellites, aircraft equipped with high-altitude point cloud scanning systems, and low-altitude drones equipped with high-definition cameras, as well as... Figure 2 The modeling subsystem is exemplified by mesh A6 in the modeling subsystem.

[0201] The platform's multi-data tile processing encryption system is divided into subsystems in each grid. Figure 2 The example provides a multi-data tile processing encryption system subsystem in grid A6, which can be used to perform multi-data tile processing on underground pipeline data in grid A6, store the data, encrypt the data based on the grid and the static multi-parameter adjustable encryption matrix to form encrypted data, and transmit the encrypted data to the sub-server in this grid.

[0202] The diagram illustrates the partial distribution of oil pipes, water pipes, cables, and optical cables in grid A6.

[0203] This embodiment 1 will describe a method for abnormal data analysis and monitoring of monitoring device data. The example of grid A6 in embodiment 1 will be used for further explanation.

[0204] like Figure 3 As shown, Figure 2 This is a magnified view of the partial distribution of oil pipes, water pipes, cables, and optical fibers within the rectangular area shown in grid A6. For oil pipes, a, b, c, and d represent the midpoint numbers along the oil flow direction of four interconnected oil pipes. Figure 3 The oil pipes near the small and medium circles are equipped with electronic noses and second pressure gauges as monitoring devices at every other oil pipe. The acquired fuel gas molecule and pressure data are used as monitoring data for the oil pipes. Electronic noses and second pressure gauges are essential at bends because bends are prone to erosion from the changing fluid, requiring extra monitoring.

[0205] Similarly, for the water pipes, where ei represents the number of the midpoint of the five interconnected pipes along the direction of water flow. A fiber optic hydrophone is installed at one-third of the length of each pipe, and a first pressure gauge serves as the monitoring device. For the east-west oriented cables and optical fibers in the diagram, a surge generator and a reflected wave receiver are used as monitoring devices, respectively. A WDM multiplexer, filter, and OTDR are used to detect underground optical fibers, and one and two anomalies were detected, respectively.

[0206] Taking oil pipes and water pipes as examples, the distribution areas of both are divided into rectangular subdivision grids (divided in an approximately uniform manner), and the monitoring devices are all located within a subdivision grid. The upper right vertex of the subdivision grid is used as the scanning point on which the rectangular pattern arrangement order is based to determine the data acquisition order of the monitoring devices.

[0207] Taking a scan from east to west, then from north to south as an example, for oil pipes, the grid sequence for the first pressure gauge containing a fiber optic hydrophone is subdivided into grids numbered c and a, and a1, b1, c1, d1, e1, f1, g1; for water pipes, it is h, g, f, e, i. Therefore, color values ​​are assigned to data from various monitoring devices, including abnormal data, to obtain the color values ​​corresponding to the real-time monitoring data. According to... Figure 4 As shown, the abnormal data of oil pipes, water pipes, cables, and optical cables are assigned color values ​​to pixels in the order shown, and then arranged into rectangular patterns. As indicated by the arrow, the arrangement is from left to right and then from top to bottom, and blank (i.e., white) pixels are filled to form a 500*500 rectangular pattern.

[0208] Based on the above-described rectangular pattern formation method, the method for abnormal data analysis and monitoring of monitoring device data in this embodiment 1 includes the following steps:

[0209] S11: Divide each monitoring device into different subdivided grids according to the map;

[0210] S12: Acquire monitoring data from the monitoring device, assign different colors or grayscale values ​​to the rectangular pattern, and store them in the historical matrix pattern;

[0211] S13: Construct the convolutional neural network model ResNET and train the convolutional neural network model ResNET based on the historical matrix pattern;

[0212] S13: Construct the convolutional neural network model ResNET and train the convolutional neural network model ResNET based on the historical matrix pattern;

[0213] For the monitoring devices of the aforementioned oil pipes, water pipes, cables, and optical fibers, abnormal data monitoring values ​​are defined. 0-11 abnormal data points are selected, and the corresponding pixel values, grayscale values, or color values ​​are randomly assigned to a historical rectangular pattern. Each abnormal data point is selected from values ​​exceeding the normal value by 5%, and values ​​that exceed the normal value by 10%-100% in 10 percentage point increments. That is, as... Figure 5 As shown, values ​​exceeding the normal value by 5%, and values ​​exceeding the normal value by 10%-100% at 10% intervals, are selected as follows: 0 (if the selected historical matrix pattern itself contains abnormal values), 1, 2, 3, and 4 very positive data points. A total of [number missing] such data points are present. There are four selection methods (the first is that all abnormal values ​​are different, the second is that two are the same, the third is that three are the same, and the fourth is that four are the same). In each selection method, the pixel value corresponding to the selected abnormal value is randomly assigned to a blank pixel in the historical rectangular pattern. Then, the historical rectangular pattern with the randomly assigned pixel values ​​is divided into a training set and a validation set (ratio of 3:1). The training set is input into the input terminal of ResNet, and the output terminal is input into a fully connected FC network and then into SOFTMAX to form a binary classification prediction result. A loss function L is constructed, and the prediction accuracy acc is verified using the validation set. The ResNet network parameters are optimized until the values ​​of acc and L stabilize, thus completing the training.

[0214] The random assignment method is as follows:

[0215] The first step is to select a historical rectangular pattern, number its blank pixels, and assign the corresponding pixel value of an abnormal value selected by the random machine to the blank pixel selected by the random machine.

[0216] The second step involves randomly selecting one number from the remaining numbers, and then randomly selecting another abnormal value from the remaining abnormal values ​​in the corresponding selection method. The corresponding pixel value is then assigned to the blank pixel of the number randomly selected by the machine.

[0217] The third step is to repeat step two until all abnormal values ​​in the corresponding selection method have been assigned pixel values.

[0218] Fourth step: Select the same historical rectangular pattern before assigning pixel values ​​in the first step, and then change to another selection method. Repeat steps one through three until all selection methods have been changed.

[0219] Step 5: Repeat steps 1 through 4 until all historical rectangular pattern pixel values ​​have been assigned. The order in which the selection methods are changed and the initial selection method are not strictly defined; simply iterate through all selection methods. For example... Figure 6 The diagram shows a process flow that first selects three abnormal values, then two abnormal values, and then iterates through all selection methods and all historical rectangular patterns using a randomly assigned method. The numbers 1-4 represent the process guidance sequence numbers, each indicating the flow path from the arrow pointing to one type of arrow to the next type of arrow after completing the current one.

[0220] The purpose of the above-mentioned random assignment method is to construct a simulation set for training and validation. In reality, the possibility of abnormal pipeline states is small, and the number of historical rectangular patterns cannot be obtained in a short time for training. The simulation set is used to solve the problem of the difficulty in obtaining a large amount of data in a short time.

[0221] This embodiment 1 will describe a data encryption method based on a statically multi-parameter adjustable encryption matrix, including the following steps:

[0222] S21: Determine the geographic coordinate system E of the mountainous area in Implementation 1, which is divided by grids; for each grid, use the modeling subsystem to create a composite model, and for the raster data layer therein, determine the reference plane passing through the origin of the coordinate system and the smooth surface whose projection on the reference plane completely covers the reference plane, and divide the smooth surface into multiple regions.

[0223] S22: According to the data transmission request, set a multi-parameter static random number generation algorithm based on random geographical coordinates, and generate encryption matrices in the multiple regions based on the algorithm, so that the boundaries of the encryption matrix in each region are continuous and smooth;

[0224] S23: According to the request, the raster data layer and the projection of random geographic location coordinates onto the reference plane or smooth surface are encrypted using an encryption matrix to form first-level encrypted data, and an offset matrix is ​​formed.

[0225] S24: After performing secondary encryption on the primary encrypted data using the offset matrix, a secondary encrypted map is generated and transmitted to the at least one server.

[0226] S22 includes S221: Let the multi-parameter random number generation algorithm rand(x,y,w)∈(-1,1), where x and y are the random geographical coordinates of the projection of the raster data layer on the smooth surface under E when at least one server requests it, and w is a preset parameter. The random number is related to x, y, and w. For a given w value, the generated random values ​​are the same for given x and y.

[0227] Still using grid A6 as an example, such as Figure 2 The actual width W and height H corresponding to the area R, x, y of the rectangle are used as the width W and height H of the encryption matrix, with an allowable offset s. x ,s y All are 2m. If the smooth surface is divided into multiple rectangular regions, and the average length and average width of each region projected onto the smooth surface are also 120m, and the preset coordinates Xo and Yo in the projection are one or more resolutions of 2m offset from the geometric center of each region in the X-north direction, and the registered user's ID code is 000000001, then w is the arithmetic sum of the above 10 data points, where s x ,s y satisfy When k = 1801, To obtain w, the average of the derivatives of each region across multiple regions is calculated using the formula: (1) the derivatives of each region at three points (boundary endpoints and midpoint).

[0228] S222: Select the arithmetic and combination methods of Cg and the NC value, and obtain the result based on the NC value. Constantly changing x s y , Xo, Yo, obtain the k value mun D For based on R and The number of determined regions ( [] indicates rounding), so that on the boundary b of the plurality of regions Both exist, and Thus, the intermediate w is obtained. proc The value is then calculated by averaging the derivatives at three points (the two endpoints and the midpoint) on each boundary of each region across multiple regions.

[0229] S223: Calculate s x s y , Xo, Yo, and all partitioned regions and Arithmetic and w D To form an encryption matrix

[0230] S23 specifically includes:

[0231] S231: The raster data layer created in Example 1 and the user's location coordinates (x) ap ,y ap The projection onto the smooth surface, i.e., the reference plane itself, is used for refinement:

[0232]

[0233] X1 and Y1 are the coordinates of the raster data layer after projection onto the smooth surface, respectively, while x and y are the coordinates of the raster data layer before projection onto the smooth surface, respectively. ap and Y ap These are the user's location coordinates projected onto the smooth surface, x and x. ap and y ap These are the user's location coordinates before being projected onto the smooth surface.

[0234] S232: Superimpose X1 and Xap, and superimpose Y1 and Yap to obtain the first-level encrypted data:

[0235]

[0236] Obtain the offset matrix:

[0237] And at this time, the coordinate system E becomes E (1) ;

[0238] S233: Translate the raster data layer before encryption along with the coordinate system E so that E and E (1) Data on the first-level encrypted smooth surface is merged and deleted to form a first-level encrypted three-dimensional digital map M1.

[0239] S24: Specifically includes:

[0240] S241: Using the offset matrix M s The first-level encrypted data J1 is then subjected to second-level encryption to generate second-level encrypted data:

[0241]

[0242] And at this time, the coordinate system E becomes E (2) ;

[0243] S242: Connect the first-level encrypted raster data layer M1 with the coordinate system E (1)The coordinate system E is translated together. (1) With E (2) Data on the secondary smooth surface is merged and deleted to form a secondary encrypted raster data layer M2;

[0244] S243: Preset additional code A = 0.6, calculate E (2) With E (1) The length L of the line segment connecting the origins of coordinates E is used to connect the secondary encrypted raster data layer M2 along the line segment with the coordinate system E. (2) They were translated together by a distance of 0.6L so that E (2) The origin of the coordinate system is close to E. (1) The origin of the coordinate system is used to obtain the final encrypted raster data layer M. f ;

[0245] S244: The positions of the encrypted data and the original data are offset, but the visual distortion of the geometric shape is very small, which meets the offset requirement. M f Return to the user's device for decryption and display.

[0246] This embodiment 1 provides a typical interface display of some functional modules in the application layer. Figures 7-11 The interface displays the data upload interface, service registration interface, mountain oil pipeline display interface, App offline data loading interface, and database monitoring interface of the AI ​​underground pipeline anomaly data analysis and monitoring system.

[0247] It is understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of this disclosure, and this disclosure is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and substance of this disclosure, and these modifications and improvements are also considered to be within the scope of protection of this disclosure.

Claims

1. A method for analyzing and monitoring abnormal data of underground pipelines based on artificial intelligence, characterized in that, The analytical monitoring methods include: S11: Divide each monitoring device into different subdivided grids according to the map; S12: Acquire monitoring data from the monitoring device, assign different colors or grayscale values ​​to the rectangular pattern, and store them in the historical matrix pattern; S13: Construct the convolutional neural network model ResNET and train the convolutional neural network model ResNET based on the historical matrix pattern; S14: Obtain monitoring data from the real-time monitoring device, input the real-time monitoring data into the convolutional neural network model ResNET, and output the corresponding color or grayscale to the real-time matrix pattern.

2. The method for analyzing and monitoring abnormal data of underground pipelines based on artificial intelligence according to claim 1, characterized in that, The construction of the ResNET convolutional neural network model, and the training of the ResNET convolutional neural network model based on historical matrix patterns, specifically includes: S131: Construct the ResNET convolutional neural network model; S132: Select anomaly monitoring data from the historical matrix pattern, randomly assign pixel values ​​of the anomaly monitoring data to blank pixels in the historical matrix pattern, and divide the historical matrix pattern into a training set and a validation set. S133: Optimize the ResNET convolutional neural network model based on the training and validation sets.

3. The method for analyzing and monitoring abnormal data of underground pipelines based on artificial intelligence according to claim 2, characterized in that, The optimization of the ResNET convolutional neural network model based on the training and validation sets specifically includes: S1331: Construct the loss function L, and train the convolutional neural network model ResNET based on the training set and the loss function L; S1332: Optimize the ResNET convolutional neural network model based on the validation set; S1333: Based on the stable binary classification prediction results and loss function L values ​​on the validation set, the convolutional neural network model ResNET is determined.

4. The method for analyzing and monitoring abnormal data of underground pipelines based on artificial intelligence according to claim 2, characterized in that, The random assignment method includes: S31: Select different abnormal monitoring data by using the target selection method to randomly assign different pixel values ​​to the different blank pixel numbers in the target historical rectangular pattern; S32: Traverse all selection methods to select different anomaly monitoring data, and randomly assign different pixel values ​​to different blank pixel numbers; S33: Traverse all historical rectangular patterns of anomaly monitoring data, select different anomaly monitoring data, and randomly assign different pixel values ​​to different blank pixel numbers.

5. The method for analyzing and monitoring abnormal data of underground pipelines based on artificial intelligence according to claim 1, characterized in that, The monitoring method further includes a method for encrypting static data, wherein the method for encrypting static data includes: S21: Based on the geographic coordinates of the geographic regions that have been divided into grids, establish a composite model for each grid and form a raster data layer. Divide the smooth surface of the origin reference plane that projects over the raster data layer into multiple regions. S22: Generate encryption matrices in multiple regions based on a multi-parameter static random number generation algorithm using random geographic location coordinates; S23: Encrypt the raster data layer and the projection of random geographic location coordinates onto the reference plane or smooth surface using an encryption matrix, and form first-level encrypted data and an offset matrix; S24: Perform secondary encryption on the primary encrypted data using an offset matrix, and generate a secondary encrypted map.

6. The method for analyzing and monitoring abnormal data of underground pipelines based on artificial intelligence according to claim 5, characterized in that, The multi-parameter static random number generation algorithm based on random geographical coordinates generates an encryption matrix in multiple regions, specifically including: S221: Set the random number generation algorithm: rand(x,y,w)∈(-1,1); Where x and y are the random geographic coordinates of the projection of the raster data layer onto the smooth surface in coordinate system E, and w is determined according to the following formula: Where g = 1, 2, 3, st is a status indicator; st not displayed indicates the presence of the parameter, st = 0 indicates its absence, R is the area of ​​the encrypted geographic data range, W and H are the width and height of the encryption matrix corresponding to randomly selected geographic coordinates other than those on the aforementioned servers, ID is the additional code, and S... x S is the allowable offset on the X-axis. y This is the allowable offset for the Y-axis. The average length of the projection of each region onto the reference plane. The average width of each region projected onto the reference plane, where Xo and Yo are preset coordinates in the projection; S x S y Satisfy the following formula: Where r is the spatial resolution of the raster data layer, k is negatively correlated with the number of regions, k = NC(num), num is the number of regions, k ∈ [1, K], K > 1, and when the number of regions is 1, k = K. The number of regions is such that the average area projected onto the reference plane by each region is qr. 2 When q∈(1,2], k=1; S222: Fix Cg and NC values, and change s x s y , The values ​​of Xo and Yo are used to obtain the value of k. d To obtain the intermediate value w of w. proc ; Where, k d The following conditions must be met: According to w proc To obtain the average of the derivatives at p points on each boundary of each region: Where i = 1, 2, 3, 4, p ∈ [1, +∞), b j Let w be a point on boundary b, and j = 1, 2, ..., p. proc =C gproc ; S223: Calculate Wd using the following formula and form the encryption matrix JM: JM=k D ·rand(x,y,w D ); Where g = 1, 2, 3; G = 11, 22, 33; C G Similarly, for linear and / or nonlinear combinations, G = 11, 22, 33 represent linear, nonlinear, and mixed linear and nonlinear combinations, respectively, abbreviated as C. 11 C 22 and C 33 ,and 7. The method for analyzing and monitoring abnormal data of underground pipelines based on artificial intelligence according to claim 6, characterized in that, The encryption process involves using an encryption matrix to encrypt the raster data layer and the projection of random geographic location coordinates onto a reference plane or smooth surface, forming first-level encrypted data and an offset matrix. Specifically, this includes: S231: Projecting the raster data layer and the user's location coordinates onto a smooth surface yields the following formula: Where X1 and Y1 are the coordinates of the raster data layer after projection onto the smooth surface, and x and y are the coordinates of the raster data layer before projection onto the smooth surface, respectively. ap and Y ap These are the user's location coordinates projected onto the smooth surface, x and x. ap and y ap These are the user's location coordinates before being projected onto the smooth surface; S232: Connect X1 and X ap Superimpose Y1 and Y ap By superimposing the data, we obtain the first-level encrypted data: Offset matrix: The coordinate system changes from E to E(1); S233: Translate the raster data layer before encryption along with the coordinate system E so that E and E (1) Data on the first-level encrypted smooth surface is merged and deleted to form a first-level encrypted raster data layer M1.

8. The method for analyzing and monitoring abnormal data of underground pipelines based on artificial intelligence according to claim 7, characterized in that, The process of performing secondary encryption on the first-level encrypted data using an offset matrix to generate a second-level encrypted map specifically includes: S241: Using the offset matrix M s The first-level encrypted data J1 is then subjected to second-level encryption to generate second-level encrypted data: S242: Connect the first-level encrypted raster data layer M1 to the coordinate system E (1) Translation, so that coordinate system E (1) With E (2) Data on the secondary smooth surface is merged and deleted to form a secondary encrypted raster data layer M2; S243: Preset additional key A∈(0,1), calculate E (2) With E (1) The length L of the line segment connecting the origins of coordinates E is used to connect the secondary encrypted raster data layer M2 along the line segment with the coordinate system E. (2) Translate together by a distance AL such that E (2) The origin of the coordinate system is close to E. (1) The origin of the coordinate system is used to obtain the final encrypted raster data layer M. f ; S244: Based on M1, M2, M f If the offset of the encrypted matrix area corresponding to the location requested by the user is within a preset range, then at least one of them will be returned to the user for display. According to M1, M2, M f If the offset in the encryption matrix region corresponding to the location requested by the user is outside the preset range, repeat the encryption steps until M1, M2, and M are obtained. f If the offset of at least one of the encrypted matrix regions corresponding to random positions x and y is within a preset range, select at least one of these at least one and return it to the user for decryption and display.

9. An artificial intelligence-based underground pipeline anomaly data analysis and monitoring system, wherein the artificial intelligence-based underground pipeline anomaly data analysis and monitoring system is used to execute the artificial intelligence-based underground pipeline anomaly data analysis and monitoring method according to any one of claims 1 to 8, wherein the artificial intelligence-based underground pipeline anomaly data analysis and monitoring system comprises: Infrastructure layer, operating environment, data layer, support layer, service layer, application layer, and user layer; The application layer includes an intelligent detection module, which is used to divide each monitoring device into different subdivided grids according to the map; the intelligent module is used to acquire monitoring data from the monitoring devices, assign different colors or grayscale values ​​to rectangular patterns for different monitoring data, and store them in a historical matrix pattern; the intelligent module is used to construct a convolutional neural network model ResNET, and train the convolutional neural network model ResNET according to the historical matrix pattern; the intelligent module is used to acquire monitoring data from real-time monitoring devices, input the real-time monitoring data into the convolutional neural network model ResNET, and output the corresponding colors or grayscale values ​​to the real-time matrix pattern.

10. The underground pipeline anomaly data analysis and monitoring system according to claim 9, characterized in that, The application layer also includes a data encryption module. This module is used to establish a composite model for each grid and form a raster data layer based on the geographic coordinates of the divided geographic regions. It then divides the smooth surface of the origin reference plane projected onto the raster data layer into multiple regions. The data encryption module generates encryption matrices in these regions using a multi-parameter static random number generation algorithm based on random geographic coordinates. Finally, it encrypts the raster data layer and the projections of random geographic coordinates onto the reference plane or smooth surface using the encryption matrix, forming primary encrypted data and an offset matrix. The data encryption module further encrypts the primary encrypted data using the offset matrix, generating a secondary encrypted map.

Images