Detection modeling method for urban pipe network

Abstract

The invention relates to a detection modeling method for an urban pipe network. The method comprises the following steps: collecting pipe network data; carrying out field shooting; decomposing the picture; deep learning; analyzing data; establishing an industry algorithm model; and adjusting and optimizing the statistical model. According to the method, the vector coding technology is utilized, 3D is generated from 2D of the external surface, meanwhile, downward probing detection can be carried out on invisible internal structures or shields, and three-dimensional reconstruction is carried out on the invisible internal parts. According to the method, a visible 2D image is used for generating a 3D image, and particularly, a 3D vector model with physical attributes can be generated through a vector coding down-probing technology under the condition that excavation is not needed, so that detailed conditions of an underground pipeline or a covered pipeline on the ground are clearly known, and an accurate and rigorous scientific basis is provided for pipeline repair and state evaluation.

Description

Technical Field

[0001] The invention relates to a detection modeling method for a city pipe network, which is a method for modeling and digital processing of a city's above-ground and underground pipe networks. Background Art

[0002] The urban underground pipeline system is an important part of the urban infrastructure, covering a variety of pipelines such as water supply, drainage, gas, electricity, and heat. Effective pipeline detection and management can not only reduce resource waste, but also improve the safety of urban operations. However, traditional pipeline detection methods such as manual detection and ground radar often face problems such as limited detection range, complex data processing, and high costs. Therefore, a new detection technology is urgently needed to solve these problems. Summary of the invention

[0003] In order to overcome the problems of the prior art, the present invention proposes a detection modeling method for urban underground pipe networks. The method uses vector coding technology to generate 3D from 2D external surfaces, and can also detect invisible internal structures or shielding objects, establish accurate pipelines without excavation, and establish a solid scientific foundation for pipeline repair and maintenance.

[0004] The object of the present invention is achieved by: a detection modeling method for urban pipe network, the steps of the method are as follows:

[0005] Step 1, collect pipe network data: collect pipe network data of the studied area, including the composition, direction, connection status, pipe material, installation age, and corrosion status of various surface and underground pipe networks;

[0006] Step 2, on-site photography: Use cameras and positioning equipment to take on-site photography of the pipeline network. For underground grids, locate and photograph the pipeline network path, including the location of underground pipeline manhole covers, and photograph the covered and uncovered pipeline network directions on the ground;

[0007] Step 3: Image decomposition to form a 3D pipe network model: First, perform image preprocessing, denoising and normalization; then perform feature extraction, including edge detection and region segmentation, reduce the image resolution, and expand the image pixels into a vector matrix; then perform data integration to synthesize the pipe network 3D model;

[0008] Step 4, deep learning: deep learning, optimization and correction of the formed 3D model through neural network, data analysis: vector encoder with photon signal conversion variable encoding algorithm, cyclic update; update equation timely projects the current state variable as a priori estimate to the measurement update equation, and the measurement update equation corrects the priori estimate to obtain the posterior estimate of the state;

[0009] Model update equation:

[0010]

[0011]

[0012] State update equation:

[0013] K k =P - H T (HP - H T +R) -1 (2.3)

[0014]

[0015] P k+1 =(IK k H)P k (2.5)

[0016] Where: A is the state transition matrix, which describes the change relationship of the system state from one time period to the next time period. The current state is X K , then the next state X k+1 , then through X k+1 =A·X k +B·U k It is represented by , where B is the control input matrix; P is the state covariance matrix, which represents the uncertainty of the system state estimate. The diagonal elements of the covariance matrix represent the variance of each state variable, while the off-diagonal elements represent the covariance between state variables. In the measurement update process, the covariance matrix will be adjusted with the introduction of new observation data; X is the state vector, which contains the estimated values ​​of all state variables of the system at a certain moment. In an underground pipe network model, the state vector includes position and trend; U: control input vector, which represents the influence of the extracted photon signal input converted into an electrical signal on the system state. In some cases, the state change of the system depends not only on the state of its own encoder, but also on the control input; H: observation matrix, which describes how to generate the observation value Z from the state vector X. Specifically, the measurement equation can be expressed as Z = H·X+V, where V is the measurement noise. The measurement matrix maps the state space to the observation space, and is usually used to convert the dimension of the state vector to the dimension of the observation vector; Q: process noise covariance matrix, which represents the uncertainty noise in the system process, and is usually used to describe the inaccuracy of the model. Process noise refers to random disturbances in the state transition process, and the covariance matrix Q describes the statistical characteristics of these disturbances. It is used to update the state covariance matrix P in the prediction step;

[0017] Step 5, data analysis: extract useful information by sorting, processing and interpreting pipeline data. The data includes: pipe diameter, burial depth, main pipeline direction, location, pipeline physical deformation analysis, pipeline corrosion detection, anti-corrosion layer detection, leak point screening, pipeline material analysis, pipeline pressure analysis;

[0018] Step 6: Establish an industry algorithm model: Input industry professional data and transform the model into a model for an industry;

[0019] Step 7, tune the statistical model: repeatedly output the data collected during operation, establish a professional database, and debug the model until a satisfactory state is reached.

[0020] The advantages and beneficial effects of the present invention are as follows: the present invention utilizes vector coding technology to generate 3D from 2D of the external surface, and can also detect invisible internal structures or shielding objects, and perform three-dimensional reconstruction of the invisible interior. Generate 3D images from visible 2D images, especially for generating 3D vector models with physical attributes through vector coding exploration technology without excavation, so as to clearly understand the details of underground pipelines or covered pipelines on the ground, and provide a precise and rigorous scientific basis for pipeline emergency repair and status assessment.

[0021] The innovative features of the present invention are as follows:

[0022] 1. Through the collection and extraction of pipeline industry data, the dynamic statistical model is developed, and the prior knowledge and real-time geographic data are combined to dynamically update the nonlinear characteristics of the pipeline state, adapt to the changes in pipeline operating conditions, and provide a more flexible and adaptable detection solution, which is an effective supplement to the existing detection technology.

[0023] 2. Use statistical models for anomaly detection, identify abnormal patterns in pipeline operation, and predict faults based on historical data, so that maintenance measures can be taken in advance to reduce risks.

[0024] 3. Innovative research and development of key technologies for dynamic and rapid reconstruction of multi-dimensional and multi-scenario models of integrated pipeline corridors

[0025] In view of the various types of equipment and facilities in the integrated pipe corridor, the complexity of modeling various elements, and the difficulty of monitoring throughout the life cycle, we study the multi-dimensional dynamic connection and spatiotemporal correlation of the "geometry-physics-behavior-rules" of the digital model of the integrated pipe corridor, and build a multi-dimensional, multi-scenario, high-precision digital model of the pipe corridor that includes all elements of the pipe corridor pipeline facilities and multiple factors of the human environment, so as to achieve dynamic control and synchronization of the status of the urban pipe network system and the physical pipe corridor. In view of the dynamic reconstruction and assembly fusion of the multi-level and multi-scale pipe corridor digital model, the pipe corridor digital model fusion technology is formed by using unified interface calls, feature association mapping, and model dynamic coupling, so as to realize the feedback and trusted mapping of multi-modal data collection in the pipe corridor, and finally integrate and develop a set of modular and universal pipe corridor digital model construction management tool component library. The above achievements have important technical innovation value. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0027] Figure 1 is a flow chart of the method described in an embodiment of the present invention;

[0028] Figure 2 is a cyclic update view of the vector encoder according to an embodiment of the present invention;

[0029] Figure 3 is a deformation diagram of the pipeline under vertical load calculated by the standard value of the vertical earth pressure on the pipe top according to the embodiment of the present invention;

[0030] Figure 4 It is a mechanical model diagram of a pipeline for calculating the standard value of the vertical earth pressure on the pipe top according to an embodiment of the present invention; DETAILED DESCRIPTION

[0031] Example:

[0032] This embodiment is a detection modeling method for urban underground pipe networks. The research object, research process, and research points of the method described in this embodiment are as follows:

[0033] 1. Key scientific issues to be addressed:

[0034] There are many types of underground pipelines, with various materials, different burial times and methods, and pipelines often cross and overlap vertically and horizontally, which increases the difficulty of detection. In the case of trenchless, the image reconstruction algorithm is used to measure the propagation time of light particle signals and light field imaging data in the image data to achieve three-dimensional reconstruction of pipeline scenes. Then, through computer machine learning technology, the accuracy and efficiency of imaging are improved, providing theoretical evidence for non-metallic pipelines, parallel pipelines, long-distance, deep-buried pipelines, etc., improving the accuracy of judgment and reducing blind construction.

[0035] (1) Detection of accident signs and quantitative characterization of integrated pipeline corridors using digital modeling;

[0036] Comprehensive utility corridors have many types of pipelines and coupled and intertwined safety risks, which makes it difficult to accurately identify signs of utility corridor accidents. Integrating the multimodal real-time data of the utility corridor digital twin in the form of digital-analog linkage and studying the characteristic indicators of typical accident scenarios of utility corridor pipelines and their explicit and implicit quantitative correlations is the basis for identifying signs of utility corridor accidents and quantifying risks.

[0037] (2) Holographic simulation reduction method for real-time modeling and risk evolution warning;

[0038] The application of multi-dimensional models has problems such as high computing load and low prediction efficiency. Innovative research and development of non-visual field exploration methods for multi-scenario models of pipeline corridors, further integration of real-time monitoring data, and quantitative prediction of the evolution path and key nodes of pipeline corridor accident risks are the key to achieving self-correcting pipeline corridor accident risk evolution preview and accurate early warning.

[0039] 2. Key technical issues to be solved:

[0040] Technical issues: research in application direction.

[0041] In the detection of long-distance and deeply buried pipelines, there are few mapping relationships of light particle signals on PE material pipelines, and the feature mapping of pipelines without data accumulation is missing, which leads to detection errors. The study uses computer simulation and optical simulation algorithms to analyze the impact of different materials on light quantum reflection and optimize the generation scheme of the prediction model.

[0042] (1) High-precision model construction technology for integrated utility corridors;

[0043] There are numerous pipelines and ancillary facilities inside the tunnel, so high-precision modeling is achieved. The multi-dimensional dynamic connection and spatiotemporal correlation of the "geometry-physics-behavior-rules" of the tunnel digital model are studied, and a high-precision digital model module with correlation and coupling of multiple factors such as the tunnel itself, pipelines inside the tunnel, ancillary facilities, personnel, and environment is constructed.

[0044] (2) Multi-dimensional and multi-scene dynamic three-dimensional reconstruction technology for integrated pipeline corridors;

[0045] The structural layout, node network, and time and space scales of the integrated pipeline corridor are complex and diverse, and it is difficult to assemble and reconstruct model scenes. We develop modular and universal pipeline corridor three-dimensional and even four-dimensional digital model scene dynamic reconstruction technology and tool components to achieve multi-modal data collection and feedback of all elements of the pipeline corridor's "man-machine-environment-pipeline" and credible mapping with the actual pipeline network.

[0046] 3. Research Methods:

[0047] 1. Data collection and preprocessing:

[0048] 1.1 Ensure that the collected pipeline-related data (such as flow, pressure, temperature, material properties, etc.) are representative and accurate.

[0049] 1.2 Data cleaning: remove noise and outliers, and fill in missing values ​​to improve data quality.

[0050] 2. Feature engineering: transform the features of image data to improve the regularity of linear functions.

[0051] 3. Hyperparameter tuning: Use the spatial grid to tune the hyperparameters of the existing model to find the best parameter combination.

[0052] 4. Linear regression analysis: Use the pipeline industry algorithm model to fit the pipeline data model and analyze the linear laws of the existing model.

[0053] 5. Repeat steps 2-4 to infinitely approach the predicted expectation, that is, the actual pipeline data.

[0054] 6. Data acquisition and information mining. In view of the multi-source, heterogeneous, massive, and real-time application characteristics of urban underground pipe networks, a temporal data storage model based on the base state correction model is constructed, and a unified data model of "static library + industry algorithm model + dynamic library" is established. Rapid acquisition and in-depth data mining of characteristic information of urban pipe network emergencies are achieved, providing data support at three scales: local (laboratory), regional (urban) and national (typical area) for emergency prevention assessment and decision-making and emergency response.

[0055] 4. Research objectives;

[0056] In order to solve the problems of interference from magnetic fields such as electricity and metal, difficulty in detecting deep burials, and PE materials caused by existing detection technologies, three-dimensional reconstruction of pipeline scenes is achieved through image reconstruction algorithms and measurement of the propagation time of light particle signals and light field imaging data in image data. In combination with existing detection methods, accurate positioning and identification of underground pipelines can be achieved, including information such as pipeline type, size, and material, for subsequent management and maintenance.

[0057] (1) Research objectives;

[0058] The multi-dimensional dynamic connection and spatio-temporal correlation of "geometry-physics-behavior-rules" of the comprehensive digital model of the pipeline corridor are studied, and a multi-dimensional, multi-scenario, high-precision digital model of the pipeline corridor is constructed, which includes all elements of the pipeline corridor facilities and multiple factors such as the human environment. The technology of constructing the digital model of the pipeline corridor and dynamically reconstructing the scene is formed, and a modular and universal digital model construction and management tool component for the pipeline corridor is formed, which provides a reliable model foundation for subsequent research content.

[0059] (2) Main research contents;

[0060] Based on the experience of underground pipeline corridor planning and design, safety risk prevention and control, and digital modeling technology achievements, we study the relationship between the multi-dimensional dynamic connection and spatiotemporal correlation of the "geometry-physics-behavior-rules" of the integrated pipeline corridor, and develop high-precision underground pipeline corridor non-field of view digital model construction technology that is coupled with multiple factors such as the pipeline corridor itself, pipelines in the corridor, ancillary facilities, personnel, and environment. We study the dynamic reconstruction of digital models and data collection and feedback technology of various pipeline corridors (such as gas, heat, tap water, etc.) at multiple scales and levels, and realize the multi-modal data collection, feedback, and trusted mapping of all elements of the integrated pipeline corridor "man-machine-environment-pipeline". We develop a component library of management tools for the construction of digital models of integrated pipeline corridors to provide new technical support for the establishment of a digital system platform for accident risk prevention and control in integrated pipeline corridors.

[0061] (3) Major scientific and technical issues to be solved. Identification and quantitative characterization of signs of accidents in urban integrated pipe corridors. Integrated pipe corridors have many types of pipelines and safety risks are coupled and intertwined, which leads to pipe corridor accidents. It is difficult to accurately identify signs. The integration of multi-modal real-time data in the form of digital-analog linkage and the study of characteristic indicators of typical accident scenarios of pipe corridors and their explicit-implicit quantitative correlations are the basis for the identification of signs of pipe corridor accidents and the quantitative characterization of risks.

[0062] The basic idea of ​​the detection modeling method of the urban underground pipe network is to customize a vector encoder architecture, and after end-to-end training, map the instantaneous image to a depth map / model for representation. Three transient renderers are used during training to complete reflective indirect light transmission, which can quickly generate a large amount of training data. Although the system structure design is complex, this software system can make meaningful predictions on experimental data. Even if the scene contains reverse reflection objects that violate the (pure diffusion) forward model assumption, the iterative prediction model is still valid-this is the deep learning in the vector encoder software module that has been successfully applied to invisible underground pipe network measurements. Another benefit is that the handover is parameter-free, so the average time for predicting and generating a 3D model is only 60 minutes, and this method is also the fastest technology currently available. This embodiment is based on the underlying supporting graphics engine of the application software and model platform to solve the core technology of detecting urban underground pipe networks under non-excavation conditions-enhancing the ability to build primitives and realizing the deep fusion of three types of algorithms: surface, mesh and entity. The vector encoder has built-in capabilities such as parametric modeling, interactive modeling, and genetic algorithm parent-child modeling, provides good human-computer interaction functions such as multi-screen multi-channel linkage, floating and label marking menus, supports UV mapping to achieve more accurate effect presentation, and fundamentally guarantees the information security of digital applications. It supports cross-industry applications and fully assists the high-quality construction of the digital industry.

[0063] Digital construction solutions, digital collaborative management platform.

[0064] The digital collaborative management platform is based on multi-level application scenarios of cloud architecture, providing a comprehensive solution from construction to operation for engineering project management, covering the entire process from project planning, execution, monitoring, detection to asset operation. With business collaboration, element collaboration, and data collaboration as the bus, it closely links project launch, engineering management, quality inspection and completion delivery, realizes the interaction and analysis of multi-dimensional data, and provides strong support for digital design, intelligent construction, digital delivery, and asset management.

[0065] Vector Encoder Technology Overview:

[0066] Simply put, when light shines on an object, it will cause it to emit electrons. At the same time, light is not a continuous wave, but a stream of moving particles, so light has the dual nature of wave and particle. The vector encoder uses the characteristic point of light having the dual nature of wave and particle to combine the transition electron data generated when the object under test has the photoelectric effect with the algorithm model of the pipeline industry for deep learning, and after N iterations, a statistical analysis data model that meets the requirements of the industry is obtained. The vector encoder is an efficient multimodal analysis technology that can be used for urban pipe network signal processing and data analysis. Its core is to extract useful factors by decomposing, coupling and reconstructing the modes of photoelectric signals. In pipeline detection, the vector encoder technology can calculate and analyze the field energy signals of underground pipelines to obtain the spectral frequency and vector intensity, and identify and calculate the position, material density, pressure, depth (elevation), corrosion, stress and other chemical and physical states of the pipeline under test based on parameter information such as kinetic energy, potential energy and force.

[0067] Vector encoder principle:

[0068] The basic principle of the vector encoder is to collect invisible light signals around underground pipelines and field energy density information at different locations, and process the signals through modal analysis algorithms. The vector encoder can effectively distinguish different types of signals, thereby improving the dimension and accuracy of detection.

[0069] At the same time, 3D vision algorithms and AI machine learning algorithms quickly construct three-dimensional space, and support the export of VR links, mesh models, and rich data formats such as CAD and BIM engineering files. At the same time, it can achieve global coupling of underground-ground-outdoor-indoor, and build a global visualized 3D real-life space. This embodiment is also compatible with third-party laser scanners, oblique photography, panoramic photos, GIS maps, 3Dmax model files, etc., to achieve common sharing of data, and truly realize visible and invisible (underground, hidden space, etc.) real-life three-dimensional models from collection to use.

[0070] 1. Use the virtual wall model built in the main program to establish a data analysis library, conduct data mining and algorithm model training on historical data information such as steel pipes and PE pipes, and establish map management specifically for urban pipelines. In the future, predict the potential location and trend of new data pipelines and evaluate the health of pipelines.

[0071] 2. At the same time, combined with the characteristics of the virtual wall that can give multiple data model results each time, fusion algorithm analysis is carried out simultaneously, which can also improve the accuracy of underground detection.

[0072] 3. The main component of the experimental results on metal pipeline corrosion is iron oxide, and the main causes are two types of corrosion:

[0073] (1) Electrochemical corrosion: The current state (trend + position) of the pipeline surface can be reversed by the different degrees of electrochemical corrosion.

[0074] (2) Stress corrosion: Stress corrosion accelerates with the increase of longitudinal strain and can reverse the state of welds (interfaces) and valves.

[0075] 4. Determine the location and trend of underground pipe network by the density of light particles:

[0076] Use vector encoders to monitor the location and health of underground gas, heat and other pipelines:

[0077] (1) Connecting pipe network sensors: Sensors can be used to monitor parameters such as temperature, pressure, and strain in underground pipes. Vector encoders can detect small changes in pipes by analyzing the propagation characteristics of light and the above parameters, thereby providing early warning of corrosion and potential leaks or failures.

[0078] (2) Measuring photon density: Photon density is related to the intensity and propagation characteristics of light. By adjusting the intensity and wavelength of the light source, the vector encoder can optimize the sensitivity of the sensor, thereby improving the accuracy of 3D modeling and the reliability of pipeline health assessment.

[0079] 5. Multi-physics coupling matrix:

[0080] Multi-scale: The urban underground pipeline network is a combination of macroscopic, mesoscopic and microscopic phenomena.

[0081] Multi-physical and chemical mechanisms: coupling of mathematical algorithm models of multiple physical and chemical processes.

[0082] From nonlinear discretization to linear regression: from complex underground pipe networks to visualized statistical data models.

[0083] High precision: From two-dimensional to three-dimensional and then to four-dimensional, higher space-time dimensions and model accuracy.

[0084] Advantages of vector encoder application:

[0085] High efficiency: Vector encoders can quickly process large amounts of data, significantly improving detection efficiency.

[0086] Accuracy: Modal analysis can effectively reduce noise interference and improve the accuracy of detection results.

[0087] Cost-effectiveness: Compared with traditional methods, vector encoders have obvious advantages in equipment and labor costs.

[0088] 6. Vector encoding method

[0089] 1) Data collection

[0090] In urban pipeline detection, it is first necessary to conduct surface surveys on the target area to collect invisible light signals and field energy data information around underground pipelines. Signal information suitable for the underground environment is screened out to ensure the accuracy of data collection.

[0091] 2) Data processing

[0092] The collected signal data will be processed by vector coding technology. The specific steps include---

[0093] Signal preprocessing: denoising and normalizing the original signal.

[0094] Modal analysis: Use vector coding algorithm to perform modal decomposition on the signal and extract features related to the pipeline.

[0095] Location identification: Based on the extracted features, machine learning algorithms are used to identify the location, depth, and direction of the pipeline.

[0096] Pipeline health status: Establish a pipeline health assessment model and conduct a comprehensive assessment based on test data and environmental factors.

[0097] The urban pipeline penetration measurement method based on vector encoders demonstrates its advantages in improving detection efficiency and accuracy. Future research can further explore the application potential of vector encoders in detection in different industries (military, intelligence, aerospace) to promote visualization and intelligent management.

[0098] The algorithm fusion of the vector encoder performs deep learning on the data model and iteratively fits the health status of the pipeline network. The health status detection of urban underground pipeline networks is an important part of ensuring the safety and efficient operation of urban infrastructure. With the acceleration of urbanization, underground pipeline systems are facing problems such as aging, corrosion, and sedimentation. Therefore, it is particularly important to detect and evaluate the health status of pipelines in a timely and accurate manner. The following are some key points and method principles for detecting the health status of urban underground pipelines.

[0099] (1) Importance of testing:

[0100] Safety: timely detect pipeline damage or leakage, prevent accidents and ensure public safety.

[0101] Economical: Through regular inspections, problems can be discovered in advance, reducing maintenance costs and economic losses caused by emergencies.

[0102] Environmental protection: Prevent pipeline leakage from polluting soil and water sources and protect the ecological environment.

[0103] (2) Detection method:

[0104] Using the distribution of invisible light to detect the location and status of underground pipelines, it is possible to provide high-resolution images, 2D drawings and 3D models.

[0105] The specific steps of the detection modeling method of the urban underground pipe network are as follows: Figure 1 As shown:

[0106] Step 1, collect pipe network data: collect pipe network data of the studied area, including the composition, direction, connection status, pipe material, installation age, and corrosion status of various surface and underground pipe networks;

[0107] The studied area refers to the area where the pipeline network is to be renovated or repaired, such as a building or structure of one or several houses, or a compound of a certain unit, or a residential area. The pipeline network data include: drawings and text materials used in the construction of the studied pipeline network facilities and corresponding buildings or structures, including the pipe diameter, burial depth, pipeline trend, location positioning, etc. The literature drawings and materials should also involve changes in the pipeline network itself or the surrounding environment during use. These changes include: corrosion of the pipeline network itself or damage caused by external forces, whether other pipeline networks are built around the pipeline network, whether the underground or ground part of the pipeline network is covered or occupied by other foreign objects (for example, a house is built on the path of the buried underground pipeline, or the riser is covered by the installation of decorative panels or mortar), resulting in various obstacles such as difficulties in maintenance or excavation.

[0108] Step 2, on-site photography: Use cameras and positioning equipment to take on-site photography of the pipeline network. For underground grids, locate and photograph the pipeline network path, including the location of underground pipeline manhole covers, and photograph the covered and uncovered pipeline network directions on the ground;

[0109] The pipeline network is usually invisible underground, but it is possible to capture the ground conditions corresponding to the direction of the pipeline network, such as the occupying objects on the ground above the pipeline network, manhole covers on the ground, etc. Some pipes on the ground are visible, but most of them are installed in buildings and it is impossible to take pictures. The main focus is on taking pictures according to the direction of the pipeline network.

[0110] Step 3: Image decomposition to form a 3D pipe network model: First, perform image preprocessing, denoising and normalization; then perform feature extraction, including edge detection and region segmentation, reduce the image resolution, and expand the image pixels into a vector matrix; for example, reduce a photo to a resolution of 28×28 pixels, with 784 pixels, and expand it into a vector matrix of 1×784. Then perform data integration to synthesize the pipe network 3D model;

[0111] 2D images usually contain complex visual information, and decomposition can simplify this information into more manageable parts. Through decomposition, image data can be processed more efficiently, reducing the computational burden and speeding up the generation. Decomposition helps to better understand the content of the image, thereby generating a more realistic 3D model. The decomposed information can be used as training data to help statistical models learn how to generate 3D structures from 2D images. The purpose of decomposing images: By decomposing 2D images, important information in the image can be extracted, such as the shape, grayscale, texture, color and spatial relationship of the object, filtering out visible light, retaining the inverted invisible light, etc.

[0112] The decomposed information can be used to build a 3D primary model to help the data model understand the three-dimensional structure and characteristics of the object. Image decomposition can improve the accuracy of the generated 3D model, making the final result more consistent with the real-world object. The data after image decomposition can be used in the algorithm model to achieve the regularization of the data statistical model.

[0113] Through photo decomposition, the following effects can be achieved: the 3D model generated from the 2D image has a high degree of restoration and detailed texture, and can accurately reflect the shape and characteristics of the object. The generated 3D model can be widely used in the field of urban underground pipe networks to enhance users' health detection of pipelines. The decomposed data can be used for further analysis and processing. Such as object recognition, scene understanding, physical and mechanical analysis, electrochemical corrosion, and stress corrosion. Use the generated 3D model for expected evaluation, so that the post-emergency repair work of the underground pipe network can be transformed into pre-emptive prevention.

[0114] Photo decomposition is an important step in achieving three-dimensional reconstruction of underground pipeline networks. By extracting and processing image information, high-quality three-dimensional models can be generated, promoting the development of various application scenarios including urban underground pipeline networks.

[0115] Step 4, deep learning: deep learning, optimization and correction of the formed 3D model through neural network, data analysis: vector encoder with photon signal conversion variable encoding algorithm, cyclic update; update equation timely projects the current state variable as a priori estimate to the measurement update equation, and the measurement update equation corrects the priori estimate to obtain the posterior estimate of the state;

[0116] Model update equation:

[0117]

[0118] State update equation:

[0119] K k =P - H T (HP - H T +R) -1 (2.3)

[0120]

[0121] p k+1 =(IK k H)P k (2.5)

[0122] Where: A is the state transition matrix, which describes the change relationship of the system state from one time period to the next time period. If the current state is X K , then the next state X k+1 , then you can use X k+1 =A·X k +B·U k It is represented by , where B is the control input matrix; P is the state covariance matrix, which represents the uncertainty of the system state estimate. The diagonal elements of the covariance matrix represent the variance of each state variable, while the off-diagonal elements represent the covariance between state variables. In the measurement update process, the covariance matrix will be adjusted with the introduction of new observation data; X is the state vector, which contains the estimated values ​​of all state variables of the system at a certain moment. In an underground pipe network model, the state vector includes position and trend; U: control input vector, which represents the influence of the extracted photon signal input converted into an electrical signal on the system state. In some cases, the state change of the system depends not only on the state of its own encoder, but also on the control input; H: observation matrix, which describes how to generate the observation value Z from the state vector X. Specifically, the measurement equation can be expressed as Z = H·X+V, where V is the measurement noise. The measurement matrix maps the state space to the observation space, and is usually used to convert the dimension of the state vector to the dimension of the observation vector; Q: process noise covariance matrix, which represents the uncertainty noise in the system process, and is usually used to describe the inaccuracy of the model. Process noise refers to random disturbances in the state transition process, and the covariance matrix Q describes the statistical characteristics of these disturbances. It is used to update the state covariance matrix P in the prediction step.

[0123] The first thing to do in the state update equation is to calculate the gain K k , followed by measuring the output to obtain Z k, then generate the posterior estimate of the state according to formula 2.4, and finally estimate the posterior covariance of the state according to formula 2.5.

[0124] The calculation of the model update equation needs to be based on the state update equation, and the whole process is repeated again. The posterior estimate obtained from the previous calculation is used as the prior estimate for the next calculation. This recursive extrapolation is one of the characteristics of the measurement pipeline, and it is easier to implement than other algorithms. Because each time the current state estimate is recursively calculated based on the previous measurement variables. Figure 2 The entire operation process is shown.

[0125] Regardless of whether there is a reasonable standard model to select the final model, better performance parameters can be obtained by continuously fitting the statistical mathematical model and adjusting the coefficients of the self-developed encoder. When adjusting the location of the pipeline network in the three-dimensional environment, the degree of overfitting of the model will gradually decrease, and finally the optimal solution will be selected based on the user's usage intention and the uncertain and diverse solutions of the model. The distribution of discrete random variables (the density of continuous random variables) is asymptotically linear and normal after being transformed by a nonlinear system. The vector encoder is a special state estimator that achieves asymptotically optimal Bayesian decision-making through nonlinear to linearization. The vector encoder, an algorithmic variant, allows random variables after nonlinear transformation to still have normal distribution characteristics.

[0126] Calculation of vector encoder:

[0127] In order to estimate a process with nonlinear difference and measurement relationship, a new linearization representation is first given:

[0128]

[0129] Where: X k and Z k are the true values ​​of the state vector and the observation vector;

[0130] and are approximations of the state vector and the observation vector;

[0131] is the posterior estimate of the k-state vector;

[0132] Random variable w k and V k represents process excitation noise and observation noise;

[0133] A is the partial derivative matrix of f with respect to x:

[0134]

[0135] W is the partial derivative matrix of f with respect to w:

[0136]

[0137] H is the partial derivative matrix of h with respect to x:

[0138]

[0139] V is the partial derivative matrix of h with respect to v:

[0140]

[0141] For convenience, the expressions do not include the subscript k in A, W, H, V, but they actually vary with time. Define a new expression for the prediction error:

[0142]

[0143] and the residual of the observed variables:

[0144]

[0145] In practice, it is not possible to directly obtain X in formula 2.7 k , which is the true value of the state vector, that is, the object to be estimated. Similarly, it is impossible to obtain Z in Equation 2.8 k , which is used to estimate X k The true value of the observation vector. From Equations 2.7 and 2.8, the expression of the error process can be written:

[0146]

[0147] E k and η k represents a matrix with zero mean and covariance WQW T and V RV T 2.9 and 2.10 are linear, which are the state difference equation and measurement equation in the encoder. and the encoder to estimate the prediction error in Equation 2.9 The estimated result is recorded as Combined with Equation 2.7, the posterior state estimate of the initial nonlinear process can be obtained:

[0148]

[0149] The random variables in Equation 2.9 and Equation 2.10 have the following probability distribution:

[0150]

[0151] P(E k )~N(0,WQk W T )

[0152] P(ηk)~N(0,VR k V T )

[0153] make The estimated value of is zero. Based on the above approximation, we can write the estimated The encoder expression is:

[0154]

[0155] Substituting equation 2.8 back into equation 2.11, the actual encoder expression is:

[0156]

[0157] So the encoder equations and parameters are expressed as:

[0158] Model update equation:

[0159]

[0160] P - =P k k -1 +Q (2.14)

[0161] Measurement update equation:

[0162]

[0163] P k+1 =(IK k ) - k(2.16)

[0164] The above is the vector encoder observation variable update, where the encoder gain K can be obtained by appropriately replacing the observation error covariance k。 , after the matrix is ​​transposed, it can correctly transmit or weight the useful part of the observed light quantum information. When the full expression of the encoder is given, it can be used replace To express the prior probability, the matrices A, W, H, and V will eventually converge the encoder in any case. This means that they have changing values ​​at different locations and time nodes, and they need to be recalculated each time. This ensures the effectiveness of the final number of iterations and the best authenticity of the fitted model.

[0165] The three-dimensional model of the pipeline is obtained through the physical field energy density ranging technology to analyze the geometric shape and deformation of the pipeline. Figure 3 , Figure 4 The calculation method of the standard value of the vertical earth pressure on the top of the pipe is given, where: Figure 3 is the deformation diagram of the pipeline under vertical load, Figure 4 The mechanical model diagram of the pipeline is as follows:

[0166] q ck =y s H (2.17)

[0167] Where: q ck is the standard value of vertical soil pressure on the top of the pipe (KN / m 2 );r s is the gravity density of backfill soil (KN / m 2 ); H is the depth from the ground to the top of the pipe.

[0168] Under this experimental condition, the influence of vehicle load is not considered, and deformation calculation is only considered according to the accumulation load. The compaction degree of backfill soil is the dominant factor affecting the safe operation of buried pipelines. Therefore, in terms of pipeline series selection, pipelines with a small diameter-to-thickness ratio should be used to enhance the pipeline's ability to resist vertical deformation and annular stability resistance. At the same time, the buried pipeline should select a suitable covering depth (burial depth). When the local burial depth is too large or too small, the three aspects of annular bending stress, vertical deformation and position stability should be modeled and verified. If necessary, the pipeline diameter-to-thickness ratio can be partially increased to ensure the safe operation of the underground pipeline network. Therefore, through spectral analysis and the mirror change of kinetic potential energy, a mathematical model can be established in real time to monitor parameters such as pipeline pressure, temperature and humidity, electrochemical corrosion, and stress corrosion, and statistical models and data analysis techniques can be used to evaluate the pipeline network position trend status and pipeline health status.

[0169] Deep learning is a branch of machine learning that mainly uses neural network models to learn and process data. The main body of deep learning is the urban underground pipe network system, especially the model based on deep neural network. These models can be pre-trained or trained on the fly according to specific tasks. Deep learning relies on the computing power of computers (CPU), especially graphics processing units (GPU) and tensor processing units (TPU) to process large amounts of survey and collection data and complex calculations. In a statistical sense, data sets such as photos and videos can also be regarded as part of "learning" because they provide the input information required for model training.

[0170] The content of deep learning includes: Feature extraction: The deep learning model automatically extracts features from images and video data through multi-layer neural networks. These features can be edges and textures in images, or the content of each frame in a video. Pattern recognition: The model learns how to recognize patterns in data, such as specific patterns in tasks such as image classification, dimensional tolerance, shape tolerance, and position tolerance. Mapping relationship: The deep learning model learns the mapping relationship between input data and output results. For example, in an image classification task, the model learns to map images to corresponding category labels. Optimization strategy: Through the back-propagation algorithm, the algorithm model learns how to adjust its parameters to minimize the error between the predicted results and the actual results. Iterative optimization of numerical and analog algorithms and reduction of error losses for deep neural networks.

[0171] Neural network is one of the important algorithms for deep inversion, and has many applications in graph classification, detection, and secondary modeling. Let’s explain it from the perspective of optimization and error in urban underground pipe network detection.

[0172] Learning objectives achieved:

[0173] Automated decision-making: The deep learning image factor feature set can automatically make primary model predictions and decisions based on the input data. It is widely used in the field of underground pipeline network diagnosis.

[0174] Improved accuracy: By combining theoretical specifications of the learning pipeline with actual work mapping, deep learning models can achieve high accuracy on specific tasks, surpassing traditional machine learning methods in both implementation means and results.

[0175] Processing complex data: Deep learning can process high-dimensional and complex data of known pipeline models, such as the direction and depth of pipeline pulling and PE pipe damage, and output useful information to form a decision tree.

[0176] Generate new content: Some deep learning models (such as the generative underground pipeline adversarial network) can simulate and analyze data content such as electrostatic corrosion and safe distance between different types of pipelines to promote process improvements and digital early warning management during underground pipeline laying.

[0177] Enhanced human-computer interaction: The application of deep learning of urban underground pipeline networks in fields such as computer vision has improved the naturalness and intelligence of human-computer interaction.

[0178] Deep learning is a process in which an artificial intelligence system learns by processing a large amount of urban underground pipe network data. The learning content includes feature extraction, pattern recognition and mapping relationships, and the construction of pipe network location trends and health status models. The ultimate goal is to achieve multiple applications such as automated decision-making of prediction models, improve accuracy, pipe network health detection and early warning analysis, and handle difficult pipeline data.

[0179] Step 5, data analysis: extract useful information by sorting, processing and interpreting pipeline data. The data includes: pipe diameter, burial depth, main pipeline direction, location, pipeline physical deformation analysis, pipeline corrosion detection, anti-corrosion layer detection, leak point screening, pipeline material analysis, pipeline pressure analysis;

[0180] Through data analysis, we provide digital drive for urban underground pipeline networks, help management make wise decisions, and reduce the operational risks of urban pipeline networks. At the same time, we use industry algorithm models to continuously identify potential pipeline health risks (such as leaks) to help predict future pipeline network operation conditions and repair and maintenance arrangements. By analyzing current workflows and pipeline operation data, we can identify bottlenecks and improvement points, improve efficiency, and reduce costs. Through data analysis, we can identify risks and problems in potential pipeline networks, and develop corresponding strategy models through artificial intelligence and data analysis to reduce the operational risks of pipeline companies and change post-repair to pre-prevention.

[0181] Data analysis includes:

[0182] Structured data:

[0183] Tabular data: Databases of urban pipe network management departments, such as pipe model tables, operating spreadsheets, etc., contain row and column formats to facilitate the establishment of statistical data models.

[0184] Time series data: Pipeline data arranged in chronological order, establishing a trend analysis algorithm model and a prediction decision model with the time T axis as the tie point.

[0185] Image and video data: such as product pictures, surveillance videos, etc., require computer vision technology to establish BIM for analysis and thus form a predictive model.

[0186] Semi-structured data:

[0187] JSON, XML and other formats: These data formats contain tags and structures, which do not fully conform to traditional table formats. They are suitable for data exchange and storage when building statistical models.

[0188] User behavior data:

[0189] Application usage data: such as the operation records of the pipeline network company in the application, which is used to instantly analyze the existing data statistical model and determine the functional usage of the improved algorithm model.

[0190] Data fusion: Data obtained by the vector encoder detection method is used to fuse the data model to provide a more comprehensive pipeline health assessment.

[0191] Machine Learning: Use machine learning algorithms to analyze historical data and couple them with algorithm models adapted to the pipeline industry to predict the health status and potential risks of pipelines.

[0192] Health assessment model: Establish a pipeline health assessment model and conduct a comprehensive assessment based on test data and environmental factors.

[0193] Data analysis and evaluation

[0194] Data fusion: Data obtained by the vector encoder detection method is used to fuse the data model to provide a more comprehensive pipeline health assessment.

[0195] Machine Learning: Use machine learning algorithms to analyze historical data and couple them with algorithm models adapted to the pipeline industry to predict the health status and potential risks of pipelines.

[0196] Health assessment model: Establish a pipeline health assessment model and conduct a comprehensive assessment based on test data and environmental factors.

[0197] In order to ensure the safe operation of various urban buried pipelines, the focus should be on assessing their annular bending stress, deformation and corrosion stability. The stress conditions of the 3D model of the buried pipeline and the mathematical geometric method of the load are given first, and then the health assessment of the buried pipeline stress, deformation and corrosion stability is analyzed. Considering different pipe series, burial depth, backfill density and other factors, a comparative analysis is conducted to understand the many factors that affect the safe operation of buried pipelines, which can provide technical support for the safe operation of various buried pipeline networks.

[0198] The location measurement and health status detection of urban underground pipelines is a complex and important scientific research task. Fully utilizing the advantages of the vector encoder (artificial intelligence and data engineering) technology can effectively improve the accuracy and efficiency of detection. In the future, with the continuous advancement of technology, intelligent and automated detection methods will play an increasingly important role in urban pipeline management.

[0199] Step 6: Establish an industry algorithm model: Input industry professional data and transform the model into a model for an industry;

[0200] For example: if the research object is a gas pipeline, the professional data of the gas pipeline is input into the model to form a professional model of the gas pipeline.

[0201] Step 7, tune the statistical model: repeatedly output the data collected during operation, establish a professional database, and debug the model until a satisfactory state is reached.

[0202] Usually, the model needs to be verified repeatedly, and a large amount of data is needed so that the neural network can carry out a large amount of training and learning. The learning process is the process of data processing and large-scale accumulation.

[0203] Finally, it should be noted that the above is only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred arrangement scheme, a person skilled in the art should understand that the technical solution of the present invention (such as changes or refinements of the digital calculation system used, the use of various formulas, the sequence of steps, etc.) can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A detection modeling method for an urban pipe network, characterized in that: The steps of the method are as follows: Step 1, collect pipe network data: collect pipe network data of the studied area, including the composition, direction, connection status, pipe material, installation age, and corrosion status of various surface and underground pipe networks; Step 2, on-site photography: Use cameras and positioning equipment to take on-site photography of the pipeline network. For underground grids, locate and photograph the pipeline network path, including the location of underground pipeline manhole covers, and photograph the covered and uncovered pipeline network directions on the ground; Step 3: Image decomposition to form a 3D pipe network model: First, perform image preprocessing, denoising and normalization; then perform feature extraction, including edge detection and region segmentation, reduce the image resolution, and expand the image pixels into a vector matrix; then perform data integration to synthesize the pipe network 3D model; Step 4, deep learning: deep learning, optimization and correction of the formed 3D model through neural network, data analysis: vector encoder with photon signal conversion variable encoding algorithm, cyclic update; update equation timely projects the current state variable as a priori estimate to the measurement update equation, and the measurement update equation corrects the priori estimate to obtain the posterior estimate of the state; Model update equation: State update equation: K k =P - H T (HP - H T +R) -1 (2.3) p k+1 =(I-K k H)P k (2.5) Where: A is the state transition matrix, which describes the change relationship of the system state from one time period to the next time period; the current state is X K , then the next state X k+1 , then through X k+1 =A·X k +B·U k It is represented by, where B is the control input matrix; P is the state covariance matrix, which represents the uncertainty of the system state estimate; the diagonal elements of the covariance matrix represent the variance of each state variable, while the non-diagonal elements represent the covariance between state variables; during the measurement update process, the covariance matrix will be adjusted with the introduction of new observation data; X is the state vector, which contains the estimated values ​​of all state variables of the system at a certain moment; in an underground pipe network model, the state vector includes position and trend; U: control input vector, which represents the influence of the extracted photon signal input converted into electrical signal on the system state; in some cases, the state change of the system depends not only on It is affected by the state of its own encoder and the control input; H: Observation matrix, which describes how to generate observation value Z from state vector X; Specifically, the measurement equation can be expressed as Z = H · X + V, where V is the measurement noise; The measurement matrix maps the state space to the observation space and is usually used to convert the dimension of the state vector to the dimension of the observation vector; Q: Process noise covariance matrix, which represents the uncertainty noise in the system process and is usually used to describe the inaccuracy of the model; Process noise refers to random disturbances in the state transfer process, and the covariance matrix Q describes the statistical characteristics of these disturbances; It is used to update the state covariance matrix P in the prediction step; Step 5, data analysis: extract useful information by sorting, processing and interpreting pipeline data; data includes: pipe diameter, burial depth, main pipeline direction, location, pipeline physical deformation analysis, pipeline corrosion detection, anti-corrosion layer detection, leak point screening, pipeline material analysis, pipeline pressure analysis; Step 6: Establish an industry algorithm model: Input industry professional data and transform the model into a model for an industry; Step 7, tune the statistical model: repeatedly output the data collected during operation, establish a professional database, and debug the model until a satisfactory state is reached.

Images