A method for predicting the corrosion rate of a natural gas pipeline

Abstract

The application relates to a natural gas pipeline corrosion rate prediction method, in particular to the natural gas pipeline corrosion prediction field, and realizes high-precision dynamic early warning through four-step cooperative technology, first, visible light and radar data fusion are utilized to generate a pipeline full-coverage three-dimensional corrosion base film, and blind areas caused by complex environments are effectively eliminated; then, a physically constrained generative adversarial network is used to infer the corrosion probability distribution of the blocked area, and the prediction difficulty of the hidden area is solved; then, in combination with a stress field and historical data, a graph neural network is used to deduce a corrosion space-time diffusion path, and a future corrosion heat map is output; finally, model parameters are dynamically adjusted according to measured data, a closed-loop optimized corrosion risk decision matrix is formed, and the transformation from passive protection to intelligent prevention of pipeline corrosion is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of natural gas pipeline corrosion prediction, more particularly, the present application relates to a natural gas pipeline corrosion rate prediction method. BACKGROUND

[0002] As an important energy transportation facility, natural gas pipelines face challenges in inspection under complex surface environments. In mountainous, marshy, and urban dense areas, the pipeline surface is often disturbed by terrain undulations, vegetation coverage, and artificial buildings, making it difficult to accurately monitor some areas. In addition, there may be hidden coating damage areas and diverse corrosion patterns on the pipeline surface, which makes it difficult for traditional inspection methods to obtain comprehensive and accurate corrosion data. Especially in areas that are blocked or difficult to access, traditional manual inspection often cannot provide complete corrosion information, leading to missed detection. Current inspection methods lack the ability to cope with complex geographical conditions and harsh environments, especially in rainy and snowy weather, making it difficult to obtain clear images and increasing the risk of major safety accidents.

[0003] Existing inspection technologies mainly rely on visible light and infrared cameras, which can detect some corrosion areas, but in complex terrain, these devices have limited visibility and are difficult to penetrate obstacles, making it difficult to fully cover the pipeline surface. Existing image stitching techniques have large errors when dealing with non-continuous damage areas, and the stitched images cannot accurately reflect the corrosion conditions. In addition, environmental factors such as bad weather can increase image noise, further affecting the accuracy of corrosion feature extraction. These problems limit the effectiveness of existing technologies in complex environments, significantly increasing the missed detection rate, thereby increasing the risk of undetected corrosion areas causing leaks or explosions and other safety accidents. Therefore, there is an urgent need for a new multi-source occlusion penetration imaging and three-dimensional reconstruction technology that can overcome the limitations of traditional technology, achieve millimeter-level accurate monitoring of corrosion areas, and ensure efficient coverage in various environmental conditions, minimizing missed detection and potential risks. SUMMARY

[0004] The present application provides a natural gas pipeline corrosion rate prediction method to solve the problems in the background art by using four-step cooperative technology to achieve high-precision dynamic early warning.

[0005] The technical solution of the present application to solve the above technical problems is as follows: specifically including the following steps:

[0006] Step S1, when receiving visible light images, synthetic aperture radar data and environmental sensing data, performing registration operation of cross-modal deformation energy function on the texture gradient of the visible light images and the polarization scattering matrix of the synthetic aperture radar, generating a three-dimensional corrosion feature base film covering the pipeline surface;

[0007] Step S2, in the presence of an environment shelter area of a three-dimensional corrosion feature base film, taking the observed area features of the base film as input conditions, combining environmental sensing parameters and pipeline material characteristic parameters, generating a shelter area corrosion probability distribution map through a physically constrained generative adversarial network;

[0008] Step S3, when acquiring pipeline stress distribution data and historical corrosion data, taking the corrosion probability distribution map as an initial field, using a graph neural network to couple a corrosion-stress-diffusion equation to perform a space-time deduction operation, and outputting a corrosion expansion path thermodynamic map in a future preset period;

[0009] Step S4, when new inspection data is input, a multi-objective optimization function is constructed according to the difference between the predicted value and the measured value, the weight parameters of the generative adversarial network and the graph neural network are dynamically updated, and a corrosion risk decision matrix is generated;

[0010] In a preferred embodiment, the visible light image, synthetic aperture radar data and environmental sensing data in step one are synchronously collected by the following method:

[0011] The visible light image is collected by a patrol unmanned aerial vehicle equipped with a multi-spectral gimbal camera on a preset track, and its imaging waveband covers the visible light and near-infrared spectral range of 400-900 nm. The synthetic aperture radar data is obtained by penetrating through the ground cover through the X-band polarized interference radar of the same platform carrier, wherein the radar incidence angle adjustment range is set to 25°-50° to adapt to the change of the elevation angle of different terrains. The environmental sensing data is monitored in real time by a sensor node network buried in the soil around the pipeline, and the sampling parameters at least include three key physical quantities of temperature, conductivity and pH value.

[0012] In a preferred embodiment, the specific process of the registration operation is as follows:

[0013] First, the feature tensor is constructed; for the visible light image, the spatial gradient tensor is calculated to enhance the edge features of the image, and the Sobel operator is used for convolution operation to obtain the gradient information of each pixel point. For the synthetic aperture radar data, the polarization scattering entropy is calculated to describe different polarization features by extracting the scattering features;

[0014] Next, a cross-modal deformation energy function is established, which includes three parts: feature consistency constraint, thin plate spline constraint and thermal deformation compensation;

[0015] In the optimization stage, an improved alternating direction multiplier method algorithm is used to iteratively solve the two-dimensional non-rigid transformation field to optimize the cross-modal deformation matching process;

[0016] Finally, the optimized transformation field is used to generate a three-dimensional corrosion feature base film that combines the real and imaginary parts of the complex scattering features of the SAR data and the curvature features of the visible light images.

[0017] In a preferred embodiment, the step S2 of generating the occlusion area corrosion probability distribution map comprises the following specific operations:

[0018] A1, first, the observed area features of the base film are taken as input, including the depth information and spatial coordinates in the image, and second, the occlusion areas in the image are identified, specifically those areas where the depth features are not reliable, by calculating the gradient of the base film in the depth dimension, if the gradient value is less than a preset threshold, these areas are marked as occlusion areas;

[0019] A2, the environmental sensor parameters and the characteristic parameters of the pipeline material are fused;

[0020] A3, then, a dual-channel network is constructed for generating a corrosion feature map, which includes three modules:

[0021] Feature encoder: fuse the input base film features and environmental sensor data to generate a feature vector containing spatial correlation information;

[0022] Physical constraint: generate a physically constrained output by processing the material characteristic parameters and the environmental sensitivity of the corrosion rate;

[0023] Corrosion generator: combine random noise, encoded feature vector and output of the physical constraint to generate the final corrosion feature map;

[0024] During the training process, the output of the generator is optimized through adversarial training, and the training target is to minimize two loss functions:

[0025] Dual-discriminator loss: this loss function optimizes the difference between the generated image and the real image, encouraging the generator to generate more realistic corrosion features;

[0026] Physical loss: this loss measures the difference between the generated corrosion features and the actual corrosion driving force field, which is related to the iron ion concentration gradient and pH change;

[0027] A4, after training and optimization, an accurate corrosion probability distribution map is finally output.

[0028] In a preferred embodiment, the specific steps of performing the space-time deduction operation are as follows:

[0029] A1, take the corrosion probability distribution map output by the step S2 as the initial field of spatial corrosion depth, load the pipeline stress tensor and the historical corrosion time sequence library at the same time;

[0030] A2, establish a graph structure composed of corrosion feature nodes, stress feature nodes and diffusion channel edges, wherein:

[0031] The corrosion feature node binds the spatial coordinates and carries the corrosion probability value and its historical change gradient of the corresponding position;

[0032] The stress feature node carries an equivalent stress value calculated according to the horizontal principal stress, vertical stress and shear stress component;

[0033] The diffusion channel edge weight is determined by the curvature weighting function of the distance between adjacent nodes and the equivalent stress difference;

[0034] A3, input the graph structure into a graph neural network, generate a corrosion depth prediction field of a future time period by iterative solving of the corrosion-stress-diffusion coupling equation, spatially discretize and encode the corrosion depth prediction field of the current time step, superimpose the exponential decay memory effect of the historical prediction field, and output a corrosion propagation path heat map.

[0035] In a preferred embodiment, the corrosion-stress-diffusion coupling equation comprises:

[0036] A stress corrosion acceleration term based on a material stress sensitivity coefficient;

[0037] A non-local spatial diffusion integral term containing an anisotropic kernel function;

[0038] A stress-assisted diffusion term driven by hydrogen ion concentration gradient.

[0039] In a preferred embodiment, in the step S4, the specific operation of constructing a multi-objective optimization function is:

[0040] A1, first receive new inspection data and synchronously generate a corrosion propagation path heat map, then compare the current corrosion heat map with historical data, calculate the spatio-temporal alignment difference tensor and use gradient calculation to optimize the model prediction;

[0041] A2, define an innovative three-order optimization objective function, which includes three main sub-items:

[0042] Geometric topological loss: measures the change of the corrosion morphology of the material, and uses gradient and tensor operations to describe the change of the material;

[0043] Corrosion risk functional: calculates the weighted sum of the corrosion risk of the material, and predicts the future corrosion risk based on the stress depth change at different time points;

[0044] Memory entropy: used to estimate the learning ability of the neural network in the past experience, reflecting the memory and adaptive ability of the model.

[0045] In a preferred embodiment, the dual-network coupling mechanism is specifically:

[0046] Two neural networks work together to update parameters through a gradient projection operator, and a Jacobian response tensor is introduced to further optimize the model by calculating the relationship between network weights and the predicted response of corrosion.

[0047] In a preferred embodiment, the trigger condition for dynamic updating is:

[0048] If the Frobenius norm of weight update is greater than the product of the set threshold and the failure factor, the network parameter update is triggered, otherwise it is not triggered.

[0049] In a preferred embodiment, the generated corrosion risk decision matrix is constructed by:

[0050] A1, Decision Matrix Framework: Construct a two-dimensional decision matrix composed of m rows and n columns of decision units, where each decision unit is bound to a preset spatial position coordinate;

[0051] A2, Decision Unit Calculation: The numerical value of each decision unit is generated by a core decision function, and the function execution sequence includes:

[0052] Calculate the third-order mixed partial derivative of the corrosion propagation thermodynamic map with respect to the specified spatial coordinates, time variables and horizontal stress;

[0053] Multiply the above partial derivative by the risk accumulation factor and input it into the activation function for normalization processing;

[0054] A3, Risk Factor Construction: The risk accumulation factor is composed of two multiplications:

[0055] Stress-depth coupling term: Perform a continuous multiplication operation on the first, second and third order gradient modulus of the equivalent stress field, and take the output value of the hyperbolic tangent function;

[0056] Failure time gain term: Natural exponential function with critical failure time as the power index, where the exponential coefficient is a preset material sensitivity constant;

[0057] A4, Matrix Output: The numerical values of all decision units form the corrosion risk decision matrix, and the row and column dimensions of the matrix are consistent with the spatial segmentation division of the pipeline.

[0058] The natural gas pipeline corrosion rate prediction method has the beneficial effects that: the method realizes high-precision dynamic early warning through four-step synergistic technology, first, a three-dimensional corrosion base film covering the pipeline is generated by fusing visible light and radar data, effectively eliminating the blind area caused by complex environment; then, the corrosion probability distribution of the hidden area is inferred based on the physically constrained generative adversarial network, solving the prediction problem of the hidden area; then, the space-time diffusion path of the corrosion is deduced by combining the stress field and the historical data through the graph neural network, and the future corrosion heat map is output; finally, the model parameters are dynamically adjusted according to the measured data, and a closed-loop optimized corrosion risk decision matrix is formed, thereby realizing the transformation of the pipeline corrosion from passive protection to intelligent prevention. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 The method flowchart of the present application is provided. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0061] In the description of the present application, the terms "first", "second" are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0062] In the description of the present application, the term "for example" is used to indicate "as an example, illustration or explanation". Any embodiment described as "for example" in the present application is not necessarily interpreted as more preferred or more advantageous than other embodiments. The following description is given in order to enable any person skilled in the art to implement and use the present application. In the following description, details are listed for the purpose of explanation. It should be understood that those skilled in the art can recognize that the present application can be implemented without using these specific details. In other examples, well-known structures and processes will not be described in detail to avoid unnecessary details making the description of the present application obscure. Therefore, the present application is not intended to be limited to the shown embodiments, but is consistent with the broadest scope in accordance with the principles and characteristics disclosed in the present application.

[0063] Embodiment 1

[0064] The present embodiment provides a method for predicting the corrosion rate of a natural gas pipeline, which comprises the following steps: Figure 1The illustrated natural gas pipeline corrosion rate prediction method specifically comprises the following steps:

[0065] Step S1, when receiving visible light images, synthetic aperture radar data and environmental sensing data, performing a registration operation of a cross-modal deformation energy function on the texture gradient of the visible light images and the polarization scattering matrix of the synthetic aperture radar to generate a three-dimensional corrosion feature base film covering the pipeline surface;

[0066] Step S2, when there is an environmental occlusion area in the three-dimensional corrosion feature base film, taking the observed area features of the base film as input conditions, combining environmental sensing parameters and pipeline material characteristic parameters, and generating an occlusion area corrosion probability distribution map through a physically constrained generative adversarial network;

[0067] Step S3, when acquiring pipeline stress distribution data and historical corrosion data, taking the corrosion probability distribution map as an initial field, and using a graph neural network to couple a corrosion-stress-diffusion equation to perform a spatiotemporal inference operation to output a corrosion propagation path heat map for a future preset time period;

[0068] Step S4, when new inspection data is input, constructing a multi-objective optimization function according to the difference between the predicted value and the measured value, dynamically updating the weight parameters of the generative adversarial network and the graph neural network, and generating a corrosion risk decision matrix.

[0069] In this embodiment, it is specifically noted that in step one, the visible light images, synthetic aperture radar data and environmental sensing data are synchronously collected by the following methods:

[0070] The visible light images are collected by a multi-spectral gimbal camera mounted on a patrol unmanned aerial vehicle at a sub-meter positioning accuracy on a preset track, and the imaging wavelength band covers the visible light and near-infrared spectral range of 400-900 nm. The synthetic aperture radar data is obtained by penetrating through the ground cover through an X-band polarized interference radar on the same platform, and the radar incidence angle adjustment range is set to 25°-50° to adapt to the change of the elevation angle of different terrains. The environmental sensing data is monitored in real time by a sensor node network buried in the soil around the pipeline, and the sampling parameters at least include three key physical quantities of temperature, conductivity and pH value. All collected data streams are aligned to a unified time stamp and geographic coordinate frame through the Beidou space-time reference system, and the fusion preprocessing including POS information association, radar echo intensity correction and environmental parameter outlier filtering is completed when the data is transmitted to the edge computing node. Finally, a three-dimensional space-time synchronous data package meeting the receiving conditions of step one is formed;

[0071] The specific process of the registration operation is as follows:

[0072] First, the feature tensor construction is performed; for the visible light image, the spatial gradient tensor is calculated to enhance the edge features of the image, and the Sobel operator is used for convolution operation on the image to obtain the gradient information of each pixel point, and the calculation formula is:

[0073] ;

[0074] wherein, represents the spatial gradient tensor of the visible light image, represents the gradient of the image on the color channel (red, green or blue), represents the edge strength of the image, represents the convolution operator, represents the Sobel operator kernel, which is used to calculate the gradient of the image and emphasize the edges of the image, represents the square operation, which is used to calculate the intensity of the gradient and enhance the edge features in the image;

[0075] For synthetic aperture radar data, the polarization scattering entropy is calculated to describe different polarization features by extracting scattering features, which are used to further analyze the physical information in the image, and the calculation formula is:

[0076] ;

[0077] wherein, represents the scattering feature (such as scattering entropy) in the synthetic aperture radar data, which is calculated at the position , represents the tensor representation of the polarization scattering matrix of the synthetic aperture radar data, which contains multiple polarization channels (such as HH, HV, VH, VV), represents the conjugate transpose operation, which is to transpose the matrix and take the conjugate, represents the operator or weight, which depends on the polarization decomposition model, and is used to map the scattering matrix to certain polarization features (such as scattering entropy), and wherein, represents the probability distribution of the polarization component in the scattering matrix (for example, HH, HV, VH, etc.), represents the logarithm operation with base 3, which is usually used in polarization entropy calculation, and is used to measure the uncertainty or mixing degree of the scattering matrix, and the formula calculates the entropy value of the scattering matrix, which reflects the complexity and information amount of the scattering features under different polarization states;

[0078] Next, the cross-modal deformation energy function is established, which includes three parts: feature consistency constraint, thin plate spline constraint and thermal deformation compensation, and the expression is:

[0079] ;

[0080] wherein, is a function of cross-modal deformation energy, is a two-dimensional non-rigid transformation field, which is used to describe the deformation of aligning two data sources (optical image and PolSAR image), is a function of mapping the scattering characteristics of SAR data to an optical feature space, which is realized by a Wishart likelihood classifier, which is used to convert the polarization characteristics of SAR to the features in the optical image space, is a function of encoding the gradient direction of the optical image, which is usually represented by a histogram of oriented gradient (HOG) to capture the edge features of the image, is a function of the polarization scattering characteristics (such as scattering entropy, etc.) extracted from synthetic aperture radar data, which is registered with the features of the optical image, is a function of the spatial gradient features extracted from the optical image, which is used to represent the edge and texture information in the image, is a function of applying the transformation field to the HOG features, which is used to describe the result of non-rigid transformation of the features of the image, is a function of the weight of the feature consistency constraint, which is used to balance the matching accuracy of different data source features, is a function of the thin plate spline constraint, which requires the transformation field to be as smooth as possible in space, i.e. the curvature of the deformation, which helps to avoid excessive discontinuity or sharp deformation in the registration process, is a function of the weight of the thin plate spline constraint, which determines the priority of the smoothness of the transformation field, is a function of the thermal deformation compensation term, which represents the effect of temperature on the deformation field , which is used to compensate for the thermal expansion and contraction effect of metal or material caused by temperature changes, so as to ensure accurate modeling of deformation, especially in the deformation process of material or metal, is a function of the amount of temperature change, which indicates the compensation effect of temperature change on the deformation length, is a function of the weight of thermal deformation compensation, which is used to control the contribution of temperature compensation; the feature consistency constraint ensures the alignment of features between PolSAR and optical image, the thin plate spline constraint is used to smooth the transformation field to avoid excessive deformation, and the thermal deformation compensation improves the accuracy by compensating for the metal expansion effect caused by temperature gradient;

[0081] In the optimization phase, an improved Alternating Direction Method of Multipliers (ADMM) algorithm is employed to iteratively solve the two-dimensional non-rigid transformation field, to optimize the cross-modal deformation matching process; this optimization process updates the transformation field continuously, ultimately achieving accurate fusion of data, and its expression is:

[0082] ;

[0083] wherein, represents the value of the transformation field at the th iteration, which is the solution of the objective function, represents the deformation of image registration in the current iteration, represents the transformation of image data, represents the target data in the th iteration, represents the regularization parameter, which is used to balance different terms in the optimization objective, helping to stabilize the optimization process, represents the Lagrange multiplier, which is used for the processing of the constraint term, and is updated at each iteration to ensure the satisfaction of the constraint condition, represents the transformation field in the th iteration, which represents the transformation effect of the current iteration step, represents the change of the transformation field in the iteration process, which is used to update the Lagrange multiplier;

[0084] Finally, the optimized transformation field is used to generate a three-dimensional erosion feature base film, and its expression is:

[0085] ;

[0086] wherein, represents the output three-dimensional base film, i.e. the base film generated at the spatial position , which is established by the fusion of deformation and image features, represents the real part of the complex scattering feature of the sar image, here represents the inverse transformation of the transformation field to the sar feature , represents taking the real part of the complex number, sar images usually contain scattering features in the form of complex numbers, reflecting different scattering and polarization information, represents the imaginary part of the complex scattering feature of the sar image, represents taking the imaginary part of the complex number, the imaginary part in sar data reflects the phase information, which is crucial for deformation modeling, represents the curvature feature of the visible light image, which is used to capture the changes of edges or textures in the image, the curvature can be calculated by the determinant (det(Hessian)) of the Hessian matrix of the image, The curvature feature represents the morphological details of the image, such as curves and surface changes, and is combined with the real part and the imaginary part of the complex scattering feature of the SAR data and the curvature feature of the visible light image, so as to effectively represent the corrosion condition; the core innovation of this method is to introduce a thermal deformation Jacobian term to compensate for the deformation caused by the temperature gradient, thereby improving the accuracy of corrosion detection and evaluation.

[0087] In this embodiment, the specific operation of step S2 is to generate a specific operation of the occlusion area corrosion probability distribution map:

[0088] A1, first, the observed area feature of the base film (i.e. the three-dimensional corrosion feature base film ) is taken as input, which includes depth information and spatial coordinates in the image , which helps to determine which areas in the image have unreliable depth information or belong to occlusion areas, second, identify the occlusion area in the image, specifically, the area whose depth feature is unreliable, by calculating the gradient of the base film in the depth dimension, if the gradient value is less than a preset threshold (0.05), mark these areas as occlusion areas, the expression is:

[0089] ;

[0090] Wherein, the expression defines an occlusion area, wherein is the gradient of the base film in the depth dimension (indicating the degree of depth change), if the gradient is less than a threshold , it means that the depth information of the area is unreliable and belongs to the occlusion area, represents a preset threshold, which is 0.05, used to judge the reliability of the depth information, if the gradient is less than 0.05, it is considered that the depth information is unreliable, represents the depth dimension feature of the base film, i.e. the data in the depth direction of the three-dimensional corrosion base film;

[0091] A2, fuse the environmental sensor parameters (such as temperature, corrosion medium density, pH value) and the characteristic parameters of the pipeline material (such as electrochemical potential, iron ion activity coefficient, corrosion rate constant), wherein, the set of environmental sensor parameters is used to describe the influence of environmental conditions on corrosion, and the set of characteristic parameters of the pipeline material is used to describe the influence of the characteristics of the material itself on corrosion, These parameters provide necessary physical constraints for subsequent corrosion feature generation;

[0092] A3、Then, a dual-channel network is constructed for generating the corrosion feature map, which includes three modules:

[0093] Feature encoder: The input base film features (including coordinate depth data) and environmental sensor data are fused to generate a feature vector containing spatial correlation information, expressed as:

[0094] ;

[0095] wherein, represents the spatial correlation feature vector output by the feature encoder, with a dimension of , i.e., a 256-dimensional feature vector, represents a convolutional gated recurrent unit, a model combining convolutional neural networks and gated recurrent units, used to extract spatio-temporal features, respectively represent three-dimensional corrosion feature base films observation data in directions, represents an environmental sensing parameter vector, containing temperature, corrosion medium density and pH value environmental information;

[0096] Physical constraint: By processing material characteristic parameters and environmental sensitivity of corrosion rate, a physically constrained output is generated, expressed as:

[0097] ;

[0098] wherein, represents the physical constraint information output by the physical constraint, represents a multi-layer perceptron, a feedforward neural network used to map the input features to the output space, represents a pipe material characteristic parameter vector, including electrochemical potential, iron ion activity coefficient and corrosion rate constant, represents a concatenation operation that concatenates two vectors or matrices together, represents a diagonal matrix of matrix , wherein is the corrosion rate environmental sensitivity matrix, defined as: , indicating the sensitivity of environmental sensing parameters (temperature, density, pH value) to the corrosion rate constant ;

[0099] Corrosion generator: Combining random noise, encoded feature vector and output of the physical constraint, the final corrosion feature map is generated, expressed as:

[0100]

[0101] wherein, represents the final generated corrosion probability distribution map, the corrosion prediction on the occluded region , represents the generator function for generating the final corrosion image, represents a noise vector sampled from the standard normal distribution for generating randomness in the adversarial network, represents the feature vector output by the feature encoder, represents the physical constraint information output by the physical constraint, represents the vector splicing operation;

[0102] In the training process, the output of the generator is optimized through adversarial training, and the training target is to minimize two loss functions:

[0103] Dual discriminant loss: This loss function optimizes the difference between the generated image and the real image, encouraging the generator to generate more realistic corrosion features, and its expression is:

[0104] ;

[0105] where, represents the adversarial loss function for training the adversarial between the generator and the discriminator, represents the expected value, i.e. the discriminant result of the discriminator for the input corrosion generated image , the goal is to let the discriminator better distinguish between real images and generated images, represents the expected value, i.e. the discriminant result of the discriminator for the generated image , the goal is to let the generator generate more realistic images, represents the image generated by the generator, represents the noise vector sampled from the standard normal distribution, represents the output image of the corrosion generator (i.e. the corrosion probability distribution map);

[0106] Physical loss: This loss measures the difference between the generated corrosion features and the actual corrosion driving force field, ensuring that the generated features conform to the physical law, and the corrosion driving force field is related to the iron ion concentration gradient and pH change, and its expression is:

[0107]

[0108] where, represents the physical regularization loss for introducing physical knowledge into the training process of the generator, represents the gradient of the corrosion generated image for the input spatial position , represents the spatial variation rate of the image, represents the corrosion driving field, positioned as: , wherein, represents the gas constant, represents the temperature, represents the iron ion diffusion coefficient, used to describe the diffusion ability of iron ions in the material (related to the material), represents the corrosion feature base film In the second order gradient in spatial dimension, representing the change of corrosion feature in space;

[0109] A4, after optimization by training, finally output an accurate corrosion probability distribution map, which represents the corrosion possibility of each region (including occluded regions) in the image, with high accuracy and stability, the output form is:

[0110] ;

[0111] wherein, of represents a position point in the image, represents a specific region or set, usually refers to "occupied region", that is, the region where corrosion occurs in the image or space, which means when the position is located in the occupied region, the generated probability distribution map depends on the maximum value in the formula; represents the generated image by the generator under the given input noise , represents the noise input of each Monte Carlo sampling, represents the number of Monte Carlo sampling, that is, by sampling the noise space multiple times to ensure the probability stability of the generated image, and means using 128 times of Monte Carlo sampling to ensure the stability of the result, which is to reduce noise and random fluctuations by multiple sampling and averaging (or selecting the maximum value) to make the generated image more consistent with the actual probability distribution, represents the value of the generated corrosion feature base film at position , represents that in times of Monte Carlo sampling, the maximum value is selected as the output to enhance the stability and reliability of the result, represents when the position does not belong to the occupied region , the generated probability distribution map takes another value, which usually represents the probability distribution of non-corrosion region;

[0112] The specific steps of performing space-time deduction operation are:

[0113] A1, the corrosion probability distribution map outputted by step S2 A2, the corrosion probability distribution map outputted by step S2 A3, the stress tensor of the pipeline A4, the history corrosion time series database A5, the horizontal stress of the pipeline A6, the vertical stress of the pipeline A7, the shear stress of the pipeline A8, the time series database containing the history corrosion data A9, the corrosion feature at position A10, the corrosion feature at time A11, the initial time A12, the time step A13, the time step number A14, the total time step number A15, the total time step number

[0114] A2, the graph structure composed of corrosion feature nodes, stress feature nodes and diffusion channel edges, wherein:

[0115] A21, the corrosion feature node A22, the spatial coordinate bound corrosion feature node A23, the corrosion probability value of the corresponding position A24, the history change gradient of the corrosion probability value

[0116] A25, the stress feature node A26, the equivalent stress value of the stress feature node A27, the equivalent stress value calculated according to the horizontal principal stress, vertical stress and shear stress components

[0117] A28, the diffusion channel edge A29, the weight of the diffusion channel edge determined by the curvature weighting function of the distance between adjacent nodes and the difference value of the equivalent stress A30, the weight of the diffusion channel edge defined as: A31, the weight of the diffusion channel edge defined as:

[0118]

[0119] A32, the edge connecting nodes A33, the edge connecting nodes A34, the edge connecting nodes A35, the edge connecting nodes A36, the material-related corrosion correlation scale A37, the material-related corrosion correlation scale A38, the Euclidean distance between nodes A39, the Euclidean distance between nodes A40, the Euclidean distance between nodes denotes the position curvature coupling factor, which measures the curvature difference between positions, this factor affects how stress variation couples with diffusion process, denotes the node and between the equivalent stress difference, the equivalent stress takes into account the stress influence in different directions, affecting the progress of corrosion, denotes the activation function, which is used to compress a certain value to the range of [0, 1], here it is used to adjust the influence of stress difference on diffusion channel weight;

[0120] A3, input the graph structure into the graph neural network, generate the corrosion depth prediction field of the future time period by iterative solving of the corrosion-stress-diffusion coupling equation, spatially discretize and encode the corrosion depth prediction field of the current time step, superimpose the exponential decay memory effect of the historical prediction field, and output the corrosion propagation path heat map, the formula of the corrosion-stress-diffusion coupling equation is:

[0121] ;

[0122] where, denotes the corrosion depth or corrosion state, which is used to describe the corrosion depth at a certain position and time, denotes the effective stress, which is used to describe the stress state inside the material during corrosion, affecting the corrosion rate, denotes the constant related to stress corrosion behavior, describing the influence of stress corrosion on corrosion rate, denotes the non-local diffusion kernel function, which describes how the corrosion state at a certain point is affected by the surrounding area, and where, denotes the position vector, i.e. the distance between two position points, denotes the covariance matrix, which defines the anisotropy of diffusion, controlling the behavior of the diffusion process in different directions, and where denotes the diffusion coefficient in direction, describing the anisotropic diffusion characteristics of the material, denotes the stress-diffusion coupling factor, describing the coupling strength between stress field and diffusion process, indicating how stress affects diffusion; denotes the determinant of matrix , which is used to normalize the kernel function, denotes the inverse matrix of matrix , which is used to calculate the direction and strength of diffusion, denotes the spatial position variable, i.e. different positions of the corrosion state distribution, denotes the entire calculation region, represents the hydrogen diffusion coefficient, describing the speed of hydrogen diffusion in the material, represents the hydrogen stress field, the stress effect of hydrogen on the material, usually caused by the accumulation of hydrogen, represents the gradient operator, used to describe the spatial variation of the hydrogen diffusion or stress field, represents the constant of hydrogen-assisted diffusion, controlling the influence of hydrogen diffusion on the corrosion process,

[0123] The expression of the graph neural network spatiotemporal propagation operator is:

[0124] ;

[0125] where, represents the node in the layer feature vector, represents the node in the layer feature vector, represents the gated recurrent unit, a neural network structure for sequence data. Here it is used to capture the spatiotemporal relationship between nodes, represents the node in the neighbor node set, i.e. all nodes directly connected to , represents the edge transition function, used to describe the influence of the neighbor node features on the current node features, represents the edge weight, used to describe the connection strength between the node and the node ; represents the hyperbolic tangent function, used to compress the input value to the range [-1, 1], represents the weight matrix, used to map the combination of stress gradient and time difference to the appropriate space, represents the product of stress gradient and corrosion depth, used to describe the influence of stress change inside the material on the corrosion depth, and where, represents the stress gradient, describing how the stress changes with the change of position in the material, represents the progress of corrosion, represents the rate of change of stress on corrosion depth, used to describe how corrosion affects the stress in the material, represents the time step, representing the time interval from time , used for integral calculation; represents the time difference between the node and the node , represents the weight matrix, used for linear transformation of , represents a kernel function, used to describe the influence of the distance difference between nodes on the corrosion state, represents element-wise multiplication;

[0126] The expression form of the corrosion propagation path heat map is:

[0127]

[0128] wherein, represents the value of the corrosion propagation path heat map at position and time step , used to describe the corrosion propagation state of the position at that moment, represents the spatial position, represents the predicted time step, i.e. the future time of the grinding prediction, represents the total number of time steps (or the number of historical time steps), indicating the historical time period before the current moment , represents the corrosion depth or corrosion state of position at time step , and the corrosion depth at each time step is taken as a characteristic quantity to describe the progress of material corrosion, represents an exponential decay function, i.e. the influence of the historical corrosion state gradually decreases over time, is a decay coefficient that controls how the influence of corrosion history decays over time, the larger the value, the faster the decay, represents the corresponding time of the predicted time step , represents the corresponding time of the historical time step , represents a Sigmoid activation function, used to compress the calculation result to the range of 0 to 1, making it suitable for representing the intensity of the heat map, and is used to smooth the output and map it to the [0, 1] interval;

[0129] The corrosion-stress-diffusion coupling equation includes:

[0130] A stress corrosion acceleration term based on the material stress sensitivity coefficient;

[0131] A non-local spatial diffusion integral term containing an anisotropic kernel function;

[0132] A stress-assisted diffusion term driven by hydrogen ion concentration gradient.

[0133] In this embodiment, the specific operation of step S4 for constructing the multi-objective optimization function is:

[0134] A1, first receive new inspection data , wherein, ​represents a new inspection dataset containing observed values of materials and current time, represents the spatial position of each point in the inspection data, represents the corrosion data value observed at position , represents the current time, and synchronously generates a corrosion propagation path heat map , represents the corrosion propagation path heat map generated according to the predicted time , for the position, then, by comparing the current corrosion heat map with the historical data, a spatiotemporal alignment difference tensor is calculated, and the calculation formula of the spatiotemporal alignment difference tensor is:

[0135]

[0136] wherein, represents the spatiotemporal alignment difference tensor, used to describe the corrosion change at different time points, represents the aligned time, calculated as the current time minus the initial time , represents the gradient of time, related to the rate of time change, represents the gradient of the corrosion propagation path heat map to the position, describing the spatial distribution of corrosion change, represents the gradient of the observed corrosion data in space, represents the curvature sensitivity factor (inherited from step S3 ), affecting the response of the corrosion model in different spatial regions,

[0137] This difference tensor reflects the corrosion change of the material at different time points, and gradient calculation is used to optimize the model prediction;

[0138] A2, define an innovative third-order optimization objective function, and the expression is:

[0139] ;

[0140] wherein, represents the total optimization objective function, which is a multi-objective optimization by weighting different sub-objective functions, represents the weight coefficient of the geometric topology loss, controlling the influence of the geometric topology loss on the total optimization objective, represents the weight coefficient of the corrosion risk functional, controlling the influence of the corrosion risk on the total optimization objective, represents the weight coefficient of the memory entropy, controlling the influence of the memory entropy on the total optimization objective;

[0141] which includes three main sub-items:

[0142] Geometric-topology loss: measures the change of corrosion morphology of the material, uses gradient and tensor operations to describe the change of the material, and its expression is:

[0143] ;

[0144] wherein, represents the geometric-topology loss function, used to measure the geometric morphology change of the corrosion propagation path, represents a constant related to the loss function, used to adjust the morphology of the function, represents a topology tensor, used to quantify the difference between the predicted and measured geometric characteristics, and ; wherein represents a spatio-temporal alignment difference tensor, i.e. the spatial difference between the CPH heat map and the measured value, represents the horizontal principal stress gradient, represents the horizontal principal stress Hessian, represents the tensor product, representing the product of two tensors (such as and ), represents the norm (i.e. the size) of the cross product of two gradients, and represents the corrosion prediction heat map gradient, represents the measured corrosion depth gradient, represents a defined region, i.e. a spatial region of the material surface, represents an exponential decay function, used to describe the influence of corrosion over time, represents a decay coefficient, used to control the sensitivity of geometric differences;

[0145] Corrosion risk functional: calculates the weighted sum of the corrosion risk of the material, predicts the future corrosion risk based on the stress depth change at different time points, and its expression is:

[0146]

[0147] wherein, represents the corrosion risk functional, used to calculate the corrosion risk at different positions, represents a set of key points, i.e. high-risk location coordinates that need to be monitored, represents a stress sensitivity function, i.e. the second-order mixed partial derivative of equivalent stress with respect to time-space, represents the equivalent stress, and , wherein represents the predicted start time, represents the predicted end time, represents the stress spatio-temporal evolution rate, represents the spatio-temporal alignment difference tensor, representing the corrosion change at different time points, represents the position Risk indicator function on the upper bound, used to describe the risk under certain conditions;

[0148] Memory entropy: used to estimate the learning ability of neural network in past experience, reflecting the memory and adaptive ability of the model, its expression is:

[0149]

[0150] Where, Memory entropy, measures the learning and memory ability of neural network to past data, Trace operation of matrix, get the sum of diagonal elements of matrix, Neural network weight matrix, used to describe the connection strength of each layer in the network, Neural network historical weight matrix, used to calculate the memory of network learning, Exponential function, describes the influence of gradient on memory entropy, Memory decay factor, Difference field gradient, used to describe the spatial variation rate of corrosion depth difference, Two norm, Logarithmic operation, used to describe the weight distribution difference measure;

[0151] The specific coupling mechanism of double network is:

[0152] Two neural networks work together, update parameters through gradient projection operator, and introduce a Jacobian response tensor, further optimize the model by calculating the relationship between network weights and the response of corrosion prediction. This process involves dynamic adjustment of neural network weight matrix to optimize corrosion prediction. The formula of gradient projection operator is:

[0153] ;

[0154] Where, Weight update amount, Learning rate, value , Gradient of total loss function to weight, Hadamard product (element-wise multiplication), Adaptive mask intensity coefficient, Jacobian response tensor;

[0155] The expression of Jacobian response tensor is:

[0156]

[0157] Where, Generated adversarial network weight, denotes the graph neural network weights, denotes the norm of the corrosion propagation rate vector, and

[0158] The trigger condition for dynamic updating is:

[0159] If the Frobenius norm of weight update is greater than the product of the set threshold and the failure factor, then the network parameter update is triggered, otherwise it is not triggered, and the expression is:

[0160]

[0161] wherein, denotes the Frobenius norm of weight update, which is used to measure the degree of change of weight matrix, denotes the preset threshold value, which is used to judge whether to trigger update. When the size of weight update exceeds the threshold value, it means that the weight of the network needs to be updated, denotes the failure factor, which is used to adjust the update trigger condition. Its calculation method is to dynamically adjust the trigger condition by considering the corrosion risk , specifically, it represents a probability factor that controls when to trigger the update, and wherein, denotes the natural exponential function, denotes the accumulation of corrosion risk over time. This expression measures the rate of change of corrosion risk in a certain area , and the result of integration is the cumulative effect of risk, denotes the rate of change of risk function with respect to time, which reflects the sensitivity of corrosion risk to time;

[0162] The generated corrosion risk decision matrix is constructed by the following method;

[0163] A1, decision matrix framework: construct a two-dimensional decision matrix composed of m rows and n columns of decision units, wherein each decision unit is bound to a preset spatial position coordinate, and the expression is:

[0164]

[0165] A2, decision unit calculation: the numerical value of each decision unit is generated by the core decision function, and the function execution sequence includes:

[0166] Calculate the third-order mixed partial derivative of the corrosion propagation thermal map with respect to the specified spatial coordinates, time variable and horizontal stress;

[0167] Multiply the above partial derivative by the risk accumulation factor and input it into the activation function for normalization processing, and the expression is:

[0168] ​​​ ;

[0169] in, This represents the value of the decision matrix unit (i.e., the position of the pipeline). (corrosion risk level). This represents the third-order mixed partial derivative of the corrosion propagation thermogram with respect to spatial coordinates / time / stress. This indicates activation function normalization. This represents the optimized corrosion prediction heatmap (the spatiotemporal distribution field of corrosion depth output in step S3). Spatial coordinate index: Pipe surface index Each normalized grid position (inherited from step S1) system), Represents time variables (points on the prediction timeline). It represents the horizontal principal stress (the horizontal component of the duct stress tensor). This represents the risk accumulation factor (an enhancement coefficient composed of stress gradient and failure time). Furthermore, if... If , it indicates a safe zone (no intervention required). If it is, it indicates a high-risk area (requiring urgent maintenance);

[0170] A3. Risk Factor Construction: The risk accumulation factor consists of the product of two items:

[0171] Stress deep coupling term: After performing a series multiplication operation on the first, second and third gradient magnitudes of the equivalent force field, the hyperbolic tangent function output value is taken;

[0172] Failure time gain term: A natural exponential function with the critical failure time as the power, where the exponent coefficient is a preset material sensitivity constant, and its expression is:

[0173] ;

[0174] in, This represents the hyperbolic tangent function, used for nonlinear adjustment of risk. It smooths the accumulation of stress gradients. This represents the gradient product of effective stress, reflecting the spatial distribution of stress variations. Generally, a larger stress gradient indicates a higher risk of corrosion. This represents an exponential function, indicating the failure time gain. It is related to the failure time. The correlation reflects the cumulative risk effect of the corrosion process over time;

[0175] A4. Matrix Output: The values ​​of all decision units form a corrosion risk decision matrix, and the row and column dimensions of this matrix are consistent with the pipeline space segmentation.

[0176] It should be noted that the descriptions of the various embodiments are each given with emphasis on certain features of the embodiments. The descriptions of the other embodiments can be understood with reference to the descriptions of the other embodiments.

[0177] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, a system, or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects all generally referred to herein as a "circuit" or "module." Furthermore, the present application can take the form of a computer program product on a computer-readable storage medium having computer program code embodied in the medium.

[0178] The present application is described in reference to the drawings using a flow diagram and / or a block diagram of the method, apparatus (system) and computer program product according to embodiments of the application. It will be understood that each block of the flow diagram and / or block diagram, and combinations of blocks in the flow diagram and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions of the flow diagram and / or block diagram. Figure 1 one or more functions specified in the flow or flows and / or blocks. Figure 1 means for performing each of the functions specified in the flow or flows and / or blocks.

[0179] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flow diagram and / or block diagram. Figure 1 one or more functions specified in the flow or flows and / or blocks. Figure 1 means for performing each of the functions specified in the flow or flows and / or blocks.

[0180] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow diagram and / or block diagram. Figure 1 one or more functions specified in the flow or flows and / or blocks. Figure 1 means for performing each of the functions specified in the flow or flows and / or blocks.

[0181] While the preferred embodiments of the application have been described, additional variations and modifications can be made to these embodiments by those skilled in the art once they have the benefit of the present disclosure without departing from the spirit and scope of the application. Accordingly, it is intended that the appended claims include all such modifications and variations as fall within the scope of the present application.

[0182] It is apparent that those skilled in the art can make various changes and modifications to the application without departing from the spirit and scope of the application. It is therefore intended that the present application cover all such changes and modifications that are within its scope.

Claims

1. A method of predicting the corrosion rate of a natural gas pipeline, characterized by, Specifically comprising the following steps: Step S1, when receiving visible light images, synthetic aperture radar data and environmental sensing data, performing a registration operation of a cross-modal deformation energy function on the texture gradient of the visible light images and the polarization scattering matrix of the synthetic aperture radar to generate a three-dimensional corrosion feature base film covering the surface of the pipeline; Step S2, when there is an environmental occlusion area in the three-dimensional corrosion feature base film, taking the observed area features of the base film as input conditions, combining environmental sensing parameters and pipeline material characteristic parameters, and generating an occlusion area corrosion probability distribution map through a physically constrained generative adversarial network; Step S3, when pipeline stress distribution data and historical corrosion data are obtained, taking the corrosion probability distribution map as an initial field, and using a graph neural network to couple a corrosion-stress-diffusion equation to perform a spatiotemporal inference operation, and output a corrosion propagation path heat map in a future preset period; Step S4, when new inspection data is input, constructing a multi-objective optimization function according to the difference between the predicted value and the measured value, dynamically updating the weight parameters of the generative adversarial network and the graph neural network, and generating a corrosion risk decision matrix.

2. The method of claim 1, wherein: The visible light images, synthetic aperture radar data and environmental sensing data in step S1 are synchronously collected by the following method: The visible light images are collected by a multi-spectral gimbal camera mounted on a pipeline inspection unmanned aerial vehicle on a preset track, and the imaging wavelength band covers the visible light and near-infrared spectral range of 400-900 nm. The synthetic aperture radar data is obtained by penetrating through the ground cover through an X-band polarized interferometric radar on the same platform, and the radar incidence angle adjustment range is set to 25°-50° to adapt to the change of the elevation angle of different terrains. The environmental sensing data is monitored in real time by a sensor node network buried in the soil around the pipeline, and the sampling parameters at least include temperature, conductivity and pH value.

3. The natural gas pipeline corrosion rate prediction method according to claim 2, characterized in that: The specific process of the registration operation is as follows: First, the feature tensor is constructed; for the visible light image, the spatial gradient tensor is calculated to enhance the edge features of the image, and the Sobel operator is used for convolution operation to obtain the gradient information of each pixel point; for the synthetic aperture radar data, the polarization scattering entropy is calculated to describe different polarization features by extracting the scattering features; Next, a cross-modal deformation energy function is established, which includes three parts: feature consistency constraint, thin plate spline constraint and thermal deformation compensation; In the optimization stage, an improved alternating direction multiplier method algorithm is used to iteratively solve the two-dimensional non-rigid transformation field to optimize the matching process of cross-modal deformation; Finally, the optimized transformation field is used to generate a three-dimensional corrosion feature base film, which combines the real and imaginary parts of the SAR data complex scattering features and the curvature features of the visible light image.

4. The method of claim 3, wherein: In step S2, the specific operation of generating the occlusion area corrosion probability distribution map is as follows: A1、First, the observation area characteristics of the base film are taken as input, including depth information and spatial coordinates in the image. Second, the occluded areas in the image are identified, specifically those areas where the depth features are not reliable. By calculating the gradient of the base film in the depth dimension, if the gradient value is less than a pre-set threshold, these areas are marked as occluded areas; A2、Fusion of environmental sensor parameters and characteristic parameters of pipe materials; A3、Next, a dual-channel network is constructed to generate corrosion feature maps, which includes three modules: Feature encoder: fuse the input base film features and environmental sensor data to generate a feature vector containing spatial correlation information; Physical constraint: generate a physically constrained output by processing material characteristic parameters and the environmental sensitivity of corrosion rate; Corrosion generator: combine random noise, encoded feature vector, and physical constraint output to generate the final corrosion feature map; During training, the generator's output is optimized through adversarial training, with the goal of minimizing two loss functions: Dual-discriminator loss: This loss function optimizes the difference between the generated image and the real image, encouraging the generator to produce more realistic corrosion features. Physical loss: This loss measures the difference between the generated corrosion features and the actual corrosion driving force field, which is related to the iron ion concentration gradient and pH change. A4、After optimization through training, an accurate corrosion probability distribution map is finally output.

5. The method of claim 4, wherein: The specific steps for performing the space-time inference operation are as follows: A1、Take the corrosion probability distribution map output in step S2 as the initial field of spatial corrosion depth, and load the pipe stress tensor and historical corrosion time sequence library; A2、Establish a graph structure composed of corrosion feature nodes, stress feature nodes, and diffusion channel edges, where: The corrosion feature node is bound to a spatial coordinate and carries the corrosion probability value at the corresponding position and its historical change gradient. The stress feature node carries the equivalent stress value, which is calculated based on the horizontal principal stress, vertical stress, and shear stress components. The diffusion channel edge weight is determined by the curvature weighting function of the distance between adjacent nodes and the equivalent stress difference. A3、Input the graph structure into a graph neural network to generate the corrosion depth prediction field for the future time period through iterative solution of the corrosion-stress-diffusion coupling equation. After spatial discretization coding of the corrosion depth prediction field at the current time step and superimposing the exponential decay memory effect of the historical prediction field, the corrosion propagation path heat map is output.

6. The method of claim 5, wherein: The corrosion-stress-diffusion coupling equation includes: Stress corrosion acceleration term based on material stress sensitivity coefficient; Non-local spatial diffusion integral term containing anisotropic kernel function; Stress-assisted diffusion term driven by hydrogen ion concentration gradient.

7. A method of predicting a corrosion rate of a natural gas pipeline as claimed in claim 6, wherein: In step S4, the specific operation of constructing a multi-objective optimization function is as follows: A1、First, receive new inspection data and generate a corrosion propagation path heat map synchronously. Then, by comparing the current corrosion heat map with historical data, calculate the space-time alignment difference tensor and optimize the model prediction using gradient calculation; A2、Define a third-order optimization objective function, which includes three main sub-items: Geometric topology loss: measures the morphological change of the material corrosion, using gradient and tensor operations to describe the material change; Corrosion risk functional: calculates the weighted sum of the material corrosion risk, based on the stress-depth change at different time points to predict the future corrosion risk; Memory entropy: used to estimate the learning ability of the generative adversarial network and the graph neural network in the past experience, reflecting the memory and adaptive ability of the model.

8. The method of claim 7, wherein: The specific dual-network coupling mechanism is: The generative adversarial network and the graph neural network work together, the parameter update is carried out through the gradient projection operator, and a Jacobian response tensor is introduced, the relationship between the network weights and the corrosion prediction response are calculated, and the model is further optimized.

9. The method of claim 8, wherein: The trigger condition for dynamic updating is: If the Frobenius norm of weight update is greater than the product of the set threshold and the failure factor, the network parameter update is triggered, otherwise, it is not triggered.

10. The method of claim 9, wherein: The generated corrosion risk decision matrix is constructed by the following methods: A1, decision matrix framework: a two-dimensional decision matrix composed of m rows and n columns of decision units is constructed, and each decision unit is bound to a preset spatial position coordinate; A2, decision unit calculation: the numerical value of each decision unit is generated by the core decision function, and the function execution sequence includes: Calculate the third-order mixed partial derivative of the corrosion propagation thermodynamic map with respect to the specified spatial coordinates, time variable and horizontal stress; Multiply the above partial derivative by the risk accumulation factor and input it into the activation function for normalization processing; A3, risk factor construction: the risk accumulation factor is composed of two multiplications: Stress-depth coupling term: perform a continuous multiplication operation on the first-order, second-order and third-order gradient modulus of the equivalent stress field, and take the tangent hyperbolic function output value; Failure time gain term: a natural exponential function with critical failure time as the power index, where the exponential coefficient is a preset material sensitivity constant; A4, matrix output: the numerical values of all decision units form the corrosion risk decision matrix, and the row and column dimensions of the matrix are consistent with the spatial segmentation division of the pipeline.

Images