A pipeline corrosion failure spatiotemporal joint prediction method and system fusing multi-source physical information

Abstract

This invention discloses a spatiotemporal joint prediction method and system for pipeline corrosion failure integrating multi-source physical information, belonging to the field of pipeline corrosion prediction technology. The method includes: collecting, cleaning, spatially matching, and temporally aligning multi-source pipeline data to form a unified pipe segment-level state input; extracting physical features and, through key variable screening and feature group mapping, classifying relevant variables into electrochemical information, pressure load information, spatial location information, corrosion state information, and material structure information; constructing an adaptive multi-physics coupling graph to transform different physical relationships into graph edge relationships; using a GAT-Mamba spatiotemporal joint prediction model to output corrosion depth, corrosion rate, failure pressure, safety margin, remaining life, and initial risk score; updating multi-physics edge weights according to the degree of physical constraint violation and constructing a high-risk protection domain; and finally, correcting the initial risk score to form a pipe segment-level risk assessment result and maintenance decision recommendations.

Description

Technical Field

[0001] This invention belongs to the field of pipeline corrosion prediction technology, specifically relating to a spatiotemporal joint prediction method and system for pipeline corrosion failure that integrates multi-source physical information. Background Technology

[0002] Currently, pipeline corrosion status assessment mainly relies on internal inspection data, operating pressure data, cathodic protection data, environmental medium data, soil environmental data, and human experience. Existing pipeline corrosion failure prediction methods mainly include empirical formula methods, statistical regression methods, machine learning methods, and neural network methods. Empirical formula methods typically estimate corrosion based on corrosion rate, remaining wall thickness, defect size, or failure pressure calculation models, and have the advantages of simple calculation and strong engineering interpretability, but lack adaptability to complex operating conditions. Statistical regression and traditional machine learning methods can establish a mapping relationship between input features and corrosion status using historical inspection data, but they often fail to adequately express the coupling relationship between multiple physical factors. In recent years, methods such as BP neural networks, support vector machines, LSTM, graph neural networks, BiLSTM, and attention have been used to predict corrosion rate, corrosion depth, or remaining life, which can improve data fitting ability in some scenarios, but most methods still focus on data-driven prediction and lack explicit expression of corrosion mechanisms, spatial correlations, and pressure safety boundaries.

[0003] In practical pipeline corrosion failure prediction, the corrosion state is not determined by a single factor, but rather by the combined effects of multiple physical factors. The corrosion rate is related to electrochemical behavior, environmental medium, protection status, and material condition; the future corrosion depth is related to historical corrosion status, remaining wall thickness, and corrosion rate evolution; the failure pressure is related to remaining wall thickness, corrosion depth, pipe diameter, material strength, and operating pressure; and the remaining life and risk level are closely related to corrosion growth trends, pressure safety margin, degree of physical constraint violation, and maintenance strategies. Furthermore, different pipe sections along the pipeline are not independent of each other; adjacent sections may have spatial correlations due to similar soil environments, similar pressure conditions, similar electrochemical conditions, or the same material structure. Therefore, pipeline corrosion failure prediction not only needs to consider multi-source data fusion, spatial correlation modeling, temporal evolution prediction, and physical constraint output, but also requires further consideration of the adaptive adjustment of different physical relationship weights with changing risk status, as well as protection mechanisms for high-risk samples during the risk assessment process.

[0004] While existing data-driven or conventional graph neural network methods can predict corrosion rate, corrosion depth, or remaining life, most methods employ fixed feature relationships or graph edge weights during modeling. This makes it difficult to dynamically adjust the model's focus based on physical violations such as insufficient pressure safety margin, violation of remaining wall thickness constraints, abnormal risk prioritization, abrupt changes in corrosion state, or anomalies in spatial gradients. For pipe sections located near pressure safety boundaries, with significant physical constraint violations, abnormal corrosion rates, or prominent risk propagation characteristics, directly outputting risk levels based solely on conventional predictions may result in high-risk samples being overwhelmed by ordinary samples, affecting the reliability of maintenance priority ranking.

[0005] In summary, current technologies still lack a pipeline corrosion failure prediction method that can simultaneously achieve multi-source physical state modeling, adaptive multi-physics edge weight update, physical violation evaluation, high-risk protection domain identification, risk score correction, and segment-level maintenance priority output. Therefore, it is necessary to propose a pipeline corrosion failure prediction method based on adaptive multi-physics coupling and high-risk protection domains to improve the reliability of corrosion failure risk identification and the pertinence of maintenance decisions. Summary of the Invention

[0006] The purpose of this invention is to propose a pipeline corrosion failure prediction method based on adaptive multi-physics coupling and high-risk protection domains, addressing technical problems in pipeline corrosion failure prediction such as fixed multi-physics relationships, insufficient identification of physical anomalies, underestimation of high-risk pipe sections, lack of risk score correction, and insufficient support for pipe section-level maintenance decisions. By constructing multi-source physical data and selecting key variables, features such as pressure load, electrochemical behavior, spatial location, corrosion state, and material structure are mapped to different physical relationships. Based on this, a multi-physics coupling graph is constructed, and different physical boundary weights are adaptively updated according to the degree of physical violation. Subsequently, the GAT-Mamba spatiotemporal prediction model is used to obtain corrosion depth, corrosion rate, failure pressure, safety margin, remaining life, and initial risk score. Finally, through physical violation evaluation, high-risk protection domain construction, and risk correction mechanisms, pipe section-level risk levels, protection domain proportions, and maintenance priorities are output, thus providing technical support for pipeline integrity evaluation and maintenance decisions.

[0007] To achieve the above objectives, the present invention provides the following solution: a spatiotemporal joint prediction method for pipeline corrosion failure integrating multi-source physical information, comprising the following steps: S1: Collect multi-source data on pipeline corrosion failure, and perform unified conversion on the multi-source data to obtain multi-source status input; S2: Extract features from the multi-source state input to obtain key variables, and construct a multi-physics coupling graph based on the key variables; S3: Based on the multi-source state input and the multi-physics coupling diagram, the pipeline corrosion failure prediction results are obtained. The pipeline corrosion failure prediction results include: predicted corrosion rate, predicted future corrosion depth, predicted failure pressure, predicted remaining life, and initial risk score. S4: Physical consistency constraints are applied to the initial risk score of pipeline corrosion failure to obtain a physical violation score, and a high-risk protection domain score is obtained based on the physical violation score. S5: Based on the high-risk protection domain score, the initial score is corrected to obtain the final risk score; and based on the final risk score, a pipe section-level maintenance decision is obtained.

[0008] More preferably, in the multi-physics coupling graph, the edge relationships include: electrochemical admittance edge, pressure load edge, spatial adjacency edge, corrosion state edge, and material structure edge; For the i Pipe section and the first j Pipe section in q Characteristics under segment physical relationship and Physically associated edges are represented as: ; In the formula, Indicates the first i Pipe section and the first j Pipe section in q The strength of association under the physical relationship of segments; q = e Indicates electrochemical relationships, q = p Indicates the pressure-load relationship. q = c Indicates the corrosion state relationship. q = m Indicates the material structure relationship; Indicates the first q Scale parameters of physical relationships between segments; This indicates the avoidance of extremely small positive numbers with a denominator of zero; The spatial adjacent edge is based on the first i Pipe section and the first j The mileage distance between pipe segments is constructed such that when the mileage distance between the two segments is not greater than the preset spatial adjacency radius, the spatial adjacency edge is set to 1; otherwise, it is set to 0.

[0009] More preferably, in S3, the pipeline corrosion failure prediction result is obtained based on the GAT-Mamba spatiotemporal prediction model; The GAT-Mamba spatiotemporal prediction model includes: GAT and Mamba temporal evolution modules; The GAT is used to process the spatial relationships in the multi-physics coupling graph; firstly, the first... i Pipe segment nodes and the first j Graph attention weights for adjacent pipe segment nodes; aggregate the features of adjacent pipe segment nodes based on the graph attention weights to obtain spatially enhanced pipe segment features; The spatially enhanced pipe segment features are input into the Mamba temporal evolution module to obtain the pipeline corrosion failure prediction results: ; In the formula, Indicates the first i Pipeline segment at future time Preliminary multi-task prediction results; This indicates the predicted corrosion rate; This indicates a predicted future corrosion depth; Indicates the predicted failure pressure; Indicates the projected remaining lifespan; Indicates the first i Initial risk assessment of the pipeline segment.

[0010] More preferably, in S4, the physical violation score includes: ; In the formula, Indicates the first i Physical violation score of the pipe section; This indicates a violation of Faraday's law constraints; This indicates a violation of the remaining wall thickness constraint; This indicates a violation of the pressure safety margin constraint; This indicates a violation of the risk ranking constraint; Indicates a violation of spatial risk continuity; , , , , These represent the weight coefficients for different violations.

[0011] More preferably, in S4, the high-risk protection domain scoring includes: ; In the formula, Indicates the first i The scoring for the pipe section belonging to the high-risk protection zone; This represents the Sigmoid function or the normalized mapping function; This represents the normalized corrosion depth; This represents the normalized corrosion rate; This represents the normalized safety margin risk term. Indicates the risk propagation item in the spatial neighborhood; to This represents the weighting coefficients for different risk factors.

[0012] More preferably, in S5, the final risk score includes: ; In the formula, Indicates the first i Final risk score for the pipeline segment; This indicates an insufficient pressure safety margin, when... The value is 1 if the condition is met, otherwise it is 0. to Indicates the risk-adjusted weight; This indicates that the risk score will be limited to between 0 and 1.

[0013] This invention also provides a spatiotemporal joint prediction system for pipeline corrosion failure that integrates multi-source physical information, comprising: Multi-source data input module: used to collect multi-source data on pipeline corrosion failure, and to uniformly convert the multi-source data to obtain multi-source status input; Physical feature extraction module: used to extract features from the multi-source state input, obtain key variables, and construct a multi-physical coupling graph based on the key variables; Failure prediction module: used to obtain pipeline corrosion failure prediction results based on the multi-source state input and the multi-physics coupling diagram. The pipeline corrosion failure prediction results include: predicted corrosion rate, predicted future corrosion depth, predicted failure pressure, predicted remaining life and initial risk score. Physical constraint module: used to apply physical consistency constraints to the initial risk score of pipeline corrosion failure to obtain a physical violation score, and to obtain a high-risk protection domain score based on the physical violation score; Maintenance decision module: used to revise the initial score based on the high-risk protection domain score to obtain the final risk score; and to obtain the pipe section-level maintenance decision based on the final risk score.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: In this invention, pipeline corrosion failure prediction no longer relies solely on corrosion depth, defect size, or a single pressure parameter. Instead, it simultaneously considers multiple factors, including corrosion state, pressure load, electrochemical behavior, environmental medium, soil conditions, material structure, and historical operation and maintenance status. By selecting key variables, the interference of redundant and weakly correlated variables on model training can be reduced. Through physical feature group mapping, the selected variables can directly serve subsequent multiphysics graph construction, physical violation evaluation, and high-risk protection domain identification, thereby improving the effectiveness, physical interpretability, and engineering applicability of the model input.

[0015] The GAT-Mamba spatiotemporal joint prediction model can simultaneously utilize spatial correlation information along the pipeline and historical corrosion evolution information, avoiding the problems of individual time series models ignoring the spatial influence of pipe segments and individual graph models being unable to describe long-term evolution. Furthermore, since the edge weights in the multi-physics coupled graph can be adaptively adjusted according to the degree of physical violation, safety margin, and risk state, the model can dynamically increase the attention to relevant physical relationships in scenarios with insufficient pressure safety margin, abrupt changes in corrosion state, accumulation of spatial risks, or enhanced influence of material structure, thereby enhancing its adaptability to complex corrosion failure scenarios.

[0016] The corrosion rate prediction maintains a reasonable relationship with the driving factors of electrochemical corrosion, and the future corrosion depth is unlikely to exceed the non-physical result of the pipe wall thickness. The failure pressure prediction is consistent with the operating pressure and safety factor, and the risk score conforms to the relative relationship between corrosion state, pressure load, and protection state. Compared with prediction methods that rely solely on data fitting, this invention can reduce the occurrence of model outputs that do not conform to corrosion mechanisms, wall thickness boundaries, and pressure-bearing safety boundaries, thereby improving the engineering credibility of the prediction results.

[0017] The system not only outputs corrosion rate, corrosion depth, failure pressure, and remaining life, but also identifies which pipe sections have physical anomalies or safety boundary risks. For pipe sections with insufficient pressure safety margin, abnormal corrosion rate, small remaining wall thickness, high physical violation, or obvious risk propagation from neighboring areas, the system can include them in the high-risk protection domain, thereby preventing high-risk samples from being overwhelmed by ordinary samples due to their small number. This mechanism can reduce the probability of missing high-risk pipe sections and improve the safety conservatism in pipeline integrity management.

[0018] When a pipe section has an initial risk score that is low but a high degree of physical violation, insufficient safety margin, or significant risk propagation from neighboring areas, the system can improve its final risk score through a risk correction mechanism, making the risk level classification more in line with engineering safety requirements. Compared with methods that directly use neural networks to output risk levels, this invention can enhance the ability to identify boundary conditions, high-risk samples, and abnormal corrosion states, and improve the rationality of maintenance priority ranking.

[0019] The prediction results no longer stop at the corrosion depth, corrosion rate, or remaining life of a single corrosion point, but can be transformed into pipe-section level risk results and maintenance actions that are of greater concern to engineering operations and maintenance. Through a complete chain of "multi-source physical state input - key variable screening and feature group mapping - adaptive multi-physics coupling graph construction - GAT-Mamba spatiotemporal prediction - physical constraint joint output - physical violation evaluation - high-risk protection domain identification - risk correction - pipe-section level maintenance decision-making", this invention can provide more complete, interpretable, and implementable technical support for pipeline integrity assessment, risk classification, re-inspection cycle determination, and maintenance and reinforcement prioritization. Attached Figure Description

[0020] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of a spatiotemporal joint prediction method for pipeline corrosion failure that integrates multi-source physical information, provided in an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] Example 1: like Figure 1 As shown in the figure, this embodiment provides a spatiotemporal joint prediction method for pipeline corrosion failure that integrates multi-source physical information, including the following steps: S1: Collect multi-source data on pipeline corrosion failure, and perform unified conversion on the multi-source data to obtain multi-source status input.

[0025] In this embodiment, the multi-source data includes: internal corrosion defect detection data, operating pressure data, electrochemical data, environmental media data, soil environment data, material structure data, and historical operation and maintenance data. Specifically, the internal corrosion defect detection data includes: corrosion defect location, corrosion depth, defect length, defect width, wall thickness, and detection time; the operating pressure data includes: operating pressure, peak pressure, pressure fluctuation, safe operating pressure, and pressure cycle intensity; the electrochemical data includes: stray current density, pipe-to-ground potential, cathodic protection potential, and protection status; the environmental media data includes: temperature, humidity, pH value, CO2 or H2S content, flow rate, and media type; the soil environment data includes: soil resistivity, moisture content, and soil corrosivity level; the material structure data includes: pipe diameter, wall thickness, material strength, steel grade, anti-corrosion coating status, and weld status; and the historical operation and maintenance data includes: detection cycle, maintenance records, re-inspection records, protection status change records, and historical risk status.

[0026] Spatial matching of multi-source pipeline data is performed according to pipeline mileage or segment number, and time alignment is performed according to detection time or operation time. After data cleaning, normalization, outlier handling, and missing value completion, candidate physical variables are extracted or calculated from various multi-source data. These candidate physical variables are then organized according to pipe segment and time, with each pipe segment treated as a node, and a corresponding node state vector is constructed. The node state vector constitutes the multi-source state input, used for subsequent GAT-Mamba spatiotemporal prediction, adaptive multi-physics coupling graph construction, physical violation evaluation, high-risk protection domain identification, and risk score correction.

[0027] No. i Pipeline section at time t The node state vector is: (1) In the formula, Indicates the first i Pipeline section at time t The multi-source state input vector; This represents the feature vector of internally detected corrosion defects; This represents the characteristic subvector of operating pressure and pressure fluctuation. Represents the electrochemical eigenvectors; Represents the characteristic sub-vectors of environmental media and soil environment; Represents the material structure feature vector; This represents the historical operation and maintenance feature vector.

[0028] To describe the trend of corrosion status of the pipe segment over time, a historical state sequence is further constructed based on the node state vector: (2) In the formula, Indicates the first i The pipe section is of length L The multi-source state sequence formed within the historical window.

[0029] This leads to the multi-source state input for all pipe segments: (3) In the formula, N Indicates the number of nodes in the pipe segment; H ( t ) indicates the time of the entire pipeline. t Input of historical status of multiple pipe sections.

[0030] S2: Extract features from multi-source state inputs to obtain key variables, and construct a multi-physics coupling graph based on the key variables.

[0031] First, key variables are screened for the above-mentioned multi-source state inputs. Key variables that are strongly correlated with pipeline corrosion failure state are selected from physical variables such as operating pressure data, electrochemical data, environmental medium data, soil environment data, material structure data, internal detection corrosion defect data, and historical operation and maintenance data. This reduces the interference of redundant variables and weakly correlated variables on the prediction model. The selected key variables are further mapped to different physical feature groups. After completing the key variable screening and physical feature group mapping, an adaptive multi-physics coupling graph is constructed based on the selected key variables.

[0032] Specifically, this embodiment employs the VMD-GRA method for key variable selection. VMD is used to decompose multi-source states and extract feature components at different frequency bands or scales of change; GRA is used to calculate the correlation between each component and the corrosion failure target, thereby obtaining the contribution of each candidate variable to the corrosion failure prediction task. It should be noted that VMD-GRA is a preferred variable selection method of this invention. Without changing the overall technical concept of this invention, methods such as grey relational analysis, mutual information selection, random forest feature importance, and LASSO selection can also be used as alternatives.

[0033] For the j Time series of candidate physical variables After VMD decomposition, it is represented as: (4) In the formula, Indicates the first j One candidate physical variable; Indicates the first j The candidate physical variable obtained by VMD decomposition is the _th ... k One modal component; K Indicates the number of modal components; This indicates the remaining component or error term.

[0034] The original physical variables are decomposed into multiple variation components at different scales, avoiding the need to judge the correlation of variables solely from the original sequence. For non-stationary information such as operating pressure fluctuations, stray current changes, environmental disturbances, and corrosion state changes, VMD decomposition can extract more stable local variation features.

[0035] To evaluate the correlation between candidate variables and corrosion failure targets, at least one of corrosion rate, future corrosion depth, failure pressure, or remaining lifetime was used as a reference sequence. , No. j The first candidate physical variable k Modal components As a comparison sequence, the grey correlation coefficient between the comparison sequence and the reference sequence is expressed as: (5) In the formula, Represents the grey relational coefficient; This represents the minimum difference among all compared sequences; This represents the maximum difference among all compared sequences; This represents the resolution coefficient, which is usually set to 0 < <1.

[0036] The grey relational coefficient is used to measure the consistency between the modal components of candidate physical variables and the corrosion failure target. The closer a modal component is to the trend of corrosion failure target, the larger its grey relational coefficient, indicating that the candidate physical variable contributes more significantly to the prediction of corrosion failure.

[0037] Further, calculate the first j The overall correlation score of the candidate physical variables: (6) In the formula, Indicates the first j The overall association score of the candidate variables; T Indicates the length of the time series.

[0038] By combining the correlations of the same candidate variable across different modal components, the overall contribution of that candidate physical variable to the corrosion failure prediction task can be obtained. This avoids judgment based solely on a single time point or a single frequency component, thus improving the stability of key variable selection.

[0039] Based on the comprehensive correlation score, variables that meet the threshold conditions are selected as the set of key variables: (7) In the formula, This represents the set of key variables after filtering. This indicates the threshold for variable selection.

[0040] When the comprehensive correlation score of a candidate physical variable is greater than or equal to the threshold, it indicates that the candidate physical variable has a strong correlation with the corrosion failure prediction target and can be used as an input variable for subsequent models. After obtaining the set of key variables, physical feature group mapping is further performed on the selected key variables.

[0041] Through the above processing, not only can key variables closely related to pipeline corrosion failure be screened from multi-source physical variables, but the physical characteristic group to which each key variable belongs can also be identified. This mapping result is used for subsequent adaptive multiphysics coupling graph construction: pressure load variables are used to construct pressure load edges, electrochemical variables are used to construct electrochemical admittance edges, spatial location variables are used to construct spatial adjacency edges, corrosion state variables are used to construct corrosion state edges, material structure variables are used to construct material structure edges, and environmental auxiliary variables are used to supplement node states and risk assessments.

[0042] Compared to directly inputting all variables, this invention, through the VMD-GRA key variable screening and physical feature group mapping process, can reduce the impact of redundant features and noisy variables on model training. This allows subsequent adaptive multiphysics coupling graphs and GAT-Mamba spatiotemporal prediction models to focus more on physical factors strongly correlated with corrosion failure states. Simultaneously, the feature group mapping results also provide physical classification basis for subsequent physical violation assessment, high-risk protection domain construction, and risk score correction. This enables the risk correction process to no longer rely solely on neural network output but to comprehensively assess factors such as pressure load, electrochemical behavior, corrosion state, material structure, and environmental factors.

[0043] After completing the screening of key variables and mapping of physical feature groups, an adaptive multi-physics coupling graph is constructed based on the selected key variables. The multi-physics coupling graph is used to express the correlation between different pipe segments due to electrochemical behavior, pressure load, spatial adjacency, corrosion state, and material structure similarity, and further dynamically updates the weights of different physical edge relationships based on physical violation degree, safety margin, and risk state.

[0044] In this embodiment, each pipe segment or corrosion defect aggregation unit within a pipe segment is treated as a graph node, and the node characteristics are composed of the aforementioned key variables after screening. The edges between different nodes are no longer established solely based on spatial distance, but simultaneously consider electrochemical admittance edges, pressure load edges, spatial adjacency edges, corrosion state edges, and material structure edges. Specifically, electrochemical admittance edges reflect the similarity of electrochemical states between pipe segments; pressure load edges reflect the similarity of operating pressure and pressure fluctuation states; spatial adjacency edges reflect the natural adjacency relationships along the pipeline; corrosion state edges reflect the similarity of corrosion depth, corrosion rate, and remaining wall thickness; and material structure edges reflect the similarity of structural attributes such as pipe diameter, wall thickness, material strength, and anti-corrosion coating condition.

[0045] To provide a unified description of node relationships under different physical factors, let the first... i Pipe section and the first j Pipe section in q The characteristics under the segment physical relationship are respectively and The corresponding physical associated edge is represented as: (8) In the formula, Indicates the first i Pipe section and the first j Pipe section in q The strength of association under the physical relationship of segments; q = e Indicates electrochemical relationships, q = p Indicates the pressure-load relationship. q = c Indicates the relationship of corrosion state. q = m Indicates the material structure relationship; Indicates the first q Scale parameters of physical relationships between segments; This indicates the avoidance of extremely small positive numbers with a denominator of zero.

[0046] Spatial adjacent edges according to the first i Pipe section and the first j The mileage distance between pipe segments is constructed such that when the mileage distance between the two segments is not greater than the preset spatial adjacency radius, the spatial adjacency edge is set to 1; otherwise, it is set to 0.

[0047] This method describes multiple types of physical similarities between different pipe segments using a unified approach. The more similar two pipe segments are in terms of electrochemical state, pressure load, corrosion state, or material structure, the larger their corresponding edge weights; conversely, the greater the differences, the smaller the edge weights. This approach allows for the simultaneous perception of electrochemical correlations, pressure condition correlations, corrosion state correlations, and material structure correlations between different pipe segments.

[0048] For spatial adjacency relationships along the pipeline, spatial adjacency edges are constructed based on the pipeline segment mileage location: (9) In the formula, Indicates the first i Pipe section and the first j Spatial adjacency relationships between pipe segments; and These represent the mileage positions of the two pipe sections respectively; Indicates the spatial adjacency radius.

[0049] By preserving the natural adjacency relationships along the pipeline route, it is possible to identify the continuity of corrosion status, local risk accumulation characteristics, and spatial risk propagation trends that may exist between adjacent pipe sections.

[0050] Before multi-physics edge fusion, different types of edge relationships are scaled to ensure that electrochemical relationships, pressure-load relationships, spatial adjacency relationships, corrosion state relationships, and material structure relationships are within a comparable range. Subsequently, the above five types of edge relationships are weighted and fused to obtain an adaptive multi-physics coupling adjacency matrix. (10) In the formula, Indicates the first r The adaptive multi-physics coupling adjacency matrix after round-updating; Represents the electrochemical admittance side matrix; Represents the pressure load side matrix; Represents the spatial adjacency matrix; Represents the edge matrix of the erosion state; Represents the edge matrix of the material structure; , , , , They represent the first r Weight coefficients for different physical edge relationships during round updates.

[0051] By setting various edge weights as adaptive weights that can be updated with changes in iteration rounds, physical violation degree, and risk state, the importance of multi-physical relationships can be dynamically adjusted according to different risk scenarios. Specifically, when a pipe segment has insufficient pressure safety margin, the weight of the pressure load edge can be increased accordingly; when the spatial gradient of the corrosion state is abnormal or there is a significant local risk clustering, the weights of the spatially adjacent edges and the corrosion state edges can be increased accordingly; when the impact of remaining wall thickness constraints or material structure differences on failure risk is enhanced, the weight of the material structure edge can be increased accordingly.

[0052] Therefore, the set of pipe segment nodes, the set of edges, node features, and the multi-physical coupling adjacency matrix can be represented as follows: (11) In the formula, G represents the multi-physical coupling graph; V represents the set of pipe segment nodes; E represents the set of edges between nodes; X represents the pipe segment node feature matrix; and A represents the multi-physical coupling adjacency matrix.

[0053] Through the above processing, the multi-physics coupled graph constructed in this embodiment is no longer a simple spatial topology graph, but an adaptive multi-physics coupled graph that integrates electrochemical admittance relationships, pressure load relationships, spatial adjacency relationships, corrosion state relationships, and material structure relationships. Compared with methods that establish graph structures based solely on distance or single similarity, this invention can more fully reflect the multi-source physical correlations in the pipeline corrosion failure process; compared with fixed edge weight graph structures, this invention can dynamically adjust the importance of different physical edge relationships according to physical violation degree, safety margin, and risk state, providing a spatial relationship foundation with engineering physical significance for subsequent GAT-Mamba spatiotemporal prediction, physical violation degree evaluation, high-risk protection domain identification, and risk score correction.

[0054] S3: Based on multi-source state input and multi-physics coupling diagram, the pipeline corrosion failure prediction results are obtained. The pipeline corrosion failure prediction results include: predicted corrosion rate, predicted future corrosion depth, predicted failure pressure, predicted remaining life and initial risk score.

[0055] After completing the construction of the multi-physics coupling graph, the multi-source state of the pipe segment is input. and multi-physics coupling diagram Input the GAT-Mamba spatiotemporal prediction model. The GAT-Mamba spatiotemporal prediction model is used to simultaneously learn the spatial correlation between pipe segments and the evolution of corrosion state over time, thereby obtaining preliminary prediction results of pipeline corrosion failure.

[0056] Specifically, GAT is used to handle spatial relationships in multi-physics coupled graphs. For the first... i Pipe segment nodes and the first j For adjacent pipe segment nodes, the graph attention weights are calculated based on their node characteristics and multi-physical coupling adjacency relationships. (12) In the formula, Indicates the first j Pipe segment node pair i The influence weight of pipe segment nodes; and They represent the first i and the j Pipeline segment node characteristics; This represents the edge weights in the multi-physical coupling adjacency matrix; N( represents the attention scoring function;) i) represents the adaptive multiphysics coupling graph with the th i The set of neighboring node indices of a pipe segment node with associated edges.

[0057] Based on adaptive multi-physics coupling relationships, it automatically determines which adjacent pipe segments, pipe segments with similar operating conditions, pipe segments with similar electrochemical states, or pipe segments with similar corrosion states are more important for the prediction of the current pipe segment. Compared with a fixed graph structure, the graph attention weight can combine the aforementioned adaptive edge weight results to give higher attention to areas with insufficient pressure safety margins, abnormal corrosion states, or obvious spatial risk propagation.

[0058] Based on the graph attention weights, the features of adjacent pipe segment nodes are aggregated to obtain the spatially enhanced pipe segment features: (13) in, This represents the first [unit / item] after graph attention aggregation. i Pipeline segment spatial enhancement characteristics; Represents the characteristic transformation matrix of the graph; This represents a non-linear activation function.

[0059] The first i By fusing the condition of a single pipe segment with that of surrounding segments, the GAT-Mamba spatiotemporal prediction model can utilize spatial correlation information such as localized corrosion accumulation along the pipeline, similar pressure loads, similar electrochemical conditions, and similar material structures. This spatial enhancement not only reflects the condition of an individual pipe segment but also the risk propagation impact of adjacent or physically similar segments.

[0060] After obtaining the spatial enhancement features, they are input into the Mamba temporal evolution module to describe the dynamic process of corrosion state changes with detection cycle and operating conditions. For the first... i The time status update of the pipe segment is represented as follows: (14) In the formula, Indicates the first i Pipeline section at time t Temporal hidden state; This indicates the hidden state of time in the previous moment; This represents the Mamba selective state space update function.

[0061] The Mamba temporal evolution module preserves historical evolution information of corrosion states and selectively updates time-hidden states based on current changes in multiple physical states, thereby better describing the impact of corrosion growth, pressure disturbance, electrochemical changes, and environmental changes on future corrosion failure states.

[0062] Finally, based on the temporal hidden states output by the Mamba temporal evolution module, the pipe segment corrosion failure prediction results are generated: (15) In the formula, Indicates the first i Pipeline section at future time Preliminary multi-task prediction results; This indicates the predicted corrosion rate; This indicates a predicted future corrosion depth; Indicates the predicted failure pressure; Indicates the projected remaining lifespan; Indicates the first i Initial risk assessment of the pipeline segment.

[0063] In this embodiment, the GAT-Mamba spatiotemporal prediction model does not perform time series fitting or graph structure aggregation alone. Instead, it first learns the adaptive multi-physics spatial correlation between pipe segments using GAT, and then learns the evolution of corrosion state over time using Mamba, thereby obtaining preliminary predictions of corrosion rate, future corrosion depth, failure pressure, remaining lifetime, and initial risk score. The subsequent physical constraint joint output module will then perform physical consistency constraints, physical violation evaluation, high-risk protection domain identification, and final risk correction based on the above preliminary predictions.

[0064] The GAT-Mamba spatiotemporal prediction module outputs preliminary prediction results such as corrosion rate, future corrosion depth, failure pressure, remaining lifetime, and initial risk score. The physical constraint joint output does not construct new prediction targets; instead, it applies physical consistency constraints to the aforementioned multi-task prediction results, ensuring that the corrosion rate conforms to electrochemical corrosion laws, the corrosion depth meets the remaining wall thickness boundary, the failure pressure meets the operational pressure safety margin requirements, and the risk score conforms to the ranking relationship between corrosion state, pressure load, and protection state. Based on this, the present invention further constructs a physical violation score according to the degree of physical constraint violation, and combines this with a high-risk protection domain and risk correction mechanism to form the final risk level and maintenance recommendations.

[0065] S4: Physical consistency constraints are applied to the initial risk score of pipeline corrosion failure to obtain a physical violation score, and a high-risk protection domain score is obtained based on the physical violation score.

[0066] In this invention, the physical constraint joint output is constrained using a joint loss function: (16) In the formula, Represents the joint loss function; This represents the data error loss between the multi-task prediction results and the reference value; Indicates physical constraint loss; This represents the weight of the physical constraint loss.

[0067] in, (17) In the formula, This represents the constraint loss due to Faraday's law; This represents the remaining wall thickness constraint loss; This indicates the loss due to pressure safety margin constraints; This represents the loss due to risk ranking constraints. , , , These represent the weight coefficients of each constraint term.

[0068] In this embodiment, the physical constraints are not a single penalty term, but rather a combination of electrochemical corrosion constraints, wall thickness boundary constraints, pressure bearing constraints, and risk ranking constraints. Through this joint constraint, the model output must not only satisfy the data fitting relationship, but also conform to the pipeline corrosion mechanism and engineering safety boundaries.

[0069] First, in order to constrain the relationship between the predicted corrosion rate and the electrochemical corrosion behavior, according to Faraday's law, the first... i The equivalent corrosion current, metallic material parameters, and equivalent corrosion area of ​​the pipe section are converted into an equivalent corrosion rate. Based on the normalized deviation between the predicted corrosion rate and the equivalent corrosion rate, the Faraday law constraint loss is constructed and presented as follows: (18) In the formula, This represents the constraint loss due to Faraday's law; N Indicates the number of pipe segments involved in the loss calculation; Indicates the first i Predicted corrosion rate of the pipe section; Indicates the first i Equivalent corrosion current of the pipe section; Indicates the molar mass of the metal; n Indicates the number of electrons transferred in an electrochemical reaction; F Denotes Faraday's constant; Indicates metal density; Indicates the first i The equivalent corrosion area of ​​the pipe section; This indicates the avoidance of extremely small positive numbers with a denominator of zero.

[0070] Equation (18) is directly used as the Faraday law constraint loss in equation (17). When the predicted corrosion rate output by the GAT-Mamba spatiotemporal prediction model is close to the equivalent corrosion rate calculated based on Faraday's law, the constraint loss is small; when the deviation between the two increases, the constraint loss increases accordingly, so that the predicted corrosion rate not only satisfies the data fitting relationship, but also maintains a reasonable physical consistency with corrosion current, electrochemical reaction parameters and material properties.

[0071] Pressure safety margin constraint loss is used to ensure that the predicted failure pressure meets the operating pressure safety requirements: (19) In the formula, Indicates the first i Predicted failure pressure of the pipe section; Indicates operating pressure; This represents the safety factor.

[0072] The operating pressure under conditions where the failure pressure exceeds the safety factor is predicted by using pressure safety margin constraints and loss constraints. When the predicted failure pressure is lower than the safe operating requirements, this constraint term increases, making the model pay more attention to pipe sections with higher pressure bearing risks.

[0073] Risk ranking constraint loss is used to ensure a reasonable ranking between risk level and corrosion state, pressure load, and protection state: (20) In the formula, Indicates the first j Initial risk assessment of pipe sections. High-risk pipe sections can be identified based on factors such as greater corrosion depth, higher pressure load, smaller remaining wall thickness, or poorer protection status.

[0074] By using the above formula, pipe sections with more severe corrosion, higher pressure loads, or poorer protection conditions should not be assigned lower risk scores, thereby improving the rationality of risk level classification and maintenance priority ranking.

[0075] The remaining wall thickness constraint loss is used to prevent the predicted corrosion depth from exceeding the physical wall thickness of the pipe. ;(twenty one) In the formula, Indicates the first i Predicted corrosion depth for each pipe section; This indicates the pipe wall thickness.

[0076] When the predicted corrosion depth is less than or equal to the pipe wall thickness, this constraint term is zero or small; when the predicted corrosion depth exceeds the pipe wall thickness, this constraint term increases, thereby suppressing prediction results that do not conform to the engineering physical boundaries.

[0077] After combining the above physical constraints, the final prediction result is formed: ;(twenty two) In the formula, Indicates the first i The final combined output of the pipe section; Indicates the final corrosion rate; Indicates the final corrosion depth; Indicates the final failure pressure; Indicates the final safety margin; Indicates the final remaining lifespan; Indicates the first i Final risk score for the pipeline segment; This indicates the final maintenance recommendation.

[0078] In formula (22) and It is not obtained directly from the initial output of the neural network, but is obtained after further physical violation evaluation, high-risk protection domain identification, risk score correction and pipe segment-level aggregation based on the joint output of physical constraints.

[0079] Specifically, to evaluate the degree of anomaly in the joint output of physical constraints, the system constructs a physical violation score based on Faraday's law constraint, remaining wall thickness constraint, pressure safety margin constraint, risk ranking constraint, and spatial risk continuity constraint: ;(twenty three) In the formula, Indicates the first i Physical violation score of the pipe section; This indicates a violation of Faraday's law constraints; This indicates a violation of the remaining wall thickness constraint; This indicates a violation of the pressure safety margin constraint; This indicates a violation of the risk ranking constraint; Indicates a violation of spatial risk continuity; , , , , These represent the weight coefficients for different violations; .

[0080] Formula (23) serves to quantify different types of physical constraint violations into a single physical violation score. If a pipe segment simultaneously exhibits abnormal electrochemical corrosion, insufficient remaining wall thickness, insufficient pressure safety margin, abnormal risk ranking, or abnormal spatial risk gradient, its physical violation score will increase, indicating that the pipe segment needs to receive more attention in subsequent risk identification and maintenance decisions.

[0081] Among them, the Faraday law constraint violation terms are: ;(twenty four) (25) In the formula, Indicates the first i Equivalent corrosion current of the pipe section; Indicates the molar mass of the metal; n Indicates the number of electrons transferred in an electrochemical reaction; F Denotes Faraday's constant; Indicates metal density; Indicates the equivalent corrosion area; Indicates the first i The final predicted corrosion rate of the pipe section vF , i Indicates the first i Equivalent corrosion rate of the pipe section.

[0082] The remaining wall thickness constraint violation term is used to characterize the degree of violation when the predicted corrosion depth exceeds the pipe wall thickness. It is calculated based on the normalized excess amount of the final predicted corrosion depth exceeding the pipe wall thickness. (26) In the formula, Indicates the first i The final predicted corrosion depth of the pipe section; Indicates the first i Pipe section wall thickness.

[0083] Risk ranking constraint violations are calculated based on the risk score ranking violation amount between high-risk and low-risk pipe sections: (27) In the formula, Indicates the first i Risk prioritization constraint violations in pipeline segments; Indicates the relationship with the first i The pipe segments constitute a set of low-risk pipe segments based on a ranking comparison. When the... i Compared to the first section j When a pipe section has greater corrosion depth, lower safety margin, shorter remaining life, or higher physical violation, the first i The initial risk score of the pipeline segment should not be lower than the first. j If the pipe segment is not cleared, a sorting violation will occur.

[0084] The pressure safety margin constraint violation term is used to characterize the degree of violation when the predicted failure pressure is lower than the safe operating requirement. It is calculated based on the excess amount between the operating pressure under the safety factor and the final predicted failure pressure. (28) In the formula, Indicates the first i Pipeline operating pressure; Indicates the safety factor; Indicates the first i The final predicted failure pressure of the pipeline segment is used to characterize the degree of violation when the predicted failure pressure is lower than the safe operating requirements.

[0085] Spatial risk continuity constraint violation terms are used to characterize abnormal deviations exceeding the allowable threshold between the current segment risk score and the neighborhood weighted average risk score, including: (29) (30) In the formula, Indicates the first i Neighborhood-weighted average risk of the pipe segment; This indicates the threshold for permissible spatial risk differences.

[0086] After obtaining the physical violation score, a high-risk protection domain score is constructed based on the initial risk score, physical violation degree, pressure safety margin, corrosion state, and spatial neighborhood risk: (31) In the formula, Indicates the first i The scoring for the pipe section belonging to the high-risk protection zone; This represents the Sigmoid function or the normalized mapping function; zd,i This represents the normalized corrosion depth; This represents the normalized corrosion rate; This represents the normalized safety margin risk term. Indicates the risk propagation item in the spatial neighborhood; to This represents the weighting coefficients for different risk factors.

[0087] when If the risk level exceeds the preset protection threshold, or if the pipe segment exhibits insufficient pressure safety margin, excessively high physical violation rate, abnormal corrosion rate, or excessively low remaining life, the system will classify the pipe segment into the high-risk protection domain. The purpose of the high-risk protection domain is to prevent a small number of high-risk samples that are of significant safety importance from being overwhelmed by ordinary samples, thereby increasing the identification priority of high-risk pipe segments.

[0088] S5: The initial score is corrected based on the high-risk protection domain score to obtain the final risk score; and the segment-level maintenance decision is obtained based on the final risk score.

[0089] After the high-risk protection domains are identified, the initial risk score is conservatively revised to obtain the final risk score: (32) In the formula, Indicates the first i Final risk score for the pipeline segment; This indicates an insufficient pressure safety margin, when... The value is 1 if the condition is met, otherwise it is 0. to Indicates the risk-adjusted weight; This indicates that the risk score will be limited to between 0 and 1.

[0090] When a pipe segment exhibits high physical violations, a high score in the high-risk protection zone, insufficient pressure safety margin, or significant risk propagation from neighboring areas, the system conservatively modifies its initial risk score to better align with engineering safety requirements. This reduces the likelihood of high-risk pipe segments being underestimated.

[0091] Furthermore, the point-level or corrosion node-level output results are aggregated into pipe section-level maintenance decision indicators. For the first... g A collection of pipe segments, at the pipe segment level P 10 Remaining lifetime is expressed as: (33) In the formula, Indicates the first g Each pipe section P 10 Remaining lifespan Represents the 10-quantile function. Indicates belonging to the first g A collection of corrosion nodes or detection points for each pipe section.

[0092] use P The remaining life of 10 can more conservatively reflect the lower limit of the life of higher-risk locations within the pipe section, making it suitable for maintenance priority ranking.

[0093] Simultaneously, calculate the average risk, maximum risk, and protection zone ratio at the pipe segment level: (34) (35) In the formula, Indicates the first g Average risk of each pipe section; Indicates the first g The greatest risk of each pipeline segment; Indicates the first g The proportion of nodes within a single management section that belong to a high-risk protection zone; This represents the set of nodes in the high-risk protection domain.

[0094] Based on the above segment-level indicators, segment-level risk levels and maintenance recommendations are determined. When the segment level... P 10. When the remaining lifespan is short, the maximum risk is high, or the proportion of protected areas is high, suggestions for early re-inspection, maintenance and reinforcement, or pressure verification are provided; when the pipeline risk is at a medium level, suggestions for increased monitoring or shortened re-inspection cycle are provided; when the pipeline risk is low, suggestions for routine inspection are provided.

[0095] Through the above processing, this invention forms a complete mechanism from initial GAT-Mamba prediction, joint output of physical constraints, evaluation of physical violations, identification of high-risk protection domains, correction of risk scores, to pipe segment-level maintenance decision output. This mechanism not only avoids the results generated by purely data-driven models that do not conform to pipeline corrosion mechanisms or pressure bearing safety boundaries, but also enables protective identification and conservative correction of potentially high-risk pipe segments, so that the final output simultaneously possesses data predictive ability, physical consistency, safety conservatism, and engineering interpretability.

[0096] In this embodiment, the adaptive multiphysics coupling graph preferably includes electrochemical admittance edges, pressure load edges, spatial adjacency edges, corrosion state edges, and material structure edges. However, in alternative solutions, two or more types of edge relationships can be selected and combined based on data availability and engineering scenarios. For example, a simplified multiphysics graph can be constructed using spatial adjacency edges, pressure load edges, and corrosion state edges, or further added based on the pipeline protection status, such as anti-corrosion layer status edges, topographic edges, or historical maintenance edges. As long as the graph structure can express the correlation between pipe segments based on multi-source physical relationships and can adjust edge weights according to physical violation degree, safety margin, or risk status, it can be used as an alternative implementation of the present invention.

[0097] The spatiotemporal prediction model preferably employs the GAT-Mamba structure. However, alternative solutions may also use GAT-LSTM, GAT-GRU, GCN-Mamba, Graph Transformer, temporal convolutional networks, or other neural network structures capable of spatial correlation modeling and time series modeling. Although these alternative solutions may differ in terms of long-sequence modeling efficiency, spatial attention expression capability, or computational complexity, as long as they can perform spatiotemporal prediction of corrosion failure based on multi-physics coupling graphs and the historical state of pipe segments, they can all serve as alternative implementations of this invention.

[0098] The preferred physical constraint output includes Faraday's law constraints, remaining wall thickness constraints, pressure safety margin constraints, and risk ranking constraints. However, alternative solutions may select one or more of these constraints based on engineering requirements. For example, only remaining wall thickness constraints and pressure safety margin constraints may be used, or further constraints such as non-negativity of corrosion rate, monotonicity of remaining life, pipe segment risk continuity, and spatial gradient constraints may be added. Any alternative implementation of this invention can be used as long as the output can provide physical consistency constraints on the model prediction results and provide a basis for evaluating physical violations.

[0099] The high-risk protection domain and risk correction mechanism are preferably constructed based on the initial risk score, physical violation degree, safety margin, corrosion state, and spatial neighborhood risk propagation item. However, in alternative solutions, the protection domain threshold, risk correction weight, and maintenance priority rules can be adjusted according to the specific engineering scenario. For example, focus can be placed on pipe sections with insufficient pressure safety margin, or on pipe sections with abnormal electrochemical corrosion and sudden increases in corrosion rate. As long as it can protectively identify potentially high-risk pipe sections and conservatively correct the initial risk results, it can be used as an alternative implementation of this invention.

[0100] In this embodiment, the final output includes corrosion rate, future corrosion depth, failure pressure, safety margin, remaining life, risk level, protection domain ratio, and maintenance recommendations. However, in alternative solutions, only some of these indicators can be output based on specific engineering needs. For example, only corrosion rate, failure pressure, and risk level can be output, or extended results such as detection priority, recommended maintenance methods, re-inspection cycle recommendations, and pressure verification recommendations can be added. As long as the pipeline corrosion failure prediction results are based on multi-source physical information, adaptive multi-physics coupling diagrams, joint output of physical constraints, high-risk protection domains, and risk correction mechanisms, they can all be used as alternative output formats of this invention.

[0101] Example 2: This invention is applicable to corrosion failure prediction, risk level assessment, and maintenance decision-making scenarios for long-distance oil and gas pipelines, urban gas pipelines, chemical medium transportation pipelines, and other pressure-bearing pipelines. It is particularly suitable for pipeline integrity management systems that possess periodic monitoring data, operating pressure data, electrochemical monitoring data, environmental medium data, soil environmental data, material structure data, and historical operation and maintenance data.

[0102] In this embodiment, an internal inspection data management system, a pressure monitoring system, a cathodic protection monitoring system, an environmental data acquisition system, and a pipeline integrity management platform are deployed along the pipeline. The internal inspection data system provides information on corrosion defect location, corrosion depth, defect size, and wall thickness. The pressure monitoring system provides operating pressure, peak pressure, pressure fluctuation, safe operating pressure, and pressure cycle intensity. The cathodic protection monitoring system provides information on pipe-to-ground potential, cathodic protection potential, and stray current. The environmental data acquisition system provides data on soil resistivity, temperature and humidity, pH value, CO2 or H2S content, and flow rate. The pipeline integrity management platform stores material strength, pipe diameter, wall thickness, anti-corrosion coating status, maintenance records, re-inspection records, and historical risk status.

[0103] In this embodiment, the aforementioned multi-source data are first spatially matched according to pipeline mileage or segment number, and then time-aligned according to detection time or operation time. Next, key variables are screened and mapped to physical feature groups, classifying key variables into pressure load, electrochemical, spatial location, corrosion state, material structure, and environmental auxiliary feature groups. Subsequently, an adaptive multiphysics coupling graph is constructed and input into the GAT-Mamba spatiotemporal prediction model. Finally, through joint physical constraint output, physical violation evaluation, high-risk protection domain identification, and risk correction mechanisms, corrosion rate, future corrosion depth, failure pressure, safety margin, remaining life, final risk score, risk level, and maintenance recommendations are generated.

[0104] The method of this invention unifies corrosion detection, pressure, electrochemical, environmental, material, and historical operation and maintenance data that were originally scattered in different systems for pipeline corrosion failure prediction, and further transforms the point-level prediction results into pipe section-level risk assessment and maintenance decision results, forming a complete output link for engineering applications.

[0105] Example 3: In long-distance oil and gas pipelines, the pipeline routes are long and the environments along the route vary significantly. Different pipeline sections may be subjected to different soil environments, pressure conditions, electrochemical states, and material structures. These types of pipelines typically have periodic internal monitoring data and operating pressure records, but it is difficult to directly and continuously observe the corrosion development status between different monitoring cycles.

[0106] In this embodiment, the pipeline is divided into multiple pipe segment nodes according to fixed length or integrity management units. Each pipe segment node corresponds to a multi-source physical state input, including information such as the historical corrosion depth, defect size, remaining wall thickness, operating pressure, pressure fluctuation, pipe-to-ground potential, cathodic protection potential, soil resistivity, material strength, anti-corrosion coating status, and historical maintenance records of the pipe segment. Physical feature groups are established based on the key variable screening results, and an adaptive multi-physics coupling graph is constructed based on the spatial adjacency relationship, electrochemical similarity, pressure load similarity, corrosion state similarity, and material structure similarity between pipe segments.

[0107] In this embodiment, the GAT-Mamba model outputs the future corrosion rate, future corrosion depth, failure pressure, remaining life, and initial risk score for each pipe segment. The physical constraint joint output further constrains the corrosion rate to be consistent with the electrochemical corrosion law, the corrosion depth to not exceed the wall thickness, and the failure pressure to meet the operating pressure safety margin requirements. Subsequently, the system corrects the initial risk score based on the physical violation score and the high-risk protection domain identification results, forming the final risk score and maintenance recommendations.

[0108] For long-distance pipelines with significant environmental differences along the route, the role of spatial adjacency edges, electrochemical admittance edges, and environmental auxiliary features in risk identification can be enhanced. For pipeline sections with localized corrosion accumulation, spatial gradient anomalies, or obvious risk propagation in the neighborhood, they should be included in the high-risk protection domain and conservatively corrected in the final risk score, thereby improving the reliability of identifying high-risk sections of long-distance pipelines.

[0109] Example 4: In urban gas pipeline scenarios, pipeline operating pressure may be affected by peak and off-peak gas consumption, the operating status of pressure regulating stations, local load changes, and pipeline network scheduling. Some pipeline sections may experience frequent pressure fluctuations or high pressure peaks. In such scenarios, whether corrosion defects develop into high-risk defects depends not only on the corrosion depth but also on the operating pressure, pressure fluctuations, and pressure safety margin.

[0110] In this embodiment, the system focuses on collecting pressure load data such as operating pressure, peak pressure, pressure fluctuation amplitude, safe operating pressure, and pressure cycle intensity. This data, along with corrosion depth, remaining wall thickness, material strength, pipe diameter, anti-corrosion coating condition, and historical test results, is input into the model. After screening key variables, the system maps pressure-related variables to pressure load feature groups and constructs pressure load edges based on the similarity of pressure conditions.

[0111] In this embodiment, the pressure load edge in the adaptive multiphysics coupling graph is used to express the correlation between pipe segments with similar pressure conditions. The GAT-Mamba model predicts future corrosion depth, failure pressure, remaining life, and initial risk score based on pressure load changes and historical corrosion status. The pressure safety margin constraint in the physical constraint joint output plays a major role; when the predicted failure pressure is lower than the operating pressure under the safety factor, or when the pressure safety margin is insufficient, the physical violation score of the pipe segment is increased, and this score is used as an important basis for identifying high-risk protection domains.

[0112] For urban gas pipeline sections with significant pressure fluctuations or high pressure peaks, the pressure load edge weight is increased during the adaptive edge weight update process, giving pressure conditions a greater impact on failure pressure prediction, risk correction, and maintenance priority ranking. When a pipeline section experiences insufficient safety margin, suggestions for pressure verification, increased monitoring, early re-inspection, or repair and reinforcement can be generated.

[0113] Example 5: In pipeline scenarios near subways, electrified railways, substations, grounding electrodes, or high-voltage power facilities, pipelines may be subject to stray current interference. Stray currents, pipe-to-ground potential fluctuations, and abnormal cathodic protection conditions can accelerate localized corrosion, increasing the corrosion rate of local pipe sections and leading to rapid corrosion depth development.

[0114] In this embodiment, the system focuses on collecting data such as stray current density, pipe-to-ground potential, cathodic protection potential, protective current, soil resistivity, anti-corrosion layer damage status, and historical corrosion status. After screening key variables, the system maps these variables to electrochemical feature groups and environmental auxiliary feature groups, and constructs electrochemical admittance edges using electrochemical state similarity.

[0115] In this embodiment, the GAT-Mamba model predicts corrosion rate, future corrosion depth, and remaining lifetime based on electrochemical state, historical corrosion state, and spatial correlation of pipe segments. Faraday's law constraints in the physical constraints joint output are used to ensure reasonable consistency between the predicted corrosion rate and the driving factors of electrochemical corrosion. When stray current density increases, cathodic protection is insufficient, pipe-to-ground potential is abnormal, or the condition of the anti-corrosion coating is poor, the system increases the physical violation score of the corresponding pipe segment and includes it in the high-risk protection domain identification range.

[0116] For pipe sections with significant electrochemical interference, the electrochemical admittance edge weight is increased during the adaptive edge weight update process, and the influence of Faraday law constraint violation terms on the physical violation score is also increased, making the model more focused on the driving factors of electrochemical corrosion. For pipe sections with high physical violation and abnormal corrosion rate, suggestions for intensified monitoring, cathodic protection review, anti-corrosion layer inspection, or early re-inspection can be output.

[0117] Example 6: In pipeline integrity management, internal inspections are typically conducted periodically, such as every few years. Between inspections, the development of corrosion defects cannot usually be directly and continuously observed; it is necessary to combine historical inspection data, operating pressure data, electrochemical data, and environmental data to predict future corrosion trends.

[0118] In this embodiment, the internal detection data from multiple detection cycles are constructed into a historical state sequence. The corrosion depth, defect size, remaining wall thickness, operating pressure, electrochemical state, environmental state, material structure state, and historical maintenance records of each detection cycle are used as time series inputs. The Mamba module is used to learn the evolution law of corrosion state between different detection cycles, and the GAT module is used to learn the adaptive multiphysics spatial correlation between different pipe segments within the same detection cycle.

[0119] In this embodiment, the corrosion rate, future corrosion depth, failure pressure, safety margin, remaining life, and initial risk score for the next detection cycle or a specified future time window are output. Subsequently, the physical constraint joint output terminal applies consistency constraints to the prediction results. The physical violation evaluation module identifies abnormal pipe sections that violate corrosion mechanisms, remaining wall thickness, safety margins, risk ranking, or spatial continuity relationships. The high-risk protection domain module includes potentially dangerous pipe sections in the scope of key attention, and the risk correction module conservatively corrects the initial risk score.

[0120] The method of this invention can predict the corrosion development trend between adjacent detection cycles and, based on the final risk score and pipe section level... P 10 The remaining lifespan, maximum risk, and protection zone ratio are used to formulate the next round of inspection plans. For pipe sections predicted to be high-risk, the system outputs maintenance recommendations such as early re-inspection, shortening the inspection cycle, increasing monitoring frequency, repairing and reinforcing, or replacing the pipe section.

[0121] Example 7: This embodiment provides a spatiotemporal joint prediction system for pipeline corrosion failure that integrates multi-source physical information, including: a multi-source data input module for collecting multi-source data on pipeline corrosion failure, uniformly converting the multi-source data to obtain multi-source state input; a physical feature extraction module for extracting features from the multi-source state input to obtain key variables, and constructing a multi-physical coupling graph based on the key variables; a failure prediction module for obtaining pipeline corrosion failure prediction results based on the multi-source state input and the multi-physical coupling graph, including: predicted corrosion rate, predicted future corrosion depth, predicted failure pressure, predicted remaining life, and initial risk score; a physical constraint module for applying physical consistency constraints to the initial risk score of pipeline corrosion failure to obtain a physical violation score, and obtaining a high-risk protection domain score based on the physical violation score; and a maintenance decision module for correcting the initial score based on the high-risk protection domain score to obtain a final risk score, and obtaining a segment-level maintenance decision based on the final risk score.

[0122] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A spatiotemporal joint prediction method for pipeline corrosion failure integrating multi-source physical information, characterized in that, Includes the following steps: S1: Collect multi-source data on pipeline corrosion failure, and perform unified conversion on the multi-source data to obtain multi-source status input; S2: Extract features from the multi-source state input to obtain key variables, and construct a multi-physics coupling graph based on the key variables; S3: Based on the multi-source state input and the multi-physics coupling diagram, the pipeline corrosion failure prediction results are obtained. The pipeline corrosion failure prediction results include: predicted corrosion rate, predicted future corrosion depth, predicted failure pressure, predicted remaining life, and initial risk score. S4: Physical consistency constraints are applied to the initial risk score of pipeline corrosion failure to obtain a physical violation score, and a high-risk protection domain score is obtained based on the physical violation score. S5: Based on the high-risk protection domain score, the initial score is corrected to obtain the final risk score; and based on the final risk score, a pipe section-level maintenance decision is obtained.

2. The spatiotemporal joint prediction method for pipeline corrosion failure integrating multi-source physical information according to claim 1, characterized in that, In the multiphysics coupling diagram, the edge relationships include: electrochemical admittance edge, pressure load edge, spatial adjacency edge, corrosion state edge, and material structure edge. For the i Pipe section and the first j Pipe section in q Characteristics under segment physical relationship and Physically associated edges are represented as: ； In the formula, Indicates the first i Pipe section and the first j Pipe section in q The strength of association under the physical relationship of segments; q = e Indicates electrochemical relationships, q = p Indicates the pressure-load relationship. q = c Indicates the relationship of corrosion state. q = m Indicates the material structure relationship; Indicates the first q Scale parameters of physical relationships between segments; This indicates the avoidance of extremely small positive numbers with a denominator of zero; The spatial adjacent edge is based on the first i Pipe section and the first j The mileage distance between pipe segments is constructed such that when the mileage distance between the two segments is not greater than the preset spatial adjacency radius, the spatial adjacency edge is set to 1; otherwise, it is set to 0.

3. The spatiotemporal joint prediction method for pipeline corrosion failure integrating multi-source physical information according to claim 1, characterized in that, In S3, the prediction results of pipeline corrosion failure are obtained based on the GAT-Mamba spatiotemporal prediction model. The GAT-Mamba spatiotemporal prediction model includes: GAT and Mamba temporal evolution modules; The GAT is used to process the spatial relationships in the multi-physics coupling graph; firstly, the first... i Pipe segment nodes and the first j Graph attention weights for adjacent pipe segment nodes; aggregate the features of adjacent pipe segment nodes based on the graph attention weights to obtain spatially enhanced pipe segment features; The spatially enhanced pipe segment features are input into the Mamba temporal evolution module to obtain the pipeline corrosion failure prediction results: ； In the formula, Indicates the first i Pipeline segment at future time Preliminary multi-task prediction results; This indicates the predicted corrosion rate; This indicates a predicted future corrosion depth; Indicates the predicted failure pressure; Indicates the projected remaining lifespan; Indicates the first i Initial risk assessment of the pipeline segment.

4. The spatiotemporal joint prediction method for pipeline corrosion failure integrating multi-source physical information according to claim 3, characterized in that, In S4, the physical violation score includes: ； In the formula, Indicates the first i Physical violation score of the pipe section; This indicates a violation of Faraday's law constraints; This indicates a violation of the remaining wall thickness constraint; This indicates a violation of the pressure safety margin constraint; This indicates a violation of the risk ranking constraint; Indicates a violation of spatial risk continuity; , , , , These represent the weight coefficients for different violations.

5. The spatiotemporal joint prediction method for pipeline corrosion failure integrating multi-source physical information according to claim 4, characterized in that, In S4, the high-risk protection domain scoring includes: ； In the formula, Indicates the first i The scoring for the pipe section belonging to the high-risk protection zone; This represents the Sigmoid function or the normalized mapping function; This represents the normalized corrosion depth; This represents the normalized corrosion rate; This represents the normalized safety margin risk term. Indicates the risk propagation item in the spatial neighborhood; to This represents the weighting coefficients for different risk factors.

6. The spatiotemporal joint prediction method for pipeline corrosion failure integrating multi-source physical information according to claim 3, characterized in that, In S5, the final risk score includes: ； In the formula, Indicates the first i Final risk score for the pipeline segment; This indicates an insufficient pressure safety margin, when... The value is 1 if the condition is met, otherwise it is 0. to Indicates the risk-adjusted weight; This indicates that the risk score will be limited to between 0 and 1.

7. A spatiotemporal joint prediction system for pipeline corrosion failure integrating multi-source physical information, used to implement the method as described in any one of claims 1-6, characterized in that, include: Multi-source data input module: used to collect multi-source data on pipeline corrosion failure, and to uniformly convert the multi-source data to obtain multi-source status input; Physical feature extraction module: used to extract features from the multi-source state input, obtain key variables, and construct a multi-physical coupling graph based on the key variables; Failure prediction module: used to obtain pipeline corrosion failure prediction results based on the multi-source state input and the multi-physics coupling diagram. The pipeline corrosion failure prediction results include: predicted corrosion rate, predicted future corrosion depth, predicted failure pressure, predicted remaining life and initial risk score. Physical constraint module: used to apply physical consistency constraints to the initial risk score of pipeline corrosion failure to obtain a physical violation score, and to obtain a high-risk protection domain score based on the physical violation score; Maintenance decision module: used to revise the initial score based on the high-risk protection domain score to obtain the final risk score; and to obtain the pipe section-level maintenance decision based on the final risk score.

Images