Intelligent fault diagnosis method for buried pipeline cathode protection system

Abstract

The invention relates to the field of intelligent fault diagnosis, and particularly discloses an intelligent fault diagnosis method for a buried pipeline cathode protection system, which comprises the following steps of: capturing a data mode hidden in multiple dimensions by fusing information (time characteristics) of test pile parameters changing along with time and position topology and spatial correlation (spatial correlation) among test piles; therefore, the identification capability of the difference between similar representation fault types is effectively improved. The space-time collaborative analysis mode provides a solid foundation for realizing intelligent fault diagnosis with high confidence, high precision and high robustness, and is an important technical path for promoting intelligent operation and maintenance management of the buried pipeline cathode protection system.

Description

Technical Field

[0001] This application relates to the field of intelligent fault diagnosis, and more specifically, to an intelligent fault diagnosis method for cathodic protection systems of buried pipelines. Background Art

[0002] Buried pipelines, as lifeline projects for transporting important resources such as oil, natural gas, and water, are crucial for safe and stable operation. However, due to being buried in a complex soil environment for a long time, pipelines are inevitably threatened by corrosion, which may lead to leakage, environmental pollution, economic losses, and even major safety accidents in severe cases. To effectively delay and prevent pipeline corrosion, cathodic protection (CP) technology is widely used. It changes the potential of pipeline metal by applying a protection current to make it in a state that is not easily corroded. Although the cathodic protection system is a key measure to ensure pipeline safety, the system itself may also experience operating failures due to various factors such as power failures, anode failures, pipeline coating damage, and stray current interference. Once the cathodic protection system fails or its effectiveness weakens, the pipeline will lose effective protection, and the corrosion risk will increase sharply. Therefore, establishing a set of solutions that can timely and accurately diagnose faults in the cathodic protection system of buried pipelines is of great significance for ensuring the long-term safe operation of pipelines and preventing accidents.

[0003] Currently, for the fault diagnosis of cathodic protection systems of buried pipelines, it usually relies on the analysis of data such as the pipe-to-soil potential (PSP) collected from test posts set along the pipeline. Existing diagnostic methods are mostly based on manual experience interpretation, simple threshold comparison, or traditional statistical analysis and signal processing techniques. For example, the protection status is judged by monitoring whether the PSP value of a single test post is within the preset protection potential range, or methods such as Fourier transform and wavelet analysis are used to process the PSP time series data to identify abnormal fluctuations. However, these methods often show limitations when dealing with complex working conditions. A significant challenge is that different types of cathodic protection system faults (such as severe local coating damage and external strong current interference) may exhibit similar characteristics in PSP data, especially when only considering the time series data of a single test post or analyzing spatial data points in isolation. Traditional analysis methods usually separate the time and space dimensions or only focus on a single dimension (such as time series analysis), making it difficult to effectively capture the complex correlation and propagation characteristics of faults in time and space. This lack of ability to mine the spatio-temporal coupling characteristics of data results in weak ability to distinguish different fault types with similar characteristics, easily causing misjudgment or missed judgment, and affecting the timeliness and accuracy of fault handling.

[0004] Therefore, an optimized intelligent fault diagnosis method for cathodic protection systems of buried pipelines is expected. Summary of the Invention

[0005] To solve the above technical problems, the present application is proposed. An embodiment of the present application provides an intelligent fault diagnosis method for a buried pipeline cathodic protection system.

[0006] According to one aspect of the present application, there is provided an intelligent fault diagnosis method for a buried pipeline cathodic protection system, which includes: Real-time collect PSP data of each test pile along the pipeline; Organize the PSP data of each test pile according to the test pile sample dimension and time dimension to obtain a set of test pile PSP time series; Perform test pile node feature time series encoding on each test pile PSP time series in the set of test pile PSP time series to obtain a set of test pile PSP time series feature encoding vectors; Construct a test pile spatial distribution topology matrix between each test pile; Input the set of test pile PSP time series feature encoding vectors and the test pile spatial distribution topology matrix into a test pile spatio-temporal feature depth extractor based on a spatio-temporal graph neural network to obtain a test pile PSP spatio-temporal joint state representation matrix; Based on the test pile PSP spatio-temporal joint state representation matrix, determine the fault type and its confidence level.

[0007] Compared with the prior art, an intelligent fault diagnosis method for a buried pipeline cathodic protection system provided by the present application captures hidden data patterns in multiple dimensions by fusing the information of the test pile parameters changing over time (time features) and the position topology and spatial correlation between each test pile (spatial association), so as to effectively improve the identification ability of the differences between similar represented fault types. This spatio-temporal collaborative analysis method provides a solid foundation for realizing intelligent fault diagnosis with high confidence, high accuracy and high robustness, and is an important technical path to promote the intelligent operation and maintenance management of buried pipeline cathodic protection systems. Brief Description of the Drawings

[0008] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0009] Figure 1 It is a flowchart of an intelligent fault diagnosis method for a buried pipeline cathodic protection system according to an embodiment of the present application; Figure 2Schematic diagram of data flow for the intelligent fault diagnosis method of the cathodic protection system for buried pipelines according to an embodiment of the present application; Figure 3 Flowchart of sub-step S3 of the intelligent fault diagnosis method for the cathodic protection system of buried pipelines according to an embodiment of the present application. Detailed implementation manners

[0010] Hereinafter, exemplary embodiments of the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0011] As shown in the present application and the claims, unless the context clearly indicates otherwise, words such as "a", "an", "one", and / or "the" are not specifically singular and may also include plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0012] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or the server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0013] In the technical solution of the present application, an intelligent fault diagnosis method for a cathodic protection system of buried pipelines is proposed.

[0014] Hereinafter, exemplary embodiments of the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0015] In the technical solution of the present application, an intelligent fault diagnosis method for a cathodic protection system of buried pipelines is proposed. Figure 1 Flowchart of the intelligent fault diagnosis method for the cathodic protection system of buried pipelines according to an embodiment of the present application. Figure 2 Schematic diagram of data flow for the intelligent fault diagnosis method of the cathodic protection system for buried pipelines according to an embodiment of the present application. As Figure 1 and Figure 2As shown in the figure, an intelligent fault diagnosis method for an underground pipeline cathodic protection system according to an embodiment of the present application includes the steps of: S1, collecting PSP data of each test pile along the pipeline in real time; S2, sorting the PSP data of each test pile according to the test pile sample dimension and time dimension to obtain a set of test pile PSP time series; S3, performing test pile node feature time series encoding on each test pile PSP time series in the set of test pile PSP time series to obtain a set of test pile PSP time series feature encoding vectors; S4, constructing a test pile spatial distribution topology matrix between each test pile; S5, inputting the set of test pile PSP time series feature encoding vectors and the test pile spatial distribution topology matrix into a test pile spatio-temporal feature depth extractor based on a spatio-temporal graph neural network to obtain a test pile PSP spatio-temporal joint state representation matrix; S6, determining the fault type and its confidence level based on the test pile PSP spatio-temporal joint state representation matrix.

[0016] Specifically, in step S1, the PSP data of each test pile along the pipeline is collected in real time. Since underground pipelines are widely distributed, have a complex environment and are vulnerable to external factors, data at a single location or at a single moment is difficult to comprehensively reflect the health status of the entire system. Only by continuously and synchronously collecting data from all test piles can raw information covering the entire line with spatio-temporal continuity be obtained, providing solid data support for subsequent intelligent diagnosis. In addition, different types of faults (such as local grounding anomalies, uneven current distribution, etc.) often exhibit unique data patterns at different spatial positions and time periods. Therefore, relying only on partial or intermittent sampling is likely to miss potential hazards, affecting the accuracy and timeliness of the diagnosis results. Therefore, in the technical solution of the present application, the PSP data of each test pile along the pipeline is collected in real time. Among them, the PSP data of the test pile refers to the pipe-to-soil potential values measured at each test pile at different time points, and these data can dynamically reflect the actual effect and state change of the pipeline under cathodic protection at this location. The PSP data is important basic information for evaluating the operation status of the cathodic protection system, judging whether there is a corrosion risk, and discovering potential faults. It is worth mentioning that in the underground pipeline cathodic protection system, the test pile is a key monitoring node arranged along the pipeline route, used to regularly or real-time collect the pipe-to-soil potential data between the pipeline metal and the reference electrode.

[0017] In an example of the present application, during the real-time acquisition process, remote monitoring terminals are usually installed at each test pile. These terminals can automatically detect and upload PSP values to the central server according to a preset period; the central platform summarizes and organizes the data from different locations and different time points, and organizes it into a complete data set according to two dimensions: space (test pile distribution) and time (continuous observation period). In this way, not only the comprehensiveness and continuity of data acquisition are ensured, but also a solid data foundation is provided for subsequent data analysis and intelligent diagnosis.

[0018] Specifically, in step S2, the PSP data of each test pile is sorted according to the test pile sample dimension and the time dimension to obtain a set of test pile PSP time series. It should be understood that the PSP data collected from each test pile along the line are originally scattered, asynchronous and in various formats of original observation values. These data directly reflect the electrochemical protection states of different positions of the pipeline at different times, and the faults in the cathodic protection system often have both dynamic characteristics evolving over time and complex relationships of mutual influence and propagation among different spatial nodes. The unorganized data is difficult to reveal these spatio-temporal interaction patterns and is not conducive to the subsequent deep learning model to effectively capture abnormal trends and spatial distribution rules. Therefore, in order to construct a structured, continuous and multi-dimensional set of test pile PSP time series to solve the problem of information fragmentation caused by inconsistent acquisition frequencies, timestamps or spatial distributions of the original data, so that the data of each test pile can be aligned in a unified time sequence, in the technical solution of the present application, the PSP data of each test pile is sorted according to the test pile sample dimension and the time dimension to obtain a set of test pile PSP time series. Through standardized sorting, each test pile has a clearly traceable potential change trajectory, which is not only convenient for longitudinally analyzing the development trend of its own state over time, but also provides a basic condition for horizontally comparing the potential correlations between different nodes. This multi-dimensional, multi-node and continuous time series set provides high-quality input samples for the intelligent diagnosis method, enabling advanced models such as spatio-temporal graph neural networks to make full use of long-term dependence relationships and spatial topology structure information to achieve accurate identification and differentiation of complex fault types and their evolution processes.

[0019] Specifically, in step S3, each test pile PSP time series in the set of test pile PSP time series is encoded with test pile node feature time series to obtain a set of test pile PSP time series feature encoding vectors. In a specific example of the present application, such as Figure 3As shown, the S3 includes: S31, performing initial timing coding of the test pile node features on each test pile PSP time series in the set of test pile PSP time series to obtain a test pile PSP initial timing feature coding vector; S32, performing timing fine-grained feature enhancement on the test pile PSP initial timing feature coding vector to obtain a test pile PSP timing feature coding vector.

[0020] Specifically, the S31 performs initial temporal coding of the test pile node features on each test pile PSP time series in the set of the test pile PSP time series to obtain the test pile PSP initial temporal feature coding vector. It should be understood that the cathodic protection state reflected by the PSP data at different time points presents a complex time dependency and nonlinear change trend. Directly processing the original time series data is not only susceptible to noise, but also difficult to accurately capture meaningful evolution patterns. The forward LSTM model is a special recurrent neural network structure that is good at processing and modeling dependencies over time in sequence data. It effectively selectively retains or forgets information through a gating mechanism (input gate, forget gate, and output gate), thereby overcoming the gradient vanishing problem that is difficult to capture in traditional recurrent neural networks for long-distance dependencies, so that the model can learn the correlation and dynamic change characteristics between long-distance moments in the time series. Therefore, in the technical solution of the present application, each test pile PSP time series is subjected to initial temporal coding of the test pile node features based on the forward LSTM model to obtain the test pile PSP initial temporal feature coding vector.

[0021] By encoding the time series of each test pile through the forward LSTM model, the deep time series features in the sequence can be automatically extracted, and the dynamic behavior feature vector of the test pile can be mined, so that the time context information can be transmitted throughout, thereby providing structured and information-rich input for subsequent fault diagnosis. In this way, not only the expression ability and information concentration efficiency of time series data are effectively improved, but also accurate, dynamic, and real-time time series feature input is provided for the intelligent fault diagnosis of the buried pipeline cathodic protection system, which significantly helps to achieve high-precision and high-confidence fault identification and location, thereby ensuring the safe and stable operation of the pipeline.

[0022] Specifically, in S32, the initial timing feature encoding vector of the test pile PSP is subjected to timing fine-grained feature enhancement to obtain the timing feature encoding vector of the test pile PSP. It should be understood that since the potential signals of different test piles along the pipeline may exhibit high dynamics, non-stationarity, and local mutation characteristics in the time dimension, the initial features extracted only by basic timing models such as unidirectional LSTM are often difficult to fully capture potential multi-scale timing patterns and local key events. These initial encodings may lose the fault precursor information in short-term fluctuations, or cannot effectively distinguish the similar timing morphological differences caused by different fault incentives. Therefore, in order to structure the feature enhancement architecture and extract more physically meaningful and discriminative fine-grained features from the original timing encoding to reveal hidden fault patterns, in the technical solution of this application, the initial timing feature encoding vector of the test pile PSP is subjected to timing fine-grained feature enhancement to obtain the timing feature encoding vector of the test pile PSP.

[0023] In this process, first, through the local feature decomposition of one-dimensional convolution, the system can simultaneously capture short-term spikes (which may correspond to instantaneous interference) and long-period slow-varying trends (which may correspond to equipment aging); then, by calculating the manifold correlation between these local primitives (such as the propagation intensity of similar potential fluctuation patterns between spatially adjacent test piles), a topological network representing the spatial distribution relationship of fault features is constructed; furthermore, through the adaptive adjustment of the gating mask, the system can automatically strengthen the correlation between test piles with close physical distance and similar electrical characteristics, and weaken the pseudo-correlation caused by geological mutations; furthermore, through dense feedback distillation, the local timing features of each test pile are incorporated into the context information of upstream and downstream nodes during the refinement process, enabling the system to comprehensively infer the thinking process of the fault source from multi-node data; finally, through self-attention-driven feature reconstruction, the system can dynamically aggregate the enhanced local primitives according to the fault propagation logic to form a high-order feature expression that can not only represent the timing evolution details of a single test pile but also reflect the cross-node fault cooperation mode, obtaining the enhanced initial timing feature encoding vector of the test pile PSP. The generated timing feature encoding vector of the test pile PSP can effectively suppress the interference of measurement noise and redundant fluctuations on the feature space, while highlighting the local timing patterns strongly related to faults (such as sudden drops in protection current, inflection points of potential drift, etc.), significantly improving the model's ability to analyze the dynamic behavior of the cathodic protection system. This fine-grained feature processing enables the system to maintain high-robustness diagnostic performance even in the face of complex scenarios such as multiple faults occurring simultaneously and environmental interference superposition, significantly reducing the misdiagnosis rate caused by feature confusion.

[0024] Specifically, first of all, it should be understood that the PSP data of the cathodic protection system often presents multi-level coupling characteristics in a complex geological environment, such as long-period trend drift caused by changes in soil resistivity, instantaneous spikes caused by lightning interference, and progressive potential decay caused by damage to the anti-corrosion layer. These physical phenomena of different time scales are superimposed on each other in the original time series signal. Although the initial LSTM encoding can capture time dependence, its hidden state representation tends to be globally smooth, and it is difficult to effectively decouple fault modes with different response rates, resulting in the inability to accurately trace the fault mechanism during subsequent spatiotemporal correlation analysis. Therefore, in the technical solution of the present application, the initial time series feature coding vector of the test pile PSP is subjected to time series feature decomposition based on one-dimensional convolutional coding to obtain a set of initial local implicit time series feature vectors of the test pile PSP.

[0025] By adopting one-dimensional convolutional coding, the sliding scanning ability of the convolution kernel on the time series can be used to effectively identify local time series fragments in the feature coding and mine multi-scale local patterns hidden in the overall features. This is particularly critical for diagnosing fault signals that are generated in local time periods and have typical manifestations. In addition, when there are periodic industrial current interference or transient lightning pulses along the pipeline, the one-dimensional convolution decomposition can effectively separate the characteristic expressions of environmental noise and equipment failures, avoiding feature confusion caused by long-term memory of LSTM hidden states. By extracting multi-scale local features in parallel, the system can simultaneously capture the millivolt-level slow offset caused by insulation layer damage and the hundreds of millivolt-level high-frequency fluctuations caused by the oscillation of the constant potentiostat, providing feature primitives with clear physical orientation for subsequent manifold correlation analysis.

[0026] In a specific example of the present application, the initial timing feature coding vector of the test pile PSP is subjected to a timing feature decomposition based on one-dimensional convolution coding using the following one-dimensional convolution formula to obtain a set of initial local implicit timing feature vectors of the test pile PSP; wherein the one-dimensional convolution formula is: , , in, is the initial timing feature encoding vector of the test pile PSP, are the first, second, and third initial timing feature coding vectors of the test pile PSP, respectively. and eigenvalues, represents one-dimensional convolutional coding, are the first, second, and third in the set of initial local implicit time series feature vectors of the test pile PSP. and The initial local implicit timing feature vector of the test pile PSP.

[0027] Next, it should be understood that when the pipeline traverses a variable soil environment or there is concealed corrosion, the PSP fluctuations of different test piles often contain non-Euclidean space correlations - for example, the potential gradient transmission driven by groundwater flow may cause non-adjacent test piles to exhibit similar characteristics, while adjacent test piles may show heterogeneity due to geological mutations. Although the initial local features capture multi-scale temporal patterns through convolutional decomposition, it is difficult to reveal the fault propagation law beyond physical distance by analyzing each feature vector in isolation, and it is easy to misjudge faults with similar local fluctuations but different causes as the same type of event. Therefore, in order to break through the inherent paradigm of the traditional adjacency matrix relying on spatial distance, in the technical solution of this application, the manifold structure correlation coefficient between any two initial local implicit temporal feature vectors of the test pile PSP is calculated to obtain a test pile PSP manifold structure correlation topology matrix composed of multiple manifold structure correlation coefficients.

[0028] In this process, the manifold correlation topology explicitly models the deep dependence relationship between test piles in a data-driven manner, enabling the spatio-temporal graph neural network to distinguish pseudo-correlations caused by environmental noise from feature couplings formed by real fault propagation. For example, when there are multi-point grounding failures in the pipeline, the manifold correlation matrix can automatically identify sub-graphs of test piles with similar decay slopes, fluctuation phases, etc., and accurately reflect the path of fault diffusion along the underground conductive medium. Compared with traditional methods, this correlation discovery mechanism based on feature similarity can effectively handle cross-regional and non-linear fault propagation scenarios in long-distance pipelines, significantly improving the system's ability to trace multi-point concurrent faults and the spatial positioning accuracy, providing a reliable correlation analysis basis for the intelligent operation and maintenance of complex pipe networks.

[0029] In a specific example of this application, the manifold structure correlation coefficient between any two initial local implicit temporal feature vectors of the test pile PSP is calculated according to the following manifold structure correlation formula to obtain a test pile PSP manifold structure correlation topology matrix composed of multiple manifold structure correlation coefficients; where the manifold structure correlation formula is: , , where represents the one-norm of the vector, represents the exponential operation with as the base, is the transpose of is the th initial local implicit temporal feature vector of the test pile PSP in the set of initial local implicit temporal feature vectors of the test pile PSP, is the and the manifold structure correlation coefficient between denotes a function is the gating mask weight matrix, is the gating mask bias matrix, is the multiple manifold structure correlation coefficients, is the test stub PSP manifold structure correlation topology matrix of the said.

[0030] Then, it should be understood that when the pipeline passes through the electromagnetic interference area or there is a sudden change in geological conditions, the PSP fluctuations between test stubs may generate pseudo-correlation signals due to external environmental disturbances - for example, the instantaneous potential synchronous fluctuations caused by lightning strikes may be misjudged as fault propagation correlations, and the characteristic coupling signals caused by real faults may be submerged by noise. At this time, although the original manifold correlation coefficient matrix can reflect the potential correlation patterns in the feature space, the redundant connections it contains will interfere with the subsequent feature distillation process, resulting in the spatio-temporal graph neural network being unable to focus on the feature learning of the real fault propagation path. Therefore, in the technical solution of this application, the test stub PSP manifold structure correlation topology matrix is input into the gating mask function to obtain the initial test stub PSP manifold structure fine-grained correlation mask topology matrix.

[0031] That is, the adaptive modulation of the correlation strength is realized through the gating mechanism to construct a dynamic attention mechanism that conforms to the physical laws of fault propagation. In a specific example, for the real fault correlation formed by the soil conductive channel, the gating unit automatically strengthens its weight through gradient backpropagation; while for the pseudo-correlation generated by environmental noise or accidental fluctuations, it is attenuated through the suppression function. The fine-grained correlation mask topology after gating modulation effectively strips the noise components in the original manifold correlation, enabling the dense feedback distillation process to accurately capture the spatial propagation mode of the fault signal. When there are multi-point grounding faults in the pipeline, this mechanism can suppress the global weak correlations caused by common-mode interference and highlight the strong correlation features inside the local fault cluster, thus preventing the subsequent classifier from misjudging the distributed faults as systematic anomalies. This dynamic correlation focusing ability significantly improves the recognition sensitivity of the model to hidden faults.

[0032] Furthermore, it should be understood that due to the polarization and strengthening effect of the gating mask function on the correlation relationship, the geometric correlation distribution of the original topological matrix may lead to the compression of the stability of the local structure field due to the accumulation of non-linear dependencies. Specifically, it is manifested as the unevenness or discontinuous mutation of the spatial distribution of the correlation intensity. This geometric distortion will weaken the characterization ability of the fine-grained correlation mask topological matrix of the manifold structure for the potential fault propagation path between test piles. For example, it is impossible to accurately depict the spatial attenuation gradient of the corrosion current along the pipeline or the difference in the propagation range of transient interference, thus affecting the physical interpretability of subsequent feature distillation and reconstruction. Therefore, in the preferred example of the present application, the geometric correlation of the test pile PSP manifold structure is optimized based on gradient field tuning for each eigenvalue in the fine-grained correlation mask topological matrix of the initial test pile PSP manifold structure to obtain the fine-grained correlation mask topological matrix of the test pile PSP manifold structure.

[0033] That is, the adaptive smoothing and equalization of the geometric correlation structure field are realized through the gradient field tuning mechanism. Specifically, a gradient field term is introduced for each eigenvalue in the topological matrix, aiming to correct the local geometric structure distortion caused by non-linear non-saturation. For example, eliminating the false correlation peaks generated by the over-strengthening of the similarity of the potential fluctuation modes of adjacent test piles, or filling the correlation break regions caused by too far spatial distance. By using the gradient field term as an external field driving term in the mean field tuning process of the eigenvalue, the correlation intensity distribution between the high gradient response region and the low gradient region can be dynamically balanced. The fine-grained correlation mask topological matrix of the manifold structure optimized by the gradient field tuning can more accurately reflect the potential spatial correlation characteristics of the potential dynamics between test piles. The optimized topological matrix not only enhances the cooperative response characteristics of adjacent test piles under the same fault mode (such as the progressive correlation of the potential drift caused by grounding failure along the pipeline), but also can effectively distinguish the non-structural correlation fluctuations caused by environmental noise or isolated interference. This equalization of the geometric correlation structure provides a reliable weight basis for subsequent dense feedback distillation, making the cross-node information fusion of local temporal features more conform to the propagation mechanism of real faults. Finally, this optimization step improves the modeling ability of the spatio-temporal graph neural network for complex fault scenarios and lays a high-fidelity correlation topology foundation for fault type discrimination and confidence evaluation.

[0034] In this example, there will be non-linear non-saturation in the geometric correlation distribution of the test pile PSP manifold structure correlation topological matrix on the potential low-dimensional manifold structure, so that the overall correlation topological distribution of the test pile PSP manifold structure correlation topological matrix is compressed due to non-linear dependencies, and this will become more significant due to the correlation polarization and strengthening of the gating mask function, affecting the substantial geometric fine-grained correlation structure expression effect of the fine-grained correlation mask topological matrix of the test pile PSP manifold structure.

[0035] Based on this, for the fine-grained correlation mask topology matrix of the test pile PSP manifold structure for each eigenvalue , first introduce the gradient field term: , wherein represents each eigenvalue of the fine-grained correlation mask topology matrix of the test pile PSP manifold structure.

[0036] which is used to correct the non-uniformity of the local non-linear geometric structure so as to realize the fine-grained smoothing of the geometric correlation structure field.

[0037] Then, use the gradient field term as the external field driving term to perform the mean field tuning of each eigenvalue: , where is the eigen mean value of all eigenvalues of the fine-grained correlation mask topology matrix of the test pile PSP manifold structure , and is the scaling weight for modulating the over-large external field driving term, represents each eigenvalue of the fine-grained correlation mask topology matrix of the test pile PSP manifold structure.

[0038] In this way, under the action of the external field driving term as the high-order gradient, it reversely promotes the non-linear saturation of the geometric correlation distribution under the mean field (i.e., the strong gradient responsiveness decreases), so as to compensate for the discretization of the sub-geometric correlation structure caused by the strengthening of the correlation polarization through the mean harmonic response under the mean field, thereby improving the substantial geometric fine-grained correlation structure expression effect of the fine-grained correlation mask topology matrix of the test pile PSP manifold structure.

[0039] Furthermore, it should be understood that since the PSP time series characteristics of a single test pile only reflect its local potential dynamics and cannot characterize the spatial propagation effect of faults in the pipeline network (such as the collaborative drift of the potentials of adjacent test piles caused by current leakage), and the independent analysis mode relied on by traditional methods will ignore this cross-node correlation. In addition, although the manifold structure fine-grained correlation mask topology matrix quantifies the potential correlation strength between test piles, the original local features without information fusion still have semantic ambiguities (such as similar fluctuation patterns caused by environmental noise and real faults), and there is an urgent need to enhance the characterization consistency of features through structured information interaction. Therefore, in the technical solution of this application, based on the fine-grained correlation mask topology matrix of the test pile PSP manifold structure, time series feature dense feedback distillation is performed on each test pile PSP initial local implicit time series feature vector in the set of test pile PSP initial local implicit time series feature vectors to obtain the set of test pile PSP distilled initial local implicit time series feature vectors.

[0040] That is, global semantic alignment and collaborative enhancement of local features are achieved through dense information interaction across nodes. Specifically, based on the spatial weight distribution provided by the fine-grained association mask topology matrix of the manifold structure, neighborhood-aware information distillation is performed on the initial local implicit temporal feature vectors of each test stub. This process dynamically filters the feature segments of other nodes that have strong spatial dependence with the current test stub (such as the sudden potential drop pattern of the upstream test stub) through the association strength defined by the mask matrix, and integrates them into the feature expression of the current node through weighted aggregation. This dense feedback mechanism not only compensates for the local perspective defects of single-node features, but also reconstructs the physical meaning of the features by introducing spatial context information (such as the attenuation law of faults along the pipeline), enabling the distilled feature vectors to encode both the abnormal evolution in the time dimension and the fault propagation path in the space dimension. The set of test stub PSP distilled initial local implicit temporal feature vectors generated effectively suppresses local noise interference (such as instantaneous potential fluctuations caused by lightning) by fusing multi-node association information, while strengthening the real fault features with spatial propagation characteristics (such as potential gradient diffusion caused by grounding failure). This feature enhancement enables subsequent self-attention reconstruction to more accurately capture key fault patterns across spatio-temporal dimensions, providing high-discriminative input representations for spatio-temporal graph neural networks.

[0041] In a specific example of this application, based on the fine-grained association mask topology matrix of the test stub PSP manifold structure, the following distillation formula is used to perform temporal feature dense feedback distillation on each test stub PSP initial local implicit temporal feature vector in the set of test stub PSP initial local implicit temporal feature vectors to obtain the set of test stub PSP distilled initial local implicit temporal feature vectors; where, the distillation formula is: , , where, represents element-wise multiplication by position, represents matrix multiplication, represents function, is the distillation weight matrix, represents scale of, is corresponding test stub PSP distilled initial local implicit temporal feature vector, is the set of the test stub PSP distilled initial local implicit temporal feature vectors, are respectively the 1st, 2nd, th, and th test stub PSP distilled initial local implicit temporal feature vectors in the set of the test stub PSP distilled initial local implicit temporal feature vectors.

[0042] Subsequently, it should be understood that although the distilled feature vectors enhance the spatial correlation through cross-node information fusion (such as the collaborative pattern of potential drifts of adjacent test piles), they still exist in a distributed form and lack the ability to uniformly model long-distance spatio-temporal dependence relationships. For example, the slow potential shift of a test pile at the far end of a pipeline and the transient current fluctuation at the near end may belong to different stages of the same fault propagation chain, but local features without global coordination are difficult to capture such potential correlations across spatio-temporal scales. In addition, there are differences in the contribution degrees of different local feature segments to fault discrimination. Simple weighted averaging or linear superposition may submerge key feature components, resulting in the diagnostic model being unable to focus on the core characteristics of the fault. Therefore, in the technical solution of this application, feature reconstruction based on the self-attention mechanism is performed on the set of initial local implicit temporal feature vectors distilled from the test pile PSP to obtain the test pile PSP temporal feature encoding vector.

[0043] That is, intelligent screening and high-order semantic fusion of local feature segments are achieved through the self-attention mechanism. Specifically, the self-attention mechanism automatically identifies spatio-temporal patterns with strong discriminability (such as the causal relationship between the sudden potential drop feature of a certain test pile during a specific period and the oscillation feature of another node in the subsequent period) by calculating the dynamic weights between the distilled local feature vectors, and quantitatively evaluates their importance based on the global context. This reconstruction process can not only capture the propagation path of fault signals in the pipeline network (such as the direction of potential gradient diffusion caused by corrosion points), but also adaptively suppress local features dominated by redundancy or noise, thereby refining the distributed local information into a globally interpretable state representation. The generated test pile PSP temporal feature encoding vector forms a hierarchical description of complex fault scenarios by dynamically aggregating key local evidences across nodes (such as the starting point feature of potential drift caused by grounding failure and the response feature on the propagation path). Through this mechanism, the model can break through the shackles of the local perspective and establish a coupling relationship between time evolution and spatial propagation in a higher dimension.

[0044] In a specific example of this application, the following feature reconstruction formula is used to perform feature reconstruction based on the self-attention mechanism on the set of initial local implicit temporal feature vectors distilled from the test pile PSP to obtain the test pile PSP temporal feature encoding vector; where the feature reconstruction formula is: , where , and are learnable weight matrices respectively, is the test pile PSP query feature vector, is the test pile PSP key vector, is the test pile PSP value vector, denotes the length of the set of initial local implicit temporal feature vectors of the test stake PSP distillation denotes feature reconstruction of the set of initial local implicit temporal feature vectors of the test stake PSP distillation denotes function is the temporal feature encoding vector of the test stake PSP

[0045] Specifically, in S4, a test stake spatial distribution topology matrix between each test stake is constructed. It should be understood that there are clear and non-negligible geographical and environmental correlations in the spatial distribution of the PSP data collected by each test stake in the buried pipeline system. Many faults or abnormal phenomena do not occur in isolation, but show a trend of diffusion, linkage or transmission within a certain range. If only relying on time series features and ignoring the true physical distance relationship between nodes, it will be difficult to reveal and capture these complex spatial synergy effects, resulting in limitations in aspects such as the positioning accuracy and type discrimination of the fault diagnosis model. In addition, each test stake, as an important monitoring node arranged along the pipeline, has a clear spatial position relationship, and these spatial information is of great significance for comprehensively understanding the cathodic protection state of the pipeline and the potential fault propagation. Therefore, in the technical solution of this application, a test stake spatial distribution topology matrix between each test stake is constructed, where the value of each position in the non-diagonal position of the test stake spatial distribution topology matrix is the spatial distance between the corresponding two test stakes.

[0046] Here, by clearly marking the actual distance between each pair of test stakes, the model can identify which nodes are geographically closely connected, so as to reasonably infer the possible location range where abnormal signals may occur and have an impact, and more effectively distinguish different regions or characterize location-based faults that are similar but have different causes. In this way, the perception ability of the entire intelligent diagnosis scheme of the cathodic protection system for complex fault types, their propagation paths, influence ranges, etc. is significantly improved. On the one hand, based on the data structure established by accurate spatial association, it effectively prevents the problems of misjudgment and missed judgment caused by ignoring physical proximity, and improves the accuracy and stability of the diagnosis results; on the other hand, by deeply mining the multi-point collaborative change law along the pipeline, more refined and forward-looking risk early warning and positioning are realized.

[0047] In particular, in step S5, the set of timing feature encoding vectors of the test pile PSP and the test pile spatial distribution topology matrix are input into the test pile spatio-temporal feature depth extractor based on the spatio-temporal graph neural network to obtain the test pile PSP spatio-temporal joint state representation matrix. It should be understood that although the previous steps extracted the dynamic potential features of a single test pile through timing encoding and feature enhancement, these features are still limited to the independent analysis of local time windows and cannot represent the spatial propagation characteristics of faults along the pipeline network (such as the gradient change of corrosion current decay with distance or the potential drift diffusion path caused by grounding failure). In addition, the physical spatial distance between test piles directly affects the propagation intensity and delay characteristics of fault signals. If only relying on the isolated analysis of time series, the model will not be able to distinguish the collaborative anomalies of spatially adjacent nodes from isolated interference events (such as the synchronous potential shift caused by multi-point corrosion in a certain section of the pipeline and the random fluctuations caused by lightning interference). Therefore, in the technical solution of this application, the set of timing feature encoding vectors of the test pile PSP and the test pile spatial distribution topology matrix are further input into the test pile spatio-temporal feature depth extractor based on the spatio-temporal graph neural network to obtain the test pile PSP spatio-temporal joint state representation matrix.

[0048] That is, the deep fusion of time dynamics and spatial topology structure is realized through the spatio-temporal graph neural network. Specifically, based on the spatial distance information provided by the test pile spatial distribution topology matrix (such as non-diagonal elements representing the actual physical distance between nodes), the STGNN performs spatial aggregation on the timing features of each test pile through graph convolution operations. For example, for a certain test pile node, the model will dynamically fuse the timing features of neighboring nodes (such as the timing pattern of potential mutation of adjacent test piles on the fault propagation path) according to the spatial distance weight with its neighboring nodes (the closer the distance, the higher the weight), so as to generate a joint representation vector that contains both its own time evolution law and embedded spatial correlation information. This fusion mechanism can explicitly model the propagation dynamics characteristics of faults in the pipeline network (such as the spatial progressive influence of current leakage caused by corrosion points along the pipeline direction), while suppressing non-structural fluctuations caused by environmental noise or isolated events. The generated test pile PSP spatio-temporal joint state representation matrix can not only capture the timing anomalies of a single test pile (such as a slow potential drop lasting for several hours), but also identify the collaborative anomaly patterns across nodes through spatial topology correlation (such as the potential drift sequence presented by multiple adjacent test piles according to the distance gradient). This spatio-temporal coupling analysis enables the model to distinguish fault types with similar appearances but different causes: for example, stray current interference may cause randomly distributed potential fluctuations, while corrosion caused by coating damage shows spatial correlation anomalies along the pipeline direction.

[0049] In a specific example of the present application, the set of timing feature encoding vectors of the test pile PSP and the spatial distribution topology matrix of the test pile are input into a deep extractor of the spatio-temporal features of the test pile based on a spatio-temporal graph neural network to obtain a spatio-temporal joint state representation matrix of the test pile PSP, including: extracting a first test pile PSP timing feature encoding vector from the set of timing feature encoding vectors of the test pile PSP; determining neighbor nodes of the first test pile PSP timing feature encoding vector based on the spatial distribution topology matrix of the test pile to obtain a subset of neighbor node test pile PSP timing feature encoding vectors; performing spatial aggregation based on graph convolution encoding on the first test pile PSP timing feature encoding vector and the subset of neighbor node test pile PSP timing feature encoding vectors to obtain a first test pile PSP spatio-temporal joint state representation vector, where the first test pile PSP spatio-temporal joint state representation vector is the first row vector of the spatio-temporal joint state representation matrix of the test pile PSP.

[0050] Specifically, in S6, based on the spatio-temporal joint state representation matrix of the test pile PSP, the fault type and its confidence are determined. It should be understood that although the spatio-temporal joint state representation matrix has integrated the timing dynamics and spatial topological correlation of the test pile (such as the gradient distribution of the corrosion current decaying along the pipeline distance), there may be a mixed representation of multiple fault modes in its high-dimensional feature space (such as the superposition effect of ground failure and stray current interference). It is difficult for a single classification model to directly analyze such high-dimensional heterogeneous features, and the similarity of the representations of different fault types in the spatio-temporal dimension (such as the potential drift caused by multi-point corrosion and the transient fluctuation caused by electromagnetic interference) requires a hierarchical classification strategy to improve the discrimination accuracy. Therefore, in the technical solution of the present application, first, the spatio-temporal joint state representation matrix of the test pile PSP is input into a fault type identifier based on a first classifier to obtain the test pile PSP fault type; then, based on the test pile PSP fault type, a second classifier for identifying a specific test pile PSP fault type is extracted from the classification model library; furthermore, the spatio-temporal joint state representation matrix of the test pile PSP is input into the second classifier to obtain the confidence of the specific test pile PSP fault type.

[0051] That is, precise fault type localization and quantification of the credibility of diagnostic results are achieved through a hierarchical classification framework. First, the first classifier conducts coarse-grained fault type identification based on the spatio-temporal joint state representation matrix, such as distinguishing corrosion-related faults from environmental interference anomalies. In this stage, by extracting the global spatio-temporal patterns in the matrix (such as the continuity characteristics of the fault propagation along the pipeline), the fault candidate range is quickly narrowed down. Subsequently, based on the preliminary classification results, a specific second classifier is called to conduct in-depth analysis of the fine-grained features of this type of fault (such as local potential gradient anomalies caused by coating damage and insufficient global protection current caused by anode failure). The second classifier outputs a confidence level to quantify the reliability of the diagnostic result by focusing on the spatio-temporal fingerprints of specific faults (such as the combined pattern of spatial correlation strength and time evolution rate), thus solving the ambiguity problem of similar fault types. Through the rough screening of the first classifier, the system can efficiently eliminate obviously irrelevant interference scenarios (such as instantaneous noise caused by lightning), reducing computational resource consumption; while the refined analysis of the second classifier effectively distinguishes easily confused faults (such as the potential response differences between local corrosion and aging of cathodic protection equipment) through confidence level evaluation. This hierarchical strategy makes the diagnostic result have both the accuracy of type discrimination and the interpretability of confidence level quantification, providing double guarantees for operation and maintenance decisions. Finally, through the integration of spatio-temporal features and hierarchical reasoning, this method achieves high-precision diagnosis in complex fault scenarios, providing intelligent support for the preventive maintenance and risk control of pipeline systems.

[0052] In a specific embodiment, the first classifier adopts a neural network structure with a softmax as the output layer. Its input is the spatio-temporal joint state representation matrix of the test pile PSP. First, the spatio-temporal joint state representation matrix of the test pile PSP is flattened into a one-dimensional feature vector and input into the fully connected layer of the neural network (in other examples of this application, it may also include multiple hidden layers combined with non-linear activation functions). In the output layer, the softmax function is used to generate the probability distribution of various main fault types, and the category corresponding to the maximum probability is used as the fault type of the test pile PSP. To improve the discrimination accuracy for secondary, easily confused, or specific types of faults to be discriminated, the second classifier is further extracted from the classification model library. The second classifier is also a deep neural network classifier with a softmax as the output layer. Its input is the same as that of the first classifier. It is retrained for the dataset corresponding to the target category determined by the first classifier, and discriminates the sub-categories or subtype faults under this first-level category, outputs the probability distribution of each subtype category, and the category corresponding to the maximum probability is the specific fault type of the test pile PSP, and the probability is the confidence of the specific fault type of the test pile PSP. Specifically, during the training process, for the first classifier, training samples containing multiple main category annotations are used, and the cross-entropy loss function is used as the objective function, and the parameters are trained using the backpropagation algorithm to complete the model convergence. For the second classifier, a sub-sample set is constructed separately for each first-level category. Similarly, the network weights are trained and the parameters are optimized to ensure that the distribution of the softmax output can accurately reflect the fine-grained category differences.

[0053] In summary, the intelligent fault diagnosis method for the buried pipeline cathodic protection system according to the embodiments of the present application is clarified. By fusing the information of the test pile parameters changing over time (time features) with the position topology and spatial correlation (spatial association) between each test pile, it captures the hidden data patterns in multiple dimensions to effectively improve the discrimination ability of the differences between similar representation fault types. This spatio-temporal collaborative analysis method provides a solid foundation for realizing intelligent fault diagnosis with high confidence, high accuracy, and high robustness, and is an important technical path to promote the intelligent operation and maintenance management of the buried pipeline cathodic protection system.

[0054] The embodiments of the present disclosure have been described above. The above description is exemplary and not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art in the technical field without departing from the scope and spirit of the described embodiments. The selection of the terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technologies in the market, or to enable other ordinary technical personnel in the technical field to understand the embodiments disclosed herein.

Claims

1. An intelligent fault diagnosis method for the cathodic protection system of buried pipelines, characterized in that, Including: Real-time collect the PSP data of each test pile along the pipeline; Organize the PSP data of each test pile according to the test pile sample dimension and time dimension to obtain a set of test pile PSP time series; Perform test pile node feature time series encoding on each test pile PSP time series in the set of test pile PSP time series to obtain a set of test pile PSP time series feature encoding vectors; Construct a test pile spatial distribution topology matrix between each test pile; Input the set of test pile PSP time series feature encoding vectors and the test pile spatial distribution topology matrix into a test pile spatio-temporal feature depth extractor based on a spatio-temporal graph neural network to obtain a test pile PSP spatio-temporal joint state representation matrix; Based on the test pile PSP spatio-temporal joint state representation matrix, determine the fault type and its confidence level.

2. The intelligent fault diagnosis method for the cathodic protection system of buried pipelines according to claim 1, wherein Perform test pile node feature time series encoding on each test pile PSP time series in the set of test pile PSP time series to obtain a set of test pile PSP time series feature encoding vectors, including: Perform initial test pile node feature time series encoding on each test pile PSP time series in the set of test pile PSP time series to obtain a test pile PSP initial time series feature encoding vector; Perform time series fine-grained feature enhancement on the test pile PSP initial time series feature encoding vector to obtain a test pile PSP time series feature encoding vector.

3. The intelligent fault diagnosis method for the cathodic protection system of buried pipelines according to claim 2, wherein Perform initial test pile node feature time series encoding on each test pile PSP time series in the set of test pile PSP time series to obtain a test pile PSP initial time series feature encoding vector, including: Perform initial test pile node feature time series encoding based on a forward LSTM model on each test pile PSP time series to obtain a test pile PSP initial time series feature encoding vector.

4. The intelligent fault diagnosis method for the cathodic protection system of buried pipelines according to claim 3, wherein, Perform time series fine-grained feature enhancement on the test pile PSP initial time series feature encoding vector to obtain a test pile PSP time series feature encoding vector, including: Perform time series feature decomposition based on one-dimensional convolutional encoding on the test pile PSP initial time series feature encoding vector to obtain a set of test pile PSP initial local hidden time series feature vectors; Based on the manifold structure correlation relationship between any two test pile PSP initial local hidden time series feature vectors in the set of test pile PSP initial local hidden time series feature vectors, perform test pile PSP time series feature enhancement on the set of test pile PSP initial local hidden time series feature vectors to obtain a test pile PSP time series feature encoding vector.

5. The intelligent fault diagnosis method for the cathodic protection system of buried pipelines according to claim 4, characterized in that, Based on the manifold structure correlation relationship between any two test pile PSP initial local hidden time series feature vectors in the set of test pile PSP initial local hidden time series feature vectors, perform test pile PSP time series feature enhancement on the set of test pile PSP initial local hidden time series feature vectors to obtain a test pile PSP time series feature encoding vector, including: Based on the manifold structure correlation relationship between any two test pile PSP initial local hidden time series feature vectors in the set of test pile PSP initial local hidden time series feature vectors, calculate a test pile PSP manifold structure fine-grained correlation mask topology matrix; Based on the fine-grained association mask topology matrix of the test stub PSP manifold structure, perform temporal feature dense feedback distillation on each test stub PSP initial local implicit temporal feature vector in the set of test stub PSP initial local implicit temporal feature vectors to obtain a set of test stub PSP distilled initial local implicit temporal feature vectors; Perform feature reconstruction based on the self-attention mechanism on the set of test stub PSP distilled initial local implicit temporal feature vectors to obtain test stub PSP temporal feature encoding vectors.

6. The intelligent fault diagnosis method for the cathodic protection system of buried pipelines according to claim 5, characterized in that, Based on the manifold structure association relationship between any two test stub PSP initial local implicit temporal feature vectors in the set of test stub PSP initial local implicit temporal feature vectors, calculate the test stub PSP manifold structure fine-grained association mask topology matrix, including: Calculate the manifold structure correlation coefficients between any two test stub PSP initial local implicit temporal feature vectors in the set of test stub PSP initial local implicit temporal feature vectors to obtain a test stub PSP manifold structure association topology matrix composed of multiple manifold structure correlation coefficients; Input the test stub PSP manifold structure association topology matrix into the gated mask function to obtain the initial test stub PSP manifold structure fine-grained association mask topology matrix; Perform test stub PSP manifold structure geometric association optimization based on gradient field tuning on each eigenvalue in the initial test stub PSP manifold structure fine-grained association mask topology matrix to obtain the test stub PSP manifold structure fine-grained association mask topology matrix.

7. The intelligent fault diagnosis method for the cathodic protection system of buried pipelines according to claim 6, characterized in that, The values at each non-diagonal position in the test stub spatial distribution topology matrix are the spatial distances between the corresponding two test stubs.

8. The intelligent fault diagnosis method for the cathodic protection system of buried pipelines according to claim 7, characterized in that Based on the test stub PSP spatio-temporal joint state representation matrix, determine the fault type and its confidence level, including: Input the test stub PSP spatio-temporal joint state representation matrix into the fault type recognizer based on the first classifier to obtain the test stub PSP fault type; Based on the test stub PSP fault type, extract the second classifier for identifying the specific test stub PSP fault type from the classification model library; Input the test stub PSP spatio-temporal joint state representation matrix into the second classifier to obtain the confidence level of the specific test stub PSP fault type.

Images