Leak detector data intelligent analysis system and method

Abstract

The present application discloses a leak detector data intelligent analysis system and method, which uses a deep learning-based neural network model to extract the time series pattern features of the vibration signal of the first monitoring position to dig out the implicit time series features of the vibration signal, and based on this, preliminarily identify potential leakage anomalies. In response to the abnormal signal, the system further obtains the vibration signal of the adjacent monitoring point, and similarly extracts the time series pattern coding features of the vibration signal of the adjacent monitoring point, and then calculates the spatiotemporal correlation of the vibration pattern between the two. Then, by comparing the correlation with the preset threshold, multi-dimensional verification of the leakage anomaly event is achieved. In this way, the limitations of traditional reliance on single-point static threshold detection are broken through, the influence of environmental noise and non-leakage transient interference is effectively reduced, the recognition accuracy of leakage signals is improved, and the reliability of operation and maintenance decisions is enhanced.

Description

Technical Field

[0001] The present application relates to the field of intelligent analysis, and more specifically, to a leak detector data intelligent analysis system and method. Background Art

[0002] In the field of pipeline leakage monitoring, traditional leak detection technology primarily relies on single-point vibration signal analysis and threshold alarm mechanisms, but this has significant limitations in practical applications under complex working conditions. Existing systems typically use fixed-band filtering methods, which makes it difficult to dynamically adapt to signal spectral characteristics. This results in the frequency domain overlap between interference sources such as mechanical vibration and water flow noise and tiny leakage signals, making it difficult to effectively separate them. Especially in scenarios with strong ambient noise (such as close-range construction or equipment startup and shutdown), the weak energy of leakage signals is easily overwhelmed by the noise. Conventional energy threshold detection algorithms, lacking in-depth analysis of signal timing patterns, often lead to missed detections or false alarms. Furthermore, transient waveforms generated by non-leakage events (such as rapid valve actuation or heavy load shock) have similar burst characteristics to leakage signals, making it difficult for existing technologies, relying on static models at a single monitoring point, to accurately distinguish them.

[0003] Therefore, an optimized intelligent analysis method for leak detector data is expected. Summary of the Invention

[0004] In order to solve the above-mentioned technical problems, the present application is proposed. The embodiment of the present application provides a leak detector data intelligent analysis system and method, which uses a neural network model based on deep learning to extract the time series pattern features of the vibration signal of the first monitoring position to mine the time series implicit features of the vibration signal, and based on this, preliminarily identify potential leakage anomalies. In response to the abnormal signal, the system further obtains the vibration signal of the adjacent monitoring point, and similarly extracts the time series pattern coding features of the vibration signal of the adjacent monitoring point, and then calculates the spatiotemporal correlation of the vibration patterns between the two, and then, by comparing the correlation with the preset threshold, realizes multi-dimensional verification of leakage anomaly events. In this way, the limitations of traditional reliance on single-point static threshold detection are broken through, the influence of environmental noise and non-leakage transient interference is effectively reduced, the recognition accuracy of leakage signals is improved, and the reliability of operation and maintenance decisions is enhanced.

[0005] According to one aspect of the present application, a method for intelligently analyzing leak detector data is provided, comprising:

[0006] It uses a deep learning-based neural network model to extract the time series pattern features of the vibration signal at the first monitoring position to dig out the implicit time series features of the vibration signal, and based on this, preliminarily identify potential leakage anomalies. In response to the abnormal signal, the system further obtains the vibration signal of the adjacent monitoring point. Similarly, it extracts the time series pattern coding features of the vibration signal of the adjacent monitoring point, and then calculates the spatiotemporal correlation of the vibration patterns between the two. Then, by comparing the correlation with the preset threshold, it realizes multi-dimensional verification of the leakage anomaly event. In this way, it breaks through the limitations of traditional reliance on single-point static threshold detection, effectively reduces the impact of environmental noise and non-leakage transient interference, improves the recognition accuracy of leakage signals, and thus enhances the reliability of operation and maintenance decisions.

[0007] According to another aspect of the present application, a leak detector data intelligent analysis system is provided, comprising:

[0008] A first leakage vibration monitoring signal acquisition module is used to acquire a first leakage vibration monitoring signal collected by a leak detector deployed at a first monitoring position of the pipeline;

[0009] a filtering processing module, configured to apply a bandpass filter to filter the first leakage vibration monitoring signal to obtain a filtered first leakage vibration monitoring signal;

[0010] A first vibration time series pattern feature extraction module is used to extract the vibration time series pattern feature of the filtered first leakage vibration monitoring signal to obtain a first vibration time series pattern feature encoding vector;

[0011] an anomaly detection module, configured to input the first vibration time series pattern feature encoding vector into an anomaly detection model to obtain an anomaly detection result;

[0012] a second leakage vibration monitoring signal acquisition module, configured to acquire, in response to the abnormality detection result being a potential abnormality, a second leakage vibration monitoring signal collected by a leak detector deployed at a second monitoring position of the pipeline, the first monitoring position being adjacent to the second monitoring position;

[0013] A second vibration time series pattern feature extraction module is used to extract the vibration time series pattern feature of the second leakage vibration monitoring signal to obtain a second vibration time series pattern feature encoding vector;

[0014] a vibration pattern spatiotemporal correlation coding module, configured to calculate a spatiotemporal correlation degree of the vibration pattern between the second vibration time sequence pattern feature coding vector and the first vibration time sequence pattern feature coding vector;

[0015] The leakage result analysis module is used to determine whether to adjust the abnormal detection result to a high-confidence leakage event based on the comparison between the spatiotemporal correlation of the vibration pattern and a preset threshold.

[0016] Compared with the prior art, the present application provides a leak detector data intelligent analysis system and method, which uses a deep learning-based neural network model to extract the time series pattern features of the vibration signal at the first monitoring position to mine the implicit time series features of the vibration signal, and based on this, preliminarily identify potential leakage anomalies. In response to the abnormal signal, the system further obtains the vibration signal of the adjacent monitoring point, and similarly extracts the time series pattern coding features of the vibration signal of the adjacent monitoring point, and then calculates the spatiotemporal correlation of the vibration patterns between the two. Then, by comparing the correlation with the preset threshold, multi-dimensional verification of the leakage anomaly event is achieved. In this way, the limitations of traditional reliance on single-point static threshold detection are broken through, the influence of environmental noise and non-leakage transient interference is effectively reduced, the recognition accuracy of leakage signals is improved, and the reliability of operation and maintenance decisions is enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0018] Figure 1 Flowchart of a method for intelligently analyzing leak detector data according to an embodiment of the present application;

[0019] Figure 2 Schematic diagram of data flow of the leak detector data intelligent analysis method according to an embodiment of the present application;

[0020] Figure 3 Flowchart of sub-step S3 of the leak detector data intelligent analysis method according to an embodiment of the present application;

[0021] Figure 4 4 is a block diagram of a leak detector data intelligent analysis system according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0023] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0024] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0025] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0026] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0027] In the technical solution of the present application, a method for intelligent analysis of leak detector data is proposed. Figure 1 Flowchart of the method for intelligent analysis of leak detector data according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the intelligent analysis method of leak detector data according to an embodiment of the present application. Figure 1 and Figure 2As shown, the leak detector data intelligent analysis method according to the embodiment of the present application includes the following steps: S1, obtaining a first leakage vibration monitoring signal collected by a leak detector deployed at a first monitoring position of a pipeline; S2, applying a bandpass filter to filter the first leakage vibration monitoring signal to obtain a filtered first leakage vibration monitoring signal; S3, extracting the vibration time series pattern characteristics of the filtered first leakage vibration monitoring signal to obtain a first vibration time series pattern feature coding vector; S4, inputting the first vibration time series pattern feature coding vector into an anomaly detection model to obtain an anomaly detection result; S5, in response to the anomaly detection result being a potential anomaly, obtaining a second leakage vibration monitoring signal collected by a leak detector deployed at a second monitoring position of the pipeline, the first monitoring position being adjacent to the second monitoring position; S6, extracting the vibration time series pattern characteristics of the second leakage vibration monitoring signal to obtain a second vibration time series pattern feature coding vector; S7, calculating the vibration pattern spatiotemporal correlation between the second vibration time series pattern feature coding vector and the first vibration time series pattern feature coding vector; S8, determining whether to adjust the anomaly detection result to a high-confidence leakage event based on a comparison between the vibration pattern spatiotemporal correlation and a preset threshold.

[0028] Specifically, S1 acquires a first leakage vibration monitoring signal collected by a leak detector deployed at a first monitoring location in the pipeline. It should be understood that leakage vibration monitoring signals reflect the characteristics of weak mechanical vibrations caused by leakage within the pipeline. These signals contain direct physical evidence of a leak and are crucial for determining pipeline status and locating leaks. It is worth noting that during the acquisition of the first leakage vibration monitoring signal, mechanical vibration signals of a specific frequency band are generated by the interaction between fluid pressure waves and the pipe wall during a pipeline leak. These weak vibration waves propagate along the pipe wall at a specific velocity, forming physical characteristics with temporal and spatial correlations. Leak detectors, using highly sensitive sensors (such as accelerometers or piezoelectric vibration sensors) mounted directly on the outer surface of the pipeline, collect the tiny vibration signals generated by the pipeline and convert the physical vibration energy into electrical signals. These electrical signals undergo preliminary amplification and conversion to form the raw data collected by the leak detector, reflecting the vibrational manifestation of a leak or abnormal event on the pipeline structure. The leak detector's high-frequency response and anti-interference capabilities enable it to capture subtle vibration waveforms caused by leaks, effectively supporting subsequent intelligent filtering and pattern recognition algorithms. By deploying a leak detector at the first monitoring location, the system can obtain real-time vibration information of the pipeline at this key node, providing a reliable data basis for subsequent time series feature extraction and anomaly identification, and ensuring the overall accuracy and stable operation of the intelligent analysis system.

[0029] In particular, the S2 applies a bandpass filter to filter the first leakage vibration monitoring signal to obtain a filtered first leakage vibration monitoring signal. It should be understood that in a complex pipeline leakage monitoring environment, the collected first leakage vibration monitoring signal often contains a large amount of noise and interference signals in non-target frequency bands, such as low-frequency components generated by mechanical vibration, water turbulence noise and other environmental noise. These interference signals not only mask the weak leakage vibration characteristics, but also greatly affect the accuracy of subsequent feature extraction and anomaly detection. In order to improve the effectiveness of the signal, it is necessary to filter the original vibration signal to remove the interference components of irrelevant frequency bands, thereby highlighting the leakage signal characteristics. Therefore, in the technical solution of the present application, a bandpass filter is applied to process the first leakage vibration monitoring signal, that is, by setting the upper and lower cutoff frequency ranges, only the signal components within the frequency band are retained, and the low-frequency interference below the cutoff frequency and the high-frequency noise above the cutoff frequency are effectively filtered out.

[0030] It is worth mentioning that as an electronic filter, the core working principle of the bandpass filter is to combine the functions of the low-pass filter and the high-pass filter, while allowing the vibration signal in the center frequency band to pass through, while blocking other frequency bands, thereby achieving dynamic focusing of the spectrum of the first leakage vibration monitoring signal. By accurately selecting the common frequency band range of the leakage vibration signal through the bandpass filter, it not only protects the integrity of the effective leakage signal, but also significantly reduces the confusion caused by mechanical vibration and environmental noise. Among them, the filtering process usually includes signal sampling, the application of digital filtering algorithms and the output of the filtered signal, so that the subsequent vibration time series pattern feature extraction can be based on purer and more representative signal data for analysis. In this way, the signal-to-noise ratio of the first leakage vibration monitoring signal is improved, and the leakage feature representation of the first leakage vibration monitoring signal is significantly enhanced, thereby improving the detection model's recognition accuracy for leakage anomalies and enhancing the reliability and practicality of the entire leakage detection system under complex working conditions.

[0031] In particular, the step S3 extracts the vibration time series pattern characteristics of the filtered first leakage vibration monitoring signal to obtain a first vibration time series pattern characteristic encoding vector. In a specific example of the present application, Figure 3 As shown, the S3 includes: S31, extracting the vibration local timing pattern characteristics of the first leakage vibration monitoring signal after filtering to obtain a time series of the first vibration local timing pattern feature coding vector; S32, performing vibration pattern timing context coding on the time series of the first vibration local timing pattern feature coding vector to obtain a first vibration timing pattern feature coding vector.

[0032] Specifically, the S31 extracts the vibration local temporal pattern features of the filtered first leakage vibration monitoring signal to obtain the time sequence of the first vibration local temporal pattern feature encoding vector. In the technical solution of the present application, the filtered first leakage vibration monitoring signal is subjected to a vibration local temporal pattern feature extraction based on one-dimensional convolution coding to obtain the time sequence of the first vibration local temporal pattern feature encoding vector. It should be understood that although the filtered first leakage vibration monitoring signal has effectively filtered out many irrelevant noises, the signal itself still contains a large number of complex and weak temporal features, which reflect the dynamic change pattern of the leakage event in the time dimension. In order to deeply explore the key leakage information in the signal, the system adopts a vibration local temporal pattern feature extraction method based on one-dimensional convolution coding to convert the continuous vibration signal into a feature representation with greater discriminative power and expressiveness. Among them, the one-dimensional convolution coding performs local perception and feature response on the input vibration time series signal by designing multiple convolution kernels to slide along the time axis to capture the local patterns of the vibration signal in different time windows, such as amplitude changes, frequency components and burst signal characteristics.

[0033] Specifically, in the process of extracting the vibration local time series pattern features based on one-dimensional convolution coding of the first leakage vibration monitoring signal after filtering, the filtered signal is first subjected to a one-dimensional convolution operation to extract the vibration local related features in the time series, and then the vibration local time series signal is compressed and abstracted after processing by a nonlinear activation function and a pooling layer to obtain the time series of the first vibration local time series pattern feature coding vector. In the technical solution of the present application, the powerful local feature extraction capability of the convolutional neural network is utilized to effectively capture the subtle changes in the leakage signal in the time dimension and suppress the interference caused by environmental noise or non-leakage events, thereby forming a high-dimensional and information-rich feature expression, providing accurate and expressive input for the subsequent structured recursive coding and anomaly detection model. Through one-dimensional convolution coding, the system can realize the transformation from the original vibration monitoring signal to high-level features, significantly improve the recognition ability of weak leakage features and the discrimination effect of abnormal events, further enhance the sensitivity and reliability of the leak detection system under complex working conditions, and promote the precision and practicality of intelligent pipeline status monitoring.

[0034] Specifically, the S32 performs vibration pattern time sequence context encoding on the time series of the first vibration local time sequence pattern feature encoding vector to obtain the first vibration time sequence pattern feature encoding vector. It should be understood that the real leakage vibration signal will form an attenuated wavefront with spatiotemporal continuity along the pipe wall, and its waveform has dynamic correlations such as phase delay and energy diffusion between adjacent time windows, while interference signals such as local mechanical shocks often appear as isolated transient waveforms. Although traditional one-dimensional convolutional coding can capture local features within a short time window, it cannot model the dynamic evolution law across time steps that is unique to leakage signals. For example, the vibration wave caused by leakage will show quasi-periodic fluctuations due to changes in fluid pressure during propagation, and its spectral energy distribution will show an exponential decay trend over time. This contextual dependency across time steps requires more sophisticated time series modeling capabilities. Therefore, in order to construct a feature representation system with spatiotemporal collaborative perception capabilities, so as to realize the joint modeling of the vibration local temporal dependency and spatial correlation hidden in the time series of the first vibration local temporal pattern feature coding vector, in the technical solution of the present application, the time series of the first vibration local temporal pattern feature coding vector is subjected to vibration pattern temporal context encoding to obtain the first vibration temporal pattern feature coding vector.

[0035] That is, by decoupling time series encoding from structured message passing, the deep spatiotemporal patterns of vibration signals are jointly modeled. Here, a recurrent neural network (RNN) is used to perform preliminary temporal encoding on the input vibration local temporal pattern feature encoding vector time series, fully leveraging its inherent temporal memory capacity to capture the dynamic characteristics of the signal over time, thereby generating the initial encoding vector for sequence transmission. This step enhances the model's understanding of signal continuity and contextual relationships in the temporal dimension, laying the foundation for subsequent multi-dimensional information fusion. Furthermore, by introducing a spatiotemporal confidence constraint factor based on an attention mechanism, it adaptively focuses on time intervals with significant leakage signal characteristics (such as persistent low-frequency vibration segments caused by leakage), while suppressing the interference of transient interference events (such as isolated pulses caused by valve action) on model judgment. At the same time, the introduction of a spatial confidence constraint factor integrates knowledge of the pipeline network's topological structure into the model, quantifying the information propagation potential between adjacent monitoring nodes and laying the foundation for subsequent multi-node collaborative verification. In particular, the spatiotemporal synergy constraint factor achieves nonlinear fusion of temporal and spatial dimensions through a constant curvature space representation. This not only avoids the representation bias caused by single-dimensional weight allocation, but also ensures the geometric rationality of the fusion space through a curvature compensation mechanism, making the evolution trajectory of the vibration mode in the spatiotemporal joint space more closely resemble the propagation characteristics of the real leakage signal. The resulting first vibration temporal pattern feature encoding vector combines local temporal details with global contextual associations, enabling the model to accurately distinguish subtle differences in the time and frequency domains between real leakage events and interference sources such as environmental noise and equipment startup and shutdown. In this way, the system can fully explore and integrate the complex relationships between pipeline leakage vibration signals in the dynamic temporal development and multi-node spatial structure, enabling the system to carefully distinguish leakage signals from non-leakage transient interference in complex noisy environments, improving the confidence and robustness of anomaly identification, thereby effectively reducing the false alarm rate and the risk of missed alarms, significantly enhancing the safety and maintenance efficiency of pipeline network operations, and promoting the development of intelligent leak detection technology to a higher level.

[0036] Specifically, first, the time series of the first vibration local temporal pattern feature encoding vectors is input into a sequence encoder based on a recurrent neural network to obtain a time series of the first vibration sequence transmission initial encoding vectors. It should be understood that while one-dimensional convolutional coding can extract local vibration features within a short time window, the limited receptive field of its convolution kernel makes it difficult to capture the persistent temporal evolution patterns unique to leakage signals (such as the attenuation characteristics of pressure waves transmitted along the pipe wall or the gradual low-frequency oscillation process caused by the expansion of the leakage hole). In the technical solution of the present application, the time series of the first vibration local temporal pattern feature encoding vectors is input into a sequence encoder based on a recurrent neural network to obtain a time series of the first vibration sequence transmission initial encoding vectors. Among them, the recurrent neural network, with its internal feedback mechanism, can break through the limitations of the fixed sliding window and continuously accumulate the hidden state information of historical time steps during the transmission process, thereby establishing an association chain between short-term local features and long-term temporal context, so as to reweave discrete local feature fragments into a dynamic representation sequence with time continuity, so that the self-similar vibration mode unique to the leakage signal in the continuous leakage stage (such as quasi-periodic pressure fluctuations), the transient oscillation caused by sudden interference events and the temporal differences between system noise can be explicitly encoded.

[0037] Specifically, mechanical vibration noise may exhibit short, repetitive, impact-like waveforms, while vibrations caused by actual leaks typically form a low-frequency vibration envelope with phase continuity due to the continuous leakage of fluid. RNNs gradually construct the dynamic evolution trajectory of vibration signals by iteratively processing the time series of feature encoding vectors. During this process, the initial encoding vector transmitted by the first vibration sequence output at each time step not only contains the local temporal characteristics of that moment but also implicitly incorporates the accumulated vibration state information of the previous time step. For example, when a small leak occurs in a pipeline, its vibration signal initially appears as low-energy, broadband noise. As the leakage accumulates, the energy in specific frequency bands gradually increases. By transferring information between time steps, RNNs can capture this gradual change in energy distribution, forming a sequence of encoding vectors that represents the continuous change in state, thereby laying a high-quality temporal feature foundation for downstream spatiotemporal collaborative analysis. In this way, the anomaly detection model's sensitivity and ability to distinguish leakage vibration patterns are significantly improved, reducing the false alarm and missed alarm rates, ensuring that the leak detection system can still stably and accurately monitor pipeline leakage events under noise superposition and variable working conditions, thereby enhancing pipeline network operation safety and maintenance efficiency.

[0038] In a specific example of the present application, the time sequence of the first vibration local temporal pattern feature encoding vector is input into a sequence encoder based on a recurrent neural network using the following formula to obtain the time sequence of the first vibration sequence transmission initial encoding vector; wherein, the formula is:

[0039] in, is the time series of the first vibration local temporal pattern feature encoding vector, are the first, second, and third in the time series of the first vibration local temporal pattern feature encoding vector. and The first vibration local temporal pattern feature encoding vector, represents a recurrent neural network, The first, second, and third time series of the initial coding vector are respectively transmitted for the first vibration sequence. and The first vibration sequence delivers the initial code vector.

[0040] Next, the first vibration timing confidence constraint factor of each first vibration sequence transmitting the initial coding vector in the time series of the first vibration sequence transmitting the initial coding vector is calculated. It should be understood that the vibration caused by a real leak usually has a continuous phase coherent feature (such as the attenuation law when the pressure wave propagates along the pipe wall), while the vibration generated by transient events such as rapid valve movement often presents isolated pulse characteristics. However, in the hidden state transmission process of a conventional RNN, the coding vectors of all time steps are assigned the same importance weight, which makes it difficult for the model to suppress information interference during the burst noise period. Therefore, in the technical solution of the present application, the first vibration timing confidence constraint factor of each first vibration sequence transmitting the initial coding vector in the time series of the first vibration sequence transmitting the initial coding vector is calculated.

[0041] Here, by calculating the first vibration time series confidence constraint factor, the system can adaptively focus on continuous vibration pattern periods with high information value (such as the quasi-periodic fluctuation segments unique to leakage signals), while weakening the influence of the encoding vectors corresponding to isolated pulses or random noise. For example, in the early stages of a small pipeline leak, its vibration signal may intermittently be submerged in the equipment operation noise. The time series confidence constraint factor can enhance the information contribution of weak but continuous leakage feature periods through dynamic weight adjustment, and suppress the interference of random noise-dominated time steps on model decision-making. This dynamic focus shifting capability enables the model to maintain sensitivity to real leakage signals in the face of non-stationary interference, significantly reducing the risk of misjudgment caused by data contamination in local time periods, and providing a high signal-to-noise ratio time series feature foundation for subsequent spatiotemporal collaborative analysis.

[0042] In a specific example of the present application, the first vibration timing confidence constraint factor of each first vibration sequence transmitting the initial coding vector in the time series of the first vibration sequence transmitting the initial coding vector is calculated using the following formula; wherein, the formula is:

[0043]

[0044] in, express function, and denote the trainable weighted hyperparameters, and is the weight matrix, represents vector multiplication, is the weight vector of the key nodes of the first vibration local timing pattern, is the time queue of the first vibration timing confidence metric factor, express function, is the first vibration timing confidence constraint factor.

[0045] Then, the first vibration spatial confidence constraint factor of each first vibration sequence transmitting the initial coding vector in the time series of the first vibration sequence transmitting the initial coding vector is calculated. It should be understood that the traditional single-point detection model cannot distinguish whether the vibration signal is caused by local noise interference or the propagation of a remote real leakage event because it ignores the spatial topological relationship of the pipeline layout. Therefore, in the technical solution of the present application, the first vibration spatial confidence constraint factor of each first vibration sequence transmitting the initial coding vector in the time series of the first vibration sequence transmitting the initial coding vector is calculated. That is, by modeling the structured spatial properties of the pipeline network, the geographical weight and topological influence of different monitoring nodes in information transmission are quantified. For example, in a pipeline network with densely arranged sensors, the monitoring point at the elbow of the main pipeline carries more fluid pressure wave reflection signals, and its vibration sequence transmitting initial coding vector is more likely to contain key leakage features than the straight segment node. The calculation of the spatial confidence constraint factor is essentially an explicit expression of these implicit spatial structural knowledge, so that the model can automatically identify key node data with high information transmission potential.

[0046] Specifically, by extracting the spatial structure of the characteristic distribution of the initial coding vector transmitted by the first vibration sequence (such as the connection weights formed by modeling the physical distance between sensor nodes and the direction of the pipeline through a graph neural network), the system can dynamically evaluate the structural centrality of each node in the message propagation network. For example, when the vibration wave propagates along the pipeline, the node with higher centrality is more likely to capture the attenuation propagation path characteristics of the leakage signal, and the spatial confidence constraint factor of the corresponding coding vector will be assigned a higher weight. This dynamic weight allocation mechanism effectively strengthens the vibration mode representation capability that conforms to the physical propagation model of the pipeline network, while suppressing the interference of abnormal fluctuations in the coding vector caused by local random vibration (such as transient impact of construction machinery) on the model decision.

[0047] It is worth mentioning that because the spatial propagation of real leakage signals follows specific acoustic attenuation laws (such as exponential energy decay with increasing distance), characteristic correlations with consistent directions will be generated between adjacent monitoring nodes. The spatial confidence constraint factor can guide the model to focus on vibration feature combinations that conform to the physical laws of spatial propagation by quantifying the strength of structural correlations between nodes (such as the propagation loss coefficient calculated based on the distance between measuring points and the pipe material). At the same time, for isolated vibration signals generated by non-leakage events (such as instantaneous pressure shocks near a single sensor), due to the lack of spatial propagation consistency across nodes, the corresponding spatial confidence constraint factor will automatically reduce the propagation weight of the relevant coding vector, fundamentally avoiding the problem of system-level false alarms caused by false triggering of a single sensor. This dynamic modulation mechanism of spatial confidence enables the system to achieve robust suppression of local noise interference and precise focusing on real leakage events in complex pipe network scenarios.

[0048] In a specific example of the present application, the first vibration space confidence constraint factor of each first vibration sequence transmitting initial coding vector in the time series of the first vibration sequence transmitting initial coding vector is calculated using the following formula; wherein, the formula is:

[0049]

[0050] in, represents exponential operation, represents the square of the norm, express The corresponding first vibration space confidence score value, is the first vibration space confidence constraint factor.

[0051] Then, the first vibration spatial confidence constraint factor and the first vibration temporal confidence constraint factor are fused to obtain the initial first vibration mode transmission spatiotemporal synergy constraint factor. It should be understood that while real leak events and interference signals often exhibit similar characteristics in a single dimension (time or space), they exhibit essential differences in the joint spatiotemporal domain. For example, while transient vibrations generated by rapid valve actuation exhibit bursty characteristics in the temporal dimension (high temporal confidence), their energy propagation is constrained by the pipeline structure and is difficult to diffuse to adjacent nodes (low spatial confidence). In contrast, the vibration waves caused by a real leak not only exhibit a persistent temporal evolution pattern (high temporal confidence) but also exhibit attenuation characteristics along the pipeline topology consistent with acoustic propagation models (high spatial confidence). Relying solely on the independent effects of temporal or spatial confidence constraints can lead the model into a single-dimensional misjudgment trap. Focusing solely on the temporal dimension may misjudge localized transient interference as leaks, while relying solely on the spatial dimension can easily overlook early, weak leak signals at isolated nodes. Therefore, in order to construct the synergistic mechanism of spatiotemporal dual-domain confidence, the first vibration spatial confidence constraint factor and the first vibration temporal confidence constraint factor are fused to obtain the initial first vibration mode transmission spatiotemporal synergistic constraint factor.

[0052] That is, by nonlinearly fusing the first vibration temporal confidence constraint factor (characterizing the persistence of the vibration mode along the time axis) with the spatial confidence constraint factor (reflecting the information dissemination value of the monitoring node in the pipeline network topology), the system can generate a joint weight factor with physical interpretability. For example, a gated fusion mechanism is used to dynamically adjust the contribution ratio of temporal and spatial constraints: when the monitoring node is located at a key pipeline hub, the spatial confidence factor dominates the fusion process, strengthening vibration characteristics that conform to the laws of pipeline network propagation; while in the initial weak signal stage of a leak, the weight of the temporal confidence factor is increased, focusing on persistent low-energy vibration modes. This dynamic collaborative mechanism effectively solves the problem of feature matching misalignment caused by the separation of temporal analysis and spatial analysis in traditional methods, enabling the model to capture the implicit coupling correlation of vibration signals in both temporal and spatial domains.

[0053] In a specific example of the present application, the first vibration spatial confidence constraint factor and the first vibration temporal confidence constraint factor are fused using the following formula to obtain the initial first vibration mode transmission spatiotemporal synergy constraint factor; wherein the formula is:

[0054]

[0055] in, express function, and is the fusion weight parameter, A spatiotemporal cooperation constraint factor is imparted to the initial first vibration mode.

[0056] Furthermore, the initial first vibration mode transmission spatiotemporal coordination constraint factor is optimized by compensating for spatiotemporal curvature to obtain the first vibration mode transmission spatiotemporal coordination constraint factor. In particular, it should be understood that the vibration waves generated by a real leak event, when propagating along the pipe wall, are affected by factors such as fluid pressure and pipe damping, resulting in a nonlinear attenuation pattern in their spatiotemporal characteristics. Interference signals such as mechanical shock often have a localized spatiotemporal distribution. Traditional methods that directly linearly fuse temporal confidence constraints with spatial confidence constraints can produce non-Euclidean curvature in the fused spatiotemporal characteristics due to the independent modeling of the temporal dimension (e.g., the quasi-periodicity of the leakage signal) and the spatial dimension (e.g., the matching of sensor spacing and wave velocity). For example, when a section of the pipeline has an uneven weld, the propagation velocity of the vibration wave will exhibit local variations. If this spatiotemporal curvature is not compensated for, the fused feature vector may misrepresent the leak direction or intensity, resulting in failure of collaborative verification of adjacent sensors. Therefore, in a preferred example of the present application, the initial first vibration mode transfer spatiotemporal cooperation constraint factor is optimized by spatiotemporal curvature compensation to obtain the first vibration mode transfer spatiotemporal cooperation constraint factor.

[0057] Specifically, a constant curvature space representation and a spherical coordinate approximation mapping are constructed, projecting the initial spatiotemporal synergy constraint factors onto a manifold space consistent with the pipeline acoustic propagation model. During this process, a curvature compensation weight parameter is introduced to dynamically adjust the curvature contribution ratios of the time and space dimensions, so that the fused synergy constraint factors approximate the characteristics of a Euclidean plane space. For example, when vibration signals consistent with the pipeline attenuation model are detected at adjacent monitoring nodes, spatiotemporal curvature compensation reduces the curvature weight of the time dimension, strengthening the geometric constraints of the spatial propagation path. This mapping of such signals into high-confidence straight trajectories in the fused space is significantly distinguished from the distorted trajectories of localized noise. After curvature compensation, the system can identify the temporal synchronization and spatial propagation inconsistencies of such events through the reconstructed plane space, correcting their characteristic trajectories to isolated scattered points. Conversely, although the low-frequency vibrations generated by real, small leaks may have low energy, their cross-node phase delay characteristics are enhanced into coherent propagation paths in the compensated flat space. This geometric correction mechanism enables the model to break through the limitations of traditional threshold algorithms, accurately capture the spatiotemporal evolution pattern of leakage signals that conform to physical laws under a strong noise background, significantly reduce the false alarm rate caused by geometric representation distortion, and improve the reliability of multi-node collaborative verification.

[0058] In this preferred example, the initial first vibration mode transmission spatiotemporal synergy constraint factor is optimized by performing spatiotemporal curvature compensation to obtain the first vibration mode transmission spatiotemporal synergy constraint factor according to the following formula; wherein, the formula is:

[0059]

[0060] in, is a constant curvature space representation, is the spherical coordinate approximation, A spatiotemporal cooperation constraint factor is delivered to the first vibration mode.

[0061] Specifically, the initial encoding vector is first transferred based on each first vibration sequence Corresponding first vibration timing confidence constraint factor and the first vibration space confidence constraint factor , to construct a constant curvature space representation and spherical coordinates approximation :

[0062]

[0063] Then, the initial first vibration mode after fusion is transferred to the spatiotemporal cooperative constraint factor As the fusion space metric benchmark, the weight parameter in the fusion formula is determined by calculating the following formula, for example and :

[0064] That is, when When , the single-mode correlation representation in time and space dimensions can be made and The space-time coupling attraction that approaches flatness reflects the Euclidean property in the fusion space, that is, the plane maintenance in the fusion space is achieved by compensating the generation of single-mode negative curvature, thereby improving the space-time synergy constraint factor of the initial first vibration mode transmission. fusion expression effect.

[0065] Furthermore, based on the spatiotemporal synergistic constraint factor of the first vibration mode transmission, the first vibration sequence message transmission structure modulation is performed on each initial code vector of the first vibration sequence transmission to obtain a time sequence of the first vibration sequence transmission structural modulation code vector. Considering that the leakage vibration wave will form a spatiotemporal continuous feature with directionality and attenuation law when propagating along the pipeline axial direction, while local interference such as mechanical shock is manifested as a spatially isolated and temporally transient waveform. For example, when there is periodic construction vibration near the pipeline, although its energy may cover the frequency band of the leakage signal, the propagation delay of its vibration wave between adjacent sensors does not conform to the fluid mechanics model, and there is a lack of quasi-periodic attenuation pattern in the time dimension. The traditional method directly treats the features of all time windows with equal weights, resulting in excessive amplification of the noise features. Therefore, in order to achieve dynamic purification and enhancement of the feature space, in the technical solution of the present application, based on the spatiotemporal synergistic constraint factor of the first vibration mode transmission, the first vibration sequence message transmission structure modulation is performed on each initial code vector of the first vibration sequence transmission to obtain a time sequence of the first vibration sequence transmission structural modulation code vector.

[0066] Here, by element-by-element multiplication of the spatiotemporal coordination constraint factor for the first vibration mode transmission with the initial encoding vector for the first vibration sequence transmission, dual-domain spatiotemporal coordination control of information flow is achieved. For example, vibration signals captured by monitoring nodes at pipeline bends may exhibit high-frequency oscillations due to fluid impact. However, the spatiotemporal coordination constraint factor, optimized through curvature compensation, can identify the short-term burstiness (low temporal confidence) and locality (low spatial confidence) of such signals, suppressing invalid information transmission by reducing their weight. Conversely, for low-frequency vibration signals that conform to the pipeline sound velocity propagation model, their high spatiotemporal coordination weight drives the enhanced propagation of feature codes along the pipeline network topology, forming a cross-node phase-delay correlation chain. This modulation mechanism essentially constructs a vibration feature propagation pathway based on physical laws, spatially mapping the information transmission path to the diffusion trajectory of the actual leakage signal. This organic propagation mechanism enables the system to deeply purify and directionally enhance the value of vibration feature information while preserving the details of the original vibration signal.

[0067] Subsequently, the position-wise sum of the time series of the structural modulation coding vector transmitted by the first vibration sequence is calculated to obtain the first vibration time series pattern feature coding vector. Considering that when the structural modulation coding vector modulated by the spatiotemporal synergy constraint factor still maintains the time series form, the leakage features contained therein may be scattered in different time steps, making it difficult to be accurately captured by the subsequent anomaly detection model. Therefore, in order to break through the information dilution defect of the traditional sliding window averaging method, the high-order features of the vibration signal after optimization in the spatiotemporal dual domain are directionally concentrated. In the technical solution of the present application, the position-wise sum of the time series of the structural modulation coding vector transmitted by the first vibration sequence is calculated to obtain the first vibration time series pattern feature coding vector. In this way, the complexity of subsequent model processing is reduced, while maintaining and strengthening the expression of the overall characteristics and important information of the vibration signal on the time axis, effectively avoiding redundancy and noise interference, and improving the stability and robustness of the feature through the integration of the time dimension. In addition, the position-wise summation operation can alleviate the influence of short-term anomalies or occasional noise in the time series data, ensuring that the vibration time series coding feature finally generated has strong generalization ability while representing the overall vibration pattern. This provides an efficient, compact and information-rich input for the anomaly detection model that integrates information from multiple time points and covers dynamic evolution and structural correlations, greatly improving the model's accuracy and confidence in identifying leakage anomaly events.

[0068] In a specific example of the present application, the positional sum of the time series of the structural modulation coding vector transmitted by the first vibration sequence is calculated using the following formula to obtain the first vibration time sequence pattern feature coding vector; wherein, the formula is:

[0069] in, the scale of the time sequence of the structural modulation code vectors conveyed by the first vibration sequence, The first vibration timing pattern feature encoding vector.

[0070] In particular, in S4, the first vibration time-series pattern feature encoding vector is input into an anomaly detection model to obtain an anomaly detection result. In the technical solution of the present application, the first vibration time-series pattern feature encoding vector is input into a classifier-based anomaly detection model to obtain an anomaly detection result, which is used to indicate whether the first leakage vibration monitoring signal is abnormal. It should be understood that after multi-level signal processing and feature extraction, the first vibration time-series pattern feature encoding vector obtained represents deep, comprehensive spatiotemporal dynamic information in the pipeline vibration signal. However, this high-dimensional feature vector alone is not sufficient to directly determine whether a pipeline leak anomaly exists. Therefore, in the technical solution of the present application, the first vibration time-series pattern feature encoding vector is input into an anomaly detection model to obtain an anomaly detection result. The anomaly detection model refers to an intelligent prediction system based on machine learning or deep learning technology, which can learn the distribution pattern of signal features under normal vibration conditions and then identify abnormal vibration behavior that deviates from the normal pattern. The model classifies or scores the input vibration time-series feature encoding vector and outputs an anomaly detection result for the discovered abnormal pattern, such as whether there is a potential leak and the confidence level of the anomaly. In a specific example, a classifier-based anomaly detection model can be used to learn the boundary between normal and abnormal vibration samples to achieve classification and identification of abnormal vibration signals.

[0071] Specifically, by inputting the first vibration time-series pattern feature encoding vector into a classifier-based anomaly detection model, the system can map complex multi-source spatiotemporal features into clear anomaly discrimination results. This not only distinguishes leakage signals from non-leakage events such as environmental noise and equipment operation interference, but also assesses the confidence and severity of anomalies, greatly improving the accuracy and practicality of the monitoring system's response. This enables intelligent and automated anomaly identification of vibration characteristics, providing a solid basis for rapid early warning of pipeline leakage incidents.

[0072] In particular, the S5, in response to the abnormality detection result being a potential abnormality, obtains a second leakage vibration monitoring signal collected by a leak detector deployed at a second monitoring position of the pipeline, and the first monitoring position is adjacent to the second monitoring position. It should be understood that relying on a single-point vibration signal for abnormality judgment is often interfered with by environmental noise and non-leakage events, resulting in frequent false alarms and missed reports, especially in complex working conditions or strong noise environments. In the technical solution of the present application, a response mechanism is designed to address this limitation, that is, when the abnormality detection model judges the first vibration time series pattern feature coding vector collected at the first monitoring position as a potential abnormality, the system immediately obtains the leakage vibration monitoring signal of the second monitoring point deployed at an adjacent position. That is, with the help of the second leakage vibration monitoring signal collected by the adjacent leak detector, multi-point spatiotemporal correlation analysis is performed, thereby improving the accuracy and credibility of the abnormality judgment.

[0073] Specifically, acquiring vibration monitoring signals from a leak detector deployed at a second, adjacent monitoring location provides the system with independent yet correlated leakage information from different spatial locations. Because vibration signals generated by leaks have certain propagation characteristics and spatial continuity, true leaks typically produce correlated vibration patterns at adjacent monitoring points simultaneously or at similar times. In contrast, localized disturbances or transient non-leakage events likely affect only a single point and are unlikely to produce consistent anomaly signals across multiple adjacent monitoring locations. Therefore, by incorporating vibration data from a second monitoring point, the system can effectively distinguish between the two points based on the spatiotemporal correlation of their vibration patterns, thus enabling collaborative verification. This approach not only overcomes the shortcomings of single-point anomaly detection models, eliminating the need for isolated localized signal determination, but also significantly reduces the risk of false positives and missed positives through a multi-node cross-validation mechanism, resulting in more accurate identification of anomalies and higher-confidence leak alarm outputs. This not only improves the stability and reliability of the leak detection system in complex industrial field environments, but also provides more reliable decision support for operators and maintenance personnel, significantly optimizing the intelligence level of field monitoring and the cost-effectiveness of operation and maintenance.

[0074] In particular, the S6 extracts the vibration time sequence pattern characteristics of the second leakage vibration monitoring signal to obtain the second vibration time sequence pattern characteristic coding vector. That is, consistent with the processing method of the first leakage vibration monitoring signal, in the technical solution of the present application, first, a bandpass filter is applied to filter the second leakage vibration monitoring signal to obtain the filtered second leakage vibration monitoring signal; then, the filtered second leakage vibration monitoring signal is subjected to vibration local time sequence pattern feature extraction based on one-dimensional convolution coding to obtain the time series of the second vibration local time sequence pattern characteristic coding vector; and then, the time series of the second vibration local time sequence pattern characteristic coding vector is subjected to vibration pattern time sequence context coding to obtain the second vibration time sequence pattern characteristic coding vector. In this way, the original, possibly noisy second leakage vibration monitoring signal can be converted into a coding vector with important features that characterize the leakage event, ensuring that the multi-point signal is expressed in a unified and efficient representation space, facilitating subsequent comparison with the signal characteristics of the first monitoring position and spatiotemporal correlation analysis, and laying an accurate and reliable data foundation for spatiotemporal correlation calculation. It is worth mentioning that the generation of the second vibration time-series pattern feature encoding vector not only improves the system's ability to capture signal details, but also achieves feature consistency and comparability across monitoring points, providing accurate and stable input for the subsequent calculation of the spatiotemporal correlation of vibration patterns. Ultimately, it helps the system significantly improve the recognition rate of real leakage events and the confidence level of alarms, reduce the false alarm rate, and provide solid technical support for pipeline safety in complex environments.

[0075] In particular, the S7 calculates the spatiotemporal correlation of the vibration modes between the second vibration time sequence pattern characteristic coding vector and the first vibration time sequence pattern characteristic coding vector. It should be understood that in a complex pipeline monitoring environment, the vibration signal collected by the single-point sensor may be affected by a variety of interference factors, such as mechanical vibration, water flow noise, and transient waveforms generated by non-leakage events. These interferences often have a high degree of overlap with the real leakage signal in the frequency domain and time domain. It is difficult to effectively distinguish between true and false anomalies by relying solely on anomaly detection at a single monitoring position. This has led to traditional leak detection methods being prone to missed detection or false alarms, seriously affecting the reliability and practicality of the system. In response to this difficult problem, in the technical solution of the present application, the spatiotemporal correlation of the vibration modes between the second vibration time sequence pattern characteristic coding vector and the first vibration time sequence pattern characteristic coding vector is calculated.

[0076] That is, by comparing the vibration signal feature codes collected at adjacent monitoring points, the similarity and synchronization between the two point signals in the time series pattern are revealed, and the fact that the vibration signal caused by the leakage event has spatial propagation continuity is utilized to improve the confidence of the abnormality judgment. Among them, compared with the potential abnormal samples, the signal caused by the real leakage often shows highly correlated waveform characteristics and time series structure between adjacent nodes, while local environmental interference or random noise is difficult to appear synchronously at multiple spatial locations. Therefore, by calculating and quantifying the spatiotemporal correlation between the second vibration time series pattern feature coding vector and the first vibration time series pattern feature coding vector, the system can effectively filter out isolated anomalies and false signals and achieve accurate identification of real leakage events. In a specific example of the present application, the spatiotemporal correlation of the vibration pattern can be obtained by performing a spatiotemporal correlation difference calculation on the second vibration time series pattern feature coding vector and the first vibration time series pattern feature coding vector. In this way, the intrinsic connection between the signals of the first monitoring position and the second monitoring position is established, and the deep mining and intelligent identification of weak leakage signals in complex environments are achieved, laying a technical foundation for the practicality and reliability of the leak detector data intelligent analysis system.

[0077] In particular, the S8 determines whether to adjust the abnormality detection result to a high-confidence leakage event based on the comparison between the spatiotemporal correlation of the vibration pattern and a preset threshold. In a specific example of the present application, when the spatiotemporal correlation exceeds the threshold, the system will automatically adjust the "potential abnormality" result originally judged based on single-point data to a "high-confidence leakage event", that is, confirming that the abnormal signal is continuous and consistent in space, which greatly eliminates the possibility of single-point sporadic noise or interference; on the contrary, if the correlation does not meet the preset standard, the cautious judgment of the abnormality is maintained to prevent false alarms. This threshold comparison and result adjustment mechanism based on spatiotemporal correlation effectively ensures the rapid and accurate response of the leak detector data intelligent analysis system to pipeline leakage events, improves the accuracy of abnormality detection and the credibility of alarms, thereby solving the misjudgment problem caused by the lack of multi-node spatiotemporal correlation verification in traditional leak detection systems.

[0078] In summary, the method for intelligent analysis of leak detector data according to the embodiment of the present application is explained, which uses a neural network model based on deep learning to extract the time series pattern features of the vibration signal of the first monitoring position to dig out the implicit time series features of the vibration signal, and based on this, preliminarily identify potential leakage anomalies. In response to the abnormal signal, the system further obtains the vibration signal of the adjacent monitoring point, and similarly extracts the time series pattern coding features of the vibration signal of the adjacent monitoring point, and then calculates the spatiotemporal correlation of the vibration patterns between the two, and then, by comparing the correlation with the preset threshold, realizes multi-dimensional verification of the leakage anomaly event. In this way, the limitations of traditional reliance on single-point static threshold detection are broken through, the influence of environmental noise and non-leakage transient interference is effectively reduced, the recognition accuracy of leakage signals is improved, and the reliability of operation and maintenance decisions is enhanced.

[0079] Furthermore, a leak detector data intelligent analysis system is also provided.

[0080] Figure 4 FIG. 1 is a block diagram of a leak detector data intelligent analysis system according to an embodiment of the present application. Figure 4As shown, the leak detector data intelligent analysis system 300 according to the embodiment of the present application includes: a first leakage vibration monitoring signal acquisition module 310, which is used to acquire a first leakage vibration monitoring signal collected by a leak detector deployed at a first monitoring position of a pipeline; a filtering processing module 320, which is used to apply a bandpass filter to filter the first leakage vibration monitoring signal to obtain a filtered first leakage vibration monitoring signal; a first vibration time series pattern feature extraction module 330, which is used to extract the vibration time series pattern feature of the filtered first leakage vibration monitoring signal to obtain a first vibration time series pattern feature coding vector; an anomaly detection module 340, which is used to input the first vibration time series pattern feature coding vector into an anomaly detection model to obtain an anomaly detection result; a second leakage vibration monitoring signal acquisition module 310, which is used to acquire a first leakage vibration monitoring signal collected by a leak detector deployed at a first monitoring position of a pipeline; a filtering processing module 320, which is used to apply a bandpass filter to filter the first leakage vibration monitoring signal to obtain a filtered first leakage vibration monitoring signal; a first vibration time series pattern feature extraction module 330, which is used to extract the vibration time series pattern feature of the filtered first leakage vibration monitoring signal to obtain a first vibration time series pattern feature coding vector; a first vibration time series pattern feature coding vector An acquisition module 350 is used to obtain, in response to the abnormality detection result being a potential abnormality, a second leakage vibration monitoring signal collected by a leak detector deployed at a second monitoring position of the pipeline, wherein the first monitoring position is adjacent to the second monitoring position; a second vibration time series pattern feature extraction module 360 ​​is used to extract the vibration time series pattern features of the second leakage vibration monitoring signal to obtain a second vibration time series pattern feature coding vector; a vibration pattern spatiotemporal correlation coding module 370 is used to calculate the vibration pattern spatiotemporal correlation between the second vibration time series pattern feature coding vector and the first vibration time series pattern feature coding vector; a leakage result analysis module 380 is used to determine whether to adjust the abnormality detection result to a high-confidence leakage event based on a comparison between the vibration pattern spatiotemporal correlation and a preset threshold.

[0081] As described above, the leak detector data intelligent analysis system 300 according to the embodiments of the present application can be implemented in various wireless terminals, such as a server equipped with an intelligent leak detector data analysis algorithm. In one possible implementation, the leak detector data intelligent analysis system 300 according to the embodiments of the present application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the leak detector data intelligent analysis system 300 can be a software module within the wireless terminal's operating system, or an application developed specifically for the wireless terminal. Of course, the leak detector data intelligent analysis system 300 can also be one of the wireless terminal's many hardware modules.

[0082] Alternatively, in another example, the leak detector data intelligent analysis system 300 and the wireless terminal may be separate devices, and the leak detector data intelligent analysis system 300 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0083] While various embodiments of the present disclosure have been described above, the above descriptions are illustrative, non-exhaustive, and not intended to be limiting of the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to existing technologies, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A leak detector data intelligent analysis method, characterized in that: include: Acquiring a first leakage vibration monitoring signal collected by a leak detector deployed at a first monitoring position of the pipeline; Applying a bandpass filter to filter the first leakage vibration monitoring signal to obtain a filtered first leakage vibration monitoring signal; Extracting the vibration time series pattern feature of the filtered first leakage vibration monitoring signal to obtain a first vibration time series pattern feature encoding vector; Inputting the first vibration time series pattern feature encoding vector into an anomaly detection model to obtain an anomaly detection result; In response to the abnormality detection result being a potential abnormality, obtaining a second leakage vibration monitoring signal collected by a leak detector deployed at a second monitoring position of the pipeline, the first monitoring position being adjacent to the second monitoring position; extracting a vibration time series pattern feature of the second leakage vibration monitoring signal to obtain a second vibration time series pattern feature encoding vector; Calculating the spatiotemporal correlation of the vibration patterns between the second vibration time series pattern feature coding vector and the first vibration time series pattern feature coding vector; Based on a comparison between the spatiotemporal correlation of the vibration pattern and a preset threshold, it is determined whether to adjust the abnormality detection result to a high-confidence leakage event.

2. The leak detector data intelligent analysis method according to claim 1, characterized in that: Extracting the vibration time series pattern feature of the filtered first leakage vibration monitoring signal to obtain a first vibration time series pattern feature encoding vector includes: Extracting the vibration local time series pattern features of the filtered first leakage vibration monitoring signal to obtain a time series of the first vibration local time series pattern feature encoding vector; Vibration pattern temporal context coding is performed on the time series of the first vibration local temporal pattern feature coding vector to obtain a first vibration temporal pattern feature coding vector.

3. The method for intelligent analysis of leak detector data according to claim 2, characterized in that: Extracting the vibration local time series pattern feature of the filtered first leakage vibration monitoring signal to obtain a time series of the first vibration local time series pattern feature encoding vector includes: The filtered first leakage vibration monitoring signal is subjected to vibration local time series pattern feature extraction based on one-dimensional convolution coding to obtain a time series of first vibration local time series pattern feature coding vectors.

4. The method for intelligent analysis of leak detector data according to claim 2, characterized in that: Performing vibration pattern temporal context coding on the time series of the first vibration local temporal pattern feature coding vector to obtain the first vibration temporal pattern feature coding vector includes: Calculating the first vibration mode spatiotemporal cooperation confidence constraint factor of the time series of the first vibration local temporal pattern feature encoding vector to construct the first vibration mode transmission spatiotemporal cooperation constraint factor; Based on the first vibration mode transfer spatiotemporal synergy constraint factor, a first vibration sequence transfer aggregation analysis is performed on the time series of the first vibration local temporal sequence pattern feature coding vector to obtain the first vibration temporal sequence pattern feature coding vector.

5. The method for intelligent analysis of leak detector data according to claim 4, characterized in that: Calculating the first vibration mode spatiotemporal cooperation confidence constraint factor of the time series of the first vibration local temporal pattern feature encoding vector to construct the first vibration mode transmission spatiotemporal cooperation constraint factor, including: Inputting the time sequence of the first vibration local temporal pattern feature encoding vector into a sequence encoder based on a recurrent neural network to obtain the time sequence of the first vibration sequence transmission initial encoding vector; Calculating a first vibration timing confidence constraint factor of each first vibration sequence transmitting the initial coding vector in the time series of the first vibration sequence transmitting the initial coding vector; Calculating a first vibration space confidence constraint factor of each first vibration sequence transmitting an initial coding vector in a time series of the first vibration sequence transmitting an initial coding vector; Based on the first vibration space confidence constraint factor and the first vibration time sequence confidence constraint factor of each first vibration sequence transmitting initial coding vector, the first vibration mode transmission time and space coordination constraint factor of each first vibration sequence transmitting initial coding vector is constructed.

6. The leak detector data intelligent analysis method according to claim 5, characterized in that: Based on the first vibration spatial confidence constraint factor and the first vibration temporal confidence constraint factor of each first vibration sequence transmitting the initial coding vector, a first vibration mode transmission spatiotemporal coordination constraint factor of each first vibration sequence transmitting the initial coding vector is constructed, including: The first vibration spatial confidence constraint factor and the first vibration temporal confidence constraint factor are integrated to obtain an initial first vibration mode transmission spatiotemporal synergy constraint factor; The initial first vibration mode transmission spatiotemporal cooperation constraint factor is optimized by performing spatiotemporal curvature compensation to obtain the first vibration mode transmission spatiotemporal cooperation constraint factor.

7. The method for intelligent analysis of leak detector data according to claim 6, characterized in that: Based on the first vibration mode transfer spatiotemporal synergy constraint factor, performing first vibration sequence transfer aggregation analysis on the time series of the first vibration local temporal pattern feature coding vector to obtain the first vibration temporal pattern feature coding vector, including: Based on the first vibration mode transmission spatiotemporal synergy constraint factor, performing first vibration sequence message transmission structural modulation on each first vibration sequence transmission initial code vector to obtain a time sequence of first vibration sequence transmission structural modulation code vectors; The position-wise sum of the time series of the structural modulation coding vector transferred by the first vibration sequence is calculated to obtain the first vibration time sequence pattern feature coding vector.

8. The method for intelligent analysis of leak detector data according to claim 7, characterized in that: Inputting the first vibration time series pattern feature encoding vector into the anomaly detection model to obtain an anomaly detection result includes: The first vibration time series pattern feature encoding vector is input into a classifier-based anomaly detection model to obtain an anomaly detection result, and the anomaly detection result is used to indicate whether the first leakage vibration monitoring signal has an anomaly.

9. The method for intelligent analysis of leak detector data according to claim 8, characterized in that: Calculating the spatiotemporal correlation of the vibration pattern between the second vibration time sequence pattern feature coding vector and the first vibration time sequence pattern feature coding vector, comprising: A vibration pattern spatiotemporal correlation difference calculation is performed on the second vibration time sequence pattern feature coding vector and the first vibration time sequence pattern feature coding vector to obtain a vibration pattern spatiotemporal correlation degree.

10. A leak detector data intelligent analysis system, characterized in that: include: A first leakage vibration monitoring signal acquisition module is used to acquire a first leakage vibration monitoring signal collected by a leak detector deployed at a first monitoring position of the pipeline; a filtering processing module, configured to apply a bandpass filter to filter the first leakage vibration monitoring signal to obtain a filtered first leakage vibration monitoring signal; A first vibration time series pattern feature extraction module is used to extract the vibration time series pattern feature of the filtered first leakage vibration monitoring signal to obtain a first vibration time series pattern feature encoding vector; an anomaly detection module, configured to input the first vibration time series pattern feature encoding vector into an anomaly detection model to obtain an anomaly detection result; a second leakage vibration monitoring signal acquisition module, configured to, in response to the abnormality detection result being a potential abnormality, acquire a second leakage vibration monitoring signal collected by a leak detector deployed at a second monitoring position of the pipeline, the first monitoring position being adjacent to the second monitoring position; A second vibration time series pattern feature extraction module is used to extract the vibration time series pattern feature of the second leakage vibration monitoring signal to obtain a second vibration time series pattern feature encoding vector; a vibration pattern spatiotemporal correlation coding module, configured to calculate a spatiotemporal correlation degree of the vibration pattern between the second vibration time sequence pattern feature coding vector and the first vibration time sequence pattern feature coding vector; The leakage result analysis module is used to determine whether to adjust the abnormal detection result to a high-confidence leakage event based on the comparison between the spatiotemporal correlation of the vibration pattern and a preset threshold.

Images