Pipeline leakage point detection method, device and equipment fusing thermal response-seismic waves

Abstract

The invention provides a thermal response-seismic wave fused pipeline leakage point detection method, device and equipment, and relates to the technical field of underground pipeline leakage detection. The method comprises the following steps: acquiring seismic wave signals and time-space thermal response data obtained by detecting an underground pipeline; inputting the space-time thermal response data into a space-time convolutional neural network for space-time feature extraction to obtain a thermal field change mode of the underground pipeline; inputting the thermal field change mode into a physical information neural network for inversion to obtain first leakage source information of the underground pipeline; wherein the physical information neural network takes a heat conduction equation as a hard constraint; performing inversion on the seismic wave signal to obtain second leakage source information of the underground pipeline; and determining a leakage point detection result of the underground pipeline based on the first leakage source information and the second leakage source information. According to the invention, omnibearing accurate sensing of the deep buried leakage point from the spatial position to the leakage state can be realized, and the detection depth and the positioning precision are improved.

Description

Technical Field

[0001] This invention relates to the field of underground pipeline leak detection technology, and in particular to a pipeline leak detection method, device and equipment that integrates thermal response and seismic wave. Background Technology

[0002] Accurately detecting leaks in deeply buried pipelines is a long-standing technical challenge in the safe operation and maintenance of urban underground pipe networks.

[0003] Currently, mainstream detection methods each have significant limitations. Take acoustic vibration methods as an example; they rely on high-frequency sound signals generated by leaks for detection. However, high-frequency components attenuate rapidly in soil, resulting in an effective detection depth typically less than two meters, rendering them largely ineffective for pipelines buried at depths exceeding three meters. Furthermore, this method has a short signal propagation distance in non-metallic pipelines (such as HDPE pipes) and struggles to handle scenarios with multiple leak points or complex branch networks, limiting its location capabilities. Another commonly used technique, thermal imaging, identifies leaks by capturing surface temperature anomalies. While somewhat intuitive, it is essentially a static, passive observation method. This method is greatly affected by complex weather conditions such as ambient temperature changes, sunlight, and rainfall. The lack of a quantitative physical correlation between the detected surface thermal anomalies and the underground leak source leads to a high false alarm rate. Moreover, it cannot determine the specific depth and intensity of the leak, providing only a vague anomaly indication. Ground-penetrating radar technology has a drastically reduced penetration capability in strata with high groundwater levels or clay, while distributed fiber optic sensing technology, although highly accurate, requires pre-installation and is costly, making it unsuitable for rapid, large-scale inspection of existing pipeline networks.

[0004] Therefore, existing technologies generally suffer from problems such as insufficient detection depth for deeply buried pipelines, limited positioning accuracy, weak environmental resistance, and inability to quantitatively diagnose leakage conditions, making it difficult to meet the growing demand of modern cities for precise operation and maintenance and safety management of underground pipelines. Summary of the Invention

[0005] This invention provides a method, apparatus, and equipment for detecting pipeline leaks by integrating thermal response and seismic waves, in order to solve the problems of insufficient detection depth, limited positioning accuracy, and weak environmental interference resistance for deeply buried pipelines.

[0006] In a first aspect, embodiments of the present invention provide a pipeline leak detection method integrating thermal response and seismic waves, comprising: Acquire seismic wave signals and spatiotemporal thermal response data obtained from underground pipeline detection; Spatiotemporal thermal response data are input into a spatiotemporal convolutional neural network for spatiotemporal feature extraction to obtain the thermal field change pattern of underground pipelines. The thermal field change pattern is input into the physical information neural network for inversion to obtain the first leakage source information of the underground pipeline; the physical information neural network uses the heat conduction equation as a hard constraint. By inverting the seismic wave signal, information about the second leakage source of the underground pipeline was obtained; Based on the information from the first and second leakage sources, the leak detection results of the underground pipeline were determined.

[0007] In one possible implementation, the spatiotemporal thermal response data is a sequence of time-series infrared thermograms, and the spatiotemporal convolutional neural network includes an input layer, a spatiotemporal convolutional layer, and an output layer. The spatiotemporal thermal response data is input into the spatiotemporal convolutional neural network for spatiotemporal feature extraction to obtain the thermal field change pattern of the underground pipeline, including: The time-series infrared thermal image sequence is input into a spatiotemporal convolutional neural network. The data dimensionality is reduced through the input layer, and the temperature change features in the spatial and temporal dimensions are extracted through the spatiotemporal convolutional layer. The spatiotemporal feature vector and / or classification labels are generated through the output layer to form the thermal field change pattern of the underground pipeline. The classification labels include leakage hot spots, solar interference, rainfall interference, and no anomalies.

[0008] In one possible implementation, acquiring spatiotemporal thermal response data obtained from underground pipeline detection includes: Multi-time-series infrared scanning of the target area was performed under different ambient temperature backgrounds to obtain time-series thermal images; A modulated infrared radiation source is introduced to apply a thermal excitation signal with a specific temporal and spatial pattern to the ground to enhance the thermal anomaly response caused by the leak point.

[0009] In one possible implementation, the thermal field change pattern is input into a physical information neural network for inversion to obtain information about the first leakage source of the underground pipeline, including: The thermal field change pattern and the spatiotemporal sampling points of the area where the underground pipeline is located are input into the physical information neural network for inversion to obtain the first leakage source information; wherein, the first leakage source information includes the horizontal position coordinates of the leakage point, the burial depth and the leakage intensity.

[0010] In one possible implementation, before inputting the thermal field change pattern into a physical information neural network for inversion to obtain the first leakage source information of the underground pipeline, the following steps are included: Construct a training dataset containing multiple sets of thermal field change patterns and their corresponding leakage source information; Based on the training dataset and the loss function, the initial physical information neural network is trained to obtain the trained physical information neural network. The loss function includes a data fitting term and a physical constraint term, which is used to force the output of the physical information neural network to satisfy the heat conduction equation.

[0011] In one possible implementation, the leak detection results of the underground pipeline are determined based on the first leak source information and the second leak source information, including: If the spatial location indicated by the first leakage source information and the second leakage source information is consistent within the preset tolerance range, then the spatial location is confirmed as the actual leakage point.

[0012] In one possible implementation, the seismic wave signal is inverted to obtain information about the second leakage source of the underground pipeline, including: Constructing an underground 3D scanning grid; For each grid point, the theoretical time delay of the seismic wave signal is calculated using a beamforming algorithm, and the signal is aligned. A three-dimensional spatial energy distribution map is generated by calculating the energy value corresponding to each grid point through weighted superposition. The location corresponding to the energy peak is extracted as the second leakage source information.

[0013] Secondly, embodiments of the present invention provide a pipe leak detection device that integrates thermal response and seismic wave analysis, comprising: The acquisition module is used to acquire seismic wave signals and spatiotemporal thermal response data obtained from the detection of underground pipelines; The feature extraction module is used to input spatiotemporal thermal response data into a spatiotemporal convolutional neural network to extract spatiotemporal features and obtain the thermal field change pattern of the underground pipeline. The first inversion module is used to input the thermal field change pattern into the physical information neural network for inversion to obtain the first leakage source information of the underground pipeline; wherein, the physical information neural network uses the heat conduction equation as a hard constraint; The second inversion module is used to invert the seismic wave signal to obtain information on the second leakage source of the underground pipeline; The information fusion module is used to determine the leak detection results of underground pipelines based on the first leak source information and the second leak source information.

[0014] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation thereof.

[0015] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect or any possible implementation thereof.

[0016] The pipeline leak detection method, device, and equipment fused with thermal response and seismic waves provided in this invention integrate low-frequency acoustic wave-seismic wave joint imaging with dynamic heat conduction inversion. Leveraging the high penetration of low-frequency acoustic waves, it solves the problem of weak and difficult-to-capture signals from deeply buried pipelines. Simultaneously, combining the spatial positioning capability of seismic wave arrays with the quantitative diagnostic capability of thermal inversion, it achieves comprehensive and accurate perception of deeply buried leaks from spatial location to leakage status, improving detection depth and positioning accuracy. Utilizing spatiotemporal convolutional neural networks and physical information neural networks for intelligent inversion of spatiotemporal thermal response data, it not only automatically extracts subtle leak features that are difficult for the human eye to distinguish but also embeds the physical laws of heat conduction as hard constraints into the inversion process, ensuring the physical rationality of the results and reducing the false alarm rate. The entire system, through the fusion and cross-validation of multimodal data, ultimately outputs a reliable and quantitative diagnostic report, providing strong technical support for the accurate location and status assessment of leaks in deeply buried pipelines, effectively overcoming the shortcomings of traditional single methods in terms of detection depth, anti-interference capability, and information dimension. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the implementation of a pipeline leak detection method that integrates thermal response and seismic waves, according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the implementation of a pipeline leak detection method that integrates thermal response and seismic waves, provided in another embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of a pipe leak detection device that integrates thermal response and seismic waves according to an embodiment of the present invention; Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] Detecting leaks in deeply buried pipelines is a significant challenge in urban infrastructure maintenance. For a long time, the field of underground pipeline leak detection has relied primarily on acoustic vibration methods, thermal imaging technology, and other auxiliary techniques. However, with the increasing number of deeply buried pipelines (≥3 meters) in urban development, the limitations of these traditional technologies have become increasingly apparent, making it difficult to meet the demands for accurate and efficient detection.

[0019] Acoustic vibration methods, represented by correlation leak detectors, rely on the 300-2500Hz sound wave signals generated by leaks for detection. However, this technology has significant limitations: firstly, the high-frequency components of the sound waves attenuate rapidly in soil, limiting the detection depth to within 2 meters, rendering it ineffective for deeply buried pipelines; secondly, traditional correlation location methods can only identify single leak points, failing when encountering multiple leak points or complex branch networks. Furthermore, in plastic pipes (such as HDPE pipes), the sound wave propagation distance is less than 50 meters, further restricting its applicability.

[0020] Thermal imaging technology uses static infrared scanning to capture surface temperature anomalies to determine leaks, but it is greatly affected by environmental factors. Changes in ambient temperature and complex weather conditions can interfere with the detection results, making it impossible to determine the specific depth and intensity of the leak source, and resulting in a false alarm rate of over 30%. This causes many problems for actual detection work and makes it difficult to guarantee the accuracy and reliability of the detection.

[0021] Ground-penetrating radar (GPR) has a certain detection capability in conventional strata, but its penetration depth drops sharply when it enters clay or high-water-bearing strata, making it unable to effectively detect deeply buried pipelines. Distributed optical fiber technology (DTS / DAS) has advantages in monitoring accuracy, but it requires pre-installation and has high overall costs, making it difficult to widely apply in large-scale pipeline inspection.

[0022] To address the aforementioned issues, this solution innovatively combines low-frequency acoustic-seismic wave joint imaging technology with dynamic heat conduction inversion analysis, forming a precise diagnostic system that integrates multi-physics field coupling and active and passive detection. This solution aims to overcome the shortcomings of traditional methods in terms of detection depth, positioning accuracy, and anti-interference capabilities, enabling precise location and condition assessment of leaks in deeply buried pipelines. Furthermore, the penetrating advantage of low-frequency acoustic waves in deeply buried pipelines and the positioning capability of high-precision seismic wave arrays, along with the paradigm shift in infrared thermal imaging technology, achieve direct and quantitative inversion diagnosis of pipeline leaks and defects, significantly improving the system's reliability, anti-interference capability, and information output dimensions.

[0023] The application scenarios of this technical solution include: For leak detection in ultra-deep buried thermal pipelines, it can accurately locate leak points up to 10 meters deep, achieving sub-meter accuracy, far exceeding traditional methods. Its anti-interference capability is greatly enhanced, effectively suppressing environmental noise interference through dynamic scanning and active excitation mechanisms. It provides rich information output dimensions, not only accurately locating leak points but also quantitatively assessing leakage losses, such as leakage intensity. For non-metallic pipelines that are difficult to identify with seismic waves, dynamic thermal inversion can effectively detect changes in soil thermal properties caused by leaks. It is suitable for periodic automated inspections of important pipelines, enabling predictive maintenance. It is important to note that in engineering detection, the deployment of infrared thermal imager arrays and active excitation sources needs to be planned to cover the target area.

[0024] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0025] See Figure 1 The diagram illustrates the implementation flowchart of the pipeline leak detection method integrating thermal response and seismic waves provided in this embodiment of the invention, which is described in detail below: Step 101: Obtain seismic wave signals and spatiotemporal thermal response data obtained from the detection of underground pipelines.

[0026] In this embodiment, as Figure 2 As shown, the first step is to synchronously acquire two types of heterogeneous physical field data through multi-sensor collaborative deployment. The hardware foundation for this detection includes an acoustic-seismic wave detection unit, an infrared thermal imaging upgrade unit, and a central processing and control unit. The acoustic-seismic wave detection unit consists of a high-power hydraulic servo exciter (0.1Hz - 100Hz) and a high-precision seismic detector array (at least 4 nodes), capable of sensing nanoscale micro-vibrations and elastic waves induced by fluid injection or continuous scouring at leak points in deeply buried pipelines. The central processing and control unit includes a multi-channel synchronous data acquisition system and built-in advanced algorithm software (including beamforming, ST-CNN, PINN, etc.). The infrared thermal imaging upgrade unit includes a high frame rate, high thermal sensitivity (NETD < 20mK) infrared thermal imager array and a modulated infrared radiation source (active thermal excitation device), which performs multi-time-series scanning of the target area under different ambient temperature backgrounds to form a sequence of images reflecting the spatiotemporal evolution of surface temperature.

[0027] The principle behind this is that underground pipeline leakage is a multiphysics coupling event: during the escape of the leaking medium (such as hot water, oil, or gas), it rubs and impacts the surrounding soil, generating vibrations that propagate primarily as low-frequency sound waves and seismic waves; simultaneously, the heat it carries gradually alters the temperature field of the surrounding soil through heat conduction, ultimately forming a dynamic thermal anomaly on the surface that can be detected by infrared equipment. Therefore, simultaneously collecting these two types of data provides a data foundation for subsequent joint inversion and cross-validation based on multiphysics coupling.

[0028] During implementation, a high-precision clock source from the central processing and control unit is used to unify the timestamps of all sensors, and the raw data is time-stamped and spatially registered to construct a unified data framework. This step overcomes the shortcomings of traditional single-method approaches that suffer from limited information dimensions, providing multi-dimensional information support for solving the problems of weak signals and strong environmental interference in deeply buried pipelines.

[0029] Step 102: Input the spatiotemporal thermal response data into a spatiotemporal convolutional neural network to extract spatiotemporal features and obtain the thermal field change pattern of the underground pipeline.

[0030] In this embodiment, the spatiotemporal thermal response data preprocessed in step 101 is input into a pre-trained spatiotemporal convolutional neural network, which can automatically extract subtle thermal field change patterns caused by leakage that are difficult for the human eye to discern. This network is specifically designed for processing spatiotemporal sequence data, and its core contains a spatiotemporal convolutional layer composed of multiple layers of three-dimensional convolutional kernels. It can simultaneously perform convolution operations on local spatial regions and adjacent time frames in a single operation, effectively capturing the spatial distribution and temporal evolution of the temperature field.

[0031] The network consists of an input layer, a spatiotemporal convolutional layer, a pooling layer, a batch normalization layer, a fully connected layer, and an output layer. The input layer receives tensor data of a specific dimension. Through the hierarchical abstraction of the spatiotemporal convolutional layer, it automatically learns and distinguishes between subtle hotspot patterns caused by leakage and environmental interference patterns. Environmental interference typically manifests as globally uniform changes or rapid abrupt changes, while leakage signals exhibit spatiotemporal characteristics of locality, persistence, and slow diffusion.

[0032] The output layer can generate two thermal field change patterns: one is a low-dimensional dense feature vector that encodes key spatiotemporal parameters such as the diffusion rate and shape changes of thermal anomalies; the other is classification labels, including "leaking hotspots," "solar interference," "rainfall interference," and "no anomalies." This step realizes the transformation from massive, noisy raw data to refined, robust, high-order features, providing denoised input for subsequent quantitative inversion and significantly reducing computational complexity.

[0033] Step 103: Input the thermal field change pattern into the physical information neural network for inversion to obtain the first leakage source information of the underground pipeline; wherein, the physical information neural network uses the heat conduction equation as a hard constraint.

[0034] In this embodiment, the thermal field change pattern extracted in step 102 is used as the core clue and input into the Physical Information Neural Network (PINN) for inversion calculation to obtain the first leakage source information. This network integrates data-driven learning with hard constraints of physical laws. The inversion principle constructs leakage detection as an optimization problem under physical constraints: based on the dynamic temperature field characteristics observed on the surface, leakage source parameters that conform to both the observed phenomena and the laws of heat conduction are inferred in reverse. Through inversion calculation, the location, intensity, and even approximate depth of the underground leakage point can be directly inferred from the dynamic temperature field observed on the surface, achieving a leap from "phenomenon observation" to "mechanism diagnosis".

[0035] The network input includes feature vectors of the thermal field change pattern, as well as spatial coordinates (x, y, z) and temporal coordinates (t) of the underground target area and the sampling within the observation period. The sampling points cover surface points matching the observation data and underground points that enforce physical laws. Through nonlinear mapping of the hidden layer, the output leakage source state parameters θ = [X, Y, Z, Q] are obtained, corresponding to the horizontal location, burial depth, and leakage intensity of the leakage point, respectively.

[0036] The network loss function includes a data loss term and a physical loss term: the data loss term measures the difference between the predicted temperature field and the observation model; the physical loss term calculates whether the output temperature field satisfies the Fourier heat conduction equation, and embeds this equation as a hard constraint into the loss calculation, evaluating and minimizing the residuals on a large number of random sampling points. The network parameters and leakage source parameters are optimized through backpropagation and gradient descent algorithms, eventually converging to the optimal solution. This achieves a quantitative and mechanistic inversion from surface thermal phenomena to underground leakage sources, outputting high-confidence first leakage source information.

[0037] Step 104: Invert the seismic wave signal to obtain information on the second leakage source of the underground pipeline.

[0038] In this embodiment, the seismic wave signals acquired in step 101 are independently inverted, and beamforming technology is used to spatially locate the leak point. First, based on the pipeline's designed burial depth and the detection area, a fine three-dimensional scanning grid is constructed in the possible underground space, with each grid point representing a hypothetical potential leak source. .

[0039] For each hypothetical point, based on the geophone array's geometric coordinates and a pre-established subsurface wave velocity model, the seismic wave velocity is calculated from... Theoretical time delay propagation to each detector The time-domain signal of each detector Time shift alignment is performed according to the corresponding theoretical time delay to obtain Then, the signals after all channels are aligned are weighted and superimposed to calculate the power (energy) of the superimposed signal.

[0040] Its core principle is spatial coherent superposition: the signal corresponding to the real leak source will be significantly amplified due to phase alignment, while environmental noise will cancel each other out or slightly amplify due to randomness. This process is repeated for all grid points to generate a three-dimensional spatial energy distribution map; the peak position formed by energy focusing is the most probable leak source location. The three-dimensional coordinates of the peak are extracted to obtain second leak source information based on independent seismic wave imaging, primarily providing the precise spatial coordinates of the leak point. This method utilizes the spatial diversity capability of the sensor array to synthesize distributed weak signals, effectively suppressing incoherent noise and achieving high-precision localization of deep, weak leak signals.

[0041] Step 105: Based on the information of the first and second leakage sources, determine the leak detection results of the underground pipeline.

[0042] In this embodiment, by verifying the consistency and complementing the evidence from multiple physics fields, the first leakage source information in step 103 and the second leakage source information in step 104 are combined to determine the final highly reliable leak detection result.

[0043] The basic fusion strategy is to determine the consistency of spatial location: if the horizontal coordinates of the leak point obtained by thermal inversion match the coordinates of the seismic wave imaging within the preset tolerance range, it indicates that the two detection methods with different physical mechanisms point to the same location, which greatly increases the confidence that the location is the real leak point, and the system can confirm it as the final leak point.

[0044] Furthermore, the output information from both sources can be integrated, combining the leakage intensity and burial depth obtained from thermal inversion with the energy peak value from seismic wave imaging, to comprehensively and quantitatively assess the severity and status of the leakage. If the results from both sources are inconsistent, the system will issue an early warning, prompting operators to verify or initiate an adaptive adjustment process, such as adjusting the excitation parameters and re-detecting.

[0045] This decision fusion process avoids the limitations or occasional errors of a single method, and outputs a detection and diagnostic report that combines accurate spatial location and quantitative condition assessment, providing a technical basis for precise pipeline maintenance and safety decisions.

[0046] This invention integrates low-frequency acoustic-seismic wave joint imaging with dynamic thermal conduction inversion. Leveraging the high penetration of low-frequency acoustic waves, it solves the problem of weak and difficult-to-capture signals from deeply buried pipelines. Simultaneously, combining the spatial positioning capabilities of seismic wave arrays with the quantitative diagnostic capabilities of thermal inversion, it achieves comprehensive and accurate perception of deep-buried leaks from spatial location to leakage status, improving detection depth and positioning accuracy. Utilizing spatiotemporal convolutional neural networks and physical information neural networks for intelligent inversion of spatiotemporal thermal response data, it not only automatically extracts subtle leak features that are difficult for the human eye to discern but also embeds the physical laws of thermal conduction as hard constraints into the inversion process, ensuring the physical rationality of the results and reducing the false alarm rate. The entire system, through the fusion and cross-validation of multimodal data, ultimately outputs a reliable and quantitative diagnostic report, providing strong technical support for the accurate location and status assessment of leaks in deeply buried pipelines, effectively overcoming the shortcomings of traditional single methods in terms of detection depth, anti-interference capability, and information dimension.

[0047] In one possible implementation, the spatiotemporal thermal response data is a sequence of time-series infrared thermograms, and the spatiotemporal convolutional neural network includes an input layer, a spatiotemporal convolutional layer, and an output layer. The spatiotemporal thermal response data is input into the spatiotemporal convolutional neural network for spatiotemporal feature extraction to obtain the thermal field change pattern of the underground pipeline, including: The time-series infrared thermal image sequence is input into a spatiotemporal convolutional neural network. The data dimensionality is reduced through the input layer, and the temperature change features in the spatial and temporal dimensions are extracted through the spatiotemporal convolutional layer. The spatiotemporal feature vector and / or classification labels are generated through the output layer to form the thermal field change pattern of the underground pipeline. The classification labels include leakage hot spots, solar interference, rainfall interference, and no anomalies.

[0048] In this embodiment, the spatiotemporal thermal response data consists of multiple frames of infrared thermal images collected at different time points and arranged chronologically. The input layer of the spatiotemporal convolutional neural network receives a normalized sequence tensor, which can be preliminarily adjusted or normalized. The core spatiotemporal convolutional layer uses a three-dimensional convolutional kernel, sliding convolution in two spatial dimensions and one temporal dimension respectively, sensing the temperature correlation between spatially adjacent pixels and the temperature change trend of temporally adjacent frames, effectively capturing the diffusion and intensity change patterns of leakage "hot spots".

[0049] After multiple convolutional and pooling layers, high-level features are passed to the output layer. The output layer generates a fixed-length spatiotemporal feature vector, which serves as the continuous input to the subsequent inversion network; simultaneously or optionally, it outputs classification probabilities through the Softmax function to determine the pattern category of the input sequence. This dual-output design enhances the system's flexibility and interpretability; the feature vector provides refined input for quantitative inversion, while the classification labels provide a user-friendly preliminary judgment for on-site engineers.

[0050] Specifically, the structure and functions of each part of the spatiotemporal convolutional neural network are shown in Table 1.

[0051] Table 1

[0052] Before inputting the time-series infrared thermal image sequence into the spatiotemporal convolutional neural network, the spatiotemporal convolutional neural network needs to be trained. The training process includes: The dataset construction requires the collection of both real-world and simulated data. Real-world data should cover pipeline leakage scenarios with varying burial depths, leakage intensities, and environmental conditions, accumulating ≥1000 sequences, each with 12 frames. Burial depths should range from 3-10m, leakage intensities from 0.1-1m³ / h, and environmental conditions should include sunny, rainy, cloudy, and day / night cycles. Simulated data is generated using COMSOL Multiphysics to simulate heat transfer processes, accumulating ≥5000 virtual sequences, including extreme environmental interference such as strong sunlight and heavy rain. The dataset is divided into training, validation, and test sets at a ratio of 70%, 20%, and 10% respectively to ensure consistent scenario distribution across the three data types.

[0053] The training parameters are set as follows: the optimizer uses the Adam optimizer with an initial learning rate of 1e-4, which decays to 0.5 every 10 epochs; for the loss function, the feature vector pattern uses MSE loss to fit the true thermal field features, and the classification pattern uses cross-entropy loss to distinguish the pattern type; the training epochs are 50 epochs long, and training stops when the validation set loss does not decrease for 5 consecutive epochs, using an early stopping mechanism; the hardware requirement is a GPU with ≥16GB of video memory, and the training time based on 5000 sets of data is approximately 24 hours.

[0054] The model evaluation metrics include feature extraction accuracy and pattern classification accuracy. Feature extraction accuracy requires an error of ≤5% in thermal anomaly diffusion rate and ≤0.5m in peak position; pattern classification accuracy requires a test set accuracy of ≥95% and a leak hot spot identification recall rate of ≥98% to avoid missed detections.

[0055] The significance of using spatiotemporal convolutional neural networks for spatiotemporal feature extraction for subsequent physical information neural networks is as follows: Reducing inversion complexity and improving efficiency: The original spatiotemporal temperature field data has extremely high dimensionality, with 12×256×256=786432 pixels. Directly inputting it into PINN would lead to an explosion in computational load. ST-CNN, through feature extraction, compresses the high-dimensional data into a 512-dimensional feature vector, reducing the computational load of PINN inversion by more than 99% and shortening the inversion time from hours to minutes.

[0056] Filtering noise and focusing on effective information: ST-CNN learns typical spatiotemporal patterns of leaking hotspots through training, such as slow diffusion, higher central temperature than the surrounding area for a long period of time, which can effectively filter environmental interference, such as the uniform rise in global temperature caused by sunlight and the sudden drop in local temperature caused by rainfall. This provides PINN with the core features after noise reduction, avoiding PINN being misled by noise.

[0057] Providing physical prior clues to constrain the inversion process: The feature vector output by ST-CNN contains key parameters of thermal anomalies, such as diffusion rate and shape. These parameters can serve as soft constraints for PINN. For example, the thermal diffusion rate of the leak point should be consistent with the feature vector to help PINN quickly lock the inversion range. If the diffusion rate is slow, the leak point is buried at a large depth, thus avoiding the inversion results from deviating from physical laws.

[0058] Improved inversion accuracy and stability: Without ST-CNN, PINN may experience fluctuations in inversion results due to noise in the original data. After adding ST-CNN, the stability of the feature vectors reduces the inversion error of PINN, and the consistency of repeated detections remains at a high level.

[0059] In one possible implementation, acquiring spatiotemporal thermal response data obtained from underground pipeline detection includes: Multi-time-series infrared scanning of the target area was performed under different ambient temperature backgrounds to obtain time-series thermal images; A modulated infrared radiation source is introduced to apply a thermal excitation signal with a specific temporal and spatial pattern to the ground to enhance the thermal anomaly response caused by the leak point.

[0060] In this embodiment, two active strategies can be used to improve the quality and signal-to-noise ratio of thermal response data. The first strategy is multi-temporal infrared scanning: instead of a single image capture, the same target area is repeatedly scanned during periods of significant environmental temperature change, such as before sunrise, noon, after sunset, and late at night. Temperature changes caused by environmental factors are global and trend-based, while thermal anomalies caused by leaks are local and continuous. By analyzing the temperature change curves of pixels through time-series data, characteristic signals at different time scales can be separated, and background thermal noise can be suppressed. The second strategy is active thermal excitation: a modulated infrared radiation source is deployed to apply thermal excitation of a known pattern to the surface, such as periodic pulse heating. The actively injected thermal signal is conducted downward through the soil. The soil above the leak point undergoes thermal property changes due to additional heat exchange with the medium, resulting in a difference in its thermal response to active excitation compared to the non-leakage area, thereby amplifying the leak signal and making it stand out from the environmental background. Combining multi-temporal passive observation and active excitation can systematically eliminate environmental interference and construct a high-quality, high-signal-to-noise-ratio spatiotemporal temperature field sequence.

[0061] Specifically, the thermal imaging process can include multi-temporal scanning of thermal field data. Instead of taking a single shot, the infrared thermal imager repeatedly scans the target area at different times (such as before sunrise, midday, and after sunset) to acquire a spatiotemporal temperature field sequence dataset. This helps to separate environmental thermal noise and capture the evolution of real thermal anomalies caused by leaks.

[0062] In this process, a modulated infrared radiation source is introduced as an active thermal excitation unit. Based on algorithmic decisions, this unit applies thermal excitation with specific temporal and spatial patterns (such as pulses or periodic modulation) to the ground, stimulating the thermal response of underground pipelines and greatly enhancing the signal-to-noise ratio.

[0063] In one possible implementation, the thermal field change pattern is input into a physical information neural network for inversion to obtain information about the first leakage source of the underground pipeline, including: The thermal field change pattern and the spatiotemporal sampling points of the area where the underground pipeline is located are input into the physical information neural network for inversion to obtain the first leakage source information; wherein, the first leakage source information includes the horizontal position coordinates of the leakage point, the burial depth and the leakage intensity.

[0064] In this embodiment, the input to the physical information neural network consists of two parts: first, the thermal field change pattern derived by the spatiotemporal convolutional neural network, which serves as a condensed expression of observational evidence; and second, the spatiotemporal sampling points defined within the inversion domain, including surface points that fit the observation data and underground points that enforce physical laws, whose coordinates (x, y, z, t) inform the network of the location and time of evaluating the physical equations.

[0065] The network outputs leak source parameters through an internal nonlinear function: the horizontal coordinates (X, Y) of the leak point's projection onto the Earth's surface, the vertical burial depth (Z), and the leak intensity (Q). These parameters not only pinpoint the leak location but also quantify the severity and depth of the leak, providing rich support for engineering decision-making and enabling a leap from phenomenological description to condition assessment.

[0066] The Physical Information Neural Network (PIN) combines data-driven approaches with physical constraints to infer leak point parameters from the dynamic surface temperature field by solving an inverse problem. First, the inverse problem and parameterized leak source are defined, assuming a known spatiotemporal temperature field sequence at the surface. State parameters including the leak point's surface projection coordinates, burial depth, and leak intensity are solved, assuming the leak source as a point source with a known medium temperature. The initial soil temperature is obtained from background data obtained through multiple time-series scans. Next, a network structure incorporating physical constraints is constructed, designing a loss function composed of data loss and physical loss. The data loss is obtained by calculating the mean square error between the predicted and observed temperatures at randomly sampled key points. The physical loss is based on the heat conduction equation, calculating the sum of squared residuals at randomly sampled points. The total loss is a combination of both types of losses, ensuring the inversion results conform to both observational data and physical laws. Subsequently, network training and parameter optimization are performed. The leak source parameters are initialized first. The feature vector and sampling points are input into the network to obtain the predicted temperature and leak parameters. The gradient of the total loss with respect to the leak parameters is calculated using an automatic differentiation framework. The parameters are updated and iterated using an optimizer until the total loss converges or the maximum number of iterations is reached, outputting the optimal parameters. The inversion results need to undergo consistency verification, boundary correction, and multiple sampling verification to ensure that the results are reliable and within a reasonable range. Finally, the precise spatial location, burial depth, and leakage intensity parameters of the leak point are output.

[0067] The structure and functions of each part of the physical information neural network are shown in Table 2.

[0068] Table 2

[0069] In one possible implementation, before inputting the thermal field change pattern into a physical information neural network for inversion to obtain the first leakage source information of the underground pipeline, the following steps are included: Construct a training dataset containing multiple sets of thermal field change patterns and their corresponding leakage source information; Based on the training dataset and the loss function, the initial physical information neural network is trained to obtain the trained physical information neural network. The loss function includes a data fitting term and a physical constraint term, which is used to force the output of the physical information neural network to satisfy the heat conduction equation.

[0070] In this embodiment, the training process of the physical information neural network before it is put into use is the key to its inversion capability.

[0071] First, a high-quality training dataset is constructed, with each sample being paired data: the input is the thermal field change pattern (obtained by processing real or simulated thermal image sequences through a spatiotemporal convolutional network), and the label is the corresponding leakage source information parameters [X, Y, Z, Q]. The data can be derived from controlled field experiments or generated by simulating various leakage scenarios and environmental conditions using numerical simulation software.

[0072] The PINN model is trained on the training dataset. The core task is to define and optimize a composite loss function, which is a weighted sum of a data fitting term and a physical constraint term. The data fitting term calculates the error between the network's predicted parameters and the sample labels; the physical constraint term evaluates the degree to which the network's predicted parameters and the implicit temperature field violate the heat conduction equation. During training, the network weights are adjusted using gradient descent to minimize the total loss, enabling the network to simultaneously grasp the empirical correlation between thermal field characteristics and leakage parameters, while ensuring that the implicit physical processes conform to the heat conduction law. After training, the network can reliably infer parameters for thermal field change patterns in new scenarios.

[0073] PINN's loss function consists of two parts: data loss and physical loss, ensuring that the inversion results conform to both the observed data and the physical laws of heat conduction. Data loss ( ): To measure the difference between predicted and observed temperatures, M key points are randomly sampled from the spatiotemporal temperature field of the Earth's surface. Calculate and predict temperature With observed temperature MSE:

[0074] Physical loss ( Forced satisfaction of the heat conduction equation (Fourier's law) Heat conduction equation (three-dimensional isotropic medium):

[0075] Where: α: soil thermal diffusivity (known, obtained from geological survey data, such as clay α=1e-7 m² / s); S(x,y,z,t;θ): heat source term of the leakage source, S=Q only when (x,y,z)=(X,Y,Z). (T0-T_s) (Heat released from the leaking medium into the soil), otherwise S=0.

[0076] Physical loss is defined as the sum of squared residuals of the heat conduction equation at random sampling points:

[0077] Where N=500 (underground + surface sampling points to ensure physical compliance throughout the space).

[0078] The final loss function expression is:

[0079] The specific steps of network training and parameter optimization include: 1. Initialize the leakage source parameter θ0 (randomly generated, satisfying X∈[detection region]). ]、Y∈[detection region Z∈[3,10], Q∈[0.1,1]). 2. Forward Propagation: Input the feature vector and sampling point (x,y,t) into PINN, and output the predicted temperature. and leakage parameter θ; 3. Backpropagation: Calculate the total loss The gradient of θ is updated by the Adam optimizer using an automatic differentiation framework such as TensorFlowAutograd. 4. Iterative optimization: Repeat steps 2-3 until... Convergence (convergence threshold: If the number of iterations reaches 1000 (≤1e-6), the final output is the optimal parameter θ*=[X*,Y*,Z*,Q*).

[0080] Step 5: Verification and Correction of Inversion Results 1. Consistency Verification: Substitute θ* into the heat conduction equation and simulate the surface temperature field. , and the observed temperature field In comparison, if MSE ≤ 5%, the results are considered reliable; 2. Boundary Correction: If Z exceeds the reasonable burial depth range (e.g., Z>10m), or Q deviates too much from the actual rated flow rate of the pipeline (e.g., Q>0.5 times the rated flow rate), then adjust the physical loss weight (e.g., adjust the weight of the physical loss weight by adjusting ... (Weight increased to 2), then iterate again; 3. Multiple sampling verification: Three independent inversions were performed using different initial θ0. If the deviation of the three results was ≤0.3m (location) and ≤0.05m³ / h (intensity), the average value was taken as the final result.

[0081] Step 6: Output inversion parameters The final output shows the precise parameters of the leak point: Spatial location: X* (±0.1m), Y* (±0.1m); Burial depth: Z* (±0.3m); Leakage intensity: Q* (±0.05 m³ / h).

[0082] In one possible implementation, the leak detection results of the underground pipeline are determined based on the first leak source information and the second leak source information, including: If the spatial location indicated by the first leakage source information and the second leakage source information is consistent within the preset tolerance range, then the spatial location is confirmed as the actual leakage point.

[0083] In this embodiment, the decision fusion rule is based on the "independent evidence convergence" criterion: if the spatial locations of the leak points calculated independently by two detection technologies based on different physical principles are highly similar, then the confidence that the location is the real leak point is extremely high.

[0084] During implementation, a reasonable spatial tolerance range is predefined based on the theoretical positioning accuracy, sensor resolution, and engineering requirements of both methods. The Euclidean distance between the horizontal coordinates given by the thermal method and the seismic wave method is calculated. If the distance is less than the preset tolerance threshold, the results are deemed spatially consistent. After the consistency condition is met, the system confirms the coordinate region as the true leak point and can output the coordinate conclusion based on the weighted average or geometric center. This rule utilizes the complementarity and redundancy of multiphysics information, and reduces the probability of false alarms through cross-validation, making the detection results more robust and reliable.

[0085] In one possible implementation, the seismic wave signal is inverted to obtain information about the second leakage source of the underground pipeline, including: Constructing an underground 3D scanning grid; For each grid point, the theoretical time delay of the seismic wave signal is calculated using a beamforming algorithm, and the signal is aligned. A three-dimensional spatial energy distribution map is generated by calculating the energy value corresponding to each grid point through weighted superposition. The location corresponding to the energy peak is extracted as the second leakage source information.

[0086] In this embodiment, the data preprocessing stage preprocesses the original seismic wave signal, including removing high-frequency environmental noise (such as traffic vibration and mechanical interference) and power frequency interference. A bandpass filter is used to retain the effective frequency band of 0.1-100Hz to ensure the integrity of the target signal. At the same time, an adaptive filtering algorithm is used to suppress periodic noise.

[0087] Because the data collected by each detector in the array have slight time differences, precise time synchronization correction is required. A high-precision clock source is used to provide a unified timestamp for each sensor to ensure the accuracy of subsequent time difference calculations. The precise position (X, Y, Z coordinates) of each seismic detector in the ground coordinate system is determined, and a spatial geometric model of the sensor array is constructed. A preliminary seismic wave propagation velocity model is established based on the characteristics of the subsurface medium (such as soil type and rock structure), which is crucial for calculating the time difference.

[0088] The time difference (delay) calculation process includes: assuming a leak source exists at a certain underground point, calculating the theoretical time required for seismic waves to propagate from this potential source point to each detector in the array. The calculation formula is: in, It is the theoretical time delay of the i-th detector. is the distance from the leakage source to the i-th detector, and v is the propagation speed of the seismic wave in the current medium.

[0089] First, based on the requirements of the exploration mission, the target underground space is discretized into a fine three-dimensional grid, with each grid node having coordinates... This represents a potential leakage source to be examined. This is a crucial step in transforming a continuous-space problem into a discrete computational problem. Next, for each node in the mesh... Beamforming processing is performed: based on the known detector locations and the established subsurface wave velocity model, the seismic wave velocity is calculated from... Theoretical time to propagate to the i-th detector Then, the actual signal recorded by the i-th detector is... Move backward on the timeline This process aligns the time delays of all channels. After alignment, the signals from all channels are superimposed (different weights can be applied to optimize performance), and the energy (e.g., sum of squared amplitudes) of the superimposed signal within a specific time window is calculated. This step is essentially spatial filtering: if... The signal from the leak event is a true source; in the aligned signal, the components from the leak event will be superimposed in phase, resulting in coherent energy enhancement. Noise, on the other hand, will be incoherently superimposed, with limited energy enhancement. By traversing all grid points and repeating this operation, each point is assigned an "energy value," and the energy values ​​of all points together form a three-dimensional energy distribution map. Finally, local maxima with significantly higher energy than their surrounding neighborhoods are searched in this map. The three-dimensional coordinates of these peak points are identified as the most likely leak source locations and output as secondary leak source information for independent seismic wave detection. This method achieves high-resolution imaging and localization of subsurface microseismic sources through systematic spatial scanning and coherent processing.

[0090] Specifically, the steps performed by the beamforming algorithm include: For each potential leak source location (scanning the entire underground target area), the signals received by each detector in the array are time-shifted and aligned according to the calculated theoretical time difference, and then weighted and superimposed. The mathematical expression is:

[0091] in: It is the output power after beamforming, corresponding to the spatial direction. Energy.

[0092] It is the weight of the i-th sensor (which can be used to adjust the array sensitivity direction).

[0093] It is the time-domain signal received by the i-th sensor.

[0094] N is the total number of sensors in the array.

[0095] Spatial scanning and imaging: By changing the hypothetical location of the leak source The time difference calculation and beamforming algorithms are repeatedly executed to perform a three-dimensional energy scan of the entire target underground space. Finally, a spatial energy distribution map is generated, and the highest point of energy focusing in the map is identified as the most likely location of the leak source.

[0096] Results Validation and Optimization: The beamforming imaging results were compared with the results of dynamic heat conduction inversion analysis. When the leak source locations obtained by the two independent methods are spatially consistent, the confidence that the location is the true leak point is significantly improved.

[0097] Parameter feedback and adaptive adjustment: If the results are not consistent or the signal quality is poor, the system can feed back problems such as velocity model mismatch encountered in beamforming processing to the control unit, thereby automatically adjusting the excitation parameters (such as transmission frequency and power) or optimizing the geometric layout of the sensor array, and starting a new round of detection and processing.

[0098] Through the above steps, beamforming technology makes full use of the spatial information of the sensor array, coherently superimposes distributed weak signals, effectively suppresses incoherent noise, and ultimately achieves super-resolution location of leaks in deeply buried pipelines, enabling this solution to achieve extremely deep detection depth and extremely high positioning accuracy.

[0099] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0100] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0101] Figure 3 A schematic diagram of the pipe leak detection device integrating thermal response and seismic wave analysis provided in an embodiment of the present invention is shown. For ease of explanation, only the parts relevant to the embodiment of the present invention are shown, and are described in detail below: like Figure 3 As shown, the pipe leak detection device 3, which integrates thermal response and seismic wave, includes: The acquisition module 31 is used to acquire seismic wave signals and spatiotemporal thermal response data obtained from the detection of underground pipelines; Feature extraction module 32 is used to input spatiotemporal thermal response data into spatiotemporal convolutional neural network for spatiotemporal feature extraction to obtain the thermal field change pattern of underground pipeline; The first inversion module 33 is used to input the thermal field change pattern into the physical information neural network for inversion to obtain the first leakage source information of the underground pipeline; wherein, the physical information neural network uses the heat conduction equation as a hard constraint; The second inversion module 34 is used to invert the seismic wave signal to obtain the information of the second leakage source of the underground pipeline; The information fusion module 35 is used to determine the leak detection results of the underground pipeline based on the first leak source information and the second leak source information.

[0102] In one possible implementation, the spatiotemporal thermal response data is a sequence of time-series infrared thermal images, and the spatiotemporal convolutional neural network includes an input layer, a spatiotemporal convolutional layer, and an output layer; the feature extraction module 32 is specifically used for: The time-series infrared thermal image sequence is input into a spatiotemporal convolutional neural network. The data dimensionality is reduced through the input layer, and the temperature change features in the spatial and temporal dimensions are extracted through the spatiotemporal convolutional layer. The spatiotemporal feature vector and / or classification labels are generated through the output layer to form the thermal field change pattern of the underground pipeline. The classification labels include leakage hot spots, solar interference, rainfall interference, and no anomalies.

[0103] In one possible implementation, module 31 is specifically used for: Multi-time-series infrared scanning of the target area was performed under different ambient temperature backgrounds to obtain time-series thermal images; A modulated infrared radiation source is introduced to apply a thermal excitation signal with a specific temporal and spatial pattern to the ground to enhance the thermal anomaly response caused by the leak point.

[0104] In one possible implementation, the first inversion module 33 is specifically used for: The thermal field change pattern and the spatiotemporal sampling points of the area where the underground pipeline is located are input into the physical information neural network for inversion to obtain the first leakage source information; wherein, the first leakage source information includes the horizontal position coordinates of the leakage point, the burial depth and the leakage intensity.

[0105] In one possible implementation, the first inversion module 33 is further used for: Before inputting the thermal field change pattern into the physical information neural network for inversion to obtain the first leakage source information of the underground pipeline, a training dataset is constructed. The training dataset contains multiple sets of thermal field change patterns and their corresponding leakage source information. Based on the training dataset and the loss function, the initial physical information neural network is trained to obtain the trained physical information neural network. The loss function includes a data fitting term and a physical constraint term, which is used to force the output of the physical information neural network to satisfy the heat conduction equation.

[0106] In one possible implementation, the information fusion module 35 is specifically used for: If the spatial location indicated by the first leakage source information and the second leakage source information is consistent within the preset tolerance range, then the spatial location is confirmed as the actual leakage point.

[0107] In one possible implementation, the second inversion module 34 is specifically used for: Constructing an underground 3D scanning grid; For each grid point, the theoretical time delay of the seismic wave signal is calculated using a beamforming algorithm, and the signal is aligned. A three-dimensional spatial energy distribution map is generated by calculating the energy value corresponding to each grid point through weighted superposition. The location corresponding to the energy peak is extracted as the second leakage source information.

[0108] This invention integrates low-frequency acoustic-seismic wave joint imaging with dynamic thermal conduction inversion. Leveraging the high penetration of low-frequency acoustic waves, it solves the problem of weak and difficult-to-capture signals from deeply buried pipelines. Simultaneously, combining the spatial positioning capabilities of seismic wave arrays with the quantitative diagnostic capabilities of thermal inversion, it achieves comprehensive and accurate perception of deep-buried leaks from spatial location to leakage status, improving detection depth and positioning accuracy. Utilizing spatiotemporal convolutional neural networks and physical information neural networks for intelligent inversion of spatiotemporal thermal response data, it not only automatically extracts subtle leak features that are difficult for the human eye to discern but also embeds the physical laws of thermal conduction as hard constraints into the inversion process, ensuring the physical rationality of the results and reducing the false alarm rate. The entire system, through the fusion and cross-validation of multimodal data, ultimately outputs a reliable and quantitative diagnostic report, providing strong technical support for the accurate location and status assessment of leaks in deeply buried pipelines, effectively overcoming the shortcomings of traditional single methods in terms of detection depth, anti-interference capability, and information dimension.

[0109] Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. For example... Figure 4 As shown, the electronic device 4 in this embodiment includes a processor 40 and a memory 41. The memory 41 stores a computer program 42. When the processor 40 executes the computer program 42, it implements the steps in the various method embodiments described above. Alternatively, when the processor 40 executes the computer program 42, it implements the functions of each module / unit in the various device embodiments described above.

[0110] For example, computer program 42 may be divided into one or more modules / units, which are stored in memory 41 and executed by processor 40 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 42 in electronic device 4.

[0111] Electronic device 4 may include, but is not limited to, processor 40 and memory 41. Those skilled in the art will understand that... Figure 4 This is merely an example of electronic device 4 and does not constitute a limitation on electronic device 4. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 4 may also include input / output devices, network access devices, buses, etc.

[0112] The processor 40 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0113] The memory 41 can be an internal storage unit of the electronic device 4, such as a hard disk or RAM. The memory 41 can also be an external storage device of the electronic device 4, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory 41 can include both internal and external storage units of the electronic device 4. The memory 41 is used to store the computer program 42 and other programs and data required by the electronic device 4. The memory 41 can also be used to temporarily store data that has been output or will be output.

[0114] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.

[0115] This invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0116] This invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0117] Computer programs include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0118] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0119] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A pipeline leak detection method integrating thermal response and seismic wave analysis, characterized in that, include: Acquire seismic wave signals and spatiotemporal thermal response data obtained from underground pipeline detection; The spatiotemporal thermal response data is input into a spatiotemporal convolutional neural network for spatiotemporal feature extraction to obtain the thermal field change pattern of the underground pipeline. The thermal field change pattern is input into a physical information neural network for inversion to obtain the first leakage source information of the underground pipeline; wherein, the physical information neural network uses the heat conduction equation as a hard constraint; The seismic wave signal was inverted to obtain the information on the second leakage source of the underground pipeline; Based on the first leak source information and the second leak source information, the leak detection results of the underground pipeline are determined.

2. The pipeline leak detection method integrating thermal response and seismic waves according to claim 1, characterized in that, The spatiotemporal thermal response data is a time-series infrared thermal image sequence, and the spatiotemporal convolutional neural network includes an input layer, a spatiotemporal convolutional layer, and an output layer; the step of inputting the spatiotemporal thermal response data into the spatiotemporal convolutional neural network for spatiotemporal feature extraction to obtain the thermal field change pattern of the underground pipeline includes: The time-series infrared thermal image sequence is input into a spatiotemporal convolutional neural network. The data dimensionality is reduced through the input layer, and the temperature change features in the spatial and temporal dimensions are extracted through the spatiotemporal convolutional layer. The spatiotemporal feature vector and / or classification labels are generated through the output layer to form the thermal field change pattern of the underground pipeline. The classification labels include leakage hot spots, solar interference, rainfall interference, and no anomalies.

3. The pipeline leak detection method integrating thermal response and seismic waves according to claim 2, characterized in that, Acquire spatiotemporal thermal response data obtained from underground pipeline detection, including: Multi-time-series infrared scanning of the target area was performed under different ambient temperature backgrounds to obtain time-series thermal images; A modulated infrared radiation source is introduced to apply a thermal excitation signal with a specific temporal and spatial pattern to the ground to enhance the thermal anomaly response caused by the leak point.

4. The pipeline leak detection method integrating thermal response and seismic wave analysis according to claim 1, characterized in that, The step of inputting the thermal field change pattern into a physical information neural network for inversion to obtain the first leakage source information of the underground pipeline includes: The thermal field change pattern and the spatiotemporal sampling points of the area where the underground pipeline is located are input into a physical information neural network for inversion to obtain the first leakage source information; wherein, the first leakage source information includes the horizontal position coordinates, burial depth and leakage intensity of the leakage point.

5. The pipeline leak detection method integrating thermal response and seismic waves according to claim 4, characterized in that, Before inputting the thermal field change pattern into the physical information neural network for inversion to obtain the first leakage source information of the underground pipeline, the process includes: Construct a training dataset, which contains multiple sets of the thermal field change patterns and their corresponding leakage source information; Based on the training dataset and loss function, the initial physical information neural network is trained to obtain a trained physical information neural network; wherein, the loss function includes a data fitting term and a physical constraint term, and the physical constraint term is used to force the output of the physical information neural network to satisfy the heat conduction equation.

6. The pipeline leak detection method integrating thermal response and seismic wave analysis according to any one of claims 1 to 5, characterized in that, The step of determining the leak detection result of the underground pipeline based on the first leak source information and the second leak source information includes: If the spatial location indicated by the first leakage source information and the second leakage source information is consistent within a preset tolerance range, then the spatial location is confirmed as the actual leakage point.

7. The pipeline leak detection method integrating thermal response and seismic wave analysis according to any one of claims 1 to 5, characterized in that, The process of inverting the seismic wave signal to obtain information about the second leakage source of the underground pipeline includes: Constructing an underground 3D scanning grid; For each grid point, the theoretical time delay of the seismic wave signal is calculated using a beamforming algorithm, and the signal is aligned. A three-dimensional spatial energy distribution map is generated by calculating the energy value corresponding to each grid point through weighted superposition. The location corresponding to the energy peak is extracted as the second leakage source information.

8. A pipe leak detection device integrating thermal response and seismic wave, characterized in that, include: The acquisition module is used to acquire seismic wave signals and spatiotemporal thermal response data obtained from the detection of underground pipelines; The feature extraction module is used to input the spatiotemporal thermal response data into a spatiotemporal convolutional neural network for spatiotemporal feature extraction to obtain the thermal field change pattern of the underground pipeline. The first inversion module is used to input the thermal field change pattern into the physical information neural network for inversion to obtain the first leakage source information of the underground pipeline; wherein, the physical information neural network uses the heat conduction equation as a hard constraint; The second inversion module is used to invert the seismic wave signal to obtain the second leakage source information of the underground pipeline; The information fusion module is used to determine the leak detection results of the underground pipeline based on the first leak source information and the second leak source information.

9. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images