Real-time leakage location monitoring system and method based on electrical resistance tomography

Abstract

The application discloses a real-time leakage positioning monitoring system and method based on electrical resistance tomography, relates to the technical field of process tomography and industrial safety monitoring, and solves the technical problem that it is difficult to intelligently identify, three-dimensionally position and quantitatively evaluate leakage events; the system comprises a sensor array, a multi-channel data acquisition and excitation module, a real-time data processing and image reconstruction module, a leakage intelligent identification and positioning module, a man-machine interaction and data management module; the core lies in that a multi-layer nested annular array and a sparse measurement strategy are adopted to improve data acquisition efficiency, a physical information neural network is introduced to reconstruct an image, a measurement topological coding matrix and a physical residual correction module are used to reconstruct the image, and finally, a time-space dual-domain leakage probability field model is constructed, a virtual sensor network and an improved D-S evidence reasoning framework are combined to intelligently identify and position a leakage diffusion mode.

Description

Technical Field

[0001] This invention belongs to the field of process tomography and industrial safety monitoring technology, specifically relating to a real-time leakage location monitoring system and method based on resistivity tomography. Background Technology

[0002] Electrical resistance tomography (ERT) is a process tomography technique based on the principle of electrical sensitivity. It involves arranging an electrode array at the boundary of the test field, applying an excitation current and measuring the boundary voltage. Based on the changes in the boundary conductivity distribution, an image reconstruction algorithm is used to reconstruct the distribution image of the medium inside the test field. ERT technology has the advantages of being non-invasive, radiation-free, low-cost, and fast-responding, and has been widely used in fields such as multiphase flow detection and geological structure exploration.

[0003] However, directly applying traditional ERT technology to leakage monitoring of structures such as underground storage tanks, chemical pipelines, and water conservancy dams faces the following prominent technical challenges: Traditional ERT systems have long data acquisition and image reconstruction processes, typically taking seconds, making it difficult to meet the requirements for continuous, real-time monitoring and rapid response to sudden leaks; traditional ERT mainly provides static or quasi-static images of the internal medium distribution, and its analysis and interpretation heavily rely on operator experience. The system lacks dedicated algorithms and functional modules for feature identification, automatic alarm, and location of the specific physical event of "leakage"; the conductivity changes caused by early or minute leaks are weak and easily drowned out by measurement noise and environmental interference, making it difficult for standard imaging algorithms to effectively extract such weak anomalies, resulting in a high false alarm rate; simultaneously, interference in complex backgrounds is easily misjudged as leakage, leading to a high false alarm rate; traditional ERT reconstructions are mostly two-dimensional cross-sectional images, making it difficult to accurately locate the leakage source in three-dimensional space (i.e., provide three-dimensional coordinates), and even more difficult to quantitatively estimate the leakage intensity, limiting its application value in precise maintenance guidance.

[0004] Therefore, the industry urgently needs a dedicated ERT leakage monitoring system that integrates efficient data acquisition, intelligent image reconstruction, automatic event recognition, and precise location. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a real-time leakage location monitoring system and method based on electrical resistivity tomography, which is used to solve the following technical problems: When applied to leakage monitoring, the system struggles to achieve real-time, rapid response and to automatically and intelligently identify and alarm leakage events. Furthermore, it lacks the ability to effectively suppress noise and extract weak leakage signals, resulting in insufficient system sensitivity and reliability, and high false alarm and false alarm rates. Moreover, existing technologies lack the technical means to accurately locate leakage sources in three-dimensional space and quantitatively estimate their intensity from two-dimensional measurement data.

[0006] To address the aforementioned problems, a first aspect of the present invention provides a real-time leakage location and monitoring system based on electrical resistivity tomography, comprising the following modules: Sensor array module: contains multiple electrodes fixed to the surface of the structure under test in a specific spatial topology; Multi-channel data acquisition and excitation module: Outputs excitation current to selected electrode pairs according to preset excitation-measurement modes, while measuring boundary voltage signals between other electrode pairs and converting the measured analog signals into digital signals; Real-time data processing and image reconstruction module: Receives the digital signal, constructs a measurement topology coding matrix, concatenates the sparse boundary voltage vector with the measurement topology coding matrix, uses it as input to the physical information neural network, and generates a differential image sequence of conductivity changes in the measured area; Leakage intelligent identification and location module: Constructs a spatiotemporal dual-domain leakage probability field model, analyzes the dynamic evolution characteristics of abnormal regions in the differential image sequence, combines the prior geometric constraints of the measured structure, solves the inverse problem with the measurement data as boundary conditions, identifies spatiotemporal patterns that conform to the physical diffusion law of leakage, and triggers an alarm when the similarity exceeds the preset similarity threshold and calculates and outputs the three-dimensional spatial coordinates and intensity estimate of the leakage source. Human-computer interaction and data management module: A visual interface that displays real-time location results, monitoring information and alarm information.

[0007] Preferably, the specific spatial topology is a multi-layer nested concentric ring array; the multi-layer nested concentric ring array is composed of two or more ring electrode arrays with different diameters and coincident central axes, and the electrodes on the same layer of the ring array are distributed at equal intervals; the ring arrays of different layers are arranged along the axial direction of the pipe or tank being tested.

[0008] Preferably, the multi-channel data acquisition and excitation module includes: a . The mode controller divides the electrode pairs in the sensor array into subsets of excitation electrode pairs according to a preset sparse measurement strategy. b. Based on a pre-calculated sensitivity matrix The optimization algorithm selects a subset of combinations from all possible stimulus-measurement pair combinations that optimize the condition number of the reconstruction matrix. c. For each selected excitation electrode pair, a constant current source outputs an excitation current, and at the same time, a multiplexer performs voltage measurement on the measurement electrode pair in the combination selected in step b. d. Output the boundary voltage dataset defined by the sparse measurement strategy.

[0009] Preferably, the real-time data processing and image reconstruction module includes: The real-time data processing and image reconstruction module performs image reconstruction based on a physical information neural network. Constructing a measurement topology coding matrix ,in, For the set of real numbers, For sparse measurement quantity, The total dimension of the encoded vector; each line Corresponding to one excitation-measurement pair ,in, As the excitation electrode, For measuring electrodes; For the sparse measurement matrix, the first... One excitation-measurement pair Its topological encoding vector Calculated according to preset encoding rules, specifically: Normalized inscribed angle difference encoding: ,in, This represents the angular difference between the two electrodes in the unfolded circular coordinate system; Normalized spatial distance coding: ,in, and Electrode and Coordinate vectors in three-dimensional space It is the L2 norm. This represents the maximum value of the Euclidean distance between any two electrodes in the array; Electrode position periodic encoding: for the first One electrode, ,in, This represents the total number of electrodes on the same-layer ring matrix. When the electrode pair spans multiple layers, add cross-layer relationship coding: ,in, For same-layer markers, if ,but ,otherwise, , This is a function representing the layer number where the electrode is located. This represents the maximum layer number difference in the sensor array. final The concatenated vector, where, and Electrodes and Periodic encoding.

[0010] Preferably, the physical information neural network is a variant of the U-Net architecture, whose encoder input is a sparse boundary voltage vector. With topological encoding matrix The flattened and concatenated feature vectors, where, The output of step d A vector composed of measured values; The physical information neural network is trained using a knowledge distillation framework: the teacher network is trained with a complete measurement dataset, and the student network is trained with the intermediate layer features of the teacher network as supervision signals. Its loss function is a weighted sum of reconstruction loss and feature alignment loss. Each upsampling layer in the network decoder is followed by a physical residual correction module, which estimates the residual based on the conductivity distribution output of the current layer. Using differentiable forward modeling as input, the boundary voltage is predicted. Calculate measurement residuals The measurement residual is obtained by using a small back-projection network. Mapped to distributed correction quantity And fused with decoder features; The output layer of the neural network has a multi-task head structure, which simultaneously outputs a conductivity distribution image and a preliminary leakage area confidence heatmap.

[0011] Preferably, the construction and updating of the spatiotemporal dual-domain leakage probability field model includes: Modeling a continuous T-frame difference image sequence as a spatiotemporal graph , where nodes For the first Superpixel regions in a frame image, edges Connect nodes in adjacent frames that are spatially close or have similar features; Each node calculates the dynamic propagation kernel. ,in, This is a normalized exponential function along the row direction. For nodes eigenvectors, Its spatiotemporal neighborhood nodes eigenmatrix and For the parameter matrix, The dimension of the feature vector; Initial leakage probability field Generated based on single-frame anomaly detection results, using iterative formulas. ,in, Let be the probability field vector. For spacetime diagram The adjacency matrix, For all The propagation kernel matrix constructed by node index. Index for iteration count, The Hadamard product of the matrices. It is the attenuation factor; After each iteration, the probability field is compared with the prior probabilities based on the structure's location. By multiplying the elements and normalizing, we obtain the probability field with physical constraints.

[0012] Preferably, solving the inverse problem with measurement data as boundary conditions includes: Each excitation-measurement pair is abstracted as a virtual sensor, which measures any spatial point within the imaging region. The proposition that a source of leakage exists Provide evidence; For the A virtual sensor, whose actual measured voltage value is... The virtual sensor for the proposition Basic probability assignment Calculated according to preset assignment rules; The preset assignment rules are as follows: ,in, In the first Under certain observations, the leakage proposition is supported. Basic probability assignment, To support normalization and confidence coefficients, For the first The measured value obtained from this measurement Assuming leakage proposition When it is established, the first term calculated by the forward model The theoretical voltage value of each sensor, For the first Standard deviation of measurement noise at each observation point; ,in, In the first Under observation, the proposition against leakage is... Basic probability assignment, To oppose the normalization constant and confidence coefficient; ,in, For the proposition Uncertainty.

[0013] Preferably, the fusion and localization of the evidence further includes: Calculate any two virtual sensors and Point The distance between the evidence and Jousselme :like >Preset distance threshold, then the sensor and Marked as at point There is a significant conflict at this point; based on the conflict relationship, each point... The sensors are dynamically divided into highly consistent subgroups. and high-conflict subgroups ; right A modified weighted Dempster rule is used for fusion, where the weight of each sensor is related to the deterministic entropy of its evidence. Proportional; to All evidence is retained as a set of competing hypotheses for that point. ; Search the entire imaging space and select Trust function after fusion highest These points are used as the candidate core point set. ;in, for Evidence of fusion; For each Its spatial neighborhood Find the point inside ,like There is evidence in and similarity If the evidence is greater than or equal to a preset similarity threshold, then the evidence will be... As support The supplementary evidence was integrated and updated. To enhance the trust function Output highest point Location of the leakage source; With point Centered on, according to The spatial gradient is used to determine the confidence region, where the mean of the similarity function at each point within the confidence region is greater than or equal to a preset confidence threshold. Mean, where, For a plausible function at a point The value, For the point The final fusion result of all available evidence at the location, A second aspect of the present invention provides a real-time leakage location and monitoring method based on resistivity tomography, comprising the following steps: S1: A sensor array consisting of multiple electrodes fixed to the surface of the structure under test in a specific spatial topology. S2: Output excitation current to the selected electrode pair according to the preset excitation-measurement mode, and at the same time measure the boundary voltage signal between other electrode pairs and convert the measured analog signal into a digital signal; S3: Receive the digital signal, construct a measurement topology coding matrix, concatenate the sparse boundary voltage vector with the measurement topology coding matrix, use it as input to the physical information neural network, and generate a differential image sequence of conductivity changes in the measured area. S4: Construct a spatiotemporal dual-domain leakage probability field model, analyze the dynamic evolution characteristics of abnormal regions in the differential image sequence, combine the prior geometric constraints of the measured structure, solve the inverse problem with the measurement data as boundary conditions, identify spatiotemporal patterns that conform to the physical diffusion law of leakage, and trigger an alarm when the similarity exceeds the preset similarity threshold and calculate and output the three-dimensional spatial coordinates and intensity estimate of the leakage source. S5: A visual interface that displays real-time location results, monitoring information, and alarm information.

[0014] The beneficial effects of this invention are: This invention employs a multi-layered nested concentric ring array arranged along the axis of the object being measured, enabling the simultaneous acquisition of information from multiple cross-sections. This lays the hardware foundation for three-dimensional positioning. Then, based on a pre-calculated sensitivity matrix, an optimization algorithm selects the excitation-measurement pair combination with the largest amount of information. Measurement and data acquisition are performed only on the selected combination. This significantly reduces the amount of data and time required for a single measurement while ensuring reconstruction quality, thus meeting the need for continuous and real-time monitoring of rapidly changing leakage processes. This invention designs a measurement topology coding matrix to encode the spatial geometric relationship of electrodes into a learnable input for a neural network. The network adopts a variant of U-Net and embeds a physical residual correction module during the decoding process. This module uses differentiable forward modeling to calculate the predicted voltage and generates a correction amount with the measured residual through a small back projection network, realizing a deep coupling between data-driven and physical laws. Through training with a knowledge distillation framework, it solves the network training problem under sparse data and realizes end-to-end, millisecond-level fast reconstruction from sparse measurement data to high-quality differential images. This invention models a continuous image sequence as a spatiotemporal graph and learns a dynamic propagation kernel for each node in the graph to simulate the diffusion and evolution of leakage probability in space and time, thus achieving intelligent modeling of leakage dynamic patterns. Simultaneously, each excitation-measurement pair is abstracted as a virtual sensor, and an evidence generation function is designed for it based on the ERT forward model. First, consistent and conflicting sensor groups are divided according to the Jousselme distance between the evidence. Consistent groups are weighted and fused, while conflicting groups are retained as competing hypotheses. Finally, utilizing the spatial continuity prior of the leakage source, evidence from different spatial points is coordinated through a spatial arbitration mechanism, transforming locally conflicting evidence into useful information supporting the globally optimal location point. Ultimately, the three-dimensional coordinates of the leakage source are output, significantly reducing the risk of mislocation due to single-point measurement anomalies. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the module flow of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 As shown, this invention is a real-time leakage location and monitoring system based on electrical resistivity tomography, comprising the following modules: Sensor array module: contains multiple electrodes fixed to the surface of the structure under test in a specific spatial topology; Multi-channel data acquisition and excitation module: Outputs excitation current to selected electrode pairs according to preset excitation-measurement modes, while measuring boundary voltage signals between other electrode pairs and converting the measured analog signals into digital signals; Real-time data processing and image reconstruction module: Receives the digital signal, constructs a measurement topology coding matrix, concatenates the sparse boundary voltage vector with the measurement topology coding matrix, uses it as input to the physical information neural network, and generates a differential image sequence of conductivity changes in the measured area; Leakage intelligent identification and location module: Constructs a spatiotemporal dual-domain leakage probability field model, analyzes the dynamic evolution characteristics of abnormal regions in the differential image sequence, combines the prior geometric constraints of the measured structure, solves the inverse problem with the measurement data as boundary conditions, identifies spatiotemporal patterns that conform to the physical diffusion law of leakage, and triggers an alarm when the similarity exceeds the preset similarity threshold and calculates and outputs the three-dimensional spatial coordinates and intensity estimate of the leakage source. Human-computer interaction and data management module: A visual interface that displays real-time location results, monitoring information and alarm information.

[0018] Specifically, the multi-layer nested concentric ring electrode array in the sensor array module is tightly attached to a designated section of the outer wall of the pipe or tank to be monitored. After the system is powered on, initial calibration is performed: under the control of the multi-channel data acquisition and excitation module, a complete excitation-measurement cycle is performed to obtain the reference boundary voltage dataset under the leak-free state. At the same time, based on the three-dimensional geometric model of the structure under test and the initial conductivity estimate, the sensitivity matrix is ​​pre-calculated, and the optimization algorithm is run to generate the initial sparse measurement strategy. The system enters a real-time monitoring loop: the mode controller of the multi-channel data acquisition and excitation module loads the current sparse measurement strategy, and then the constant current source sequentially outputs safe excitation current to the excitation electrode pairs selected in the strategy; at the same time, the high-speed multiplexer and the measurement circuit synchronously perform rapid acquisition of the boundary voltage of the corresponding measurement electrode pairs in the strategy. After one loop is completed, a sparse boundary voltage vector with a significantly reduced dimension is output. The real-time data processing and image reconstruction module receives the sparse boundary voltage vector: First, it calculates the measurement topology coding matrix in real time according to the rules based on the electrode space coordinates. Then, it concatenates the sparse boundary voltage vector with the measurement topology coding matrix and inputs it into the pre-trained physical information neural network. The network outputs the differential conductivity distribution image and preliminary leakage confidence heatmap at the current moment through a single forward propagation. The most recent T-frame images are used to construct a spatiotemporal map, and the spatiotemporal dual-domain leakage probability field is iteratively updated to predict the diffusion trend of leakage probability. All effective excitation-measurement pairs at the current moment are regarded as virtual sensors. Based on the forward model, a basic probability assignment is generated for each sensor at each spatial grid point. Subsequently, conflict detection, dynamic grouping, hierarchical fusion and spatial arbitration processes are performed to finally determine the most likely three-dimensional location of the leakage source and output its confidence region and intensity estimate. The human-computer interaction and data management module receives alarm signals and location results. On the visualization interface, the leakage location and confidence area are highlighted in the form of a 3D model overlay or cross-sectional view. At the same time, an alarm information box pops up. All raw data, intermediate images, location results and alarm events are synchronously stored in the database and can be uploaded to the cloud or central monitoring platform through the communication interface.

[0019] In one embodiment of the present invention, the specific spatial topology is a multi-layer nested concentric ring array; the multi-layer nested concentric ring array is composed of two or more ring electrode arrays with different diameters and coincident central axes, and the electrodes on the same layer of the ring array are distributed at equal intervals; the ring arrays of different layers are arranged along the axial direction of the pipe or tank being tested.

[0020] Specifically, firstly, based on the diameter of the pipe or tank being tested and the length of the area to be monitored, the number of array layers, the diameter of each layer, and the axial spacing are determined. Each layer is designed as an independent ring electrode array, with the central axis of all rings coinciding with the central axis of the pipe or tank. On the same ring array, the electrodes are welded at equal intervals or embedded, tightly fitting against the outer surface of the structure. Installation begins with the first ring array closest to the surface of the structure being tested, constructing a multi-layer nested structure layer by layer outwards. The electrodes of each ring array are led out through independent shielded cables and connected to the corresponding channel of the multi-channel data acquisition and excitation module. In the axial direction, the ring arrays of different levels maintain a fixed connection. A fixed interval is established, typically determined based on spatial resolution requirements and structural dimensions. Good insulation is ensured between adjacent electrodes on the same layer of the ring array and between ring arrays on different layers to prevent short circuits. All electrode leads must be clearly labeled before being connected to the acquisition module, and their positions must correspond one-to-one with the channel numbers of the acquisition module, establishing a clear physical location mapping. After system deployment, test current is injected into some electrode pairs on each layer through the multi-channel data acquisition and excitation module, and the response voltage is measured to verify that all electrode connections are correct, have good contact, and can synchronously and independently acquire the boundary electrical signals of each layer's cross-section. This lays the hardware foundation for subsequent synchronous 3D data acquisition and leakage source depth localization.

[0021] In one embodiment of the present invention, the multi-channel data acquisition and excitation module includes: a . The mode controller divides the electrode pairs in the sensor array into subsets of excitation electrode pairs according to a preset sparse measurement strategy. b. Based on a pre-calculated sensitivity matrix The optimization algorithm selects a subset of combinations from all possible stimulus-measurement pair combinations that optimize the condition number of the reconstruction matrix. c. For each selected excitation electrode pair, a constant current source outputs an excitation current, and at the same time, a multiplexer performs voltage measurement on the measurement electrode pair in the combination selected in step b. d. Output the boundary voltage dataset defined by the sparse measurement strategy.

[0022] Specifically, before system deployment or during the initialization phase, based on the geometric layout and initial conductivity estimation of the sensor array, a complete sensitivity matrix is ​​obtained through finite element simulation calculation. The mode controller executes an optimization algorithm (e.g., a greedy algorithm or a convex optimization algorithm) to iteratively select a fixed number (e.g., 20%-40% of the original number of combinations) of combinations from all theoretically possible excitation-measurement pair combinations. The objective of the selection is to minimize the condition number of the reconstruction matrix formed by the sensitivity submatrices corresponding to the subset, thereby ensuring the numerical stability and information efficiency of subsequent image reconstruction. After optimization, the selected combination subset is fixed as a preset sparse measurement strategy and stored in the mode controller. After entering the real-time monitoring loop, at the beginning of each measurement cycle, the mode controller reads the sequence of excitation electrode pairs to be executed in the current cycle from the preset sparse measurement strategy. Then, the controller performs the following operations in sequence: Excitation output: Controls the constant current source and output multiplexer switch to safely and accurately apply the excitation current to the currently specified excitation electrode pair; Synchronous measurement: During the stable output of the excitation current, the controller synchronously controls the measurement multiplexer switch and the high-precision differential amplifier array to sequentially switch to each measurement electrode pair specified in the strategy under the current excitation mode, and quickly and synchronously acquires the boundary induced voltage at both ends; Signal conversion and buffering: The acquired analog voltage signal is converted into a digital signal by the analog-to-digital converter (ADC) and buffered according to the order defined by the strategy. After a stimulus-measurement pattern loop (i.e., a complete traversal of all stimulus-measurement pairs specified in the strategy) ends, the pattern controller encapsulates all cached measurement values ​​according to a predetermined data structure (e.g., an M-dimensional vector, where M is the total number of measurements specified by the strategy) to form a sparse boundary voltage dataset. Subsequently, this dataset is output to the subsequent real-time data processing and image reconstruction module for processing. This process significantly reduces the number of measurements actually performed in each cycle while ensuring the integrity of key measurement information. As a result, it reduces the data acquisition time by an order of magnitude while maintaining reconstruction quality, which is the core guarantee for achieving system real-time performance.

[0023] In one embodiment of the present invention, the real-time data processing and image reconstruction module includes: The real-time data processing and image reconstruction module performs image reconstruction based on a physical information neural network. Constructing a measurement topology coding matrix ,in, For the set of real numbers, For sparse measurement quantity, The total dimension of the encoded vector; each line Corresponding to one excitation-measurement pair ,in, As the excitation electrode, For measuring electrodes; For the sparse measurement matrix, the first... One excitation-measurement pair Its topological encoding vector Calculated according to preset encoding rules, specifically: Normalized inscribed angle difference encoding: ,in, This represents the angular difference between the two electrodes in the unfolded circular coordinate system; Normalized spatial distance coding: ,in, and Electrode and Coordinate vectors in three-dimensional space It is the L2 norm. This represents the maximum value of the Euclidean distance between any two electrodes in the array; Electrode position periodic encoding: for the first One electrode, ,in, This represents the total number of electrodes on the same-layer ring matrix. When the electrode pair spans multiple layers, add cross-layer relationship coding: ,in, For same-layer markers, if ,but ,otherwise, , This is a function representing the layer number where the electrode is located. This represents the maximum layer number difference in the sensor array. final The concatenated vector, where, and Electrodes and Periodic encoding.

[0024] Specifically, when the system starts, the three-dimensional coordinates of all electrodes, their layer numbers, and the total number of electrodes in the same layer are preloaded. Global parameters are calculated and stored: the maximum Euclidean distance and the maximum layer number difference between all electrode pairs. For each sparse measurement dataset obtained in each real-time acquisition cycle, the module, based on the currently active sparse measurement strategy (i.e., which stimulus-measurement pairs were specifically used), wherein the th... Each measurement pair dynamically generates its topology encoding vector. : Calculate relative angle encoding : Reading electrodes and Circumferential angles on its respective ring array and Calculate the difference and normalize ; Computing spatial distance encoding : Obtain the three-dimensional coordinates of the two electrodes and Calculate its Euclidean distance And normalize it; Obtain position periodic encoding: based on electrode serial number and You can directly look up the table or calculate its periodic code. and ; Determine and generate inter-layer relationship codes :Compare and If they are equal, then ,otherwise, ; Calculate the normalized stratification difference , for[ The vector formed by the layer difference; The codes calculated above are concatenated in a fixed order to form the final code. ; Traverse all M stimulus-measurement pairs in the current policy, repeat the encoding vector generation step, generate the corresponding M encoding vectors, and stack these vectors as row vectors to form a measurement topology encoding matrix; Organize the sparse boundary voltage data collected in this cycle into a vector. ,Will With matrix The flattened one-dimensional vectors are concatenated to form a hybrid feature vector, which is then fed into a pre-trained physical information neural network to initiate the image reconstruction process.

[0025] In one embodiment of the present invention, the physical information neural network is a variant of the U-Net architecture, whose encoder input is a sparse boundary voltage vector. With topological encoding matrix The flattened and concatenated feature vectors, where, The output of step d A vector composed of measured values; The physical information neural network is trained using a knowledge distillation framework: the teacher network is trained with a complete measurement dataset, and the student network is trained with the intermediate layer features of the teacher network as supervision signals. Its loss function is a weighted sum of reconstruction loss and feature alignment loss. Each upsampling layer in the network decoder is followed by a physical residual correction module, which estimates the residual based on the conductivity distribution output of the current layer. Using differentiable forward modeling as input, the boundary voltage is predicted. Calculate measurement residuals The measurement residual is obtained by using a small back-projection network. Mapped to distributed correction quantity And fused with decoder features; The output layer of the neural network has a multi-task head structure, which simultaneously outputs a conductivity distribution image and a preliminary leakage area confidence heatmap.

[0026] Specifically, a finite element simulation model with geometric parameters consistent with the sensor array is constructed. Multiple leakage source locations, sizes, and conductivity distributions are set as real labels. In the simulation environment, a full combination of excitation-measurement modes is executed to obtain a complete boundary voltage dataset without missing data. A standard U-Net is constructed as the teacher network, using the complete boundary voltage dataset as input and the corresponding real conductivity distribution as the supervision label. It is trained using loss functions such as mean squared error (MSE) until convergence. This network learns a direct mapping from complete data to high-precision images. A variant of the U-Net is constructed as the student network, with its encoder input being a concatenation of sparse measurement data and its corresponding topological coding matrix. When training the student network, the parameters of the teacher network are fixed, and the loss function of the student network is designed as follows: in, Predicting images for student networks With real images The reconstruction losses between Features of the intermediate layer of the student network encoder Intermediate layer features corresponding to the teacher network Feature alignment loss between and The weighting coefficients are determined using the analytic hierarchy process (AHP) in this embodiment. and The value of is 0.5. Through this joint loss, the student network is forced to imitate the rich feature representation learned by the teacher network while directly learning the reconstruction, thereby making up for the lack of information in the sparse input. The trained student network is deployed in the real-time module. During online operation, for each input cycle, the network performs forward propagation as follows: the concatenated feature vector is downsampled by the encoder to extract features, and then upsampled by the decoder to gradually reconstruct the image; after each upsampling layer of the decoder (e.g., layer l), the following steps are performed: the preliminary conductivity distribution estimate of the layer output is input into a lightweight, differentiable ERT forward modeling layer, which internally implements fast, differentiable electric field calculations based on sensitivity matrices or finite element methods, and outputs the predicted boundary voltage; the residual between the current predicted voltage and the actual measured voltage is calculated; the residual is input into a small back-projection network that is paired with the current layer. This network is a shallow fully connected network or a convolutional network, and its training objective is to map the voltage residual to a correction of the conductivity distribution. The positive values ​​make the initial conductivity distribution estimate and the correction amount for the conductivity distribution closer to the true distribution. The network parameters are trained together with the main network in the offline stage. The calculated correction amount is transformed into features with the same number of channels as the current decoder feature map through a convolutional layer. Then, it is added element-wise or concatenated channel by channel with the original decoder features, and then subsequent upsampling operations are performed. This process injects the physical consistency constraint into the reconstruction process in a learnable way in real time. After multiple layers of decoding and correction, the output layer of the network is divided into two heads: Reconstruction head: a convolutional layer that outputs the final conductivity distribution image; Confidence head: another convolutional layer with sigmoid activation, which outputs a preliminary leakage area confidence heatmap of the same size as the reconstructed image, where each pixel value represents the probability that the location belongs to the leakage anomaly area.

[0027] In one embodiment of the present invention, the construction and updating of the spatiotemporal dual-domain leakage probability field model includes: Modeling a continuous T-frame difference image sequence as a spatiotemporal graph , where nodes For the first Superpixel regions in a frame image, edges Connect nodes in adjacent frames that are spatially close or have similar features; Each node calculates the dynamic propagation kernel. ,in, This is a normalized exponential function along the row direction. For nodes eigenvectors, Its spatiotemporal neighborhood nodes eigenmatrix and For the parameter matrix, The dimension of the feature vector; Initial leakage probability field Generated based on single-frame anomaly detection results, using iterative formulas. ,in, Let be the probability field vector. For spacetime diagram The adjacency matrix, For all The propagation kernel matrix constructed by node index. Index for iteration count, The Hadamard product of the matrices. It is the attenuation factor; After each iteration, the probability field is compared with the prior probabilities based on the structure's location. By multiplying the elements and normalizing, we obtain the probability field with physical constraints.

[0028] Specifically, for the latest T consecutive frames of differential conductivity images output by the real-time processing module, each frame is segmented using a superpixel segmentation algorithm (e.g., the SLIC algorithm) to divide the image into a group of connected regions with similar conductivity change characteristics. Each region is treated as a node, and its feature vector is composed of statistical quantities such as the average conductivity change value, area, shape factor, and spatial centroid coordinates of the region. Between two consecutive frames (frame t and frame t+1), for each node in the previous frame, the K nodes (e.g., K=3) that are closest to its spatial centroid (within a certain radius) are found in the next frame, or the K nodes with the highest feature vector similarity (cosine similarity) are found, and directed edges are established to connect them. At the same time, edges are also established between spatially adjacent superpixel nodes in the same frame to simulate spatial diffusion. The parameter matrix is ​​obtained through system pre-training or online fine-tuning. and For each node in the current spatiotemporal graph The feature vector is enhanced by passing it through a shared feature encoding network to obtain the query vector. Simultaneously, collect all its spatiotemporal neighbor nodes. Enhanced features This constructs the neighbor feature matrix and calculates the key vector matrix. Compute nodes to each of its neighbor nodes Attention score: ,in, for Corresponding neighbors OK, Feature dimensions, The vector is transposed; the scores of all neighbors are then Softmax normalized to obtain the dynamic propagation kernel vector. Its elements This indicates that in this update, from the node Spread to neighbors The probability weights; Based on the preliminary leakage confidence heatmap of the latest frame image (from the image reconstruction module), it is mapped onto the nodes of the spatiotemporal graph. For example, the average confidence of the heatmap of the region where the node is located, after being transformed by a sigmoid function, is used as the initial probability of that node, forming the initial probability field vector; a sparse adjacency matrix is ​​constructed according to the adjacency relationship of the spatiotemporal graph. (If nodes i and j are connected by an edge, then) Otherwise, ); dynamic propagation kernel of all nodes Organized into a propagation kernel matrix according to its node index. ( and Having the same sparse structure, non-zero element values ​​are corresponding ); Set the weakening factor (For example, 0.08) In each iteration, the value is updated according to the iterative formula, where... This represents the application of propagation kernel weights to adjacency relationships. The multiplication of this matrix with the probability field vector simulates the process of probability spreading along the edges of the graph according to the learned weights. This term is used to preserve initial observation information and prevent the original signal from being overly smoothed during the diffusion process; After each iteration, the updated probability field is multiplied element-wise with a predefined prior probability field. The prior probability field is generated based on the 3D CAD model of the structure and engineering knowledge. For example, higher prior probabilities are assigned to locations such as pipe welds, corrosion-prone areas at the bottom, and historical defect points, while lower probabilities are assigned to areas with intact structures. After multiplication, the probability field is normalized to ensure that the total probability sum is 1, thus obtaining the updated probability field after physical constraints. After several iterations (e.g., 3-5 times), the probability field tends to stabilize. The final probability field is used as the spatiotemporal dual-domain leakage probability estimate at the current moment. This probability field can serve as an overall situation map of leakage and provide important spatial prior information for the subsequent evidence fusion and localization module (high-probability areas are the key search areas). The system then adds the latest differential image frame to the sequence, removes the oldest frame, updates the spatiotemporal map, and begins the probability field calculation for the next cycle.

[0029] In one embodiment of the present invention, solving the inverse problem with measurement data as boundary conditions includes: Each excitation-measurement pair is abstracted as a virtual sensor, which measures any spatial point within the imaging region. The proposition that a source of leakage exists Provide evidence; For the A virtual sensor, whose actual measured voltage value is... The virtual sensor for the proposition Basic probability assignment Calculated according to preset assignment rules; The preset assignment rules are as follows: ,in, In the first Under certain observations, the leakage proposition is supported. Basic probability assignment, To support normalization and confidence coefficients, For the first The measured value obtained from this measurement Assuming leakage proposition When it is established, the first term calculated by the forward model The theoretical voltage value of each sensor, For the first Standard deviation of measurement noise at each observation point; ,in, In the first Under observation, the proposition against leakage is... Basic probability assignment, To oppose the normalization constant and confidence coefficient; ,in, For the proposition Uncertainty.

[0030] Specifically, the internal and near-surface areas of the structure to be monitored (pipe section, tank wall) are discretized into a regular grid in three-dimensional space to obtain N candidate spatial points. , : For each candidate spatial point Assuming a leakage source of standard unit strength exists at this point (represented by a local conductivity abrupt change model), the theoretical boundary voltage response of this assumed leakage source to each excitation-measurement pair (i.e., each virtual sensor) is calculated using a pre-calibrated, quickly searchable forward model (e.g., based on sensitivity matrices or pre-stored fundamental solution data). ,in, , For the total number of virtual sensors (i.e., the number of excitation-measurement pairs actually used in the sparse measurement strategy), all An M*N dimensional forward response database is constructed; through system calibration or long-term measurements under leak-free steady-state conditions, the measurement noise standard deviation of each virtual sensor and its historical reliability index are statistically obtained for dynamic determination. and (For example, a sensor with a high signal-to-noise ratio gives higher performance.) and lower ); Within each real-time monitoring cycle, after completing data acquisition and obtaining the current sparse measurement vector, for each candidate point in the imaging space, the following loop is executed: extract the corresponding point from the forward modeling database. The theoretical response vectors of all M virtual sensors For the first One virtual sensor: calculates the current measured value. Compared with theoretical value The square of the difference, according to the formula Calculate the sensor's support for "point" The quality of evidence for the proposition that "it is the source of leakage" Used for normalization to ensure the result is within a reasonable probability range, according to the formula. Calculate the quality of evidence from the sensor that opposes the proposition. This reflects the sensor's tendency towards negative propositions and can be set according to the noise level; calculation This represents the uncertainty that the sensor cannot definitively determine regarding the proposition; after completing the calculations for all M sensors, the result for the spatial point is obtained. A set of evidence , of which each It includes three focal elements Assigning a value; Traverse all N spatial candidate points Repeating the above steps, the system eventually generates a three-dimensional evidence field covering the entire imaging space, with each spatial point... The system stores the "views" (evidence) of all virtual sensors on whether a point is a leakage source. This three-dimensional evidence field serves as core intermediate data and is then passed to the subsequent evidence fusion and localization modules for further processing.

[0031] In one embodiment of the present invention, the fusion and localization of evidence further includes: Calculate any two virtual sensors and Point The distance between the evidence and Jousselme :like >Preset distance threshold, then the sensor and Marked as at point There is a significant conflict at this point; based on the conflict relationship, each point... The sensors are dynamically divided into highly consistent subgroups. and high-conflict subgroups ; right A modified weighted Dempster rule is used for fusion, where the weight of each sensor is related to the deterministic entropy of its evidence. Proportional; to All evidence is retained as a set of competing hypotheses for that point. ; Search the entire imaging space and select Trust function after fusion highest These points are used as the candidate core point set. ;in, for Evidence of fusion; For each Its spatial neighborhood Find the point inside ,like There is evidence in and similarity If the evidence is greater than or equal to a preset similarity threshold, then the evidence will be... As support The supplementary evidence was integrated and updated. To enhance the trust function Output highest point Location of the leakage source; With point Centered on, according to The spatial gradient is used to determine the confidence region, where the mean of the similarity function at each point within the confidence region is greater than or equal to a preset confidence threshold. Mean, where, For a plausible function at a point The value, For the point The final fusion result of all available evidence at the location.

[0032] Specifically, for each candidate point in the imaging space, a set of evidence from M virtual sensors is received from the previous module (evidence generation); For any two sensor evidences at that point and Calculate its Jousselme distance ,in, and It is a three-dimensional vector representing the assignment of three focal elements. It is a 3x3 matrix with 0 elements on its diagonal (same proposition). (Opposing propositions) (and uncertain propositions); Set a distance threshold (e.g., 0.5) for all sensor pairs. ,like If the distance threshold is exceeded, the sensor is marked. and At point There is a significant conflict at this point; based on the conflict marker, the point... All sensors at the location are divided into two subsets: a high-consistency subset, which contains the set of sensors with the fewest conflicting labels among themselves (this can be achieved by finding the maximum clique or a community detection algorithm based on the conflict graph), where the consensus of evidence is high; and a high-conflict subset, which contains the remaining sensors where there are significant conflicts among the evidence. A modified weighted Dempster combination rule is used for fusion: First, weights are assigned to each sensor within the group. This weight is inversely proportional to the certainty entropy of its evidence, i.e. The lower the deterministic entropy, the clearer the evidence (less uncertainty), and the higher its weight. Then, the sensor evidence is weighted and averaged using this weight, followed by a standard Dempster combination operation (pairwise recursive fusion) to obtain the fused evidence for the consensus group. ; Calculate its trust function ; For evidence in highly conflicting subgroups, no forced fusion is performed; all of them are preserved to form a point. A set of competing hypotheses, each hypothesis representing a “view” about the possible states of a point, supported by partially conflicting sensors; Search and select a trust function from N candidate points in the entire imaging space. The top K points (K=3) with the highest values ​​constitute the candidate core point set; for each candidate core point... Define its spatial neighborhood, for example, with Centered on, All other candidate points within a spherical region of radius ; traverse each point in the neighborhood. Examine each piece of evidence in its competing hypothesis set; calculate the similarity between this evidence and the fusion result of the current core point's consensus group. The similarity can be obtained by calculating the cosine similarity or 1-normalized Euclidean distance between the two evidence vectors (support, opposition, uncertainty); if there exists evidence such that the similarity between the fusion results is ≥ a preset similarity threshold (e.g., 0.8), then the evidence is considered valid even though it originates from the conflict group and is within the consensus set. Point calculation, but its "view" is actually different from... The mainstream consensus on this point is highly consistent; therefore, the arbitration procedure should be performed: remove this evidence from its competing hypothesis set and use it as supplementary evidence, in conjunction with... The original consensus group fusion results are combined again by Dempster to update the enhanced fusion evidence, and a new enhanced trust function is calculated. The pre-set reliability threshold is determined based on a fixed value derived from experience or theoretical analysis.

[0033] Please see Figure 2 As shown, this invention is a real-time leakage location and monitoring method based on electrical resistivity tomography, comprising the following methods: S1: A sensor array consisting of multiple electrodes fixed to the surface of the structure under test in a specific spatial topology. S2: Output excitation current to the selected electrode pair according to the preset excitation-measurement mode, and at the same time measure the boundary voltage signal between other electrode pairs and convert the measured analog signal into a digital signal; S3: Receive the digital signal, construct a measurement topology coding matrix, concatenate the sparse boundary voltage vector with the measurement topology coding matrix, use it as input to the physical information neural network, and generate a differential image sequence of conductivity changes in the measured area. S4: Construct a spatiotemporal dual-domain leakage probability field model, analyze the dynamic evolution characteristics of abnormal regions in the differential image sequence, combine the prior geometric constraints of the measured structure, solve the inverse problem with the measurement data as boundary conditions, identify spatiotemporal patterns that conform to the physical diffusion law of leakage, and trigger an alarm when the similarity exceeds the preset similarity threshold and calculate and output the three-dimensional spatial coordinates and intensity estimate of the leakage source. S5: A visual interface that displays real-time location results, monitoring information, and alarm information.

[0034] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A real-time leakage location monitoring system based on electrical resistivity tomography, characterized in that, Includes the following modules: Sensor array module: contains multiple electrodes fixed to the surface of the structure under test in a specific spatial topology; Multi-channel data acquisition and excitation module: Outputs excitation current to selected electrode pairs according to preset excitation-measurement modes, while measuring boundary voltage signals between other electrode pairs and converting the measured analog signals into digital signals; Real-time data processing and image reconstruction module: Receives the digital signal, constructs a measurement topology coding matrix, concatenates the sparse boundary voltage vector with the measurement topology coding matrix, uses it as input to the physical information neural network, and generates a differential image sequence of conductivity changes in the measured area; Leakage intelligent identification and location module: Constructs a spatiotemporal dual-domain leakage probability field model, analyzes the dynamic evolution characteristics of abnormal regions in the differential image sequence, combines the prior geometric constraints of the measured structure, solves the inverse problem with the measurement data as boundary conditions, identifies spatiotemporal patterns that conform to the physical diffusion law of leakage, and triggers an alarm when the similarity exceeds the preset similarity threshold and calculates and outputs the three-dimensional spatial coordinates and intensity estimate of the leakage source. Human-computer interaction and data management module: A visual interface that displays real-time location results, monitoring information and alarm information.

2. The real-time leakage location and monitoring system based on resistivity tomography according to claim 1, characterized in that, The specific spatial topology is a multi-layer nested concentric ring array; the multi-layer nested concentric ring array is composed of two or more ring electrode arrays with different diameters and coincident central axes, and the electrodes on the same layer of the ring array are distributed at equal intervals; the ring arrays of different layers are arranged along the axial direction of the pipe or tank being tested.

3. The real-time leakage location monitoring system based on resistivity tomography according to claim 1, characterized in that, The multi-channel data acquisition and excitation module includes: a . The mode controller divides the electrode pairs in the sensor array into subsets of excitation electrode pairs according to a preset sparse measurement strategy. b. Based on a pre-calculated sensitivity matrix The optimization algorithm selects a subset of combinations from all possible stimulus-measurement pair combinations that optimize the condition number of the reconstruction matrix. c. For each selected excitation electrode pair, a constant current source outputs an excitation current, and at the same time, a multiplexer performs voltage measurement on the measurement electrode pair in the combination selected in step b. d. Output the boundary voltage dataset defined by the sparse measurement strategy.

4. The real-time leakage location and monitoring system based on resistivity tomography according to claim 1, characterized in that, The real-time data processing and image reconstruction module includes: The real-time data processing and image reconstruction module performs image reconstruction based on a physical information neural network. Constructing a measurement topology coding matrix ,in, For the set of real numbers, For sparse measurement quantity, The total dimension of the encoded vector; each line Corresponding to one excitation-measurement pair ,in, As the excitation electrode, For measuring electrodes; For the sparse measurement matrix, the first... One excitation-measurement pair Its topological encoding vector Calculated according to preset encoding rules, specifically: Normalized inscribed angle difference encoding: ,in, This represents the angular difference between the two electrodes in the unfolded circular coordinate system; Normalized spatial distance coding: ,in, and Electrode and Coordinate vectors in three-dimensional space It is the L2 norm. This represents the maximum value of the Euclidean distance between any two electrodes in the array; Electrode position periodic encoding: For the first One electrode, ,in, This represents the total number of electrodes on the same-layer ring matrix; When the electrode pairs span multiple layers, add cross-layer relationship coding: ,in, For same-layer markers, if ,but ,otherwise, , This is a function representing the layer number where the electrode is located. This represents the maximum layer number difference in the sensor array. final The concatenated vector, where, and Electrodes and Periodic encoding.

5. The real-time leakage location and monitoring system based on resistivity tomography according to claim 4, characterized in that, The physical information neural network is a variant of the U-Net architecture, whose encoder input is a sparse boundary voltage vector. With topological encoding matrix The flattened and concatenated feature vectors, where, The output of step d A vector composed of measured values; The physical information neural network is trained using a knowledge distillation framework: the teacher network is trained with a complete measurement dataset, and the student network is trained with the intermediate layer features of the teacher network as supervision signals. Its loss function is a weighted sum of reconstruction loss and feature alignment loss. Each upsampling layer in the network decoder is followed by a physical residual correction module, which estimates the residual based on the conductivity distribution output of the current layer. Using differentiable forward modeling as input, the boundary voltage is predicted. Calculate measurement residuals The measurement residual is obtained by using a small back-projection network. Mapped to distributed correction quantity And fused with decoder features; The output layer of the neural network has a multi-task head structure, which simultaneously outputs a conductivity distribution image and a preliminary leakage area confidence heatmap.

6. The real-time leakage location and monitoring system based on resistivity tomography according to claim 1, characterized in that, The construction and updating of the spatiotemporal dual-domain leakage probability field model includes: Modeling a continuous T-frame difference image sequence as a spatiotemporal graph , among which, nodes For the first Superpixel regions in a frame image, edges Connect nodes in adjacent frames that are spatially close or have similar features; Each node calculates the dynamic propagation kernel. ,in, This is a normalized exponential function along the row direction. For nodes eigenvectors, Its spatiotemporal neighborhood nodes eigenmatrix and For parameter matrices, The dimension of the feature vector; Initial leakage probability field Generated based on single-frame anomaly detection results, using iterative formulas. ,in, Let be the probability field vector. For spacetime diagram The adjacency matrix, For all The propagation kernel matrix constructed by node index. Index for iteration count, The Hadamard product of the matrices. It is the attenuation factor; After each iteration, the probability field is compared with the prior probabilities based on the structure's location. By multiplying the elements and normalizing, we obtain the probability field with physical constraints.

7. The real-time leakage location monitoring system based on resistivity tomography according to claim 1, characterized in that, The solution to the inverse problem with measurement data as boundary conditions includes: Each excitation-measurement pair is abstracted as a virtual sensor, which measures any spatial point within the imaging region. The proposition that a source of leakage exists Provide evidence; For the A virtual sensor, whose actual measured voltage value is... The virtual sensor for the proposition Basic probability assignment Calculated according to preset assignment rules; The preset assignment rules are as follows: ,in, In the first Under certain observations, the leakage proposition is supported. Basic probability assignment, To support normalization and confidence coefficients, For the first The measured value obtained from this measurement Assuming leakage proposition When it is established, the first term calculated by the forward model The theoretical voltage value of each sensor, For the first Standard deviation of measurement noise at each observation point; ,in, In the first Under observation, the proposition against leakage is... Basic probability assignment, To oppose the normalization constant and confidence coefficient; ,in, For the proposition Uncertainty.

8. The real-time leakage location monitoring system based on resistivity tomography according to claim 7, characterized in that, The fusion and localization of the evidence also includes: Calculate any two virtual sensors and Point The distance between the evidence and Jousselme :like >Preset distance threshold, then the sensor and Marked as at point There is a significant conflict at this point; based on the conflict relationship, each point... The sensors are dynamically divided into highly consistent subgroups. and high-conflict subgroups ; right A modified weighted Dempster rule is used for fusion, where the weight of each sensor is related to the deterministic entropy of its evidence. Proportional; to All evidence is retained as a set of competing hypotheses for that point. ; Search the entire imaging space and select the fused trust function. highest These points are used as the candidate core point set. ;in, for Evidence of fusion; For each Its spatial neighborhood Find the point inside ,like There is evidence in and similarity If the evidence is greater than or equal to a preset similarity threshold, then the evidence will be... As support The supplementary evidence was integrated and updated. To enhance the trust function Output highest point Location of the leakage source; With point Centered on, according to The spatial gradient is used to determine the confidence region, where the mean of the similarity function at each point within the confidence region is greater than or equal to a preset confidence threshold. Mean, where, For a plausible function at a point The value, For the point The final fusion result of all available evidence at the location.

9. A real-time leakage location monitoring method based on resistivity tomography, used to implement the real-time leakage location monitoring system based on resistivity tomography as described in any one of claims 1-8, characterized in that, Includes the following steps: S1: A sensor array consisting of multiple electrodes fixed to the surface of the structure under test in a specific spatial topology. S2: Output excitation current to the selected electrode pair according to the preset excitation-measurement mode, and at the same time measure the boundary voltage signal between other electrode pairs and convert the measured analog signal into a digital signal; S3: Receive the digital signal, construct a measurement topology coding matrix, concatenate the sparse boundary voltage vector with the measurement topology coding matrix, use it as input to the physical information neural network, and generate a differential image sequence of conductivity changes in the measured area. S4: Construct a spatiotemporal dual-domain leakage probability field model, analyze the dynamic evolution characteristics of abnormal regions in the differential image sequence, combine the prior geometric constraints of the measured structure, solve the inverse problem with the measurement data as boundary conditions, identify spatiotemporal patterns that conform to the physical diffusion law of leakage, and trigger an alarm when the similarity exceeds the preset similarity threshold and calculate and output the three-dimensional spatial coordinates and intensity estimate of the leakage source. S5: A visual interface that displays real-time location results, monitoring information, and alarm information.

Images