Landfill leakage detection method and system based on boundary double-voltage electric field

Abstract

The invention discloses a landfill leakage detection method and system based on a boundary double-voltage electric field, and the method comprises the steps: building an artificial electric field which covers the whole area of a landfill along a boundary measuring line electrode at the periphery of the landfill based on the high-resistance characteristic of an impermeable film and the conductive characteristic of leachate, and collecting the boundary potential and electric field data through the high and low double-voltage artificial electric field, the method comprises the following steps: constructing a boundary survey line data pair of physical alignment through differential denoising and normalization preprocessing, adopting an encoder-decoder architecture based on 1D-CNN, fusing spatial transformation and a Focal Loss loss function, directly learning a mapping relation between a boundary survey line potential anomaly mode and an internal leakage position, and outputting a leakage probability distribution diagram of a corresponding landfill contour. The method does not need to excavate and pre-embed the landfill pile body, the positioning precision reaches + / -2m, the single detection time consumption is less than or equal to 4h, the comprehensive cost is reduced by 70%, the electrode reuse rate is greater than or equal to 90%, the method is suitable for landfill sites with various shapes and various stages, and efficient, accurate and low-cost detection of the leakage area is realized.

Description

Technical Field

[0001] This invention discloses a method and system for detecting landfill leakage based on a boundary dual-voltage electric field, which belongs to the application of electrical detection and artificial intelligence in landfill leakage detection. Background Technology

[0002] Landfilling has always been a crucial method for the final disposal of solid waste. To control the risk of landfill leakage, solid waste landfills typically use HDPE membranes laid at the bottom as a protective barrier between the landfill and the underlying soil. However, these anti-seepage systems may leak leachate due to aging, mechanical damage, or construction defects during long-term operation, causing serious pollution to groundwater and soil, threatening the ecological environment and public health. Currently developed and deployed landfill leakage detection technologies mainly include electrochemical induction cable methods, electrode grid methods, dipole methods, and electrical methods. However, given the large area of ​​landfills and the small leakage points, traditional methods require laying electrode cables over a large area, resulting in large construction workloads, high costs, and low efficiency.

[0003] Chinese patent application CN202310461543.2 discloses a method and apparatus for locating HDPE membrane leakage points using off-site electrodes in a landfill. The method involves parallel electrode placement on the near-ground surface beneath the HDPE membrane on the landfill's outer slope; then, electro-optical imaging is performed on the natural foundation layer beneath the membrane, using the symmetrically placed electrodes as both power and detection electrodes to analyze the electrical parameters of the underlying medium; finally, based on the electro-optical imaging results, inversion imaging is performed on low-resistivity anomaly areas to pinpoint the HDPE membrane damage location. However, this method only uses parallel electrode placement on the landfill's outer slope. Leakage points in the central area of ​​the landfill are calculated extensively, resulting in significant errors in leakage point calculations and an inability to accurately determine the leakage area's extent.

[0004] Therefore, developing a technology that can be used to delineate leakage areas in landfills and improve the efficiency of traditional methods has become a problem that needs to be solved. Summary of the Invention

[0005] The technical problem solved by this invention is to provide a landfill leakage detection method and system based on a boundary dual-voltage electric field, addressing the issues of large computational load and low detection accuracy in existing HDPE membrane leakage detection methods for landfills.

[0006] This invention is achieved using the following technical solution: This invention first discloses a landfill leakage detection method based on a boundary dual-voltage electric field, comprising the following steps: S1. Establish a stable artificial electric field on both sides of the geomembrane in the landfill; S2. A rectangular boundary survey line is arranged around the outer perimeter of the landfill body. Several detection electrodes are arranged at equal intervals along the rectangular boundary survey line. All detection electrodes are connected to the detection host through shielded cables. The potential values ​​of each detection electrode are collected under low voltage artificial electric field and high voltage artificial electric field respectively. S3. Calculate the electric field strength between adjacent detection electrodes under low-voltage artificial electric field and high-voltage artificial electric field by using the potential value of each detection electrode. Combine the difference between the potential value and the electric field strength under low-voltage artificial electric field and high-voltage artificial electric field to remove the background electric field noise of the landfill. S4. The landfill outline is gridded using the detection electrodes on the rectangular boundary survey line as connection points. Smooth potential data of the detection electrodes on two intersecting boundary survey lines are selected to construct a one-dimensional sequence. The sequence index of the one-dimensional sequence is aligned with the horizontal and vertical coordinates of the internal grid of the landfill to form a physically aligned detection data pair. S5. The detection data is input into the encoder-decoder model based on 1D-CNN, and the local abnormal features of the potential data on the boundary measurement line are extracted by the dual-branch 1D-CNN to output the leakage probability distribution map. S6. Set a probability threshold and automatically delineate areas above the threshold in the leakage probability distribution map as suspected leakage areas, and output the location coordinates of the suspected leakage areas in the landfill.

[0007] In the landfill leakage detection method based on boundary dual voltage electric field of the present invention, further, in step S1, an adjustable DC power supply is used to establish a detection circuit with voltage of 100-500V and current ≤10mA on both sides of the landfill's impermeable membrane.

[0008] In the landfill leakage detection method based on boundary dual voltage electric field of the present invention, further, in step S2, the potential values ​​of each detection electrode on the boundary measuring line are first collected by a high-density electrical resistivity tomography instrument under a 100V low voltage artificial electric field, keeping the detection electrode layout unchanged, and then the potential values ​​of the same group of detection electrodes are collected under a 500V high voltage artificial electric field.

[0009] In the landfill leakage detection method based on boundary dual-voltage electric field of the present invention, further, in step S3, the electric field strength between the detection electrode and the adjacent detection electrode is calculated under both low-voltage and high-voltage artificial electric fields. and : , ; in, , Index positions , The potential values ​​of two adjacent detection electrodes under a high-voltage artificial electric field. , Index positions , The potential values ​​of two adjacent detection electrodes under a high-voltage artificial electric field. The distance between adjacent detection electrodes; Then, the potential anomaly difference between the detection electrode and the adjacent detection electrode is calculated separately. and the abnormal difference in electric field strength Removal of background electric field noise from landfills: , ; in, , index position The potential values ​​of the detection electrode under low-voltage and high-voltage artificial electric fields, respectively. , index position The electric field strength between the detection electrode and the adjacent detection electrode under low-voltage artificial electric field and high-voltage artificial electric field, respectively.

[0010] In the landfill leakage detection method based on boundary dual voltage electric field of the present invention, the smooth potential data in step S4 is further obtained by normalizing the potential value data after removing the background electric field noise of the landfill, and then obtaining the smooth potential data of each detection electrode by moving average filtering.

[0011] In the landfill leakage detection method based on boundary dual voltage electric field of the present invention, the smoothed potential data is further obtained by three-point moving average filtering: , in, index position The detection electrode smoothing potential data, , , Index positions , , The normalized potential values ​​of the three adjacent detection electrodes are used to replace the potential values ​​of the adjacent detection electrodes on the missing side by performing a three-point moving average filtering.

[0012] In the landfill leakage detection method based on boundary dual voltage electric field of the present invention, the encoder-decoder model in step S5 further includes a decoder, a feature fusion and spatial mapping unit, and a decoder. The encoder includes two 1D-CNN branches with identical structures and independent parameters. Each 1D-CNN branch contains 3-5 convolutional layers, ReLU activation function, and 1D max pooling layer, which respectively process the smooth potential data sequences corresponding to the two boundary test lines in the detection data pair, and finally convert them into fixed-dimensional feature vectors corresponding to the smooth potential data sequences of the two boundary test lines. The feature fusion and spatial mapping unit concatenates the two feature vectors converted by the encoder into a fused feature vector, and maps it into a low-resolution two-dimensional feature map through a fully connected layer. The decoder uses a transposed convolutional layer to upsample the low-resolution two-dimensional feature map, and fuses the encoder features with the decoder upsampled features through skip connections. Finally, it outputs a leakage probability distribution map with the same size as the landfill grid through a sigmoid activation function.

[0013] In the landfill leakage detection method based on boundary dual voltage electric field of the present invention, the encoder-decoder model in step S5 is further trained through the following process: A training dataset was constructed using a combination of simulated and measured data. The simulated data was used to build an electric field model of the landfill using COMSOL, setting leakage points of different locations and sizes to generate boundary potential data containing leakage points. The measured data collected boundary potential data from historical leakage detection in the landfill. Boundary potential data samples at leakage point locations were positive samples, while boundary potential data samples at non-leakage locations were negative samples. Add noise, shift, and scale the training dataset to increase its size. The training dataset is input into the initialized 1D-CNN-based encoder-decoder model for training, using Focal Loss as the loss function: , in, The loss is the prediction result of the samples in the training dataset. It is the probability that the model predicts for the target class of samples in the training dataset. γ These are the positive and negative sample mining coefficients in the loss function. The weights are positive and negative samples, with positive samples having a greater weight than negative samples. Negative samples are easier to classify. The loss is suppressed when the positive samples are difficult to classify, tending towards 1. As the value approaches zero, the loss is amplified. The training process optimizes the model using Adam; IoU, Precision, and Recall were used as evaluation metrics. When IoU ≥ 0.7, the training of the encoder-decoder model was stopped and the optimal model was saved.

[0014] This invention also discloses a system for implementing the above-mentioned landfill leakage detection method based on boundary dual voltage electric field, comprising: An electric field excitation module includes an excitation power supply, the positive and negative poles of which are respectively connected to the inner and outer sides of the landfill geomembrane to establish a stable artificial electric field on both sides of the landfill geomembrane. The boundary electrode module arranges rectangular boundary survey lines around the outer perimeter of the landfill body, and arranges several detection electrodes at equal intervals along the rectangular boundary survey lines. All detection electrodes are connected to the detection host through shielded cables, and the potential values ​​of each detection electrode are collected under low voltage artificial electric fields and high voltage artificial electric fields respectively. The detection host includes a data processing module, which calculates the electric field strength between adjacent detection electrodes under low-voltage and high-voltage artificial electric fields based on the potential values ​​of each detection electrode. It then removes background electric field noise from the landfill by performing differential calculations between the potential values ​​and electric field strengths under the low-voltage and high-voltage artificial electric fields. The landfill contour is gridded using the detection electrodes on the rectangular boundary survey line as connection points. Smooth potential data of the detection electrodes on two intersecting boundary survey lines are selected to construct a one-dimensional sequence. The sequence index of the one-dimensional sequence is aligned with the horizontal and vertical coordinates of the internal grid of the landfill to form physically aligned detection data pairs. The detection host also has a built-in encoder-decoder model based on 1D-CNN. The detection data is input to the encoder-decoder model based on 1D-CNN. The dual-branch 1D-CNN extracts local anomaly features of the potential data on the boundary measurement line and outputs a leakage probability distribution map. A probability threshold is set, and the areas in the leakage probability distribution map that are higher than the threshold are automatically delineated as suspected leakage areas. The location coordinates of the suspected leakage areas in the landfill are output.

[0015] In the landfill leakage detection system based on boundary dual voltage electric field of the present invention, the detection host collects the potential of each detection electrode through a high-density electrical resistivity meter, and converts the potential signals of all detection electrodes into digital signals through an ADC converter and a multiplexer and transmits them to the data processing module of the detection host.

[0016] In the landfill leakage detection system based on boundary dual voltage electric field of the present invention, the detection electrode is a copper electrode or an Ag / AgCl non-polarized electrode, which is inserted into the soil surrounding the landfill body at a depth of ≥0.5m and a spacing of 1-6m.

[0017] At the physical level, this invention utilizes the difference between the high resistance of the HDPE geomembrane and the high conductivity of the leachate to construct an artificial electric field inside and outside the geomembrane in the landfill. The leakage points form low-resistance channels, causing electric field distortion, and this distortion signal is significantly amplified under high voltage. Through dual-voltage acquisition and differential processing, the abnormal leakage potential signal is effectively separated from the background field, highlighting the abnormal leakage potential signal. Combined with moving average filtering and data normalization, the signal-to-noise ratio of the acquired potential data is significantly improved. Detection electrodes are arranged along the boundary survey line around the landfill outline. By setting up a ring of detection electrodes at the landfill boundary, full coverage of the landfill body can be achieved by deploying electrodes only at the landfill boundary, eliminating the need to lay electrodes inside the landfill. The electrode spacing can be flexibly adjusted to adapt to various landfill shapes, and the electrode reuse rate is ≥90%. At the data level, this invention physically aligns the potential data collected by the boundary survey line detection electrodes with the spatial coordinates inside the landfill. It strictly binds the one-dimensional sequence index of the detection electrodes on the intersecting boundary survey lines with the horizontal and vertical coordinates of the grid inside the landfill, constructing structured input detection data pairs, injecting physical prior knowledge, enhancing model interpretability and positioning accuracy, and avoiding complex physical field inversion. A dedicated 1D-CNN encoder-decoder model is designed to directly learn the nonlinear mapping of "boundary anomaly pattern → leakage location". A dual-branch 1D-CNN is used to specifically extract smooth potential data from the boundary survey line to construct local anomaly features of the one-dimensional sequence. The encoder-decoder achieves end-to-end conversion of "sequence → space" and jump connections to restore spatial details. The Focal Loss loss function is used to solve the problem of class imbalance of leakage samples during model training. By reducing the weight of easily classified samples without leakage through the modulation factor, the focus is on leakage point samples that are difficult to identify, solving the model bias problem caused by class imbalance and improving the ability to identify weak leakage signals.

[0018] The present invention, by adopting the above technical solution, has the following beneficial effects: (1) High efficiency and low cost: This invention only requires the deployment of detection electrodes at the boundary of the landfill, which can be deployed quickly. The time for a single detection is ≤4 hours, the overall cost is reduced by 70%, and the electrode reuse rate is ≥90%.

[0019] (2) It has strong non-destructive adaptability. This invention does not require excavation or pre-embedded equipment and is suitable for landfills in operation. It supports various shapes of sites such as circular and polygonal.

[0020] (3) High positioning accuracy: This invention achieves a positioning accuracy of ±2m by combining physical alignment with deep learning to fuse boundary line potential data. It is highly sensitive and can accurately identify weak leakage signals.

[0021] (4) High degree of automation: This invention achieves end-to-end model direct output of leakage probability map through the detection motor of the boundary survey line to the detection host, automatically delineates suspicious areas, and reduces manual intervention.

[0022] In summary, the landfill leakage detection method and system based on boundary dual-voltage electric field provided by this invention constructs a global electric field by deploying electrodes at the boundary, and combines dual-voltage differential denoising and end-to-end deep learning model to achieve efficient, accurate and low-cost positioning of landfill geomembrane leakage areas.

[0023] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the deployment of the landfill leakage detection system based on the boundary dual voltage electric field in the landfill in the embodiment.

[0025] Figure 2 This is a schematic diagram showing how the detection electrodes of the landfill boundary survey line are connected to the detection host in the embodiment.

[0026] Figure 3 This is a schematic diagram illustrating the detection principle of the landfill leakage detection system based on the boundary dual voltage electric field in the embodiment.

[0027] Figure 4 This is a diagram of the encoder-decoder model architecture based on 1D-CNN in the embodiment.

[0028] Figures 5a-5f The following is a chart showing the detection data from an actual testing scenario, in which... Figure 5a The potential value is measured under a low-voltage artificial electric field. Figure 5b The potential value is measured under a high-voltage artificial electric field. Figure 5c The potential data after differential denoising. Figure 5d The potential values ​​are normalized and filtered using a moving average. Figure 5e This is a masked representation of the actual leakage area in the landfill area. Figure 5f This is a probability map of the leakage area predicted by the model.

[0029] The numbers in the diagram are: 100-Geomembrane, 101-Leakage point, 200-Landfill body, 300-Excitation power supply, 301-Internal electrode of landfill, 302-External grounding electrode, 400-Boundary electrode module, 401-Detection electrode, 402-Shielded cable, 500-Detection host. Detailed Implementation Example

[0030] See Figure 1 and Figure 2The figure shows a detection system for implementing the landfill leakage detection method based on boundary dual voltage electric field of the present invention. The system includes an excitation power supply 300, an internal electrode 301 of the landfill, an external grounding electrode 302, a boundary electrode module 400, and a detection host 500. The excitation power supply 300 is connected to an internal landfill electrode 301 buried within the landfill pile 200. The negative terminal of the excitation power supply 300 is connected to an external grounding electrode 302 inserted into the surrounding soil outside the landfill, creating an artificial electric field on both sides of the geomembrane 100. The high resistance of the HDPE geomembrane prevents the transmission of the artificial electric field on both sides of the geomembrane. When the geomembrane 100 is damaged, the landfill leachate passes through the leakage point 101 of the geomembrane 100. The high conductivity of the leachate forms a low-resistance channel at the leakage point 101, causing electric field distortion on both sides of the geomembrane. The boundary electrode module 400 arranges several detection electrodes 401 around the outline of the landfill. All detection electrodes are connected to the detection host 500 through shielded cables 402. By collecting abnormal potential data of each detection electrode 401 and combining the position of the detection electrodes 401, the location and area of ​​the leakage point 101 of the geomembrane 100 are obtained.

[0031] Specifically, the steps of the landfill leakage detection method based on the boundary dual-voltage electric field in this embodiment are as follows: S1. Establish a stable artificial electric field on both sides of the geomembrane in the landfill. The excitation power supply 300 uses an adjustable DC power supply to establish a detection circuit with a voltage of 100-500V and a current ≤10mA on both the inside and outside of the geomembrane in the landfill. Alternatively, a low-frequency AC power supply of 0.1-1Hz can be used in conjunction with phase-sensitive detection (PSD) to suppress power frequency interference. Before collecting data for boundary potential detection, the excitation power supply first outputs a low voltage of 100V and provides continuous power for 30 minutes to ensure that the established artificial electric field is stably distributed in the landfill layer, HDPE membrane, and soil, avoiding the influence of polarization effects.

[0032] S2. A rectangular boundary survey line is arranged around the outer perimeter of the landfill body. Several detection electrodes 401 are arranged at equal intervals along the rectangular boundary survey line. All detection electrodes 401 are connected to the detection host 500 through shielded cables 402. The potential values ​​of each detection electrode are collected under low voltage artificial electric field and high voltage artificial electric field respectively.

[0033] When deploying the detection electrodes, each electrode, along with the corner points of the rectangular boundary survey line, is equidistantly distributed to ensure that the landfill stockpile within the rectangular boundary survey line can be uniformly gridded using the detection electrodes as connection points. The spacing between the detection electrodes is 1-6m, preferably 3m. The detection electrodes are made of copper rods or Ag / AgCl non-polarized electrodes, and are inserted vertically into the ground to a depth of ≥0.5m to ensure close contact with the soil.

[0034] Specifically, the excitation power supply 300 was first adjusted to establish a 100V low-voltage artificial electric field on both sides of the geomembrane. The potential values ​​of each detection electrode on the boundary survey line under the low-voltage artificial electric field were then collected using a GIM-12 high-density electrical resistivity tomography instrument. Record the position index of each electrode. x Then, the excitation power supply 300 was adjusted to establish a 500V high-voltage artificial electric field on both sides of the geomembrane. The arrangement of the detection electrodes remained unchanged. After stabilizing for 5 minutes, the potential values ​​of the same set of detection electrodes in the high-voltage artificial electric field were collected using a high-density electrical resistivity tomography (EMT) instrument. The loop current signal is simultaneously acquired through a current sampling resistor to assist in verifying the stability of the electric field.

[0035] S3. Calculate the electric field strength between adjacent detection electrodes under low-voltage and high-voltage artificial electric fields by using the potential values ​​of each detection electrode. Combine the difference between the potential values ​​and the electric field strength under low-voltage and high-voltage artificial electric fields to remove the background electric field noise of the landfill.

[0036] In this step, the electric field strength between each detection electrode and adjacent detection electrodes is calculated using the electric field strength calculation formula. According to electrostatic field theory, the electric field strength E is related to the potential difference. The relationship is: The negative sign indicates that the direction of the electric field is the same as the direction of the decrease in potential. The distance between adjacent detection electrodes is denoted as . The low-voltage background electric field intensity under a low-voltage artificial electric field is obtained. High voltage excitation electric field strength under high voltage artificial electric field , The position index of the i-th electrode is as follows: , .

[0037] in, , The index positions are respectively , The potential values ​​of two adjacent detection electrodes under a low-voltage artificial electric field. , The index positions are respectively , The potential values ​​of two adjacent detection electrodes under a high-voltage artificial electric field. This represents the spacing between adjacent detection electrodes.

[0038] Then, the potential anomaly difference between the detection electrode and adjacent detection electrodes between the low-voltage artificial electric field and the high-voltage artificial electric field is calculated. and the abnormal difference in electric field strength Remove background electric field noise from landfills. The specific calculations are as follows: , ; in, , index position The potential values ​​of the detection electrode under low-voltage and high-voltage artificial electric fields, respectively. , index position The electric field strength between the detection electrode and the adjacent detection electrode under low-voltage artificial electric field and high-voltage artificial electric field, respectively.

[0039] Under a low-voltage artificial electric field of 100V, the electric field mainly reflects the background field of the landfill area (the inherent resistance distribution of the soil, HDPE membrane, and waste layer), and the low-resistivity characteristics of the leakage point are not fully manifested. Includes background signal and noise : .

[0040] Under a 500V high-voltage artificial electric field, current can more easily pass through the low-resistance channels at the leak points of the geomembrane, thus detecting abnormal leakage signals. It was significantly enhanced, therefore Includes background signal Enhanced abnormal signals and noise : .

[0041] Where k is the enhancement coefficient. The conductivity and voltage level at the leakage point are determined by the leakage point's conductivity.

[0042] Subtracting the two equations gives the index position. The abnormal potential difference between the detection electrode and the adjacent detection electrode : .

[0043] Similarly, index position The abnormal difference in electric field strength between the detection electrode and the adjacent detection electrode for: .

[0044] The background signal was effectively canceled out through differential operation. and noise Only leakage anomaly signals are retained, thus achieving signal purification.

[0045] S4. The landfill outline is gridded using the detection electrodes on the rectangular boundary survey line as connection points. Smooth potential data of the detection electrodes on two intersecting boundary survey lines are selected to construct a one-dimensional sequence. The sequence index of the one-dimensional sequence is aligned with the horizontal and vertical coordinates of the internal grid of the landfill to form a physically aligned detection data pair.

[0046] The smoothed potential data of the detection electrodes are obtained as follows: the potential value data after removing the background electric field noise of the landfill is normalized, and then the smoothed potential data of each detection electrode is obtained by moving average filtering to suppress random noise.

[0047] First, the Z-score formula is used to analyze the potential anomaly difference. Normalization is performed to eliminate data magnitude differences caused by factors such as supply voltage and electrode spacing, resulting in normalized potential values. , This represents the mean of all abnormal differences in the potential of the detection electrodes. M represents the number of detection electrodes. The standard deviation of all abnormal differences in the detection electrode potentials. The potential anomaly difference data is converted into a distribution with a mean of 0 and a standard deviation of 1, which meets the input requirements of deep learning models.

[0048] Then, a three-point moving average filter is applied to the potential anomaly difference between the three adjacent detection electrodes: , in, index position The detection electrode smoothing potential data, , , Index positions , , The normalized potential values ​​of the three adjacent detection electrodes are used to perform a three-point moving average filtering. The smoothed potential data of the detection electrodes at both ends of the boundary survey line are selected by replacing the potential values ​​of the adjacent detection electrodes on the missing side with their own potential values. That is, for the detection electrode at the beginning of the boundary survey line, i=1, and... The detection electrode at the end of the boundary survey line is equal to M, and is taken as... This avoids missing boundary data, while smoothing out local fluctuations and preserving key anomalies such as peak values.

[0049] Two orthogonal boundary survey lines are selected from the rectangular boundary survey lines surrounding the landfill outline, such as the top boundary survey line and the left boundary survey line. The smoothed potential data of all detection electrodes on these lines are used as one-dimensional sequences. and The sequence indexes of each detection electrode are strictly aligned with the horizontal and vertical coordinates of the gridded area within the landfill contour. sequence index It directly corresponds to the x-coordinate of the grid inside the landfill. sequence index The detection data pairs directly correspond to the ordinate Y of the internal grid of the landfill, forming a physically aligned pair. Establish a strict mapping relationship between one-dimensional signals and two-dimensional space.

[0050] like Figure 3 As shown, the top boundary survey line of the landfill. The curve, with the locations of abnormal peaks marked as indicated by the arrows, is on the left boundary line. The curves, marked with arrows indicating the locations of abnormal peaks, are shown in blue as "suspected leakage areas" within the landfill grid, where the peak arrows of the two survey lines intersect.

[0051] S5. The detection data is input into a 1D-CNN-based encoder-decoder model, and local anomaly features of potential data on the boundary measurement line are extracted through a dual-branch 1D-CNN to output a leakage probability distribution map.

[0052] The encoder-decoder model in step S5 includes a decoder, a feature fusion and spatial mapping unit, and a decoder.

[0053] like Figure 4 As shown, the encoder includes two structurally identical but parameter-independent 1D-CNN branches. Each 1D-CNN branch contains 3-5 convolutional layers (kernel size 3-7, stride 1), a ReLU activation function, and a 1D max-pooling layer (kernel size 2, stride 2). The output formula of the convolutional layer is: , l Where k is the number of convolutional layers and k is the number of channels. For the first l The convolution kernel of the k-th channel of the layer, size , This represents the number of channels in the previous layer. This is a one-dimensional convolution operation. This is the feature map of the previous layer. For the bias term, the ReLU activation function This is used to introduce non-linearity and enhance feature representation. The output formula of the max pooling layer is: The pooling kernel size is 2, and the stride is 2. Downsampling reduces the feature map size, expands the receptive field, and captures more global anomaly patterns. Two 1D-CNN branches process the smoothed potential data sequences corresponding to the two boundary lines in the detection data pair, ultimately converting them into fixed-dimensional feature vectors of the smoothed potential data sequences corresponding to the two boundary lines. and .

[0054] The feature fusion and spatial mapping unit converts the two feature vectors from the encoder. and Concatenate into a fused feature vector This is mapped to a low-resolution two-dimensional feature map through a fully connected layer. The fused feature vector is: The fully connected layer will fuse the feature maps to obtain a low-resolution two-dimensional feature map as follows: ,in, , This refers to the size of the low-resolution feature map. For the number of channels, The operation converts flattened vectors into three-dimensional feature maps. .

[0055] The decoder uses 3-5 transposed convolutional layers to upsample the low-resolution two-dimensional feature map, and fuses the encoder features with the decoder upsampled features through skip connections. Finally, it outputs a leakage probability distribution map with the same size as the landfill grid through the Sigmoid activation function.

[0056] The decoder implements the transposed convolution output formula for upsampling as follows: .

[0057] in, k This is the size of the transposed convolution kernel. s Step size, s =2, by gradually increasing the feature map size, the final size is matched to the landfill grid size. Skip connections connect the feature maps of the encoder's l-th layer. Upsampled feature map of the corresponding layer of the decoder Fusion: ,in Interpolation upsampling is used to align feature map sizes, fuse low-level detail features with high-level semantic features, and improve boundary clarity.

[0058] Finally, the output is mapped to the [0,1] interval using the Sigmoid activation function to obtain the leakage probability: .

[0059] in This is the output feature map of the last layer of the decoder. Represents a grid The probability of leakage at that location.

[0060] The constructed detection data Given a pre-trained encoder-decoder model, the model performs the following process: The encoder has two branches respectively for... Stop and S left Perform convolution, activation, and pooling to extract local anomaly features, such as potential peaks and gradient abrupt changes, and output the features. and .

[0061] Feature fusion and splicing and To obtain splicing features .

[0062] Spatial mapping, through fully connected layers (4096 = 8 × 8 × 64) Mapped to a low-resolution feature map of 8×8×64.

[0063] Decoder upsampling involves progressively enlarging the feature map using transposed convolutional layers. H and W represent the number of rows and columns of the landfill after gridding. Each layer is connected by skip connections to fuse the low-level features of the encoder.

[0064] The probability output is obtained by applying the Sigmoid activation function to obtain the leakage probability of each pixel, and then outputting the result. The probability distribution of leakage.

[0065] S6. Set a probability threshold and automatically delineate areas above the threshold in the leakage probability distribution map as suspected leakage areas, and output the location coordinates of the suspected leakage areas in the landfill.

[0066] Set the probability threshold to 0.7, traverse the leakage probability distribution map, mark the areas with pixel values ​​≥ 0.7 in the map as suspected leakage areas, extract the horizontal and vertical X / Y coordinate range of the area, output to the host display unit, and generate a detection report on the leakage area of ​​the landfill geomembrane.

[0067] The encoder-decoder model in step S5 is trained through the following process: A training dataset was constructed using a combination of simulated and measured data. The simulated data was used to build an electric field model of the landfill using COMSOL, setting leakage points of different locations and sizes to generate boundary potential data containing the leakage points. The measured data collected boundary potential data from historical leakage detection in the landfill, with the coordinates of the leakage points marked. Boundary potential data samples at leakage point locations were positive samples, while boundary potential data samples at non-leakage locations were negative samples.

[0068] Add noise, shift, and scale the training dataset to increase its size.

[0069] The training dataset is input into the initialized 1D-CNN-based encoder-decoder model for training. The encoder, a dual-branch 1D-CNN, consists of three convolutional layers with kernels of 3×64, 3×128, and 3×256, and a 2×2 pooling layer, outputting a 256-dimensional feature vector. The feature fusion and spatial mapping unit fuses the feature vectors and outputs an 8×8×64 feature map through a fully connected layer. The decoder consists of three transposed convolutional layers with 3×3 kernels and a stride of 2, skipping connections to the corresponding layers in the encoder. The specific model structure and data processing flow are as follows: Figure 4 As shown.

[0070] The model training output predicts the leakage probability corresponding to the sample potential data. Based on a 1D-CNN encoder-decoder model, it is trained to classify and predict the "leakage point location." The model training process uses Focal Loss as the loss function. .

[0071] in, The loss is the prediction result of the samples in the training dataset. This is the model's predicted probability of the target class for samples in the training dataset. If a sample is a positive sample labeled as a leak location, then... To predict the probability that the sample is a leak location for the model, if the sample is a negative sample labeled as a non-leaking location, then... = 1 - The probability that the model predicts this sample as a leak location, which is the model's output probability of no leak. Since there are very few leak point pixels (positive samples) in the leak probability distribution map, the model may be biased towards predicting background data. This embodiment introduces a modulation factor using the Focal Loss function. and balance factor , γ This is the positive and negative sample mining coefficient in the loss function, used to adjust the loss weights for hard-to-classify positive samples and easy-to-classify negative samples, suppressing the proportion of loss for easily classified negative samples and focusing training on hard-to-classify positive samples. Its value ranges from 0 to 5 and is commonly used. γ =2, Balance the weights of positive and negative samples, with the weight of positive samples being greater than the weight of negative samples. Balance the weights of positive and negative samples, positive samples negative samples This further alleviates sample class imbalance, as negative samples are easier to classify. When it approaches 1, The loss is suppressed when positive samples are difficult to classify. Approaching zero, such as slight leakage. The loss is amplified. Through this loss function, the model training process focuses on difficult-to-classify positive samples such as leak points, significantly improving the ability to identify weak anomalies.

[0072] The training process optimizes the model using Adam with a learning rate of 0.001, a batch size of 32, and 100 iterations.

[0073] IoU, Precision, and Recall were used as evaluation metrics. When IoU ≥ 0.7, the training of the encoder-decoder model was stopped and the optimal model was saved.

[0074] In summary, combining the above... Figure 1 and Figure 2 The detection system that realizes the above landfill leakage detection method based on boundary dual voltage electric field can be summarized into three parts: electric field excitation module, boundary electrode module and detection host.

[0075] like Figure 2 As shown, the electric field excitation module includes an excitation power supply 300. The positive and negative terminals of the excitation power supply 300 are connected to the inner and outer sides of the landfill geomembrane, respectively, to establish a stable artificial electric field on both sides of the landfill geomembrane. The excitation power supply 300 is an adjustable DC power supply. The positive terminal of the adjustable DC power supply is connected to the internal electrode 301 of the landfill, and the negative terminal is connected to the external grounding electrode, located 5m outside the boundary of the landfill. At the same time, a reverse connection protection circuit is connected in series to establish a detection loop with a voltage of 100-500V and a current ≤10mA on both sides of the landfill geomembrane.

[0076] The boundary electrode module 400 arranges rectangular boundary survey lines around the outer perimeter of the landfill body. Several detection electrodes 401 are equidistantly arranged along the rectangular boundary survey lines. All detection electrodes are connected to the detection host 500 via shielded cables 402. The potential values ​​of each detection electrode are collected under both low-voltage and high-voltage artificial electric fields. During deployment, the positions of each detection electrode, along with the corner points of the rectangular boundary survey lines, are equidistantly distributed. The spacing between detection electrodes is 1-6m, preferably 3m. The detection electrodes are copper rod electrodes or Ag / AgCl non-polarized electrodes, vertically inserted into the ground to a depth ≥0.5m.

[0077] The detection host 500 includes a data processing module. The data processing module calculates the electric field strength between adjacent detection electrodes under low-voltage and high-voltage artificial electric fields based on the potential values ​​of each detection electrode. It then uses differential calculations between the potential values ​​and electric field strengths under the low-voltage and high-voltage artificial electric fields to remove background electric field noise from the landfill. The data processing module includes a GIM-12 high-density electrical resistivity meter, a multiplexer, a high-precision voltmeter (resolution 0.1mV), a current sampling resistor (0.1Ω), a differential amplifier, and a 24-bit ADC (sampling rate 1kHz) to synchronously acquire the potential and current signals of each detection electrode and convert them into digital signals. The landfill outline is meshed using the detection electrodes on the rectangular boundary survey line as connection points. Smooth potential data from the detection electrodes on two intersecting boundary survey lines are selected to construct a one-dimensional sequence. The sequence index of the one-dimensional sequence is aligned with the horizontal and vertical coordinates of the internal grid of the landfill, forming a physically aligned detection data pair. The detection host 500 also has a built-in encoder-decoder model based on 1D-CNN. The detection data pair output by the data processing module is input into the encoder-decoder model based on 1D-CNN. The dual-branch 1D-CNN extracts local anomaly features of the potential data on the boundary survey line and outputs a leakage probability distribution map. By setting a probability threshold, the area above the threshold in the leakage probability distribution map is automatically delineated as a suspected leakage area, and the location coordinates of the suspected leakage area in the landfill are output.

[0078] Figure 5a -f displays the test data results for this embodiment, where Figure 5a This is the boundary line potential curve under a 100V low-voltage artificial electric field. The curve is flat, containing only background signal. Figure 5b The potential curve of the boundary line under a 500V high-voltage artificial electric field shows a significant peak at the location corresponding to the leakage point. Figure 5c To obtain the abnormal potential curve of the boundary survey line after differential denoising, the background signal is canceled out, and only the peak abnormal potential data is retained. Figures 5a to 5c This embodiment demonstrates the potential data processing principle of leakage detection, from "low voltage reference → high voltage amplification → differential anomaly extraction". Figure 5d To normalize and denoise the abnormal potential signal, thus completing the conversion from "physical signal to model feature", Figure 5e and Figure 5f By comparing the probability map predicted by the model with the actual mask, the accuracy of the model in locating the leakage area can be intuitively verified.

[0079] To make the objectives, technical solutions, and advantages of the present invention clearer, 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, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0080] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0081] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other. It should also be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments; similar or identical parts between embodiments can be referred to interchangeably.

[0082] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting landfill leakage based on a boundary dual-voltage electric field, characterized in that... Includes the following steps: S1. Establish a stable artificial electric field on both sides of the geomembrane in the landfill; S2. A rectangular boundary survey line is arranged around the outer perimeter of the landfill body. Several detection electrodes are arranged at equal intervals along the rectangular boundary survey line. All detection electrodes are connected to the detection host through shielded cables. The potential values ​​of each detection electrode are collected under low voltage artificial electric field and high voltage artificial electric field respectively. S3. Calculate the electric field strength between adjacent detection electrodes under low-voltage artificial electric field and high-voltage artificial electric field by using the potential value of each detection electrode. Combine the difference between the potential value and the electric field strength under low-voltage artificial electric field and high-voltage artificial electric field to remove the background electric field noise of the landfill. S4. The landfill outline is gridded using the detection electrodes on the rectangular boundary survey line as connection points. Smooth potential data of the detection electrodes on two intersecting boundary survey lines are selected to construct a one-dimensional sequence. The sequence index of the one-dimensional sequence is aligned with the horizontal and vertical coordinates of the grid inside the landfill to form a physically aligned detection data pair. S5. The detection data is input into the encoder-decoder model based on 1D-CNN, and the local abnormal features of the potential data on the boundary measurement line are extracted by the dual-branch 1D-CNN to output the leakage probability distribution map. S6. Set a probability threshold and automatically delineate areas above the threshold in the leakage probability distribution map as suspected leakage areas, and output the location coordinates of the suspected leakage areas in the landfill.

2. The landfill leakage detection method based on a boundary dual-voltage electric field according to claim 1, characterized in that: In step S1, an adjustable DC power supply is used to establish a detection circuit with a voltage of 100-500V and a current of ≤10mA on both sides of the geomembrane in the landfill.

3. The landfill leakage detection method based on a boundary dual-voltage electric field according to claim 1, characterized in that: In step S2, the potential values ​​of each detection electrode on the boundary measurement line are first collected by a high-density electrical resistivity tomography instrument under a low-voltage artificial electric field of 100V, keeping the layout of the detection electrodes unchanged, and then the potential values ​​of the same group of detection electrodes are collected under a high-voltage artificial electric field of 500V.

4. The landfill leakage detection method based on a boundary dual-voltage electric field according to claim 1, characterized in that: In step S3, the electric field strength between the detection electrode and the adjacent detection electrode is calculated under both low-voltage and high-voltage artificial electric fields. and : 、 ； in, , Index positions , The potential values ​​of two adjacent detection electrodes under a high-voltage artificial electric field. , Index positions , The potential values ​​of two adjacent detection electrodes under a high-voltage artificial electric field. The distance between adjacent detection electrodes; Then, the potential anomaly difference between the detection electrode and the adjacent detection electrode is calculated separately. and the abnormal difference in electric field strength Removal of background electric field noise from landfills: 、 ； in, , index position The potential values ​​of the detection electrode under low-voltage and high-voltage artificial electric fields, respectively. , index position The electric field strength between the detection electrode and the adjacent detection electrode under low-voltage artificial electric field and high-voltage artificial electric field, respectively.

5. The landfill leakage detection method based on a boundary dual-voltage electric field according to claim 1, characterized in that: The smoothed potential data in step S4 is obtained as follows: the potential value data after removing the background electric field noise of the landfill is normalized, and then the smoothed potential data of each detection electrode is obtained by moving average filtering.

6. The landfill leakage detection method based on a boundary dual-voltage electric field according to claim 5, characterized in that: The smoothed potential data is obtained by filtering with a three-point moving average: ， in, index position The detection electrode smoothing potential data, , , Index positions , , The normalized potential values ​​of the three adjacent detection electrodes are used to replace the potential values ​​of the adjacent detection electrodes on the missing side by performing a three-point moving average filtering.

7. The landfill leakage detection method based on a boundary dual-voltage electric field according to claim 1, characterized in that: The encoder-decoder model in step S5 includes a decoder, a feature fusion and spatial mapping unit, and a decoder; The encoder includes two 1D-CNN branches with identical structures and independent parameters. Each 1D-CNN branch contains 3-5 convolutional layers, ReLU activation function, and 1D max pooling layer, which respectively process the smooth potential data sequences corresponding to the two boundary test lines in the detection data pair, and finally convert them into fixed-dimensional feature vectors corresponding to the smooth potential data sequences of the two boundary test lines. The feature fusion and spatial mapping unit concatenates the two feature vectors converted by the encoder into a fused feature vector, and maps it into a low-resolution two-dimensional feature map through a fully connected layer. The decoder uses a transposed convolutional layer to upsample the low-resolution two-dimensional feature map, and fuses the encoder features with the decoder upsampled features through skip connections. Finally, it outputs a leakage probability distribution map with the same size as the landfill grid through a sigmoid activation function.

8. The landfill leakage detection method based on a boundary dual-voltage electric field according to claim 7, characterized in that: The encoder-decoder model in step S5 is trained through the following process: A training dataset was constructed using a combination of simulated and measured data. The simulated data was used to build an electric field model of the landfill using COMSOL, setting leakage points of different locations and sizes to generate boundary potential data containing leakage points. The measured data collected boundary potential data from historical leakage detection in the landfill. Boundary potential data samples at leakage point locations were positive samples, while boundary potential data samples at non-leakage locations were negative samples. Add noise, shift, and scale the training dataset to increase its size. The training dataset is input into the initialized 1D-CNN-based encoder-decoder model for training, using Focal Loss as the loss function: ， in, The loss is the prediction result of the samples in the training dataset. It is the probability that the model predicts for the target class of samples in the training dataset. γ These are the positive and negative sample mining coefficients in the loss function. The weights are positive and negative samples, with positive samples having a greater weight than negative samples. Negative samples are easier to classify. The loss is suppressed when the positive samples are difficult to classify, tending towards 1. As the value approaches zero, the loss is amplified. The training process optimizes the model using Adam; IoU, Precision, and Recall were used as evaluation metrics. When IoU ≥ 0.7, the training of the encoder-decoder model was stopped and the optimal model was saved.

9. A system for implementing any one of the landfill leakage detection methods based on a boundary dual-voltage electric field according to claims 1-8, characterized in that... include: An electric field excitation module includes an excitation power supply, the positive and negative poles of which are respectively connected to the inner and outer sides of the landfill geomembrane to establish a stable artificial electric field on both sides of the landfill geomembrane. The boundary electrode module arranges rectangular boundary survey lines around the outer perimeter of the landfill body, and arranges several detection electrodes at equal intervals along the rectangular boundary survey lines. All detection electrodes are connected to the detection host through shielded cables, and the potential values ​​of each detection electrode are collected under low voltage artificial electric fields and high voltage artificial electric fields respectively. The detection host includes a data processing module, which calculates the electric field strength between adjacent detection electrodes under low-voltage and high-voltage artificial electric fields based on the potential values ​​of each detection electrode. It then removes background electric field noise from the landfill by performing differential calculations between the potential values ​​and electric field strengths under the low-voltage and high-voltage artificial electric fields. The landfill contour is gridded using the detection electrodes on the rectangular boundary survey line as connection points. Smooth potential data of the detection electrodes on two intersecting boundary survey lines are selected to construct a one-dimensional sequence. The sequence index of the one-dimensional sequence is aligned with the horizontal and vertical coordinates of the internal grid of the landfill to form physically aligned detection data pairs. The detection host also has a built-in encoder-decoder model based on 1D-CNN. The detection data is input into the encoder-decoder model based on 1D-CNN. The local abnormal features of the potential data on the boundary measurement line are extracted through the dual-branch 1D-CNN, and the leakage probability distribution map is output. Set a probability threshold, and automatically delineate areas above the threshold in the leakage probability distribution map as suspected leakage areas, and output the location coordinates of the suspected leakage areas in the landfill.

10. A landfill leakage detection system based on a boundary dual-voltage electric field according to claim 9, characterized in that: The detection electrodes are made of copper or Ag / AgCl non-polarized electrodes, and are inserted into the soil surrounding the landfill body to a depth of ≥0.5m, with a spacing of 1-6m.

Images