A method for detecting seepage in hydraulic structures

Abstract

The present invention discloses a method for detecting seepage in hydraulic structures, belonging to the technical field of electromagnetic detection. First, the surface of the hydraulic structure is divided into a grid, and the same kind of electromagnetic wave is emitted to each grid, and the reflected wave signal and the time interval of the signal round-trip are recorded. Then, the medium characteristic coefficient and the medium centrifugal value are calculated according to the time interval to construct a medium centrifugal matrix; the intensity centrifugal value is calculated according to the transmitted and reflected wave signals to construct an intensity centrifugal matrix; the reflected wave signal is subjected to Fourier transform, and the phase centrifugal value is calculated after hierarchical processing to construct a phase centrifugal matrix. Finally, a seepage evaluation neural network is used to fuse and enhance the features of the three centrifugal matrices, so as to obtain the seepage evaluation value of the hydraulic structure. The present invention effectively improves the accuracy of seepage detection in hydraulic structures by extracting seepage features in multiple dimensions.

Description

Technical Field

[0001] The present invention relates to the technical field of electromagnetic detection, and particularly to a method for detecting water seepage in hydraulic structures. Background Art

[0002] With the acceleration of the urbanization process, hydraulic structures play an increasingly important role in infrastructure construction. However, during the use of hydraulic structures, water seepage problems are often faced, which not only affect their structural safety but may also cause damage to the surrounding environment. Traditional water seepage detection methods mostly rely on manual inspections, but the seepage locations are usually inside, and it is difficult to detect internal water seepage using manual inspections. In addition, existing non-destructive testing technologies, such as ultrasonic testing, although able to identify water seepage problems to a certain extent, still have some technical problems in their applications.

[0003] Existing ultrasonic testing technologies mainly judge the defects and water seepage conditions inside the structure by emitting ultrasonic signals and analyzing their reflected waves. However, due to the complex materials and structures of hydraulic structures, ultrasonic signals are easily affected by various factors during propagation, such as changes in the density, humidity, and temperature of concrete. This results in different states of the reflected ultrasonic waves for different hydraulic structures, making it difficult to accurately extract the characteristics of water seepage, and there is a problem of low accuracy in detecting water seepage in hydraulic structures. Summary of the Invention

[0004] Aiming at the above deficiencies in the prior art, the method for detecting water seepage in a hydraulic structure provided by the present invention solves the problem of low accuracy in detecting water seepage in hydraulic structures existing in the prior art.

[0005] To achieve the above invention objective, the technical solution adopted by the present invention is: A method for detecting water seepage in a hydraulic structure, comprising the following steps:

[0006] S1. Perform grid division on the surface of the hydraulic structure to obtain grids, emit the same kind of electromagnetic wave in each grid, obtain the reflected wave signals of each grid, and record the signal round-trip time interval, where L is the number of grids in the horizontal direction and W is the number of grids in the vertical direction;

[0007] S2. Calculate the medium characteristic coefficient for each grid according to the signal round-trip time interval, obtain the medium centrifugal value, and construct a medium centrifugal matrix;

[0008] S3. Calculate the intensity centrifugal value for each grid according to the electromagnetic wave and reflected wave signals emitted by each grid, and construct an intensity centrifugal matrix;

[0009] S4. Perform Fourier transform on the reflected wave signals of each grid to obtain an amplitude-phase pair sequence, perform hierarchical processing on the amplitude-phase pair sequence, calculate the phase centrifugal value, and construct a phase centrifugal matrix;

[0010] S5. Use a seepage evaluation neural network to respectively fuse and enhance two types of features of the medium centrifugal matrix, the strength centrifugal matrix, and the phase centrifugal matrix to obtain the seepage evaluation value of the hydraulic structure.

[0011] Further, S2 includes the following sub-steps:

[0012] S21. Calculate the medium characteristic coefficient for each grid according to the signal round-trip time interval: , where ε i is the medium characteristic coefficient of the i-th grid, c is the speed of light, d is the detection depth, and △t i is the signal round-trip time interval in the i-th grid;

[0013] S22. Calculate the average value of the medium characteristic coefficients of each grid to obtain the medium characteristic mean value;

[0014] S23. Calculate the medium centrifugal value of each grid according to the medium characteristic coefficient and the medium characteristic mean value of each grid;

[0015] S24. Use the medium centrifugal value of each grid as an element and construct a medium centrifugal matrix according to the grid layout.

[0016] Further, the formula for calculating the medium centrifugal value in S23 is: , where η ε,i is the medium centrifugal value of the i-th grid, ε c is the medium characteristic mean value, ε max is the maximum medium characteristic coefficient of each grid, and i is a positive integer.

[0017] Further, S3 includes the following sub-steps:

[0018] S31. Respectively perform time-domain discrete sampling on the transmitted electromagnetic wave and the reflected wave signal to obtain a transmitted electromagnetic wave time-domain sequence and a reflected wave time-domain sequence;

[0019] S32. Respectively calculate the mean values of the transmitted electromagnetic wave time-domain sequence and the reflected wave time-domain sequence to obtain the transmitted intensity mean value and the reflected intensity mean value;

[0020] S33. Take the ratio of the transmitted intensity mean value and the reflected intensity mean value of the same grid as the intensity ratio of this grid;

[0021] S34. Calculate the first strength centrifugal component according to the intensity ratios of each grid;

[0022] S35. Calculate the standard deviation of the time domain sequence of the reflected wave to obtain the standard deviation of the reflection intensity;

[0023] S36. Calculate the second intensity centrifugal component according to the standard deviation of the reflection intensity of each grid;

[0024] S37. Weight the first intensity centrifugal component and the second intensity centrifugal component to obtain the intensity centrifugal value;

[0025] S38. Use the intensity centrifugal value of each grid as an element and construct an intensity centrifugal matrix according to the grid arrangement;

[0026] Further, the formula for calculating the first intensity centrifugal component in S34 is: , where η θ,i is the first intensity centrifugal component of the i-th grid, θ i is the intensity ratio of the i-th grid, θ max is the maximum intensity ratio of each grid, and i is a positive integer;

[0027] The formula for calculating the second intensity centrifugal component in S36 is: , where η σ,i is the second intensity centrifugal component of the i-th grid, σ i is the standard deviation of the reflection intensity of the i-th grid, σ max is the maximum standard deviation of the reflection intensity of each grid.

[0028] Further, S4 includes the following sub-steps:

[0029] S41. Perform Fourier transform on the reflected wave signal of each grid, arrange them in descending order of amplitude to obtain an amplitude-phase pair sequence;

[0030] S42. Divide the amplitude-phase pair sequence into 3 parts to obtain a first-level amplitude-phase pair subsequence, a second-level amplitude-phase pair subsequence, and a third-level amplitude-phase pair subsequence;

[0031] S43. Add the phases in each level of the amplitude-phase pair subsequence to obtain a first-level total phase, a second-level total phase, and a third-level total phase;

[0032] S44. Calculate the phase centrifugal component of the same level according to the total phase of the same level of each grid;

[0033] S45. Weight the phase centrifugal components of the first level, second level, and third level to obtain the phase centrifugal value;

[0034] S46. Use the phase centrifugal value of each grid as an element and construct a phase centrifugal matrix according to the grid arrangement.

[0035] Further, the formula for calculating the phase centrifugal component in S44 is: , where η φ,i is the phase centrifugal component of the same level of the i-th grid, φ i is the total phase of the same level of the i-th grid, φ max is the maximum total phase selected from the same level of each grid, and i is a positive integer.

[0036] Further, the seepage evaluation neural network in S5 includes: a first grid feature extraction unit, a second grid feature extraction unit, a third grid feature extraction unit, a first type of grid feature fusion unit, a second type of grid feature fusion unit, a first activation function unit, a second activation function unit, and a weighting unit;

[0037] The input end of the first grid feature extraction unit is used to input the medium centrifugal matrix; the input end of the second grid feature extraction unit is used to input the intensity centrifugal matrix; the input end of the third grid feature extraction unit is used to input the phase centrifugal matrix;

[0038] The input end of the first type of grid feature fusion unit is respectively connected to the first output end of the first grid feature extraction unit, the first output end of the second grid feature extraction unit, and the first output end of the third grid feature extraction unit;

[0039] The input end of the second type of grid feature fusion unit is respectively connected to the second output end of the first grid feature extraction unit, the second output end of the second grid feature extraction unit, and the second output end of the third grid feature extraction unit;

[0040] The input end of the first activation function unit is connected to the output end of the first type of grid feature fusion unit; the input end of the second activation function unit is connected to the output end of the second type of grid feature fusion unit;

[0041] The input end of the weighting unit is respectively connected to the output end of the first activation function unit and the output end of the second activation function unit, and its output end serves as the output end of the seepage evaluation neural network.

[0042] Further, the expressions of the 3 grid feature extraction units are all: , , where H Max is the first type of grid feature, H Avg is the second type of grid feature, MaxPool is the max pooling layer, AvgPool is the average pooling layer, Conv is the convolution operation, and X in is the input matrix of the grid feature extraction unit.

[0043] Further, the first type of grid feature fusion unit is used to perform Hadamard product multiplication on the first type of grid features output by the first grid feature extraction unit, the first type of grid features output by the second grid feature extraction unit, and the first type of grid features output by the third grid feature extraction unit to obtain the first fused grid features;

[0044] The second type of grid feature fusion unit is used to perform Hadamard product multiplication on the second type of grid features output by the first grid feature extraction unit, the second type of grid features output by the second grid feature extraction unit, and the second type of grid features output by the third grid feature extraction unit to obtain the second fused grid features.

[0045] The beneficial effects of the present invention are as follows:

[0046] 1. The present invention divides the surface of the hydraulic structure into grids for zonal detection. At the same time, for the same hydraulic structure with the same structure and material, after dividing it into multiple grids, it is convenient to compare the reflected signals between multiple grids to obtain the medium centrifugal value, intensity centrifugal value, and phase centrifugal value, highlighting the characteristics of water seepage and improving the detection accuracy of water seepage in hydraulic structures.

[0047] 2. Calculate the medium characteristic coefficient based on the signal round-trip time interval and construct the medium centrifugal matrix. In this way, the medium characteristics of each grid area can be quantified. Then, calculate the intensity centrifugal value according to the electromagnetic wave and reflected wave signals emitted by each grid, which reflects the intensity characteristics of electromagnetic waves in different grid areas. Finally, calculate the phase centrifugal value. The change in phase can reflect the subtle changes inside the building structure and is closely related to the water seepage situation.

[0048] 3. The present invention uses a water seepage evaluation neural network to fuse and enhance the two types of features of the medium centrifugal matrix, intensity centrifugal matrix, and phase centrifugal matrix respectively, realizing feature extraction from multiple dimensions and feature fusion enhancement, and improving the accuracy of water seepage detection. Description of the Drawings

[0049] Figure 1 It is a flowchart of a method for detecting water seepage in a hydraulic structure;

[0050] Figure 2 It is a schematic structural diagram of a water seepage evaluation neural network. Detailed Embodiments

[0051] The following describes the detailed embodiments of the present invention to facilitate those skilled in the art of the present technology to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the detailed embodiments. For those of ordinary skill in the art of the present technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are within the scope of protection.

[0052] As Figure 1 shown, a method for detecting seepage in hydraulic structures includes the following steps:

[0053] S1. Divide the surface of the hydraulic structure into a grid to obtain grids. Emit the same type of electromagnetic wave in each grid to obtain the reflected wave signal of each grid, and record the signal round-trip time interval. Here, L is the number of grids in the horizontal direction, and W is the number of grids in the vertical direction;

[0054] S2. Calculate the medium characteristic coefficient for each grid according to the signal round-trip time interval, obtain the medium centrifugal value, and construct a medium centrifugal matrix;

[0055] S3. Calculate the intensity centrifugal value for each grid according to the electromagnetic wave emitted and the reflected wave signal in each grid, and construct an intensity centrifugal matrix;

[0056] S4. Perform Fourier transform on the reflected wave signal of each grid to obtain an amplitude-phase pair sequence, perform hierarchical processing on the amplitude-phase pair sequence, calculate the phase centrifugal value, and construct a phase centrifugal matrix;

[0057] S5. Use a seepage evaluation neural network to fuse and enhance the two types of features of the medium centrifugal matrix, the intensity centrifugal matrix, and the phase centrifugal matrix respectively to obtain the seepage evaluation value of the hydraulic structure.

[0058] For example, taking an arch dam with a width of 3 meters and a length of 50 meters as an example, it can be divided at a length of 0.5 meters in the width direction and at a length of 0.5 meters in the length direction, then grids are obtained, and the size of each grid is .

[0059] In this embodiment, when the frequency of the electromagnetic wave is set to 1 GHz, the detection depth is about 5 meters.

[0060] In this embodiment, S2 includes the following sub-steps:

[0061] S21. Calculate the medium characteristic coefficient for each grid according to the signal round-trip time interval: , where ε i is the medium characteristic coefficient of the i-th grid, c is the speed of light, d is the detection depth, and △t i is the signal round-trip time interval in the i-th grid;

[0062] S22. Calculate the average value of the medium characteristic coefficients of each grid to obtain the medium characteristic mean value;

[0063] S23. Calculate the medium centrifugal value of each grid according to the medium characteristic coefficient and the medium characteristic mean value of each grid;

[0064] S24. Construct a medium centrifugation matrix with the medium centrifugation values of each grid as elements according to the grid arrangement.

[0065] In this embodiment, the formula for calculating the medium centrifugation value in S23 is: , where η ε,i is the medium centrifugation value of the i-th grid, ε c is the mean value of medium characteristics, ε max is the maximum medium characteristic coefficient of each grid, and i is a positive integer.

[0066] In the present invention, the signal round-trip time interval is the interval time from transmitting an electromagnetic wave to receiving a reflected wave, that is, the wave propagation time.

[0067] The present invention calculates the medium characteristic coefficient for each grid to reflect the medium characteristics of each grid area, takes the mean value of each medium characteristic coefficient to reflect the overall medium characteristics, and obtains the medium centrifugation value of the grid according to the deviation of the medium characteristic coefficient of each grid from the mean value of medium characteristics, reflecting the deviation of the medium characteristics of the grid.

[0068] The medium characteristic coefficient of water is about 20 times that of dry concrete. Therefore, when there is water seepage in a hydraulic structure, its medium characteristic coefficient is larger than that of a non-seeping hydraulic structure.

[0069] In this embodiment, S3 includes the following sub-steps:

[0070] S31. Perform time-domain discrete sampling on the transmitted electromagnetic wave and the reflected wave signal respectively to obtain the transmitted electromagnetic wave time-domain sequence and the reflected wave time-domain sequence;

[0071] S32. Calculate the mean value of the transmitted electromagnetic wave time-domain sequence and the reflected wave time-domain sequence respectively to obtain the transmitted intensity mean value and the reflected intensity mean value;

[0072] S33. Take the ratio of the transmitted intensity mean value to the reflected intensity mean value of the same grid as the intensity ratio of the grid;

[0073] S34. Calculate the first intensity centrifugation component according to the intensity ratios of each grid;

[0074] S35. Calculate the standard deviation of the reflected wave time-domain sequence to obtain the reflected intensity standard deviation;

[0075] S36. Calculate the second intensity centrifugation component according to the reflected intensity standard deviations of each grid;

[0076] S37. Weight the first intensity centrifugation component and the second intensity centrifugation component to obtain the intensity centrifugation value;

[0077] S38. Construct a strength centrifugal matrix with the strength centrifugal values of each grid as elements according to the grid arrangement.

[0078] In this embodiment, the formula for calculating the first strength centrifugal component in S34 is: , where η θ,i is the first strength centrifugal component of the i-th grid, θ i is the strength ratio of the i-th grid, θ max is the maximum strength ratio of each grid, and i is a positive integer.

[0079] When there is seepage in hydraulic structures, the presence of water will cause greater attenuation of electromagnetic waves during propagation. Therefore, in the present invention, the ratio of the average emission intensity to the average reflection intensity of the same grid is taken as the strength ratio. When there is seepage in the grid, the strength ratio is larger. The deviation of each strength ratio from the average strength ratio is calculated to obtain the first strength centrifugal component, highlighting the seepage characteristics.

[0080] The formula for calculating the second strength centrifugal component in S36 is: , where η σ,i is the second strength centrifugal component of the i-th grid, σ i is the standard deviation of the reflection intensity of the i-th grid, σ max is the maximum standard deviation of the reflection intensity of each grid.

[0081] When there is seepage in hydraulic structures, the medium becomes inhomogeneous, and electromagnetic waves will experience different propagation speeds and attenuations during propagation, resulting in distortion of the waveform of the reflected wave. Therefore, in the present invention, the standard deviation is calculated to reflect the distortion situation, and the deviation of the standard deviation of the reflection intensity from the mean value is calculated to obtain the second strength centrifugal component, highlighting the seepage characteristics.

[0082] In this embodiment, S4 includes the following sub-steps:

[0083] S41. Perform Fourier transform on the reflected wave signal of each grid, and arrange them in descending order of amplitude to obtain an amplitude-phase pair sequence;

[0084] S42. Divide the amplitude-phase pair sequence into three parts to obtain a first-level amplitude-phase pair subsequence, a second-level amplitude-phase pair subsequence, and a third-level amplitude-phase pair subsequence;

[0085] S43. Add the phases in each level of the amplitude-phase pair subsequence to obtain a first-level total phase, a second-level total phase, and a third-level total phase;

[0086] S44. Calculate the phase centrifugal component of the same level according to the total phase of the same level of each grid;

[0087] S45. Weight the phase centrifugal components of the first, second, and third levels to obtain the phase centrifugal value;

[0088] S46. Use the phase centrifugal value of each grid as an element and construct a phase centrifugal matrix according to the grid arrangement.

[0089] After performing Fourier transform on the reflected wave signal, amplitude and phase information will be obtained. Arrange them in descending order of amplitude to get the amplitude-phase pair sequence. Divide the amplitude-phase pair sequence into three parts. For example, divide the first three amplitude-phase pairs into the first level, divide the first three amplitude-phase pairs except those in the first level into the second level, and the rest are classified into the third level.

[0090] The dielectric properties of water will cause phase delay of electromagnetic waves at the interface, thus affecting the phase information of the reflected wave. This phase change can be used to infer the presence and degree of water seepage.

[0091] In this embodiment, the formula for calculating the phase centrifugal component in S44 is: , where η φ,i is the phase centrifugal component of the same level of the i-th grid, φ i is the total phase of the same level of the i-th grid, φ max is the maximum total phase selected from the same level of each grid, and i is a positive integer.

[0092] In the formula for calculating the phase centrifugal component, when dealing with the total phase of the first level, φ i is the total phase of the first level of the i-th grid, φ max is the maximum value selected from the total phases of the first level of each grid; when dealing with the total phase of the second level, φ i is the total phase of the second level of the i-th grid, φ max is the maximum value selected from the total phases of the second level of each grid; when dealing with the total phase of the third level, φ i is the total phase of the third level of the i-th grid, φ max is the maximum value selected from the total phases of the third level of each grid.

[0093] The presence of water will cause phase delay of the reflected wave, affecting the phase information of the reflected wave. In the present invention, the amplitude-phase pair sequence is classified into the first, second, and third levels. In the same level, according to the deviation of the total phase of each grid from the average value of the total phase of this level, the phase centrifugal component of the same level is calculated to reflect the change of phase in different grids.

[0094] In this embodiment, the weighting formula in S45 is: , where γ is the phase centrifugal value, γ 1 is the phase centrifugal component of the first level, γ 2 is the phase centrifugal component of the second level, γ3 The phase centrifugal component of the third level.

[0095] In the present invention, the medium centrifugal value, the intensity centrifugal value, and the phase centrifugal value are all arranged in a grid layout to construct a corresponding matrix, so that the positions of all features in the same grid correspond.

[0096] As Figure 2 shown, the water seepage evaluation neural network in S5 includes: a first grid feature extraction unit, a second grid feature extraction unit, a third grid feature extraction unit, a first type of grid feature fusion unit, a second type of grid feature fusion unit, a first activation function unit, a second activation function unit, and a weighting unit;

[0097] The input end of the first grid feature extraction unit is used to input the medium centrifugal matrix; the input end of the second grid feature extraction unit is used to input the intensity centrifugal matrix; the input end of the third grid feature extraction unit is used to input the phase centrifugal matrix;

[0098] The input end of the first type of grid feature fusion unit is respectively connected to the first output end of the first grid feature extraction unit, the first output end of the second grid feature extraction unit, and the first output end of the third grid feature extraction unit;

[0099] The input end of the second type of grid feature fusion unit is respectively connected to the second output end of the first grid feature extraction unit, the second output end of the second grid feature extraction unit, and the second output end of the third grid feature extraction unit;

[0100] The input end of the first activation function unit is connected to the output end of the first type of grid feature fusion unit; the input end of the second activation function unit is connected to the output end of the second type of grid feature fusion unit;

[0101] The input end of the weighting unit is respectively connected to the output end of the first activation function unit and the output end of the second activation function unit, and its output end serves as the output end of the water seepage evaluation neural network.

[0102] In this embodiment, the expressions of the 3 grid feature extraction units are all: , , where H Max is the first type of grid feature, H Avg is the second type of grid feature, MaxPool is the max pooling layer, AvgPool is the average pooling layer, Conv is the convolution operation, and X in is the input matrix of the grid feature extraction unit.

[0103] In this embodiment, the convolution operation can adopt convolution layer.

[0104] In the present invention, two types of grid features are extracted by each grid feature extraction unit, and then three first-type grid features are processed by the first-type grid feature fusion unit, and three second-type grid features are processed by the second-type grid feature fusion unit to achieve mutual fusion and enhancement of features. Then, through the activation function unit, the seepage component is obtained, and by synthesizing the two seepage components, the seepage evaluation value of the hydraulic structure is obtained.

[0105] In this embodiment, the first-type grid feature fusion unit is used to perform Hadamard product multiplication processing on the first-type grid features output by the first grid feature extraction unit, the first-type grid features output by the second grid feature extraction unit, and the first-type grid features output by the third grid feature extraction unit to obtain the first fused grid feature, that is, element-wise multiplication of three two-dimensional first-type grid features is performed to fuse them into a two-dimensional first fused grid feature.

[0106] The second-type grid feature fusion unit is used to perform Hadamard product multiplication processing on the second-type grid features output by the first grid feature extraction unit, the second-type grid features output by the second grid feature extraction unit, and the second-type grid features output by the third grid feature extraction unit to obtain the second fused grid feature, that is, element-wise multiplication of three two-dimensional second-type grid features is performed to fuse them into a two-dimensional second fused grid feature.

[0107] In this embodiment, the expression of the first activation function unit is: , and the expression of the second activation function unit is: , where r 1 is the first seepage component, r 2 is the second seepage component, Tanh is the hyperbolic tangent function, h Max,k is the k-th eigenvalue in the first fused grid feature, h Avg,k is the k-th eigenvalue in the second fused grid feature, ω Max,k is the weight of h Max,k , b Max,k is the bias of h Max,k , ω Avg,k is the weight of h Avg,k , b Avg,k is the bias of h Avg,k , k is a positive integer, and K is the number of eigenvalues.

[0108] In the present invention, the weighting unit is used to weight the first seepage component and the second seepage component to obtain the seepage evaluation value of the hydraulic structure. The expression of the weighting unit is: , where y is the seepage evaluation value of the hydraulic structure, ω r1 is the weight of the first seepage component r 1 , ω r2 is the weight of the second seepage component r 2 .

[0109] In this embodiment, the weights and biases in the seepage evaluation neural network are obtained by training with the existing gradient descent method.

[0110] The present invention divides the surface of the hydraulic structure into grids to achieve zonal detection. At the same time, for the same hydraulic structure with the same structure and material, after it is divided into multiple grids, it is convenient to compare the reflection signals between multiple grids to obtain the medium centrifugal value, strength centrifugal value, and phase centrifugal value, highlighting the characteristics of seepage and improving the detection accuracy of seepage in hydraulic structures.

[0111] Calculate the medium characteristic coefficient according to the signal round-trip time interval and construct the medium centrifugal matrix. In this way, the medium characteristics of each grid area can be quantified. Then, according to the electromagnetic wave and reflected wave signals emitted by each grid, calculate the strength centrifugal value, which reflects the intensity characteristics of electromagnetic waves in different grid areas. Finally, calculate the phase centrifugal value. The change in phase can reflect the subtle changes inside the building structure and is closely related to the seepage situation.

[0112] The present invention uses a seepage evaluation neural network to respectively fuse and enhance the two types of features of the medium centrifugal matrix, strength centrifugal matrix, and phase centrifugal matrix, realizing feature extraction from multiple dimensions and feature fusion enhancement to improve the accuracy of seepage detection.

[0113] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for detecting water seepage in hydraulic structures, characterized in that: The following steps are involved: S1. Grid the surface of the hydraulic structure to obtain grids, emit the same electromagnetic wave in each grid, obtain the reflected wave signal of each grid, and record the round-trip time interval of the signal, where L is the number of grids in the horizontal direction and W is the number of grids in the vertical direction; S2. Calculate the medium characteristic coefficient for each grid according to the signal round trip time interval, obtain the medium centrifugal value, and construct the medium centrifugal matrix; S3, according to the electromagnetic wave emission and reflected wave signals of each grid, the intensity centrifugal value of each grid is calculated to construct an intensity centrifugal matrix; S4, performing Fourier transform on the reflected wave signal of each grid to obtain an amplitude-phase pair sequence, performing hierarchical processing on the amplitude-phase pair sequence, calculating the phase centrifugal value, and constructing a phase centrifugal matrix; S5. Use the water seepage assessment neural network to fuse and enhance the two types of features of the medium centrifugal matrix, the intensity centrifugal matrix and the phase centrifugal matrix respectively to obtain the water seepage assessment value of the hydraulic structure; The water seepage assessment neural network in S5 includes: a first grid feature extraction unit, a second grid feature extraction unit, a third grid feature extraction unit, a first type of grid feature fusion unit, a second type of grid feature fusion unit, a first activation function unit, a second activation function unit and a weighting unit; The input end of the first grid feature extraction unit is used to input the medium centrifugal matrix; the input end of the second grid feature extraction unit is used to input the intensity centrifugal matrix; the input end of the third grid feature extraction unit is used to input the phase centrifugal matrix; The input end of the first type of grid feature fusion unit is respectively connected to the first output end of the first grid feature extraction unit, the first output end of the second grid feature extraction unit and the first output end of the third grid feature extraction unit; The input end of the second type of grid feature fusion unit is respectively connected to the second output end of the first grid feature extraction unit, the second output end of the second grid feature extraction unit, and the second output end of the third grid feature extraction unit; The input end of the first activation function unit is connected to the output end of the first type of grid feature fusion unit; the input end of the second activation function unit is connected to the output end of the second type of grid feature fusion unit; The input end of the weighting unit is connected to the output end of the first activation function unit and the output end of the second activation function unit respectively, and the output end thereof serves as the output end of the water seepage assessment neural network; The expressions of the three grid feature extraction units are: , , where H Max is the first type of grid feature, H Avg is the second type of grid feature, MaxPool is the maximum pooling layer, AvgPool is the average pooling layer, Conv is the convolution operation, X in It is the input matrix of the mesh feature extraction unit.

2. The method for detecting water seepage in hydraulic structures according to claim 1, characterized in that: The S2 comprises the following sub-steps: S21. Calculate the dielectric characteristic coefficient for each grid according to the signal round trip time interval: , where ε i is the dielectric characteristic coefficient of the ith grid, c is the speed of light, d is the detection depth, △t i is the round trip time interval of the signal in the i-th grid; S22, calculating the average value of the dielectric characteristic coefficients of each grid to obtain a dielectric characteristic mean value; S23, calculating the medium centrifugal value of each grid according to the medium characteristic coefficient and the medium characteristic mean of each grid; S24. Taking the medium centrifugal value of each grid as an element, a medium centrifugal matrix is ​​constructed according to the grid arrangement.

3. The method for detecting water seepage in hydraulic structures according to claim 2, characterized in that: The formula for calculating the medium centrifugal value in S23 is: , where η ε,i is the medium centrifugal value of the i-th grid, ε c is the mean value of the medium characteristic, ε max is the maximum dielectric characteristic coefficient of each grid, and i is a positive integer.

4. The method for detecting water seepage in hydraulic structures according to claim 1, characterized in that: The S3 comprises the following sub-steps: S31, performing time domain discrete sampling on the transmitted electromagnetic wave and the reflected wave signals respectively to obtain a transmitted electromagnetic wave time domain sequence and a reflected wave time domain sequence; S32, respectively calculating the mean of the time domain sequence of the transmitted electromagnetic wave and the time domain sequence of the reflected wave to obtain the mean value of the transmitted intensity and the mean value of the reflected intensity; S33, taking the ratio of the average emission intensity to the average reflection intensity of the same grid as the intensity ratio of the grid; S34, calculating a first intensity centrifugal component according to the intensity ratio of each grid; S35, calculating the standard deviation of the reflected wave time domain sequence to obtain the reflection intensity standard deviation; S36, calculating the second intensity centrifugal component according to the reflection intensity standard deviation of each grid; S37, weighting the first intensity centrifugal component and the second intensity centrifugal component to obtain an intensity centrifugal value; S38. Taking the intensity centrifugal value of each grid as an element, and constructing an intensity centrifugal matrix according to the grid arrangement.

5. The method for detecting water seepage in hydraulic structures according to claim 4, characterized in that: The formula for calculating the first intensity centrifugal component in S34 is: , where η θ,i is the first intensity centrifugal component of the i-th grid, θ i is the intensity ratio of the i-th grid, θ max is the maximum intensity ratio of each grid, i is a positive integer; The formula for calculating the second intensity centrifugal component in S36 is: , where η σ,i is the second intensity centrifugal component of the i-th grid, σ i is the standard deviation of the reflection intensity of the ith grid, σ max is the standard deviation of the maximum reflection intensity of each grid.

6. The method for detecting water seepage in hydraulic structures according to claim 1, characterized in that: The S4 comprises the following sub-steps: S41, performing Fourier transform on the reflected wave signal of each grid, arranging them from large to small in amplitude, and obtaining an amplitude-phase pair sequence; S42, dividing the amplitude-phase pair sequence into three parts, obtaining a primary amplitude-phase pair subsequence, a secondary amplitude-phase pair subsequence, and a tertiary amplitude-phase pair subsequence; S43, adding the phases in the amplitude-phase pair subsequences of each level to obtain a first-level total phase, a second-level total phase, and a third-level total phase; S44, calculating the phase centrifugal component of the same level according to the total phase of the same level of each grid; S45, weighting the first-order, second-order, and third-order phase centrifugal components to obtain a phase centrifugal value; S46. The phase centrifugal value of each grid is used as an element, and a phase centrifugal matrix is ​​constructed according to the grid arrangement.

7. The method for detecting water seepage in hydraulic structures according to claim 6, characterized in that: The formula for calculating the phase centrifugal component in S44 is: ,in, is the phase centrifugal component of the same level of the i-th grid, is the total phase of the same level of the i-th grid, is the maximum total phase selected from the same level of each grid, and i is a positive integer.

8. The method for detecting water seepage in hydraulic structures according to claim 7, characterized in that: The first-type grid feature fusion unit is used to perform Hadamard product multiplication processing on the first-type grid features output by the first grid feature extraction unit, the first-type grid features output by the second grid feature extraction unit, and the first-type grid features output by the third grid feature extraction unit to obtain a first fused grid feature; The second-type grid feature fusion unit is used to perform Hadamard product multiplication on the second-type grid features output by the first grid feature extraction unit, the second-type grid features output by the second grid feature extraction unit, and the second-type grid features output by the third grid feature extraction unit to obtain second fused grid features.

Images