Reservoir slope cushion intelligent detection system and method based on three-dimensional laser scanning

Abstract

The application provides a reservoir slope cushion intelligent detection system and method based on three-dimensional laser scanning, and relates to the technical field of water conservancy engineering detection. The method comprises the following steps: obtaining high-precision point cloud data of the slope cushion through a three-dimensional laser scanning device; then, the high-precision point cloud data is preprocessed to construct a three-dimensional point cloud model, and target point cloud data sets corresponding to the distribution area of the slope cushion are extracted from the three-dimensional point cloud model; then, the target point cloud data sets are augmented to generate an enhanced point cloud sequence; then, the enhanced point cloud sequence is subjected to plane fitting analysis to calculate the fitting planes of each local area on the surface of the slope cushion; finally, based on the spatial position relationship between the fitting planes and each data point in the enhanced point cloud sequence, the thickness deviation data, the flatness deviation data and the slope deviation data of the slope cushion are calculated, and a comprehensive quality detection result is generated based on the calculation result. The application improves the precision and automation level of the construction quality detection of the reservoir slope cushion.

Description

Technical Field

[0001] This application relates to the field of water conservancy engineering testing technology, and in particular to an intelligent testing system and method for reservoir slope cushion layer based on three-dimensional laser scanning. Background Technology

[0002] As a key component of the dam's seepage prevention system, the construction quality of the reservoir slope cushion layer directly affects the reservoir's operational safety and service life. Traditional manual inspection methods are insufficient to meet the efficiency and accuracy requirements of large-scale water conservancy projects. With the development of 3D laser scanning technology, this technology, with its advantages of non-contact operation, high precision, and high efficiency, has shown broad application prospects in the field of water conservancy project quality inspection.

[0003] Currently, methods for detecting reservoir slope subgrade based on 3D laser scanning typically involve acquiring point cloud data of the subgrade surface using scanning equipment, stitching and filtering the acquired point cloud data to construct a 3D model reflecting the surface morphology of the subgrade, and then comparing and analyzing the constructed 3D model with design benchmarks to obtain detection data on geometric parameters such as subgrade thickness, flatness, and slope. This type of method enables rapid assessment of subgrade construction quality and has been initially applied in some water conservancy projects.

[0004] However, in real-world engineering environments, the point cloud data collected during scanning often suffers from sparsity or partial missing data due to factors such as ambient lighting, weather conditions, and construction interference. This makes it difficult for the constructed 3D model to fully represent the true geometry of the subgrade surface. The thickness, flatness, and slope parameters calculated based on this incomplete data are biased, resulting in the final inspection results failing to accurately reflect the actual construction quality of the subgrade. Therefore, existing technologies suffer from the technical problem of incomplete inspection data leading to insufficient accuracy in inspection results. Summary of the Invention

[0005] This application provides an intelligent detection system and method for reservoir slope cushion layer based on three-dimensional laser scanning, which solves the problems of low accuracy and poor automation level in the existing technology for detecting the construction quality of reservoir slope cushion layer.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides an intelligent detection method for reservoir slope cushion layers based on three-dimensional laser scanning, comprising:

[0007] The surface of the slope cushion layer of the reservoir is scanned non-contactly using a three-dimensional laser scanning device to obtain high-precision point cloud data of the slope cushion layer.

[0008] The high-precision point cloud data is preprocessed to construct a three-dimensional point cloud model, and the target point cloud dataset corresponding to the slope cushion layer distribution area is automatically identified and extracted from the three-dimensional point cloud model using a feature recognition algorithm.

[0009] A conditional generative adversarial network is used to augment the target point cloud dataset, generating an enhanced point cloud sequence.

[0010] The least squares plane fitting algorithm is used to perform plane fitting analysis on the enhanced point cloud sequence to calculate the fitting plane of each local area on the surface of the slope cushion layer;

[0011] Based on the spatial relationship between the fitted plane and the data points in the enhanced point cloud sequence, the thickness deviation data, flatness deviation data, and slope deviation data of the slope cushion layer are calculated, and a comprehensive quality inspection result is generated based on the calculation results.

[0012] Optionally, based on the spatial relationship between the fitted plane and the data points in the enhanced point cloud sequence, the thickness deviation data, flatness deviation data, and slope deviation data of the slope cushion layer are calculated, and a comprehensive quality inspection result is generated based on the calculation results, including:

[0013] For each set of data in the enhanced point cloud sequence, the vertical distance from each data point in each set of data to the corresponding fitting plane is calculated, and the vertical distance of each set of data is filtered using an outlier detection algorithm. The standard deviation of all vertical distances after filtering is used as the flatness deviation data.

[0014] For each grid cell, the average elevation of the corresponding data in the enhanced point cloud sequence is calculated, and the average elevation is compared with the design elevation to obtain the thickness deviation data.

[0015] Calculate the angle between the main slope direction represented by the enhanced point cloud sequence and the design direction, and use it as slope deviation data;

[0016] The flatness deviation data, the thickness deviation data, and the slope deviation data are compared with the preset allowable deviation range, and the grid cells and deviation values ​​that exceed the allowable deviation range are identified and recorded.

[0017] Based on the identification results and combined with the fuzzy comprehensive evaluation model, a quality level score is obtained for each grid cell, and a comprehensive quality inspection report is generated based on the quality level score.

[0018] Secondly, this application provides an intelligent detection system for reservoir slope cushion layers based on three-dimensional laser scanning, comprising:

[0019] The acquisition module is used to perform non-contact scanning of the surface of the slope cushion layer of the reservoir using a three-dimensional laser scanning device to acquire high-precision point cloud data of the slope cushion layer.

[0020] The construction module is used to preprocess the high-precision point cloud data to construct a three-dimensional point cloud model, and to use a feature recognition algorithm to automatically identify and extract the target point cloud dataset corresponding to the slope cushion layer distribution area from the three-dimensional point cloud model.

[0021] The generation module is used to augment the target point cloud dataset using a conditional generative adversarial network to generate an enhanced point cloud sequence.

[0022] The analysis module is used to perform plane fitting analysis on the enhanced point cloud sequence using the least squares plane fitting algorithm to calculate the fitting plane of each local area on the surface of the slope cushion layer.

[0023] The calculation module is used to calculate the thickness deviation data, flatness deviation data, and slope deviation data of the slope cushion layer based on the spatial positional relationship between the fitted plane and each data point in the enhanced point cloud sequence, and to generate a comprehensive quality inspection result based on the calculation results.

[0024] The intelligent detection method for reservoir slope cushion layer based on three-dimensional laser scanning provided in this application has the following beneficial effects:

[0025] This application acquires point cloud data using a 3D laser scanning device, enabling rapid acquisition of high-density surface information without contacting the slope cushion layer, providing accurate basic data for subsequent analysis. The acquired data is then preprocessed and feature-recognized, accurately separating the point cloud sets corresponding to the cushion layer distribution area while removing noise, effectively improving the targeting of subsequent analysis. Next, a conditional generative adversarial network is used to augment the target point cloud dataset, simulating the surface morphology of the cushion layer under various working conditions, making the analysis data more representative.

[0026] Then, the least squares plane fitting algorithm is used to perform partition fitting on the amplified point cloud sequence to obtain the best fitting plane for each local area, accurately restoring the actual surface geometry of the subbase. Finally, based on the positional relationship between the fitting plane and the point cloud data, various deviation indicators are calculated, which can comprehensively reflect the construction quality status of key parameters such as subbase thickness, flatness, and slope.

[0027] Furthermore, this application also effectively eliminates the influence of measurement noise on the flatness assessment by calculating the vertical distance from the point cloud data points to the fitting plane and using an outlier detection algorithm for screening. The standard deviation of the filtered distance is used as the flatness deviation data. At the same time, the average elevation of the point cloud is calculated for each grid cell and compared with the design elevation to obtain the thickness deviation data. The slope deviation data is obtained by calculating the angle between the main slope direction of the point cloud and the design direction. Based on this, the various deviation data are compared with the preset range to identify the areas exceeding the standard and record the deviation values. Finally, the fuzzy comprehensive evaluation model is used to perform multi-index fusion scoring on each grid cell to generate an inspection report containing quality level distribution information.

[0028] Furthermore, through multi-parameter collaborative analysis and fuzzy comprehensive evaluation, the construction quality of the subbase can be refined from two dimensions: spatial distribution and numerical deviation, resulting in more objective and comprehensive test results.

[0029] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 A flowchart illustrating an intelligent detection method for reservoir slope cushion layer based on three-dimensional laser scanning, provided in this application embodiment;

[0032] Figure 2 This application provides a schematic diagram illustrating a specific implementation of an intelligent detection method for reservoir slope cushion layer based on three-dimensional laser scanning.

[0033] Figure 3 This is a schematic diagram of a smart detection system for reservoir slope cushion layer based on three-dimensional laser scanning, provided in an embodiment of this application. Detailed Implementation

[0034] In the field of reservoir slope cushion layer construction quality inspection, existing three-dimensional laser scanning-based technologies typically involve collecting point cloud data of the cushion layer surface through scanning equipment, stitching and filtering the data to construct a three-dimensional model, and comparing and analyzing the model with the design benchmark to obtain geometric parameters such as thickness, flatness, and slope.

[0035] However, in actual engineering environments, the point cloud data collected during the scanning process often suffers from sparse or partially missing data due to factors such as ambient lighting, weather conditions, and construction interference. This makes it difficult for the constructed 3D model to fully represent the true geometric shape of the subgrade surface. The various geometric parameters calculated based on such incomplete data are biased, making it impossible for the final detection results to accurately reflect the actual construction quality of the subgrade. This contradiction stems from the existing methods' over-reliance on the integrity of the original collected data and the indiscriminate handling of data missing issues. There is an urgent need for a quality inspection method that can effectively address the situation of incomplete data.

[0036] To address the aforementioned issues, this application proposes an intelligent detection method for reservoir slope cushion layers based on three-dimensional laser scanning. The core of this method involves acquiring high-precision point cloud data of the slope cushion layer using a three-dimensional laser scanning device. The data is preprocessed to construct a three-dimensional point cloud model, and a feature recognition algorithm is used to extract the target point cloud dataset corresponding to the cushion layer distribution area. Then, a conditional generative adversarial network is used to augment the target point cloud dataset, generating an enhanced point cloud sequence. Next, a least-squares plane fitting algorithm is used to perform plane fitting analysis on the enhanced point cloud sequence, calculating the fitting plane for each local area on the slope cushion layer surface. Finally, based on the spatial relationship between the fitting plane and the data points in the enhanced point cloud sequence, the thickness deviation, flatness deviation, and slope deviation data of the slope cushion layer are calculated, and a comprehensive quality detection result is generated based on the calculation results.

[0037] Therefore, this method abandons the traditional approach of passively relying on the integrity of the original collected data. It uses a conditional generative adversarial network to intelligently amplify the limited collected data, generating surface morphology data of the subgrade under various working conditions. This fundamentally solves the problem of insufficient accuracy of detection results due to incomplete original point cloud data in existing technologies, and improves the reliability and environmental adaptability of reservoir slope subgrade construction quality detection.

[0038] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0039] The core of this application is to provide an intelligent detection method for reservoir slope cushion layers based on three-dimensional laser scanning. A flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0040] Step 101: Use a three-dimensional laser scanning device to perform a non-contact scan on the surface of the slope cushion layer of the reservoir to obtain high-precision point cloud data of the slope cushion layer.

[0041] In step 101, the three-dimensional laser scanning device is a measuring instrument that can emit laser beams and receive reflected signals from a distance in a non-contact manner to measure the spatial position of the target surface; the slope cushion layer refers to the structural layer laid on the surface of the reservoir slope for seepage prevention and protection, and its construction quality is directly related to the stability and durability of the reservoir slope.

[0042] High-precision point cloud data refers to a set of coordinates of a large number of discrete points that can accurately reflect the three-dimensional spatial location information of the slope cushion layer surface, obtained by a three-dimensional laser scanning device. Each data point contains the precise location information of that point in space.

[0043] In this embodiment of the application, a three-dimensional laser scanning device installed in a safe area outside the reservoir slope is used to perform multi-angle non-contact scanning of the slope cushion layer surface. The laser beam emitted by the device scans and covers the entire cushion layer area and then receives the reflected signal, thereby obtaining high-precision point cloud data that reflects the precise spatial position of the slope cushion layer surface.

[0044] Step 102: Preprocess the high-precision point cloud data to construct a three-dimensional point cloud model, and use a feature recognition algorithm to automatically identify and extract the target point cloud dataset corresponding to the slope cushion layer distribution area from the three-dimensional point cloud model.

[0045] Among them, the three-dimensional point cloud model refers to the set of point cloud data that can completely reflect the spatial morphology of the slope surface after the coordinate unification and noise filtering of high-precision point cloud data. The slope cushion layer distribution area refers to the spatial range on the slope surface where the cushion layer is actually laid. The target point cloud dataset refers to the set of point cloud data extracted from the three-dimensional point cloud model that accurately corresponds to the actual paving area of ​​the slope cushion layer.

[0046] In this embodiment, step 102 includes the following process:

[0047] Step 1021: Based on the design axis of the reservoir slope, coordinate registration is performed on the high-precision point cloud data obtained from different scanning stations to form an initial point cloud model.

[0048] In step 1021, the design axis refers to the baseline determined according to the reservoir slope design drawings, which is used to describe the extension direction of the main body of the slope. The initial point cloud model refers to the preliminary point cloud set formed after unifying the point cloud data of different scanning stations into the same coordinate system, which has not yet undergone noise filtering.

[0049] In this embodiment, the design axis of the reservoir slope is first used as a spatial reference. Multiple sets of high-precision point cloud data obtained from different scanning stations are registered in coordinates. By calculating the spatial transformation relationship between the cloud data of each station, the scattered point clouds are unified into the same coordinate system to form an initial point cloud model that completely covers the working surface of the slope.

[0050] Step 1022: The initial point cloud model is processed using a spatial clustering algorithm to identify noisy data points and remove the noisy data points from the initial point cloud model to obtain a three-dimensional point cloud model.

[0051] In step 1022, noise data points refer to anomalous data points that are spatially isolated from the main slope surface due to environmental interference or scanning errors.

[0052] In this embodiment, a spatial clustering algorithm is used to process the initial point cloud model. The algorithm divides the point cloud into different clusters based on the spatial distance and neighborhood density between data points, identifies isolated data points that are spatially separated from the point cloud clusters on the main slope surface as noise data points, and removes these noise data points from the initial point cloud model to obtain a three-dimensional point cloud model after removing interference.

[0053] Step 1023: Map the design boundary line of the slope cushion layer to the spatial location corresponding to the three-dimensional point cloud model to define the design boundary area.

[0054] In step 1023, the design boundary line refers to the theoretical outline line determined according to the subbase construction drawings and used to define the subbase laying range, and the design boundary area refers to the theoretical laying range enclosed by the design boundary line in three-dimensional space.

[0055] In this embodiment of the application, the design boundary line of the slope cushion layer is mapped onto the three-dimensional point cloud model according to the same spatial coordinate system. The design boundary area is enclosed according to the position of the design boundary line in space. This area is the spatial range that the cushion layer should theoretically exist in.

[0056] Step 1024: Using a feature recognition algorithm, calculate the normal direction of each point in the three-dimensional point cloud model within the design boundary area, and select points whose normal direction deviates from the normal direction of the design subfloor surface within a preset allowable range to form a candidate point cloud set.

[0057] In step 1024, the normal direction refers to the direction vector that passes through a data point in the 3D point cloud model and is perpendicular to the local fitting plane of that point. The normal direction of the design subbase surface refers to the vertical direction calculated based on the theoretical surface of the subbase in the design drawings.

[0058] The preset allowable range refers to the angle range that is pre-set to determine whether the deviation between the actual normal direction and the design normal direction is within acceptable limits. The candidate point cloud set refers to the set of point cloud data that initially conforms to the surface characteristics of the subbase after being filtered by the normal direction.

[0059] The embodiments of this application do not specifically limit the numerical value of the preset allowable range, which can be set according to the actual situation.

[0060] In this embodiment, a feature recognition algorithm is applied to the three-dimensional point cloud model within the design boundary area. The normal direction of each data point is obtained by calculating the local plane fitted by each data point and its neighboring points. The normal direction is compared with the normal direction of the design subfloor surface determined according to the design drawings. All data points whose normal direction deviation is within a preset allowable range are selected and these points are used as a candidate point cloud set.

[0061] Step 1025: Perform morphological closing operations on the candidate point cloud set to generate the target point cloud dataset.

[0062] In this embodiment of the application, morphological closing operations are performed on the candidate point cloud set. The small gaps and discontinuous regions in the candidate point cloud set caused by occlusion or reflection are connected by the dilation operation. Then, the edge contour of the point cloud set is smoothed by the erosion operation. After the processing is completed, a continuous and complete target point cloud dataset is generated.

[0063] This application effectively removes interference data from the original point cloud and accurately extracts the point cloud set corresponding to the cushion layer area through multi-level optimization of coordinate registration, noise filtering, boundary definition, normal filtering and morphological processing, providing an accurate and reliable data foundation for subsequent quality inspection.

[0064] Step 103: Use a conditional generative adversarial network to augment the target point cloud dataset and generate an enhanced point cloud sequence.

[0065] The conditional generative adversarial network can include two sub-networks: a generator and a discriminator. The generator adopts an encoder-decoder structure. The encoder consists of four convolutional layers to extract deep features from the input conditional data, and the decoder consists of four deconvolutional layers to reconstruct the deep features into simulated point cloud data. The encoder and decoder retain multi-scale feature information through skip connections.

[0066] The discriminator can adopt a conditional discriminant structure, consisting of five convolutional layers and two fully connected layers. The convolutional layers are used to extract the spatial distribution features of the input point cloud data, and the fully connected layers are used to fuse the extracted features with the material parameters and environmental parameters in the conditional data for judgment.

[0067] The training process of this conditional generative adversarial network includes: firstly, collecting multiple sets of real slope cushion layer point cloud data and their corresponding material parameters and environmental parameters from historical engineering data as training sample sets; then, combining the material parameters and environmental parameters in the training samples with random noise vectors and inputting them into the generator, which outputs simulated point cloud data; finally, inputting the simulated point cloud data and the corresponding real point cloud data, along with the same conditional data, into the discriminator, which outputs the judgment result on the authenticity of the two sets of data.

[0068] Calculate the loss function of the generator and the loss function of the discriminator. The generator's loss function encourages it to generate simulated data that can deceive the discriminator, while the discriminator's loss function encourages it to accurately distinguish between real and simulated data. Alternately optimize the network parameters of the generator and the discriminator through the backpropagation mechanism, and repeat the above process until the network converges.

[0069] It should be noted that the above structure is exemplary. This application does not impose specific limitations on the internal structure design of the conditional generative adversarial network, and corresponding settings can be made according to the actual situation.

[0070] An enhanced point cloud sequence refers to a set of point cloud data that consists of the original target point cloud dataset and multiple sets of simulated point cloud data, and can reflect the surface morphology of the subgrade under various working conditions.

[0071] In this embodiment, step 103 includes the following process:

[0072] Step 1031: Obtain the material and environmental parameters of the slope cushion layer.

[0073] In step 1031, material parameters refer to numerical information describing the physical properties of the slope cushion material, including the density, elastic modulus, and friction angle of the cushion; environmental parameters refer to numerical information describing the environmental conditions of the slope, including temperature, humidity, and load conditions.

[0074] In this embodiment of the application, the material parameters and environmental parameters of the slope cushion layer are first obtained by consulting the construction design documents and on-site monitoring equipment. These parameters will be used as constraints for the generation of subsequent simulated point cloud data.

[0075] Step 1032: Combine the target point cloud dataset, the material parameters, and the environmental parameters into conditional data.

[0076] In step 1032, conditional data refers to the combination of multi-source information formed by integrating the target point cloud dataset with material parameters and environmental parameters, and used as input to the conditional generative adversarial network to control the generation results.

[0077] Step 1033: Input the conditional data into the generator of the conditional generative adversarial network to obtain simulated point cloud data.

[0078] In step 1033, the generator is a neural network module in the conditional generative adversarial network that is responsible for generating new data samples based on the input conditional data. The simulated point cloud data refers to the point cloud data generated by the generator based on the conditional data, which simulates the surface morphology of the cushion layer under specific working conditions.

[0079] In this embodiment of the application, based on the pre-training of the conditional generative adversarial network, the conditional data formed in step 1032 is input into the generator of the network. The generator uses the target point cloud dataset in the conditional data as the basic shape, and the material parameters and environmental parameters as constraints, and generates simulated point cloud data that matches the input conditions through internal calculation.

[0080] Step 1034: Introduce the cosine similarity algorithm to calculate the matching degree between the distribution feature vector of the simulated point cloud data and the distribution feature vector of the target point cloud dataset.

[0081] In step 1034, the distribution feature vector refers to the vector composed of the feature values ​​of the point cloud data in spatial distribution extracted by statistical methods, and the matching degree refers to the value calculated by the cosine similarity algorithm, which reflects the degree of similarity between the simulated point cloud data and the target point cloud dataset in distribution features.

[0082] In this embodiment, a cosine similarity algorithm is introduced. First, the distribution feature vectors of the simulated point cloud data and the target point cloud dataset are extracted respectively. The distribution feature vectors include features such as point cloud density distribution, elevation distribution and normal direction distribution. Then, the cosine value of the angle between the two distribution feature vectors is calculated by the cosine similarity algorithm, and this value is used as the matching degree between the simulated point cloud data and the target point cloud dataset.

[0083] Step 1035: Generate a discriminator for the adversarial network based on the conditions, and determine whether the simulated point cloud data conforms to the distribution characteristics of the target point cloud dataset based on the matching degree. Based on the judgment result, iteratively optimize the generation parameters of the generator through a backpropagation mechanism.

[0084] In step 1035, the discriminator is a neural network module in the conditional generative adversarial network responsible for determining whether the input data is a real sample or a generated sample. The backpropagation mechanism is an optimization method that updates the neural network parameters by calculating the gradient of the loss function and propagating it in the back. The generation parameters refer to the weight coefficients of the connections between each layer in the generator network. Adjusting these parameters will affect the quality of the generated data.

[0085] Furthermore, the embodiments of this application do not specifically limit the specific implementation process of the backpropagation mechanism, and can be set accordingly according to the actual situation.

[0086] In this embodiment, the simulated point cloud data and the matching degree calculated in step 1034 are input together into the discriminator of the conditional generative adversarial network. The discriminator first performs a preliminary evaluation of the quality of the simulated point cloud data based on the matching degree. Then, it combines its own comparison of the distribution characteristics of the simulated point cloud data with those of the real target point cloud dataset to comprehensively determine whether the simulated point cloud data conforms to the distribution characteristics of the target point cloud dataset. The judgment result output by the discriminator is transmitted to the generator through a backpropagation mechanism. The generator adjusts its own generation parameters based on the feedback information and optimizes through multiple iterations to make the subsequently generated simulated point cloud data closer to the distribution characteristics of the real cushion layer surface.

[0087] In addition to the matching degree calculated by the cosine similarity algorithm as an auxiliary discrimination criterion, the discriminator of the conditional generative adversarial network also uses the spatial distribution characteristics of the simulated point cloud data itself as the main discrimination criterion. The main discrimination criterion includes the local geometric structure of the point cloud data, the relative positional relationship between points, the overall density distribution pattern, and the coupling consistency with the material parameters and environmental parameters in the conditional data. The discriminator extracts these spatial distribution characteristics through its internal convolutional layers and integrates them with the auxiliary matching degree information to comprehensively judge whether the simulated point cloud data conforms to the distribution characteristics of the target point cloud dataset.

[0088] Step 1036: Based on various preset working conditions, use the optimized generator to generate multiple sets of simulated point cloud data in a loop.

[0089] In step 1036, the preset multiple working conditions refer to the combination of different environmental conditions and load conditions set in advance. These working conditions include dry working conditions, rainy working conditions, low temperature working conditions and high load working conditions, etc.; cyclic generation refers to performing a generation process once for each preset working condition to obtain a corresponding set of simulated point cloud data.

[0090] In this embodiment of the application, various working conditions are preset according to actual engineering needs, including different temperature ranges, humidity ranges and load levels. For each preset working condition, the corresponding material parameters and environmental parameters are obtained. Steps 1031 to 1035 are repeated. An optimized generator is used to generate a set of corresponding simulated point cloud data for each working condition. The process is repeated until multiple sets of simulated point cloud data covering all preset working conditions are obtained.

[0091] Step 1037: Merge the target point cloud dataset with the multiple sets of simulated point cloud data, arrange them in the order of working conditions, and form an enhanced point cloud sequence.

[0092] In this embodiment of the application, the target point cloud dataset generated in step 102 is merged with the multiple sets of simulated point cloud data generated in step 1036, and arranged according to the preset working condition order to form an enhanced point cloud sequence containing the original data and multiple simulated working condition data. This sequence not only retains the surface information of the actual collected subgrade, but also supplements the subgrade morphology change data under multiple possible working conditions.

[0093] This application uses a conditional generative adversarial network to intelligently augment a limited target point cloud dataset. By combining material and environmental parameters as constraints, it generates simulated point cloud data covering various working conditions, effectively enriching the data samples of the subbase surface morphology and providing more comprehensive data support for subsequent quality inspection and analysis.

[0094] Step 104: Perform plane fitting analysis on the enhanced point cloud sequence using the least squares plane fitting algorithm to calculate the fitting plane of each local area on the surface of the slope cushion layer.

[0095] Among them, the local area refers to the sub-blocks after the slope cushion layer surface is divided according to certain rules, and the fitting plane refers to the virtual plane that best represents the overall trend of the cushion layer surface in each local area by using the least squares plane fitting algorithm.

[0096] In this embodiment, step 104 includes the following process:

[0097] Step 1041: Based on the construction quality specifications of the slope cushion layer, the horizontal projection area of ​​the enhanced point cloud sequence is divided into multiple grid units.

[0098] In step 1041, the construction quality specification refers to the technical requirements document for the construction quality inspection of slope cushion layer in water conservancy projects. The construction quality specification includes the size requirements for the division of the inspection area; the horizontal projection area refers to the planar range formed by vertically projecting all data points in the enhanced point cloud sequence onto the horizontal plane; and the grid cell refers to multiple regular-shaped sub-regions obtained by dividing the horizontal projection area according to the size specified in the construction quality specification.

[0099] In this embodiment, firstly, according to the size requirements for the division of the detection area in the construction quality specification of the slope cushion layer, all data points in the enhanced point cloud sequence are vertically projected onto the horizontal plane to obtain the horizontal projection area of ​​the entire cushion layer. Then, the horizontal projection area is divided into multiple grid units with the same shape and without overlapping each other according to the grid size specified in the specification. Each grid unit corresponds to a local area on the surface of the slope cushion layer.

[0100] In practical applications, assuming that the construction quality specification for a reservoir slope cushion layer stipulates that the size of the inspection grid unit is 2 meters by 2 meters, firstly, all data points in the enhanced point cloud sequence are vertically projected onto the horizontal plane to obtain a horizontal projection area covering the entire cushion layer. The length of this projection area is 50 meters and the width is 30 meters. Then, the horizontal projection area is divided according to the grid size of 2 meters by 2 meters, resulting in a total of 375 grid units.

[0101] Step 1042: For each grid cell, extract all data points located in the vertical projection space of the grid cell to form a cell data point set.

[0102] In step 1042, the vertical projection space refers to the three-dimensional columnar space range that extends vertically upwards from the grid cell to cover the highest point of the cushion layer surface, and the cell data point set refers to the set of all data points extracted from the enhanced point cloud sequence whose spatial positions fall within the vertical projection space of a certain grid cell.

[0103] In this embodiment of the application, for each grid cell divided in step 1041, a columnar space is constructed in three-dimensional space with the grid cell as the bottom surface and the vertical height range covering the entire thickness of the cushion layer. All data points whose spatial coordinates are located in the columnar space are selected from the enhanced point cloud sequence, and these data points are extracted as a unit data point set. This process is repeated until all grid cells have obtained the corresponding unit data point set.

[0104] In practical applications, for each grid cell divided in step 1041, such as the grid cell located in the 5th row and 8th column, a columnar space is constructed in three-dimensional space with the 2-meter by 2-meter grid as the base and extending vertically from the bottom elevation of the foundation layer of 100 meters to the top elevation of the foundation layer of 105 meters. Data points whose spatial coordinates satisfy the following conditions are selected from the enhanced point cloud sequence: the X direction is within the X coordinate range of the grid, the Y direction is within the Y coordinate range of the grid, and the Z direction is within the elevation range of 100 meters to 105 meters. A total of 328 data points are extracted, and these data points are used as the unit data point set of the grid cell. The corresponding unit data point sets are extracted for all 375 grid cells in the same way.

[0105] Step 1043: Apply the least squares plane fitting algorithm to each unit data point set to obtain a reference plane corresponding to the grid unit, and use the reference plane as the fitting plane corresponding to the grid unit.

[0106] In step 1043, the reference plane refers to the plane calculated by the least squares plane fitting algorithm that minimizes the sum of the squares of the vertical distances from all points in the unit data point set to the plane. The fitting plane refers to the reference plane corresponding to the grid unit, which is used to represent the overall trend of the surface of the cushion layer in the local area.

[0107] In this embodiment of the application, for each set of data points in the unit formed in step 1042, the least squares plane fitting algorithm is applied to perform the calculation. The algorithm solves a mathematical optimization problem to find a plane such that the sum of the squares of the vertical distances from all data points in the unit set to this plane reaches the minimum value. The calculated plane is used as the reference plane of the grid unit, and the reference plane is determined as the fitting plane of the corresponding grid unit. This process is repeated until all grid units obtain the corresponding fitting plane.

[0108] In practical applications, for the set of cell data points located in the 5th row and 8th column of the grid cell in step 1042, this set contains 328 data points, each with three-dimensional coordinates. Applying the least squares plane fitting algorithm, let the equation of the plane to be determined be... ,in, Let represent the plane fitting parameters to be determined, for the _th ... The coordinates of the data points are: The perpendicular distance from this point to the fitting plane is ,in, Indicates the first The algorithm calculates the equation of the reference plane corresponding to each of the 328 data points by finding the perpendicular distances from each data point to the fitted plane. This is achieved by solving for the parameters a, b, and c that minimize the sum of the squared perpendicular distances to all 328 points. The reference plane is used as the fitting plane for this mesh element, and the corresponding fitting planes are calculated for all 375 mesh elements in the same way.

[0109] This application transforms discrete point cloud data into a local surface representation with clear geometric meaning by dividing the enhanced point cloud sequence into grid cells and calculating the fitting plane for each cell, thus providing a unified reference surface for the accurate calculation of subsequent subbase thickness, flatness, and slope.

[0110] Step 105: Based on the spatial relationship between the fitted plane and the data points in the enhanced point cloud sequence, calculate the thickness deviation data, flatness deviation data, and slope deviation data of the slope cushion layer, and generate a comprehensive quality inspection result based on the calculation results.

[0111] Among them, thickness deviation data refers to the difference between the actual thickness of the slope cushion layer and the designed thickness; flatness deviation data refers to the difference between the degree of undulation of a local area on the surface of the slope cushion layer and the ideal plane; slope deviation data refers to the difference between the actual tilt direction of the overall slope surface of the slope cushion layer and the designed slope direction; and comprehensive quality inspection results refer to a complete set of inspection information including all deviation data, the location of the exceeding area, and the quality grade evaluation.

[0112] In this embodiment, step 105 includes the following process, such as... Figure 2 As shown:

[0113] Step 1051: For each set of data in the enhanced point cloud sequence, calculate the vertical distance from each data point in each set of data to the corresponding fitting plane, and use the outlier detection algorithm to filter the vertical distances of each set of data. The standard deviation of all vertical distances after filtering is used as the flatness deviation data.

[0114] In step 1051, the vertical distance refers to the shortest distance from a data point in space to the specified fitting plane, that is, the distance measured along the direction perpendicular to the plane; the standard deviation is a statistic that reflects the dispersion of a set of data.

[0115] In this embodiment, firstly, for each set of data in the enhanced point cloud sequence, the vertical distance from each data point in the set of data to the fitting plane of the grid cell where the set of data is located is calculated to obtain a set of vertical distance values; then, an outlier detection algorithm is applied to process this set of vertical distance values ​​to identify and remove vertical distance values ​​that deviate significantly due to measurement errors or local anomalies; finally, the standard deviation of the remaining vertical distance values ​​after filtering is calculated, and this standard deviation is used as the flatness deviation data under the corresponding working condition of the set of data.

[0116] In practical applications, taking the fitting plane corresponding to the grid cell in the 5th row and 8th column in step 1043 as an example, the equation of this fitting plane is: The data point set of this grid cell contains 328 data points. The vertical distance from each data point to the fitted plane is calculated, resulting in 328 vertical distance values. The isolated forest outlier detection algorithm is applied to these 328 vertical distance values ​​to identify and remove 8 obviously deviated vertical distance values. The standard deviation of the remaining 320 vertical distance values ​​is calculated, and the flatness deviation of this grid cell is found to be 0.023 meters.

[0117] Step 1052: For each grid cell, calculate the average elevation of the corresponding data in the enhanced point cloud sequence, compare the average elevation with the design elevation, and obtain the thickness deviation data.

[0118] In step 1052, the average elevation refers to the arithmetic mean of the elevation coordinates of all data points corresponding to a certain grid cell, and the design elevation refers to the theoretical elevation value that the surface of the cushion layer should reach at the center point of the grid cell, as determined by the slope cushion layer construction drawings.

[0119] In this embodiment of the application, for each grid cell, all data points corresponding to the grid cell are extracted from the enhanced point cloud sequence, the average value of the elevation coordinates of these data points is calculated as the average elevation of the grid cell, and then the average elevation is compared with the design elevation of the center point of the grid cell, the difference between the two is calculated, and the difference is used as the thickness deviation data of the grid cell.

[0120] In practical applications, taking the grid cell in the 5th row and 8th column of step 1043 as an example, the data point set corresponding to this grid cell contains 328 data points. The average elevation coordinates of these 328 data points are calculated, and the average elevation of this grid cell is 102.56 meters. The design elevation of the center point of this grid cell is 102.50 meters. Subtracting the design elevation from the average elevation gives a thickness deviation of +0.06 meters, indicating that the cushion layer of this grid cell is 0.06 meters thicker than the design thickness.

[0121] Step 1053: Calculate the angle between the main slope direction represented by the enhanced point cloud sequence and the design direction, as the slope deviation data.

[0122] In step 1053, the main slope direction refers to the spatial vector representing the overall tilt direction of the subbase, obtained by performing planar fitting or principal component analysis on the enhanced point cloud sequence. The design direction refers to the theoretical tilt direction vector that the subbase surface should have, determined according to the slope subbase construction drawings.

[0123] In this embodiment of the application, a spatial vector representing the main slope direction of the entire slope cushion layer is calculated by performing overall plane fitting or principal component analysis on all data points in the enhanced point cloud sequence. The main slope direction vector is then multiplied by the design direction vector obtained from the design drawings, and the magnitude is calculated. The spatial angle between the two vectors is obtained by using the inverse cosine function, and this angle is used as the slope deviation data of the entire slope cushion layer.

[0124] In practical applications, principal component analysis is performed on all data points in the enhanced point cloud sequence generated in step 103. The first principal component direction vector is calculated to be [0.15, 0.08, 0.99]. The design direction vector obtained from the design drawings is [0.14, 0.07, 0.99]. The dot product of the two vectors is calculated to be 0.15×0.14+0.08×0.07+0.99×0.99=0.9821. The magnitudes of the two vectors are both 1.0. Therefore, the inverse cosine of the included angle is arccos(0.9821)=3.5 degrees. This 3.5 degrees is used as the slope deviation data of the slope cushion layer.

[0125] Step 1054: Compare the flatness deviation data, the thickness deviation data, and the slope deviation data with the preset allowable deviation range, and identify and record the grid cells and deviation values ​​that exceed the allowable deviation range.

[0126] In step 1054, the preset allowable deviation range refers to the numerical range of each quality indicator that is acceptable according to the construction quality specification. A grid cell that exceeds the allowable deviation range refers to a grid cell in which at least one of the flatness deviation data, thickness deviation data, or slope deviation data is not within the corresponding allowable deviation range.

[0127] Deviation values ​​refer to the differences between various quality indicators obtained through calculation and design standards. Among them, the deviation value corresponding to the flatness deviation data is the standard deviation calculated after screening the vertical distance from each data point to the fitting plane. The deviation value corresponding to the thickness deviation data is the difference obtained by subtracting the design elevation from the average elevation of the grid cells. The deviation value corresponding to the slope deviation data is the spatial angle between the main slope direction vector and the design direction vector. These deviation values ​​together constitute the basic data for evaluating the construction quality of the slope cushion layer.

[0128] In this embodiment of the application, the flatness deviation data of each grid cell calculated in step 1051, the thickness deviation data of each grid cell calculated in step 1052, and the slope deviation data of the entire slope calculated in step 1053 are compared with the preset allowable deviation range. For grid cells whose flatness deviation data or thickness deviation data exceeds the allowable range, the location number of the grid cell and the specific deviation value are recorded. For cases where the slope deviation data exceeds the allowable range, the deviation value is recorded and the overall slope surface is marked as having a problem.

[0129] In practical applications, the construction quality specifications stipulate that the allowable deviation range for flatness is ≤0.05 meters, the allowable deviation range for thickness is -0.03 meters to +0.05 meters, and the allowable deviation range for slope is ≤2 degrees. For the grid cell in the 5th row and 8th column, its flatness deviation data is 0.023 meters, which is within the allowable range, and its thickness deviation data is +0.06 meters, which exceeds the allowable upper limit of +0.05 meters. Therefore, this grid cell is identified as a grid cell with excessive thickness deviation, and its location and deviation value of +0.06 meters are recorded. The slope deviation data is 3.5 degrees, which exceeds the allowable range of 2 degrees. Therefore, the detection conclusion that the overall slope deviation exceeds the standard is recorded.

[0130] Step 1055: Based on the identification results and combined with the fuzzy comprehensive evaluation model, obtain the quality level score of each grid unit, and generate a comprehensive quality inspection report based on the quality level score.

[0131] The fuzzy comprehensive evaluation model is a multi-index comprehensive evaluation method based on fuzzy mathematics theory. Its structure includes five core components: evaluation index set, evaluation level set, membership function, weight set, and fuzzy synthesis operator. The evaluation index set consists of three indicators: flatness deviation data, thickness deviation data, and slope deviation data. The evaluation level set is set with three quality levels: qualified, warning, and unqualified. The membership function adopts a triangular distribution function to map the specific deviation values ​​of each indicator to the degree of membership to each level. The weight set assigns weight coefficients of 0.3, 0.4, and 0.3 to the flatness, thickness, and slope indicators respectively according to construction specifications. The fuzzy synthesis operator uses a weighted average operator for comprehensive calculation.

[0132] This model does not require a traditional training process. Instead, it pre-sets the form and parameters of the membership function and the allocation of weight coefficients based on expert experience and standard requirements. In practical applications, the comprehensive evaluation result can be obtained by directly substituting the deviation data of each grid cell into the model for calculation.

[0133] It should be noted that the above structure is exemplary. This application does not impose specific limitations on the internal structure design of the fuzzy comprehensive evaluation model, and corresponding settings can be made according to the actual situation.

[0134] The quality grade score refers to the quantitative result that reflects the construction quality grade of the grid unit, calculated by the fuzzy comprehensive evaluation model. The comprehensive quality inspection report is an inspection document that includes the quality grade scores of all grid units, the marking of areas exceeding the standard, and the overall evaluation conclusion.

[0135] Step 1055 may specifically include the following steps:

[0136] A1: Using the flatness deviation data, the thickness deviation data, and the slope deviation data as evaluation indicators, and according to the preset membership function, calculate the membership degree of each evaluation indicator of each grid cell in the identification result to the preset multiple quality levels, and assign preset weight coefficients to each evaluation indicator.

[0137] In step A1, the evaluation index refers to various parameters used to evaluate the construction quality of the slope cushion layer, including flatness deviation data, thickness deviation data, and slope deviation data. The membership function is a mathematical function established according to the water conservancy project cushion layer quality acceptance specification, which maps specific deviation values ​​to the degree of membership in each quality level. Membership degree refers to the degree to which a certain evaluation index of a certain grid cell belongs to a certain quality level, and its value ranges from 0 to 1. The weight coefficient is a proportional coefficient allocated according to the degree of influence of each evaluation index on the seepage prevention performance and stability of the reservoir slope cushion layer, and the sum of the weight coefficients of all indicators is 1.

[0138] It should be noted that this embodiment does not specifically limit the specific expression used for the membership function; it can be set according to the actual situation.

[0139] In this embodiment of the application, firstly, for all grid cells identified in step 1054, the flatness deviation data, thickness deviation data, and slope deviation data of each grid cell are used as three evaluation indicators. According to the preset membership function, the membership value of each evaluation indicator of each grid cell to the three quality levels of qualified, warning, and unqualified is calculated. At the same time, according to the reservoir slope cushion layer construction quality specification, preset weight coefficients are assigned to the three evaluation indicators of flatness, thickness, and slope.

[0140] In practical applications, assuming the preset membership function is a triangular distribution function, for the grid cell in the 5th row and 8th column of a reservoir slope cushion layer, its flatness deviation data of 0.023 meters yields a membership degree of 0.8 for the qualified level, 0.2 for the warning level, and 0 for the unqualified level. Its thickness deviation data of +0.06 meters yields a membership degree of 0 for the qualified level, 0.3 for the warning level, and 0.7 for the unqualified level. Its slope deviation data of 3.5 degrees yields a membership degree of 0 for the qualified level, 0.1 for the warning level, and 0.9 for the unqualified level. According to the reservoir engineering design specifications, the weight coefficients assigned to flatness, thickness, and slope are 0.3, 0.4, and 0.3, respectively.

[0141] A2: Input the membership degree and corresponding weight coefficient of each evaluation index into the fuzzy comprehensive evaluation model. The fuzzy comprehensive evaluation model calculates the membership degree and the weight coefficient according to the fuzzy synthesis rules and outputs the comprehensive evaluation result of each grid cell.

[0142] In step A2, the fuzzy synthesis rule refers to the mathematical rule that combines the membership degree of each evaluation index with the corresponding weight coefficient; the comprehensive evaluation result refers to the vector obtained by fuzzy synthesis calculation, which reflects the comprehensive membership degree of the grid unit to each quality level, and is used to characterize the overall status of the subbase construction quality of the grid unit.

[0143] In this embodiment of the application, the membership values ​​of the three evaluation indicators of each grid cell calculated in step A1, corresponding to the three quality levels, are input into the fuzzy comprehensive evaluation model along with the corresponding weight coefficients. The model adopts a weighted average fuzzy synthesis rule, which multiplies the membership of each evaluation indicator at each quality level by the corresponding weight coefficient and then sums them to obtain the comprehensive membership of the grid cell to that quality level. This forms a vector containing the comprehensive membership of the three quality levels as the comprehensive evaluation result of the grid cell.

[0144] In practical application, for the grid cell in the 5th row and 8th column of the reservoir slope cushion layer, the comprehensive membership degree for the qualified level is 0.8×0.3+0×0.4+0×0.3=0.24, the comprehensive membership degree for the warning level is 0.2×0.3+0.3×0.4+0.1×0.3=0.06+0.12+0.03=0.21, and the comprehensive membership degree for the unqualified level is 0×0.3+0.7×0.4+0.9×0.3=0+0.28+0.27=0.55. Therefore, the comprehensive evaluation result of this grid cell is the vector [0.24, 0.21, 0.55].

[0145] A3: Based on the comprehensive evaluation results, the quality level with the highest probability is determined as the quality level score of the grid cell.

[0146] In this embodiment of the application, based on the comprehensive evaluation result of each grid unit obtained in step A2, the grade with the largest comprehensive membership degree among the three quality grades of qualified, warning, and unqualified is selected as the quality grade score of the grid unit, that is, the final evaluation grade of the subbase construction quality of the grid unit.

[0147] In practical application, for the grid cell in the 5th row and 8th column of the reservoir slope cushion layer, the comprehensive evaluation results are: qualified level membership degree 0.24, warning level membership degree 0.21, and unqualified level membership degree 0.55. Among them, the unqualified level has the largest membership degree. Therefore, the quality level score of this grid cell is determined to be unqualified, indicating that the construction quality of the cushion layer in this area does not meet the design requirements.

[0148] A4: Generate a comprehensive quality inspection report based on the quality level scores and spatial location information of all grid cells.

[0149] In this embodiment of the application, the spatial location information of all grid units and the quality grade score of each grid unit determined in step A3 are collected. This information is integrated to generate a comprehensive quality inspection report. The report marks the location of each grid unit and its corresponding quality grade in the form of a reservoir slope plan projection map, so that reservoir engineering managers can intuitively understand the quality distribution of the entire slope cushion layer.

[0150] In practical applications, the row and column numbers and corresponding quality grade scores of all 375 grid units of the reservoir slope are collected. For example, the grid unit in the 5th row and 8th column is unqualified, and the grid unit in the 5th row and 9th column is qualified. A two-dimensional planar map is generated, and each grid unit is filled with different colors to represent its quality grade. Red indicates unqualified, yellow indicates warning, and green indicates qualified. A statistical summary table of various deviation data is attached to form a complete inspection report applicable to the quality acceptance of the reservoir slope cushion layer.

[0151] In this embodiment, after step 105, the method further includes the following steps:

[0152] B1: The isolated forest algorithm is used to perform global anomaly detection on the various deviation data in the comprehensive quality inspection results, and to identify abnormal detection units that deviate from the normal construction mode in terms of the combination characteristics of flatness, thickness and slope deviation.

[0153] In step B1, the anomaly detection unit refers to a grid unit that differs significantly from most other grid units of the reservoir slope cushion layer in terms of the combined characteristics of flatness deviation data, thickness deviation data, and slope deviation data.

[0154] In this embodiment of the application, the flatness deviation data, thickness deviation data and slope deviation data of all grid cells recorded in step 1054 are first combined into a multi-dimensional feature vector. These feature vectors are then input into the isolated forest algorithm. The algorithm constructs multiple isolated trees and calculates the average path length of each sample in the tree. Samples with shorter path lengths are identified as abnormal samples. The grid cells corresponding to these abnormal samples are the abnormal detection units that deviate from the normal construction mode in the combined deviation features of the reservoir slope cushion layer.

[0155] In practical applications, the flatness deviation data, thickness deviation data, and overall slope deviation data of all 375 grid units of the reservoir slope are combined into a three-dimensional feature vector for each grid unit. This vector is then input into the isolated forest algorithm, with the anomaly ratio set to 0.1. After calculation, the algorithm identifies 32 grid units as anomaly detection units, including the grid unit in the 5th row and 8th column identified in step 1054.

[0156] B2: Input the spatial coordinates of the anomaly detection unit and the corresponding multidimensional deviation data into the long short-term memory network prediction model to predict the quality degradation trend of the anomaly detection unit within a specified time window in the future.

[0157] In step B2, the quality degradation trend refers to the prediction results of the change law of various deviation data of the anomaly detection unit over a period of time, which is used to assess the potential quality problems of the reservoir slope cushion layer in the future.

[0158] This application does not impose specific limitations on the model type, internal structure design, parameter design, training process, etc. of the Long Short-Term Memory Network prediction model, and corresponding settings can be made according to the actual situation.

[0159] In this embodiment of the application, the spatial coordinates of each anomaly detection unit identified in step B1 and its corresponding flatness deviation data and thickness deviation data are used as the initial state of the multidimensional time series and input into the pre-trained long short-term memory network prediction model. Based on the changes in the quality of the cushion layer learned from the historical monitoring data of the reservoir, the model predicts the quality degradation trend of each anomaly detection unit within a specified time window in the future and outputs a sequence of predicted values ​​of various deviation data over time.

[0160] The specific implementation process is as follows: First, the spatial coordinates of each anomaly detection unit identified in step B1 are converted into grid unit numbers, and the flatness deviation data sequence and thickness deviation data sequence corresponding to the unit are arranged in chronological order. The data at each time point constitutes a two-dimensional feature vector, and the feature vectors of multiple consecutive time points constitute the multi-dimensional time series sample of the unit. Then, these time series samples are divided according to the length of the time window, for example, with a prediction period of 30 days and a time step of 5 days. The data of the first 6 time steps are used as the input sequence and the data of the last 6 time steps are used as the output sequence to construct the training dataset.

[0161] Next, the training data is input into a pre-constructed long short-term memory network for training. This network includes an input layer for receiving two-dimensional feature vectors, two long short-term memory hidden layers for extracting time-dependent features, and a fully connected output layer for generating predicted values ​​for future time steps. After training, the deviation data for six consecutive time steps before the current time point is input into the trained network. The network calculates and outputs the predicted values ​​of flatness deviation and thickness deviation for the next six time steps through forward propagation, thereby obtaining the quality degradation trend of each anomaly detection unit within a specified future time window.

[0162] In practical applications, the spatial coordinates and current deviation data of the 32 anomaly detection units identified in step B1 are input into the Long Short-Term Memory Network prediction model. This model has been trained using the reservoir's cushion layer monitoring data from the past two years. The model outputs the predicted flatness deviation and thickness deviation values ​​for each anomaly detection unit every 5 days over the next 30 days. For the grid unit in the 5th row and 8th column of the reservoir slope, the model predicts that its thickness deviation data will deteriorate from the current +0.06 meters to +0.12 meters after 30 days.

[0163] B3: Based on the predicted results of quality degradation trends, generate a quality early warning report, which includes the location of abnormal units, the current deviation status, and the future risk level.

[0164] In step B3, the quality early warning report refers to a comprehensive early warning document that includes the spatial location of the anomaly detection unit, the current deviation value, the prediction result of the future quality degradation trend, and the risk level information classified according to the degree of degradation; the future risk level refers to the classification result of the anomaly detection unit into different early warning levels according to the predicted degradation trend, which is used to guide the maintenance priority of the reservoir slope cushion layer.

[0165] In this embodiment of the application, based on the quality degradation trend of each anomaly detection unit predicted in step B2, and combined with the preset risk level classification standard, the future risk level of each anomaly detection unit is determined. Then, the spatial coordinates of the anomaly detection unit, the current deviation data, the predicted value of the quality degradation trend, and the determined risk level information are integrated to generate a quality early warning report containing text descriptions and charts.

[0166] In practical application, based on the predicted thickness deviation of the grid unit in the 5th row and 8th column of the reservoir slope after 30 days +0.12 meters, and in accordance with the reservoir engineering maintenance specifications that a thickness deviation > +0.10 meters is considered high risk, this grid unit is determined to be of high risk level. Combined with the risk level classification results of the other 31 anomaly detection units, a quality early warning report applicable to the reservoir slope cushion layer is generated. The report includes a reservoir slope plan map marking the location of the high-risk area, a list of current deviation data for each anomaly detection unit, and a quality degradation trend curve for the next 30 days.

[0167] Figure 3 A schematic diagram of a smart detection system for reservoir slope cushion layer based on three-dimensional laser scanning, provided in an embodiment of this application, is shown below. Figure 3 As shown, the system includes:

[0168] The acquisition module 31 is used to perform non-contact scanning of the surface of the slope cushion layer of the reservoir using a three-dimensional laser scanning device to acquire high-precision point cloud data of the slope cushion layer.

[0169] The construction module 32 is used to preprocess the high-precision point cloud data to construct a three-dimensional point cloud model, and to use a feature recognition algorithm to automatically identify and extract the target point cloud dataset corresponding to the slope cushion layer distribution area from the three-dimensional point cloud model.

[0170] The generation module 33 is used to augment the target point cloud dataset using a conditional generative adversarial network to generate an enhanced point cloud sequence.

[0171] Analysis module 34 is used to perform plane fitting analysis on the enhanced point cloud sequence using the least squares plane fitting algorithm to calculate the fitting plane of each local area on the surface of the slope cushion layer.

[0172] The calculation module 35 is used to calculate the thickness deviation data, flatness deviation data, and slope deviation data of the slope cushion layer based on the spatial positional relationship between the fitted plane and each data point in the enhanced point cloud sequence, and to generate a comprehensive quality inspection result based on the calculation results.

[0173] The intelligent detection system for reservoir slope cushion layer based on three-dimensional laser scanning in this application embodiment is used to implement the aforementioned intelligent detection method for reservoir slope cushion layer based on three-dimensional laser scanning. Therefore, the specific implementation of the intelligent detection system for reservoir slope cushion layer based on three-dimensional laser scanning can be found in the embodiment section of the intelligent detection method for reservoir slope cushion layer based on three-dimensional laser scanning above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0174] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described intelligent detection methods for reservoir slope cushion layers based on three-dimensional laser scanning.

[0175] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described intelligent detection methods for reservoir slope cushion layers based on three-dimensional laser scanning.

[0176] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0177] The embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the intelligent detection method for reservoir slope cushion layer based on three-dimensional laser scanning.

[0178] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0179] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0180] The above provides a detailed description of the intelligent detection system and method for reservoir slope cushion layer based on three-dimensional laser scanning provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this application. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A smart detection method for reservoir slope cushion layer based on three-dimensional laser scanning, characterized in that, include: The surface of the slope cushion layer of the reservoir is scanned non-contactly using a three-dimensional laser scanning device to obtain high-precision point cloud data of the slope cushion layer. The high-precision point cloud data is preprocessed to construct a three-dimensional point cloud model, and the target point cloud dataset corresponding to the slope cushion layer distribution area is automatically identified and extracted from the three-dimensional point cloud model using a feature recognition algorithm. A conditional generative adversarial network is used to augment the target point cloud dataset, generating an enhanced point cloud sequence. The least squares plane fitting algorithm is used to perform plane fitting analysis on the enhanced point cloud sequence to calculate the fitting plane of each local area on the surface of the slope cushion layer; Based on the spatial relationship between the fitted plane and each data point in the enhanced point cloud sequence, the thickness deviation data, flatness deviation data, and slope deviation data of the slope cushion layer are calculated, and a comprehensive quality inspection result is generated based on the calculation results. Based on the spatial relationship between the fitted plane and the data points in the enhanced point cloud sequence, the thickness deviation data, flatness deviation data, and slope deviation data of the slope cushion layer are calculated. Based on the calculation results, a comprehensive quality inspection result is generated, including: For each set of data in the enhanced point cloud sequence, the vertical distance from each data point in each set of data to the corresponding fitting plane is calculated, and the vertical distance of each set of data is filtered using an outlier detection algorithm. The standard deviation of all vertical distances after filtering is used as the flatness deviation data. For each grid cell, the average elevation of the corresponding data in the enhanced point cloud sequence is calculated, and the average elevation is compared with the design elevation to obtain the thickness deviation data. Calculate the angle between the main slope direction represented by the enhanced point cloud sequence and the design direction, and use it as slope deviation data; The flatness deviation data, the thickness deviation data, and the slope deviation data are compared with the preset allowable deviation range, and the grid cells and deviation values ​​that exceed the allowable deviation range are identified and recorded. Based on the identification results and combined with the fuzzy comprehensive evaluation model, a quality level score is obtained for each grid unit, and a comprehensive quality inspection report is generated based on the quality level score.

2. The method according to claim 1, characterized in that, Based on the identification results and combined with a fuzzy comprehensive evaluation model, a quality level score is obtained for each grid cell, and a comprehensive quality inspection report is generated based on the quality level score, including: Using the flatness deviation data, the thickness deviation data, and the slope deviation data as evaluation indicators, and according to the preset membership function, the membership degree of each evaluation indicator of each grid cell in the identification result to multiple preset quality levels is calculated, and a preset weight coefficient is assigned to each evaluation indicator. The membership degree and corresponding weight coefficient of each evaluation index are input into the fuzzy comprehensive evaluation model. The fuzzy comprehensive evaluation model calculates the membership degree and the weight coefficient according to the fuzzy synthesis rule and outputs the comprehensive evaluation result of each grid cell. Based on the comprehensive evaluation results, the quality level with the highest probability is determined as the quality level score of the grid cell; Based on the quality grade scores and spatial location information of all grid cells, a comprehensive quality inspection report is generated.

3. The method according to claim 1, characterized in that, The step of performing plane fitting analysis on the enhanced point cloud sequence using the least squares plane fitting algorithm to calculate the fitting plane for each local region on the surface of the slope cushion layer includes: Based on the construction quality specifications for slope cushion layers, the horizontal projection area of ​​the enhanced point cloud sequence is divided into multiple grid units; For each grid cell, extract all data points located in the vertical projection space of the grid cell to form a cell data point set; The least squares plane fitting algorithm is applied to each set of data points to obtain a reference plane corresponding to the grid cell, and the reference plane is used as the fitting plane corresponding to the grid cell.

4. The method according to claim 1, characterized in that, The step of using a conditional generative adversarial network to augment the target point cloud dataset and generate an enhanced point cloud sequence includes: Obtain the material and environmental parameters of the slope cushion layer; The target point cloud dataset, the material parameters, and the environmental parameters are combined into conditional data. The conditional data is input into the generator of the conditional generative adversarial network to obtain simulated point cloud data; A cosine similarity algorithm is introduced to calculate the matching degree between the distribution feature vector of the simulated point cloud data and the distribution feature vector of the target point cloud dataset; The discriminator of the Generative Adversarial Network (GAN) is generated based on the conditions. It uses the matching degree to help determine whether the simulated point cloud data conforms to the distribution characteristics of the target point cloud dataset. Based on the judgment result, the generation parameters of the generator are iteratively optimized through the backpropagation mechanism. Based on various preset working conditions, the optimized generator is used to generate multiple sets of simulated point cloud data in a loop. The target point cloud dataset is merged with the multiple sets of simulated point cloud data and arranged in the order of working conditions to form an enhanced point cloud sequence.

5. The method according to claim 1, characterized in that, The process of preprocessing the high-precision point cloud data to construct a three-dimensional point cloud model, and automatically identifying and extracting the target point cloud dataset corresponding to the slope cushion layer distribution area from the three-dimensional point cloud model using a feature recognition algorithm, includes: Based on the design axis of the reservoir slope, coordinate registration is performed on the high-precision point cloud data obtained from different scanning stations to form an initial point cloud model; The initial point cloud model is processed using a spatial clustering algorithm to identify noisy data points, and the noisy data points are removed from the initial point cloud model to obtain a three-dimensional point cloud model. The design boundary line of the slope cushion layer is mapped to the spatial location corresponding to the three-dimensional point cloud model to define the design boundary area; Using a feature recognition algorithm, the normal direction of each point in the three-dimensional point cloud model is calculated within the design boundary area, and points whose normal direction deviates from the normal direction of the design subfloor surface within a preset allowable range are selected to form a candidate point cloud set. The candidate point cloud set is subjected to morphological closing operation to generate the target point cloud dataset.

6. The method according to claim 1, characterized in that, After generating the comprehensive quality inspection results based on the calculation results, the following is also included: The isolated forest algorithm is used to perform global anomaly detection on various deviation data in the comprehensive quality inspection results, and to identify abnormal detection units that deviate from the normal construction mode in terms of the combination characteristics of flatness, thickness and slope deviation. The spatial coordinates of the anomaly detection unit and the corresponding multidimensional deviation data are input into the long short-term memory network prediction model to predict the quality degradation trend of the anomaly detection unit within a specified time window in the future. Based on the predicted results of quality degradation trends, a quality early warning report is generated, which includes the location of abnormal units, the current deviation status, and the future risk level.

7. A smart detection system for reservoir slope cushion layer based on three-dimensional laser scanning, characterized in that, include: The acquisition module is used to perform non-contact scanning of the surface of the slope cushion layer of the reservoir using a three-dimensional laser scanning device to acquire high-precision point cloud data of the slope cushion layer. The construction module is used to preprocess the high-precision point cloud data to construct a three-dimensional point cloud model, and to use a feature recognition algorithm to automatically identify and extract the target point cloud dataset corresponding to the slope cushion layer distribution area from the three-dimensional point cloud model. The generation module is used to augment the target point cloud dataset using a conditional generative adversarial network to generate an enhanced point cloud sequence. The analysis module is used to perform plane fitting analysis on the enhanced point cloud sequence using the least squares plane fitting algorithm to calculate the fitting plane of each local area on the surface of the slope cushion layer. The calculation module is used to calculate the thickness deviation data, flatness deviation data, and slope deviation data of the slope cushion layer based on the spatial positional relationship between the fitting plane and each data point in the enhanced point cloud sequence, and to generate a comprehensive quality inspection result based on the calculation results. Based on the spatial relationship between the fitted plane and the data points in the enhanced point cloud sequence, the thickness deviation data, flatness deviation data, and slope deviation data of the slope cushion layer are calculated. Based on the calculation results, a comprehensive quality inspection result is generated, including: For each set of data in the enhanced point cloud sequence, the vertical distance from each data point in each set of data to the corresponding fitting plane is calculated, and the vertical distance of each set of data is filtered using an outlier detection algorithm. The standard deviation of all vertical distances after filtering is used as the flatness deviation data. For each grid cell, the average elevation of the corresponding data in the enhanced point cloud sequence is calculated, and the average elevation is compared with the design elevation to obtain the thickness deviation data. Calculate the angle between the main slope direction represented by the enhanced point cloud sequence and the design direction, and use it as slope deviation data; The flatness deviation data, the thickness deviation data, and the slope deviation data are compared with the preset allowable deviation range, and the grid cells and deviation values ​​that exceed the allowable deviation range are identified and recorded. Based on the identification results and combined with the fuzzy comprehensive evaluation model, a quality level score is obtained for each grid unit, and a comprehensive quality inspection report is generated based on the quality level score.

Images