A holographic deformation monitoring and analyzing method for construction of a pumped storage power station

Abstract

The application discloses a kind of holographic deformation monitoring and analysis methods for pumped storage power station construction, belong to holographic deformation monitoring technical field, this method utilizes multidimensional holographic acquisition unit to continuously dynamic scanning to construction area, obtains spatial coding holographic image sequence and multispectral phase information;Holographic image sequence is input three-dimensional reconstruction network and carries out object light wavefront recovery, obtains three-dimensional holographic point cloud data;Based on the data extraction surface displacement field, depth strain distribution and key point relative displacement amount etc.Deformation characteristic parameter;Four-dimensional space-time deformation evolution model is constructed to fusion space and time dimension, identifies abnormal deformation mode and early warning security hidden danger.The present application effectively solves the problem that traditional monitoring data processing lag leads to deformation anomaly difficult to find in real time, significantly improves the spatial coverage, time continuity and early warning timeliness of pumped storage power station construction deformation monitoring.

Description

Technical Field

[0001] This invention belongs to the field of holographic deformation monitoring technology, and in particular, a holographic deformation monitoring and analysis method for the construction of pumped storage power stations. Background Technology

[0002] With the continuous expansion of pumped storage power station construction, the demand for deformation monitoring of key structures such as dams, slopes, and underground caverns during construction is becoming increasingly urgent. Existing deformation monitoring technologies mainly rely on total stations, levels, and discrete sensors for point cloud-based displacement monitoring, or use contact sensor arrays combined with traditional photogrammetry for periodic data acquisition. However, these methods have significant limitations in terms of spatial coverage, temporal continuity, deformation magnitude capture capability, and real-time early warning, making it difficult to meet the needs of structural safety management in complex construction environments.

[0003] A search revealed a patent, CN121932958A, titled "A Method and Equipment for Monitoring Surface Deformation Based on Oblique Photogrammetry." This patent utilizes an unmanned aerial vehicle (UAV) surveying system equipped with cameras and sensors to conduct aerial surveys of a target area, acquiring image data from different periods and corresponding POS data, achieving centimeter-level accuracy in monitoring horizontal surface movement. However, this technical solution has the following shortcomings: First, it primarily targets surface deformation monitoring in open-pit mine slopes, subsidence areas, and areas prone to geological disasters, failing to specifically optimize for complex conditions such as large-volume concrete pouring and underground cavern excavation during pumped-storage power station construction. Second, the method employs a periodic aerial survey mode, resulting in long data acquisition intervals, making it difficult to achieve real-time continuous monitoring under dynamic loads during construction. Third, its reliable capture capability for deep displacement and minute deformation magnitudes is limited, failing to meet the requirements of pumped-storage power stations for precise characterization of subtle structural deformations.

[0004] A search revealed a construction deviation detection system and method (publication number CN122041817A). This patent uses a combination of binocular cameras and laser scanning equipment to acquire 2D images and 3D point cloud data of construction components, and performs construction deviation analysis through image preprocessing, feature extraction, and deviation detection modules. However, this technical solution has the following shortcomings: First, it mainly targets deviation detection of construction components and fails to cover the overall deformation monitoring needs of large-scale three-dimensional spaces such as dams, slopes, and deep underground caverns in pumped storage power station construction; second, the method relies on laser scanning equipment for point cloud acquisition, and the integrity and accuracy of the point cloud data are difficult to guarantee when faced with obstructions and complex surfaces caused by large-volume concrete pouring; third, its data processing flow is relatively complex, and the lag in processing monitoring data makes it impossible to detect and warn of deformation anomalies in a timely manner, which is difficult to meet the real-time requirements of safety management in pumped storage power station construction.

[0005] The shortcomings of the existing technical solutions mentioned above in terms of the specificity of the monitoring targets, spatiotemporal continuity, ability to capture minute deformations, and real-time early warning capabilities have hindered the development of deformation monitoring technology for pumped storage power station construction. Therefore, this invention provides a holographic deformation monitoring and analysis method for pumped storage power station construction, aiming to achieve real-time, full-site, and high-precision deformation monitoring and analysis throughout the entire construction process, meeting the urgent needs of safety management and control during pumped storage power station construction. Summary of the Invention

[0006] The purpose of this invention is to provide a holographic deformation monitoring and analysis method for the construction of pumped storage power stations, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a holographic deformation monitoring and analysis method for the construction of pumped storage power stations, comprising the following steps: A multi-dimensional holographic acquisition unit was used to continuously and dynamically scan the construction area of ​​the pumped storage power station to obtain the spatially encoded holographic image sequence of the monitoring area and the corresponding multispectral phase information. The multi-dimensional holographic acquisition unit includes at least three synchronous imaging devices in different spatial locations, and each of the synchronous imaging devices performs cross-coverage imaging of the same monitoring area at a preset sampling frequency; The spatially encoded holographic image sequence is input into a three-dimensional reconstruction network for object wavefront restoration to obtain three-dimensional holographic point cloud data of the monitoring area. The 3D reconstruction network is built on a deep convolutional neural network, and its network weights are obtained through pre-training using a large amount of holographic data from hydropower engineering scenarios. Based on the three-dimensional holographic point cloud data, deformation feature parameters of the monitoring area are extracted. These deformation feature parameters include the surface displacement field, depth strain distribution, and relative displacement of key structural points. The three-dimensional holographic point cloud data at different times are registered and compared to calculate the total deformation of the monitoring area under construction load. A four-dimensional spatiotemporal deformation evolution model of the monitoring area is constructed. The four-dimensional spatiotemporal deformation evolution model integrates the three-dimensional spatial coordinates with the time dimension to characterize the dynamic law of deformation of the monitoring area as the construction process progresses. Based on the aforementioned four-dimensional spatiotemporal deformation evolution model, abnormal deformation patterns are identified, and early warning judgments are made on potential structural safety hazards.

[0008] In a preferred embodiment of this scheme, the continuous dynamic scanning of the pumped storage power station construction area using a multi-dimensional holographic acquisition unit includes: rationally deploying each of the aforementioned synchronous imaging devices around the monitoring area, such that any monitoring point is covered by at least two synchronous imaging devices at different angles; the distance between each of the synchronous imaging devices is determined according to the spatial range and accuracy requirements of the monitoring area; controlling each of the synchronous imaging devices to synchronously trigger shooting at time intervals not exceeding 0.5 seconds to ensure that the acquired holographic image sequence has sufficient temporal resolution in the time dimension; during each shooting process, each of the aforementioned synchronous imaging devices simultaneously acquires the interference pattern of the reference beam and the object beam, and recovers the object beam wavefront information of the monitoring area through a phase decoding algorithm.

[0009] In a preferred embodiment of this scheme, the step of inputting the spatially encoded holographic image sequence into a three-dimensional reconstruction network for object wavefront restoration includes: preprocessing the holographic images acquired by each of the synchronous imaging devices, including removing background noise, correcting optical system distortion, and compensating for phase errors caused by environmental vibrations; inputting the preprocessed holographic images into a trained three-dimensional reconstruction network, which adopts an encoder-decoder structure. The encoder is used to extract deep feature representations of the holographic images, and the decoder is used to reconstruct three-dimensional point cloud data from the feature representations. The three-dimensional point clouds reconstructed from each angle are fused, and the point clouds from different viewpoints are precisely registered using an iterative nearest-point algorithm to obtain complete three-dimensional holographic point cloud data of the monitoring area.

[0010] In a preferred embodiment of this scheme, the extraction of deformation feature parameters of the monitoring area based on the three-dimensional holographic point cloud data includes: using a point cloud segmentation algorithm based on normal vectors to divide the three-dimensional holographic point cloud data of the monitoring area into regions, identifying different structural regions such as the dam surface, slope surface, and surrounding rock of the cavern; calculating the surface displacement field of the point cloud data of each structural region; obtaining the displacement vector of each point in the monitoring area by performing a difference operation between the point cloud at the current time and the point cloud at the reference time; analyzing the deep point cloud data using the strain tensor calculation method to obtain the strain distribution inside the monitoring area, identifying possible stress concentration areas; extracting the three-dimensional coordinate time series of each structural key point; and calculating the displacement rate and acceleration changes of the key points using a curve fitting method.

[0011] In a preferred embodiment of this scheme, the construction of a four-dimensional spatiotemporal deformation evolution model of the monitoring area includes: using the initial three-dimensional holographic point cloud data of the monitoring area as a spatial reference, establishing a time series database containing point cloud data from all historical moments, using a spatiotemporal interpolation method to align the three-dimensional holographic point cloud data at different moments to eliminate errors caused by inconsistent sampling times, constructing a four-dimensional spatiotemporal deformation evolution model based on the processed point cloud time series, wherein the model uses spatial coordinates as the horizontal axis and time as the vertical axis to characterize the relationship between the deformation of each point in the monitoring area and time, using machine learning algorithms to perform pattern recognition on the four-dimensional spatiotemporal deformation evolution model, extracting a typical deformation evolution pattern library, and matching the current monitoring data with the pattern library.

[0012] In this preferred embodiment, the early warning judgment of potential structural safety hazards includes: setting deformation thresholds for various types of structures based on historical monitoring data and engineering experience; triggering an early warning signal when the deformation in the monitored area exceeds the corresponding threshold; combining deformation rate and acceleration for comprehensive early warning judgment; identifying abnormal trends in accelerated deformation development; providing early warning of possible instability; conducting focused and continuous monitoring of the early warning area; automatically adjusting the sampling frequency and monitoring angle to improve the monitoring accuracy of dangerous areas; and generating a monitoring and analysis report containing deformation distribution maps, time-series curves, and early warning information to provide data support for construction safety decisions.

[0013] In a preferred embodiment of this scheme, the imaging device of the multi-dimensional holographic acquisition unit adopts digital holographic technology or computational holographic technology, and the light source of the imaging device includes a laser light source or a broadband light source, wherein the wavelength of the laser light source is in the visible light band or the near-infrared band.

[0014] In this preferred embodiment, the training process of the 3D reconstruction network includes: collecting holographic image data and corresponding real 3D point cloud data of multiple pumped storage power station construction scenarios to construct a training dataset; adopting an encoder-decoder network architecture; training network parameters through supervised learning methods; using loss functions including coordinate error and geometric feature error of the reconstructed point cloud; using data augmentation techniques to expand the training data, including adding noise, simulating illumination changes, and handling random occlusion; evaluating the reconstruction accuracy of the network after training on a validation set to ensure that the network has good generalization ability for different construction scenarios.

[0015] In this preferred embodiment, the monitoring area includes key structural components such as the dam body, slope rock mass, underground caverns, water diversion pipelines, and booster stations, with different deformation thresholds and early warning strategies set for each of these key structural components.

[0016] A holographic deformation monitoring and analysis device for the construction of a pumped storage power station, the device comprising: A multi-dimensional holographic acquisition unit is used to continuously and dynamically scan the construction area of ​​a pumped storage power station to obtain a spatially encoded holographic image sequence of the monitoring area and the corresponding multispectral phase information. The multi-dimensional holographic acquisition unit includes at least three synchronous imaging devices at different spatial locations. The data processing module is used to input the spatially encoded holographic image sequence into a three-dimensional reconstruction network for object wavefront restoration to obtain three-dimensional holographic point cloud data of the monitoring area; The feature extraction module is used to extract deformation feature parameters of the monitoring area based on the three-dimensional holographic point cloud data; The modeling and analysis module is used to construct a four-dimensional spatiotemporal deformation evolution model of the monitoring area, characterizing the dynamic law of deformation change in the monitoring area as the construction process progresses; The early warning and judgment module is used to identify abnormal deformation patterns based on the four-dimensional spatiotemporal deformation evolution model and to make early warning judgments on potential structural safety hazards.

[0017] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the holographic deformation monitoring and analysis method for construction of pumped storage power stations as described above.

[0018] A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the holographic deformation monitoring and analysis method for construction of pumped storage power stations as described above.

[0019] A computer program product includes a computer program that, when executed by a processor, implements the holographic deformation monitoring and analysis method for construction of pumped storage power stations as described above.

[0020] Compared with the prior art, the technical effects and advantages of the present invention are as follows: The holographic deformation monitoring and analysis method for the construction of pumped storage power stations acquires spatially encoded holographic image sequences of the monitoring area through multi-dimensional holographic acquisition units, and restores the three-dimensional holographic point cloud data of the monitoring area using a deep learning three-dimensional reconstruction network, thus realizing full-field deformation monitoring of the construction area of ​​the pumped storage power station. By extracting multi-dimensional deformation characteristic parameters such as surface displacement field and depth strain distribution, reliable changes in minute deformation magnitudes can be captured. By constructing a four-dimensional spatiotemporal deformation evolution model, a complete characterization of the dynamic evolution law of structural deformation with the construction process can be achieved. Through abnormal deformation pattern recognition and safety early warning judgment, the problem of deformation anomalies that cannot be detected in a timely manner due to the lag in monitoring data processing can be effectively solved.

[0021] This invention significantly improves the spatial coverage, temporal continuity, and timely early warning of deformation monitoring during the construction of pumped storage power stations, providing strong technical support for construction safety management. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the overall technical solution architecture according to this application; Figure 2 This is a schematic diagram illustrating the core principle framework of multi-dimensional holographic acquisition and three-dimensional reconstruction network collaborative operation according to embodiments of this application; Figure 3 This is a flowchart illustrating the logical process of deformation feature extraction based on 3D holographic point cloud data according to an embodiment of this application. Figure 4 This is a flowchart illustrating the process of constructing a four-dimensional spatiotemporal deformation evolution model and performing safety hazard early warning judgment according to an embodiment of this application. Figure 5 This is a schematic diagram of the overall architecture according to Embodiment 1 of this application. Detailed Implementation

[0024] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.

[0025] Unless otherwise defined, the directions mentioned herein, such as up, down, left, right, front, back, inside, and outside, are based on the directions shown in the figures of this invention, and are explained here together.

[0026] This embodiment provides, for example Figures 1 to 5 This paper presents a holographic deformation monitoring and analysis method for the construction of a pumped storage power station.

[0027] Example 1

[0028] A holographic deformation monitoring and analysis method for the construction of pumped storage power stations is proposed. The technical solution of this method covers multiple core links such as data acquisition, three-dimensional reconstruction, feature extraction, model construction, and safety early warning.

[0029] In this embodiment, during the data acquisition phase, the present invention utilizes a multi-dimensional holographic acquisition unit to continuously and dynamically scan the construction area of ​​the pumped storage power station. The multi-dimensional holographic acquisition unit includes at least three synchronous imaging devices located at different spatial positions. Each synchronous imaging device performs cross-coverage imaging of the same monitoring area at a preset sampling frequency. The synchronous imaging devices are strategically deployed around the monitoring area so that any monitoring point is covered by at least two synchronous imaging devices at different angles. The distance between the synchronous imaging devices is determined based on the spatial range and accuracy requirements of the monitoring area. The synchronous imaging devices are controlled to trigger synchronous imaging at intervals not exceeding a specific time interval, ensuring that the acquired holographic image sequence has sufficient temporal resolution. During each imaging process, each synchronous imaging device simultaneously acquires the interference pattern of the reference beam and the object beam, and recovers the object beam wavefront information of the monitoring area using a phase decoding algorithm.

[0030] In this embodiment, the imaging device of the multi-dimensional holographic acquisition unit employs digital holography or computational holography. The light source of the imaging device includes a laser light source or a broadband light source. The wavelength of the laser light source is in the visible light band or near-infrared band. In practical applications, digital holography directly records the interference pattern of the object light and the reference light through an image sensor, and then uses a numerical reconstruction algorithm to recover the three-dimensional information of the object. Computational holography generates a hologram through calculation and displays it on a spatial light modulator, realizing the acquisition and reconstruction of three-dimensional information. Laser light sources have the characteristics of high coherence and good monochromaticity, enabling the acquisition of high-quality interference patterns. Broadband light sources have the advantage of strong resistance to environmental interference and are suitable for deformation monitoring in complex construction environments.

[0031] In specific deployment schemes, the spatial layout of synchronous imaging devices needs to comprehensively consider factors such as the geometry of the monitoring area, monitoring accuracy requirements, and equipment accessibility. For large structures such as dams, synchronous imaging devices are typically deployed at key locations such as the dam crest, abutment, and downstream side to form multi-angle coverage of the dam surface. For geological structures such as slopes and rock masses, the synchronous imaging devices need to be optimized based on the slope gradient, height, and rock mass distribution characteristics to ensure complete deformation information of the slope surface can be acquired. For enclosed spaces such as underground caverns, synchronous imaging devices need to be deployed at cavern entrances, cave walls, and arches to achieve comprehensive monitoring of the deformation of the surrounding rock through multi-angle cross-shooting.

[0032] In this embodiment, the baseline distance between synchronous imaging devices directly affects the accuracy and range of stereoscopic measurement. A larger baseline distance results in higher absolute measurement accuracy, but also reduces the overlapping area of ​​the field of view, increasing the measurement blind zone. In practical applications, the selection of the baseline distance needs to be comprehensively optimized based on factors such as the spatial scale of the monitoring area, the required relative measurement accuracy, and the physical constraints of the equipment. The angular interval between adjacent synchronous imaging devices is usually controlled within a specific range to ensure that any monitoring point is simultaneously covered by at least two synchronous imaging devices.

[0033] In this embodiment, regarding the time sampling strategy, the time interval for synchronous triggering of images is set based on the frequency of change of construction load and deformation response characteristics. For structures subjected to periodic construction loads, such as monitoring the compaction process of roller-compacted concrete dams, the sampling frequency needs to match the operating frequency of the compaction equipment to ensure that the instantaneous deformation caused by each compaction operation can be captured. For structures subjected to continuous construction loads, such as monitoring slope deformation during foundation pit excavation, the sampling frequency needs to be set according to the creep characteristics of the slope, ensuring sufficient monitoring timeliness while avoiding unnecessary data redundancy.

[0034] In this embodiment, the phase decoding algorithm is the core technology for digital holographic reconstruction in terms of object wavefront information recovery. This algorithm extracts the phase distribution information of the object beam by performing Fourier transform, frequency filtering, and phase unwrapping on the interference pattern. The phase unwrapping process needs to handle issues such as phase jumps and noise interference; commonly used methods include quality map-guided unwrapping algorithms, minimum norm unwrapping algorithms, and branching methods. In construction environments with significant vibrations, the phase decoding algorithm also needs to integrate a vibration compensation module to estimate and compensate for phase errors caused by environmental vibrations in real time, ensuring the accuracy of the reconstruction results.

[0035] In this embodiment, during the 3D reconstruction stage, spatially encoded holographic image sequences are input into the 3D reconstruction network for object wavefront restoration, resulting in 3D holographic point cloud data of the monitored area. The 3D reconstruction network is constructed based on a deep convolutional neural network, and its network weights are pre-trained using holographic data from a large number of hydropower engineering scenarios. The training process of the 3D reconstruction network includes: collecting holographic image data and corresponding real 3D point cloud data from multiple pumped storage power station construction scenarios to construct a training dataset; employing an encoder-decoder network architecture and training the network parameters using supervised learning methods, with the loss function including coordinate errors and geometric feature errors of the reconstructed point cloud; using data augmentation techniques to expand the training data, including adding noise, simulating illumination changes, and handling random occlusion; and evaluating the reconstruction accuracy of the network after training on a validation set to ensure that the network has good generalization ability for different construction scenarios.

[0036] In this embodiment, the holographic images acquired by each synchronous imaging device are preprocessed, including removing background noise, correcting optical system distortion, and compensating for phase errors caused by environmental vibrations. The preprocessed holographic images are then input into a trained 3D reconstruction network. This network employs an encoder-decoder structure, where the encoder extracts deep feature representations from the holographic images, and the decoder reconstructs 3D point cloud data from these feature representations. The reconstructed 3D point clouds from each angle are then fused, and an iterative nearest-point algorithm is used to precisely register the point clouds from different viewpoints, resulting in complete 3D holographic point cloud data of the monitoring area.

[0037] In this embodiment, the encoder portion of the 3D reconstruction network employs a deep convolutional neural network architecture, including multiple convolutional layers, batch normalization layers, and activation function layers. The convolutional layers are responsible for extracting local features from the holographic image, the batch normalization layers accelerate the network training process and improve its generalization ability, and the activation function layers introduce nonlinear transformations, enabling the network to learn complex feature mapping relationships. The deep feature representation of the encoder contains multi-scale information from the holographic image, including both shallow edge features and deep semantic features, providing rich input information for subsequent 3D reconstruction.

[0038] In this embodiment, the decoder employs a structure combining transposed convolutional layers and upsampling layers. It progressively upsamples the feature representations extracted by the encoder to the original image resolution and outputs the coordinates and color information of the 3D point cloud. During decoding, the network learns a nonlinear mapping relationship from 2D holographic image features to 3D spatial coordinates, achieving end-to-end 3D reconstruction. The decoder output undergoes additional point cloud post-processing modules, including outlier removal, voxel mesh downsampling, and normal vector estimation, to obtain the final 3D holographic point cloud data.

[0039] In this embodiment, the iterative nearest neighbor algorithm is used to fuse and register 3D point clouds reconstructed from different viewpoints. The algorithm first initializes the relative poses of each point cloud, then iteratively performs nearest neighbor search and pose optimization steps until it converges to the optimal registration result. During the registration process, issues such as uneven point cloud density, noise interference, and partial occlusion need to be addressed. The algorithm employs a bidirectional nearest neighbor matching strategy to reduce the impact of mismatches; it uses a random sampling consistency framework to eliminate erroneous correspondences; and it employs a hierarchical registration strategy, performing coarse registration first and then fine registration to improve registration efficiency and robustness.

[0040] In this embodiment, the 3D reconstruction network needs to be optimized for the specific characteristics of hydropower engineering scenarios in practical applications. The training dataset covers different types of hydropower engineering structures, including concrete dams, earth-rock dams, face-panel rockfill dams, underground powerhouses, tunnels, and slopes. The dataset contains monitoring data from different construction stages, including deformation data under normal construction conditions as well as monitoring cases of abnormal deformation. Data augmentation techniques are used to expand the diversity of the training data, including adding Gaussian noise to simulate environmental interference, adjusting light intensity to simulate weather changes, and adding random occlusion to simulate equipment failures. Through these measures, the 3D reconstruction network can adapt to various complex construction environments and has good generalization ability.

[0041] In this embodiment, during the feature extraction stage, deformation feature parameters of the monitoring area are extracted based on three-dimensional holographic point cloud data. These deformation feature parameters include the surface displacement field, depth strain distribution, and relative displacement of key structural points. The three-dimensional holographic point cloud data at different times are registered and compared to calculate the total deformation of the monitoring area under construction loads.

[0042] In this embodiment, a point cloud segmentation algorithm based on normal vectors is used to divide the three-dimensional holographic point cloud data of the monitoring area into regions, identifying different structural regions such as the dam surface, slope surface, and surrounding rock of the tunnel. The surface displacement field is calculated for the point cloud data of each structural region. By performing a difference operation between the point cloud at the current moment and the point cloud at the reference moment, the displacement vector of each point in the monitoring area is obtained. The strain tensor calculation method is used to analyze the deep point cloud data to obtain the strain distribution within the monitoring area and identify potential stress concentration areas. The three-dimensional coordinate time series of each key structural point is extracted, and the displacement rate and acceleration changes of the key points are calculated using curve fitting methods.

[0043] In this embodiment, the point cloud segmentation algorithm based on normal vectors first estimates the normal vectors of each point, and then divides the point cloud into regions according to the direction of the normal vectors and the local geometric features of the point cloud. This algorithm can automatically identify planar regions, curved surfaces, and regions with sharp features, and is suitable for structural surfaces with complex geometries. The segmentation result divides the 3D holographic point cloud into several homogeneous regions, with points within each region having similar geometric properties, facilitating subsequent deformation analysis.

[0044] In this embodiment, the surface displacement field is calculated using a combination of point cloud registration and differential operations. First, the point cloud at the current time is registered with the point cloud at the reference time to eliminate the influence of rigid body transformation. Then, nearest neighbor matching is performed on the two registered point clouds to calculate the three-dimensional displacement vector between corresponding points. The magnitude of the displacement vector reflects the degree of deformation at that point, and the direction of the displacement vector reflects the direction of deformation. Through interpolation, displacement values ​​at any location within the monitoring area can be obtained, constructing a complete surface displacement field distribution map.

[0045] In this embodiment, the strain tensor calculation method is based on the spatial gradient information of point cloud data and estimates the strain field of the monitoring area through numerical differentiation techniques. The method first performs surface fitting on the point cloud data to obtain a smooth displacement surface. Then, the second-order partial derivatives of the displacement surface are calculated, and combined with the basic equations of elasticity, the strain tensor distribution of the monitoring area is estimated. The strain tensor includes normal strain components and shear strain components, which can comprehensively reflect the deformation state of the monitoring area. By analyzing the eigenvalues ​​and eigenvectors of the strain tensor, stress concentration areas and potential failure modes can be identified.

[0046] In this embodiment, the selection of key structural points needs to comprehensively consider the structural stress characteristics, deformation-sensitive areas, and construction monitoring requirements. Key points typically include structural feature points, such as the dam heel, dam toe, and dam crest corners; structural control points, such as mid-span locations and variable cross-section locations; and points of application of construction loads, such as compaction locations and excavation face locations. The three-dimensional coordinate time series of key points are obtained through continuous tracking and measurement. Curve fitting methods are used to analyze the displacement trends of key points; commonly used fitting methods include polynomial fitting, exponential fitting, and Fourier fitting. The fitting results are used to calculate the displacement rate and acceleration of key points and identify acceleration or deceleration trends in deformation.

[0047] In this embodiment, during the model construction phase, the present invention constructs a four-dimensional spatiotemporal deformation evolution model of the monitoring area. This four-dimensional spatiotemporal deformation evolution model integrates three-dimensional spatial coordinates with a temporal dimension, characterizing the dynamic pattern of deformation changes in the monitoring area as the construction progresses.

[0048] In this embodiment, the initial 3D holographic point cloud data of the monitoring area is used as a spatial reference to establish a time-series database containing point cloud data from all historical moments. Spatiotemporal interpolation is used to align the 3D holographic point cloud data at different times, eliminating errors caused by inconsistent sampling times. Based on the processed point cloud time series, a four-dimensional spatiotemporal deformation evolution model is constructed. This model uses spatial coordinates as the horizontal axis and time as the vertical axis to characterize the relationship between the deformation of each point in the monitoring area and time. Machine learning algorithms are used to perform pattern recognition on the four-dimensional spatiotemporal deformation evolution model, extracting a typical deformation evolution pattern library, and matching the current monitoring data with the pattern library.

[0049] In this embodiment, the time-series database adopts a hierarchical data organization structure, indexed according to dimensions such as monitoring area, structure type, and timestamp. Each record in the database contains 3D holographic point cloud data at a specific moment, corresponding deformation characteristic parameters, and associated construction condition information. The data storage format employs an efficient point cloud compression algorithm, reducing storage space and transmission bandwidth requirements while ensuring data accuracy. The database supports incremental updates and historical backtracking queries, facilitating long-term deformation trend analysis and anomaly backtracking analysis.

[0050] In this embodiment, a spatiotemporal interpolation method is used to address the issue of incomplete alignment of point cloud sampling times at different moments. This method first establishes a mapping relationship between point cloud data and the time axis, and then estimates the virtual point cloud data at any given time using an interpolation algorithm. Commonly used spatiotemporal interpolation methods include linear interpolation, spline interpolation, and time series interpolation based on Gaussian processes. The interpolation results need to be validated to ensure that the interpolation error is within an acceptable range. For regions with high deformation rates, higher-order interpolation methods or more frequent time sampling may be required.

[0051] In this embodiment, the four-dimensional spatiotemporal deformation evolution model integrates three-dimensional spatial coordinates and the time dimension into a unified mathematical framework. This model can be represented as a four-dimensional tensor, where the three spatial dimensions correspond to the spatial coordinates of the monitoring area, and the fourth time dimension corresponds to the time series of the construction process. The core parameters of the model are learned by fitting historical monitoring data, including the spatial distribution pattern of deformation, the temporal evolution pattern, and the spatiotemporal coupling relationship. The model supports predictive functions, capable of predicting future deformation trends based on the current deformation state.

[0052] In this embodiment, a machine learning algorithm is used for pattern recognition of a four-dimensional spatiotemporal deformation evolution model. The algorithm first learns typical deformation evolution patterns from historical monitoring data. These patterns reflect deformation characteristics under different construction conditions, structural types, and geological conditions. Each pattern in the pattern library contains information such as spatial distribution characteristics, temporal evolution characteristics, and spatiotemporal coupling characteristics. In actual monitoring, the algorithm matches the current monitoring data with each pattern in the pattern library, calculates a similarity metric, and identifies the type of the current deformation state. The pattern matching results are used to assist in anomaly detection and early warning judgment.

[0053] In this embodiment, during the safety early warning stage, the present invention identifies abnormal deformation patterns based on a four-dimensional spatiotemporal deformation evolution model to make early warning judgments on potential structural safety hazards.

[0054] In this embodiment, deformation thresholds for various structural types are set based on historical monitoring data and engineering experience. An early warning signal is triggered when the deformation in the monitored area exceeds the corresponding threshold. A comprehensive early warning judgment is made by combining deformation rate and acceleration to identify abnormal trends in accelerated deformation development and provide early warnings of potential instability. The early warning area is continuously monitored with a focus on key areas, and the sampling frequency and monitoring angle are automatically adjusted to improve the monitoring accuracy of hazardous areas. A monitoring and analysis report containing deformation distribution maps, time-series curves, and early warning information is generated to provide data support for construction safety decisions.

[0055] In this embodiment, the setting of the deformation threshold needs to comprehensively consider factors such as the structural design safety factor, monitoring accuracy requirements, and construction progress. Different structural components have different deformation thresholds. For example, the deformation threshold of the dam body is usually determined based on working conditions such as the design flood level and the check flood level, while the deformation threshold of the slope rock mass needs to be determined by combining the mechanical parameters of the rock mass and the results of stability analysis. The threshold setting adopts a graded mechanism, including different levels such as attention level, warning level, and danger level, with different levels corresponding to different countermeasures and response procedures.

[0056] In this embodiment, the comprehensive early warning judgment adopts a multi-parameter fusion method, considering not only the absolute magnitude of deformation but also the deformation rate and acceleration changes. When the deformation rate exceeds a set threshold, it indicates that the structure is undergoing rapid deformation, which may pose a safety hazard. When the deformation acceleration is positive and continues to increase, it indicates that the deformation is accelerating, posing a risk of instability. The early warning algorithm comprehensively analyzes multiple parameters such as displacement, rate, and acceleration, and outputs a comprehensive early warning level and confidence level.

[0057] In this embodiment, automatically adjusting the monitoring strategy is one of the key measures for early warning response. When an early warning signal is triggered in a certain area, the system automatically increases the sampling frequency of that area, increasing the density of monitoring data collection. Simultaneously, the system automatically adjusts the monitoring angle of the synchronous imaging device, increasing the coverage angle and the number of observations of the early warning area. By dynamically adjusting the monitoring strategy, the system can concentrate monitoring resources on areas with potential safety hazards while ensuring overall monitoring efficiency, thereby improving the accuracy and timeliness of early warnings.

[0058] In this embodiment, the monitoring and analysis report uses a combination of charts and graphs to visually display the deformation status and early warning information of the monitored area. The report includes a deformation distribution map showing the magnitude and direction of deformation at each point in the monitored area; a time-series curve showing the trend of deformation at key points over time; an early warning information summary table listing the location, level, and recommended measures for all current early warning areas; and a historical comparative analysis showing the development and evolution of deformation. The report supports automatic generation and scheduled push notifications, facilitating timely monitoring of the deformation status and safety situation in the monitored area by construction management personnel.

[0059] In this embodiment, the monitoring area includes key structural components such as the dam body, slope rock mass, underground caverns, water diversion pipelines, and booster stations. Different deformation thresholds and early warning strategies are set for each key structural component.

[0060] In this embodiment, the deformation monitoring of the dam body focuses on the horizontal displacement of the dam crest, the vertical displacement of the dam heel, and the strain distribution within the dam body. The monitoring scheme is optimized for the structural characteristics of the dam body, with synchronous imaging devices mainly deployed on the upstream and downstream sides of the dam crest, the dam abutment, and within the gallery. For roller-compacted concrete dams, it is also necessary to monitor the instantaneous deformation response during the compaction process, resulting in a correspondingly higher sampling frequency.

[0061] In this embodiment, the deformation monitoring of the slope rock mass focuses on the overall displacement of the slope surface and the location of local sliding surfaces. The monitoring scheme is designed in conjunction with the geological conditions and instability mode of the slope. For slopes with weak interlayers, it is necessary to pay attention to the shear deformation along the weak interlayers; for rock slopes, it is necessary to pay attention to the opening and closing deformation of joints and fissures. The monitoring data is used for slope stability analysis and early warning judgment.

[0062] In this embodiment, deformation monitoring of the underground cavern focuses on the convergence deformation of the surrounding rock and the settlement of the arch. The monitoring scheme is optimized for the excavation method of the cavern. For caverns constructed using the New Austrian Tunneling Method (NATM), the immediate deformation response of the excavation face needs to be monitored; for tunnels constructed using the shield tunneling method, the change in ellipticity after segment installation needs to be monitored. The monitoring data is used to guide the adjustment of construction parameters and the optimization of support schemes.

[0063] In this embodiment, the deformation monitoring of the water diversion pipeline focuses on uneven settlement and axial displacement. The monitoring scheme is arranged according to the pipeline laying method. For exposed pipelines, the displacement of pipeline support points needs to be monitored; for buried pipelines, the deformation caused by changes in the soil load above the pipeline needs to be monitored. The monitoring data is used to assess the stress state and safety of the pipeline structure.

[0064] In this embodiment, deformation monitoring of auxiliary facilities such as the substation focuses on differential settlement of the building foundation and overall structural tilt. The monitoring scheme is designed according to building structure monitoring specifications, and the monitoring content includes foundation settlement observation, tilt observation, and crack observation. The monitoring data is used to assess the stability of the building structure and guide foundation treatment and superstructure construction.

[0065] Example 2

[0066] In this embodiment, during the data acquisition phase, a multi-dimensional holographic acquisition unit is used to continuously and dynamically scan the construction area of ​​the pumped storage power station, acquiring a spatially encoded holographic image sequence of the monitoring area and the corresponding multispectral phase information. The multi-dimensional holographic acquisition unit includes at least three synchronous imaging devices located at different spatial positions. Each synchronous imaging device performs cross-coverage imaging of the same monitoring area at a preset sampling frequency. The synchronous imaging devices employ computational holography technology to acquire and reconstruct three-dimensional information. The light source of the imaging devices is a broadband light source, which has broad spectral characteristics and can effectively suppress coherent noise caused by environmental vibrations, improving the stability of the reconstruction results in complex construction environments.

[0067] In this embodiment, the spatial deployment of the synchronous imaging devices adopts a dynamic optimization strategy. At the initial monitoring stage, an initial deployment plan is preset based on the geometric characteristics of the monitoring area and the construction plan. During the monitoring process, the position and orientation of each synchronous imaging device are dynamically adjusted according to the construction progress and the needs of deformation monitoring. The dynamic adjustment strategy is based on the deformation gradient distribution and structural importance assessment results of the monitoring area, tilting monitoring resources towards deformation-sensitive areas and important structural components.

[0068] In this embodiment, the sampling frequency is controlled using an adaptive mechanism. The system monitors changes in the construction load in real time. When a significant change in the construction load is detected, such as the arrival of large equipment, blasting operations, or concrete pouring, the sampling frequency is automatically increased to capture the instantaneous deformation response caused by the construction load. When the construction load remains stable, the sampling frequency is appropriately reduced to decrease the burden of data storage and processing. This adaptive sampling mechanism effectively reduces the resource consumption of the monitoring system while ensuring the timeliness of monitoring.

[0069] In the 3D reconstruction stage, this embodiment optimizes the architecture and improves the training strategy of the 3D reconstruction network. The 3D reconstruction network is built based on a deep convolutional neural network and adopts an encoder-decoder structure. The pre-training data for the network weights includes holographic data of hydropower engineering scenes and 3D measurement data from other engineering fields. By using transfer learning techniques, the pre-trained knowledge is transferred to the hydropower engineering scene, improving the network's learning efficiency and generalization ability under small sample conditions.

[0070] In this embodiment, the preprocessing of the holographic image employs an improved noise suppression algorithm. This algorithm first performs frequency domain analysis on the holographic image to identify the distribution characteristics of noise and signal in the spectrum. Then, an adaptive filter is designed based on the spectral characteristics of the noise to effectively suppress noise while preserving image detail information. The preprocessing process also includes two stages: optical system distortion correction and environmental vibration compensation. Optical system distortion correction uses a calibration board calibration method to establish a distortion model and perform compensation. Environmental vibration compensation employs a combination of multi-frame averaging and frequency domain filtering to eliminate image blurring and phase errors caused by environmental vibration.

[0071] In this embodiment, the point cloud fusion processing employs a graph optimization-based registration method. This method represents point clouds from different perspectives as nodes in a graph, and the geometric constraints between point clouds as edges. The graph optimization algorithm simultaneously optimizes the relative poses of all point clouds to obtain the globally optimal registration result. Compared to the traditional iterative nearest-point algorithm, the graph optimization-based registration method can handle more constraints and has higher registration accuracy and robustness.

[0072] In the feature extraction stage, this embodiment employs a multi-scale feature extraction method to obtain richer deformation feature information. The feature extraction process includes three levels: point-level feature extraction, region-level feature extraction, and structure-level feature extraction.

[0073] In this embodiment, point-level feature extraction calculates the local geometric properties of a single point cloud point, including normal vector, curvature, roughness, and shape index. These features reflect the geometric morphology of the point cloud at a microscale and are used to identify local deformation features.

[0074] In this embodiment, region-level feature extraction calculates the overall deformation properties of the segmented point cloud regions, including the region's average displacement, maximum displacement, deformation gradient, and strain energy. These features reflect the deformation state at the region scale and are used to identify regional deformation characteristics.

[0075] In this embodiment, structural-level feature extraction calculates the overall deformation properties of the entire monitored structure, including overall structural displacement, overall structural tilt, structural volume change, and structural modal parameters. These features reflect the deformation state at the structural scale and are used to assess the overall safety of the structure.

[0076] In this embodiment, an improved four-dimensional spatiotemporal deformation evolution model construction method is adopted in the model construction stage. This method introduces the influencing factors of construction conditions into the spatiotemporal interpolation processing. Construction conditions include construction load type, construction progress, construction method, and environmental conditions. By establishing a mapping relationship between construction conditions and deformation response, the model can more accurately predict the deformation development law under different construction conditions.

[0077] In this embodiment, the pattern recognition process of the machine learning algorithm employs a deep learning method. This method uses a long short-term memory network to model the time series of point clouds and automatically learns the temporal dependencies in the deformation evolution patterns. The pattern recognition results include not only the pattern type to which the current deformation state belongs, but also the predicted trend of pattern evolution, providing richer decision-making information for early warning judgment.

[0078] In the safety early warning stage, this embodiment employs a multi-level early warning strategy. The early warning levels include three levels: monitoring point-level early warning, regional-level early warning, and structural-level early warning. Monitoring point-level early warning targets a single monitoring point, determining whether the deformation at that point exceeds a set threshold. Regional-level early warning targets the entire monitoring area, comprehensively analyzing the deformation of each monitoring point within the area to determine the overall deformation state of the area. Structural-level early warning targets the entire monitored structure, comprehensively analyzing the deformation of each area and the overall deformation characteristics of the structure to assess its overall safety. This multi-level early warning strategy enables a comprehensive understanding of the structure's deformation state and safety situation at different scales.

[0079] Example 3

[0080] A holographic deformation monitoring and analysis device for the construction of a pumped storage power station includes a multi-dimensional holographic acquisition unit, a data processing module, a feature extraction module, a modeling and analysis module, and an early warning judgment module.

[0081] A multi-dimensional holographic acquisition unit is used to continuously and dynamically scan the construction area of ​​a pumped storage power station, acquiring spatially encoded holographic image sequences and corresponding multispectral phase information of the monitored area. The multi-dimensional holographic acquisition unit includes at least three synchronous imaging devices located at different spatial positions. These synchronous imaging devices are integrated onto an adjustable support frame, which possesses multi-degree-of-freedom attitude adjustment capabilities, enabling horizontal rotation, pitch adjustment, and vertical lifting. The adjustable support frame also integrates position and attitude sensors to monitor and record the spatial position and attitude angle of each synchronous imaging device in real time, providing initial pose estimation for subsequent point cloud registration.

[0082] The data processing module is used to input spatially encoded holographic image sequences into a 3D reconstruction network for object wavefront reconstruction, obtaining 3D holographic point cloud data of the monitored area. The data processing module is integrated into an industrial-grade computing server equipped with a multi-core CPU and GPU, possessing powerful parallel computing capabilities. The module also includes high-speed data storage units and high-speed data communication interfaces for real-time storage and transmission of large-scale holographic image data and 3D point cloud data.

[0083] The feature extraction module is used to extract deformation feature parameters of the monitoring area based on 3D holographic point cloud data. The module employs a pipelined processing architecture, connecting point cloud segmentation, displacement calculation, strain analysis, and key point extraction in a continuous process. This pipelined architecture supports parallel execution and data prefetching, significantly improving the efficiency of feature extraction. The module outputs data products including point cloud segmentation results, surface displacement fields, strain distribution maps, and key point time series data.

[0084] The modeling and analysis module is used to construct a four-dimensional spatiotemporal deformation evolution model of the monitoring area, characterizing the dynamic law of deformation changes in the monitoring area as construction progresses. The module integrates a time-series database and a model calculation engine. The time-series database adopts a distributed storage architecture, supporting efficient storage and fast querying of massive amounts of monitoring data. The model calculation engine employs high-performance computing technology, enabling the construction and updating of the four-dimensional spatiotemporal deformation evolution model within an acceptable timeframe.

[0085] The early warning and judgment module is used to identify abnormal deformation patterns based on a four-dimensional spatiotemporal deformation evolution model and to provide early warnings of potential structural safety hazards. This module integrates a rule engine and a machine learning inference engine. The rule engine performs logical reasoning based on preset deformation thresholds and early warning rules, outputting highly deterministic early warning results. The machine learning inference engine performs pattern recognition based on a trained anomaly detection model, enabling it to discover complex anomalies that are difficult for the rules to cover. The outputs of the two engines are fused to obtain the final early warning judgment conclusion.

[0086] The various modules of the device exchange data through standardized data interfaces. These interfaces employ a unified communication protocol and data format, supporting plug-and-play functionality and flexible expansion between modules. The device also integrates a system monitoring module to monitor the real-time operating status and resource usage of each module, automatically triggering fault diagnosis and recovery processes when an anomaly is detected.

[0087] This embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the holographic deformation monitoring and analysis method for pumped storage power station construction described in this invention.

[0088] The electronic device's memory uses a non-volatile storage medium, and its storage capacity is configured according to the scale of monitored data. The computer program stored in the memory consists of two parts: a system program and application programs. The system program includes low-level software such as the operating system, device drivers, and communication protocol stacks. The application programs include functional modules such as data acquisition programs, 3D reconstruction programs, feature extraction programs, model building programs, and early warning judgment programs. The processor implements the entire process of the method of this invention by loading and executing the computer program in the memory.

[0089] The electronic device is equipped with various external communication interfaces, including wired network interfaces, wireless network interfaces, and serial communication interfaces. These external communication interfaces are used for data exchange with external devices such as synchronous imaging devices, data storage servers, and monitoring centers. The electronic device also features a human-machine interface, including a display screen, keyboard, and indicator lights, for locally displaying monitoring results and receiving user operation commands.

[0090] This embodiment also provides a non-transitory computer-readable storage medium on which a computer program is stored. When executed by a processor, the computer program implements the holographic deformation monitoring and analysis method for pumped storage power station construction described in this invention.

[0091] Non-transitory computer-readable storage media utilize semiconductor storage devices such as programmable read-only memory or flash memory, offering advantages such as long data retention time, fast read / write speeds, and strong anti-interference capabilities. Computer programs stored on the storage medium employ encrypted storage and integrity verification mechanisms to prevent unauthorized copying and malicious tampering. The storage medium also supports remote upgrade functionality, enabling it to receive and install the latest program version via a network.

[0092] This embodiment also provides a computer program product, including a computer program. When executed by a processor, the computer program implements the holographic deformation monitoring and analysis method for pumped storage power station construction described in this invention.

[0093] Computer program products are provided in the form of software distribution packages, including program installers, user manuals, and technical support information. The program installer supports an automated installation process, automatically detecting the system environment and configuring the program. The user manual details the program's functions and usage methods, facilitating quick user onboarding. The technical support information provides answers to frequently asked questions and access to technical support channels, ensuring users receive assistance during use.

[0094] In specific engineering applications, the monitoring and analysis method of this embodiment needs to be coordinated with the construction organization design of the pumped storage power station. Factors such as the construction schedule, construction method selection, and construction load control in the construction organization design will affect the design of the monitoring scheme and the interpretation of the monitoring data. Therefore, when implementing the method of this invention, it is necessary to fully understand the content of the construction organization design and incorporate the construction condition information into the analysis process of the monitoring data.

[0095] The analysis results of the monitoring data need to be compared with design values, calculated values, and code limits. Design values ​​are derived from geological conditions and load assumptions during the structural design phase, calculated values ​​are derived from theoretical models during the structural analysis phase, and code limits are derived from mandatory provisions of relevant technical specifications. Through comparative analysis of multi-source data, the deformation state and safety situation of the monitored area can be more comprehensively assessed.

[0096] The monitoring results also need to be fed back into the construction organization design to optimize the construction plan and adjust construction parameters. When the monitoring data shows that the deformation exceeds expectations, it is necessary to analyze the causes and take corresponding measures, such as adjusting the construction sequence, strengthening support measures, or reducing construction loads. The monitoring data can also be used to verify the accuracy of the structural analysis model and provide a basis for subsequent design optimization.

[0097] It should be noted that, in this document, relational terms such as "one" and "two" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the phrase "comprising an element defined as..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0098] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A holographic deformation monitoring and analysis method for the construction of pumped storage power stations, characterized in that, Includes the following steps: A multi-dimensional holographic acquisition unit was used to continuously and dynamically scan the construction area of ​​the pumped storage power station to obtain the spatially encoded holographic image sequence of the monitoring area and the corresponding multispectral phase information. The spatially encoded holographic image sequence is input into a three-dimensional reconstruction network for object wavefront restoration to obtain three-dimensional holographic point cloud data of the monitoring area. Deformation feature parameters of the monitoring area are extracted based on the three-dimensional holographic point cloud data; Construct a four-dimensional spatiotemporal deformation evolution model of the monitoring area; Based on the aforementioned four-dimensional spatiotemporal deformation evolution model, abnormal deformation patterns are identified, and early warning judgments are made on potential structural safety hazards.

2. The method according to claim 1, characterized in that, The multi-dimensional holographic acquisition unit includes at least three synchronous imaging devices in different spatial locations, and each of the synchronous imaging devices performs cross-coverage imaging of the same monitoring area at a preset sampling frequency; Each of the aforementioned synchronous imaging devices is reasonably deployed around the monitoring area, so that any monitoring point is covered by at least two synchronous imaging devices at different angles. The distance between each of the synchronous imaging devices is determined according to the spatial range and accuracy requirements of the monitoring area.

3. The method according to claim 2, characterized in that, Each of the aforementioned synchronous imaging devices is controlled to synchronously trigger shooting at a time interval of no more than 0.5 seconds. During each shooting process, each of the aforementioned synchronous imaging devices simultaneously acquires the interference pattern of the reference beam and the object beam, and recovers the object beam wavefront information of the monitoring area through a phase decoding algorithm. The spatially encoded holographic image sequence is input into a three-dimensional reconstruction network for object wavefront restoration, including preprocessing the holographic images acquired by each of the synchronous imaging devices, including removing background noise, correcting optical system distortion, and compensating for phase errors caused by environmental vibrations. The preprocessed holographic image is input into the trained 3D reconstruction network; The 3D reconstruction network adopts an encoder-decoder structure, where the encoder is used to extract deep feature representations of the holographic image, and the decoder is used to reconstruct 3D point cloud data from the feature representations.

4. The method according to claim 3, characterized in that, The reconstructed 3D point clouds from various angles are fused together, and the point clouds from different viewpoints are precisely registered using an iterative nearest point algorithm to obtain complete 3D holographic point cloud data of the monitoring area. Based on the three-dimensional holographic point cloud data, deformation feature parameters of the monitoring area are extracted, including using a point cloud segmentation algorithm based on normal vectors to divide the three-dimensional holographic point cloud data of the monitoring area into regions, and identifying different structural regions such as the dam surface, slope surface, and surrounding rock of the cavern. The surface displacement field is calculated for the point cloud data of each structural region. The displacement vector of each point in the monitoring area is obtained by performing a difference operation between the point cloud at the current time and the point cloud at the reference time.

5. The method according to claim 4, characterized in that, The strain tensor calculation method is used to analyze deep point cloud data to obtain the strain distribution inside the monitoring area, identify possible stress concentration areas, extract the three-dimensional coordinate time series of each key point of the structure, and calculate the displacement rate and acceleration changes of the key points by curve fitting method. The construction of a four-dimensional spatiotemporal deformation evolution model of the monitoring area includes: using the initial three-dimensional holographic point cloud data of the monitoring area as a spatial reference, and establishing a time series database containing point cloud data from all historical moments; Spatiotemporal interpolation is used to align 3D holographic point cloud data at different times to eliminate errors caused by inconsistent sampling times. A four-dimensional spatiotemporal deformation evolution model is constructed based on the processed point cloud time series. This model uses spatial coordinates as the horizontal axis and time as the vertical axis to characterize the relationship between the deformation of each point in the monitoring area and time.

6. The method according to claim 5, characterized in that, Machine learning algorithms are used to perform pattern recognition on the four-dimensional spatiotemporal deformation evolution model, extract a typical deformation evolution pattern library, and match the current monitoring data with the pattern library to make early warning judgments on potential structural safety hazards, including: setting deformation thresholds for various types of structures based on historical monitoring data and engineering experience, and triggering an early warning signal when the deformation in the monitoring area exceeds the corresponding threshold; By combining deformation rate and acceleration for comprehensive early warning judgment, abnormal trends of accelerated deformation development can be identified, and potential instability can be predicted in advance.

7. The method according to claim 6, characterized in that, The system focuses on continuous monitoring of the warning area, automatically adjusts the sampling frequency and monitoring angle to improve the monitoring accuracy of dangerous areas, and generates a monitoring and analysis report containing deformation distribution maps, time-series curves and warning information to provide data support for construction safety decisions. The imaging device of the multi-dimensional holographic acquisition unit adopts digital holography or computational holography. The light source of the imaging device includes a laser light source or a broadband light source, and the wavelength of the laser light source is in the visible light band or the near-infrared band.

8. The method according to claim 7, characterized in that, The training process of the three-dimensional reconstruction network includes: collecting holographic image data of multiple pumped storage power station construction scenarios and corresponding real three-dimensional point cloud data to construct a training dataset; An encoder-decoder network architecture is adopted, and the network parameters are trained through supervised learning. The loss function includes the coordinate error and geometric feature error of the reconstructed point cloud. Data augmentation techniques are used to augment training data, including adding noise, simulating lighting changes, and handling random occlusion. The monitoring area includes the dam body, slope rock mass, underground caverns, water diversion pipelines, and booster station. Different deformation thresholds and early warning strategies are set for each of the key structural parts.

9. The method according to claim 8, characterized in that, The baseline distance between each of the synchronous imaging devices is comprehensively optimized based on the spatial scale of the monitoring area, the required relative measurement accuracy, and the physical constraints of the equipment. The angular interval between adjacent synchronous imaging devices is controlled within a specific range. The encoder part adopts a deep convolutional neural network architecture, including multiple convolutional layers, batch normalization layers, and activation function layers. The decoder part adopts a structure that combines transposed convolutional layers and upsampling layers.

10. The method according to claim 9, characterized in that, The point cloud segmentation algorithm based on normal vectors first estimates the normal vector of each point, and then divides the region according to the direction of the normal vector and the local geometric features of the point cloud. The segmentation result divides the 3D holographic point cloud into several homogeneous regions. The time series database adopts a hierarchical data organization structure, indexed according to monitoring area, structure type and timestamp. The database supports incremental updates and historical backtracking queries. The machine learning algorithm uses a long short-term memory network to model the time series of point clouds, automatically learns the temporal dependencies in the deformation evolution pattern, and the pattern recognition results include the pattern type to which the current deformation state belongs and the predicted trend of pattern evolution.

Images