A dynamic prediction method for earthwork quantity of a pumped storage power station

Abstract

The application discloses a kind of pumped storage power station earthwork engineering quantity dynamic prediction method, specifically relates to engineering management information processing field, for solving the problem that real terrain change is caused by misabsorption of cross-period registration and the quantity sequence is out of alignment, and plan scheduling input is unstable, is by building reference domain state record and reference domain health label limit available area index, in available range, implement limited registration and generate credible coverage map, in credible coverage, carry out elevation difference and construction partition mapping to form engineering quantity sequence, according to deviation diagnosis organization sampling review and generate reference domain update instruction and flight path fine adjustment suggestion, realize the stable transmission of data chain from measurement to plan.

Description

Technical Field

[0001] This invention relates to the field of engineering management information processing, and more specifically, to a method for dynamically predicting the earthwork volume of a pumped storage power station. Background Technology

[0002] The construction of pumped-storage power stations spans reservoir basins, dam foundations, and multiple slope levels, with the working face constantly changing as the cofferdam advances, spoil is pushed back, and excavation steps are created. Drones periodically acquire imagery and point clouds to generate 3D terrain models, forming a continuously comparable sequence of engineering quantities for rolling forecasting, plan verification, and resource allocation. Management uses this data to schedule machinery, transportation, and work shifts, aiming to guide construction pace and material movement through the measurement sequence.

[0003] However, the lack of monitoring and adaptive updates for intertemporal reference domains can easily lead to volumetric differential drift. Existing processes lack continuous assessment and dynamic revision of the reference domain status in multi-temporal registration. Intertemporal registration incorporates real-world terrain changes into the registration transformation, resulting in unidirectional shifts in the quantity sequence. These shifts permeate forecasting, scheduling, and reconciliation, causing a disconnect between planned progress and on-site development, deviations in transportation and stacking arrangements from targets, and difficulties in aligning measurement verification. Summary of the Invention

[0004] To overcome the aforementioned deficiencies in the prior art, the present invention aims to provide a dynamic prediction method for earthwork quantities in pumped storage power stations. This method involves constructing a reference domain status record and generating a reference domain health label and available area index using dual-parameter mutual verification. It also defines the scope of inter-period registration and implements point-to-plane restricted registration within the credible coverage area. After calculating the cell volume based on elevation difference, it maps the volume sequence according to construction zones. Combined with deviation diagnosis, it organizes sampling verification and generates reference domain update instructions and route fine-tuning suggestions. This addresses the problems of insufficient inter-period comparability leading to volume sequence drift and unstable planning and scheduling input in the prior art.

[0005] According to a first aspect of the embodiments of this application, a method for dynamically predicting the earthwork volume of a pumped storage power station is provided, comprising: S1: Obtain the reference domain boundary from the previous period and the UAV image point cloud from the current period, and perform coarse pose estimation based on airborne positioning and homonymous features to generate a reference domain state record; S2: Calculate the closure consistency and change compliance based on the reference domain status record and construction log, perform dual-parameter sequence synthesis, form the reference domain health label and output the available area index; S3: Perform restricted registration within the available area index to generate terrain alignment results, form a reliable coverage map based on residuals and sampling density, and implement an alternative benchmark strategy for areas of interest. S4: Calculate the difference in phased project quantities within the scope of the reliable coverage map, complete the spatial mapping according to the construction zone, generate the project quantity sequence, and output the deviation diagnosis and planning interface table; S5: Perform sampling review based on deviation diagnosis, record the source of error, generate reference domain update instructions and route fine-tuning suggestions for subsequent input.

[0006] Further, step S1 includes: Obtain the reference domain boundary from the previous period and the point cloud of the UAV imagery from the current period; Using the previous reference domain boundary as a spatial reference, the original sequence of UAV image point clouds from this period is imported. A draft of the visible range is formed by recording the positioning, coordinate unification is performed, and a reference domain grid and loop set are constructed. Based on the aforementioned visible range draft, features with the same name are matched, and the initial registration relationship and average reprojection error are solved. Based on the initial registration relationship, the visible range draft is updated to the visible range final draft. A grid cover map is generated using the reference domain grid as the carrier. The reference domain grid, the loop set, the initial registration relationship, the average reprojection residual, the visible range final draft, and the grid cover map are merged to form a reference domain state record.

[0007] Furthermore, based on the aforementioned visible range draft, corresponding features are matched, and the initial registration relationship and average reprojection error are solved, including: Within the visible draft coverage area, scale-invariant features are extracted and geometrically consistent matching between images is completed. The matching results are projected onto the image plane of the previous reference domain boundary to obtain a set of identical feature pairs. The three-dimensional coordinates of the image features are recovered based on the point cloud coordinates to form a three-dimensional identical set. Out-point suppression is performed using the random consistency method. Three pairs of three-dimensional points are taken in the sampling unit to establish candidate rigid solutions. In-point consistency is judged based on the adaptive threshold of the Euclidean distance residual. The threshold is dynamically updated with the observation noise model and baseline geometry. Out-point suppression stops when the number of in-points is stable and no longer increases. The consistency set is retained as the source of the response. A rigid transformation least squares model is established using the consensus set as input. The rotation matrix follows orthogonal constraints, and the translation vector lies in three-dimensional Euclidean space. The solution is obtained using the orthogonal-constrained least squares closed-form solution method. After obtaining the rotation matrix and translation vector, the average reprojection residual is calculated as a robustness measure. The rotation matrix is ​​denoted as The translation vector is denoted as The current three-dimensional feature points are denoted as The corresponding three-dimensional feature points in the previous period are denoted as The number of three-dimensional feature points is denoted as If the average reprojection residual exceeds the robust interval set based on the noise model, it reverts to the previous stable state of the uniform set and solves again until it enters the robust interval, thereby obtaining the initial registration relation and the average reprojection residual.

[0008] Further, step S2 includes: Obtain the structured event stream of the construction log, and combine it with the reference domain state record to generate a log change mask, i.e., an observation change mask; Based on the state records of the reference domain, the closure consistency of each grid is calculated to form a closure consistency graph; Within each grid, the spatial intersection-union ratio of the log change mask and the observation change mask is calculated to obtain the change consistency and form a consistency map; Within each grid, ordinal mapping is performed on the closure consistency and change compliance. The ordinal synthesis is completed using the quantile threshold method. Combined with the four-adjacent connectivity rule of the grid coverage map, three types of labels are marked: available, avoidable, and pending review. The available labels are summarized to form an available area index, and the avoidable and pending review labels are summarized to form an area of ​​interest.

[0009] Furthermore, in step S2: Closedness consistency is expressed as path average. The length of the loop is The arc length position variable is The displacement residual is ; Grid The degree of consistency of the change is , where grid The inner log mask area is The observation mask area is .

[0010] Further, step S3 includes: Within the range of available area index, select grids with better closure consistency as anchor sets, project the UAV image point cloud of this period based on the initial registration relationship, and construct the correspondence of each projection point; A point-to-plane iterative nearest-point algorithm is used for restricted registration. A voxel pyramid is constructed, and the algorithm is optimized layer by layer from coarse to fine. The solution of the upper layer is used as the initial value of the lower layer to obtain the fine registration relationship. The coordinate framework of the current image point cloud to the boundary of the previous reference domain is unified to obtain the terrain alignment result. The number of iterations and the convergence status are recorded as the registration log. Based on the terrain alignment results, the quantile magnitude and sampling sufficiency of the residual distribution are calculated on the reference domain grid. A reliable coverage map is generated according to the double threshold rule and excluding areas of interest. An alternative benchmark strategy is triggered for areas of interest, and stable anchored segments are used for local solutions first. If insufficient, historical slices from periods unaffected by the operation are introduced as temporary benchmark segments and incorporated into the reliable coverage map according to the connectivity rule.

[0011] Further, step S4 includes: Establish surface elevation functions for the previous and current periods within the cells marked on the reliable coverage map; Based on the surface elevation functions of the previous and current periods, the volume difference is calculated according to the cell area. Based on the volume difference, the volume of each construction zone is aggregated and combined with the current time tag to form a sequence of engineering quantities. Among them, the zone with no available cells is marked as missing measurement status. Based on the engineering quantity sequence and the current target volume of each partition, deviation diagnosis is performed. Within each partition, the median sufficiency of available cell sampling is statistically analyzed. The confidence level is composed by combining the median statistics of the residual quantile amplitude within the partition, and a planning interface table is compiled.

[0012] Further, step S5 includes: Taking deviation diagnosis as the entry point, a set of verification windows is generated within the partition by combining the credible coverage map and the region of interest. A set of verification window cells is extracted for each verification window, and an independent anchoring set and short-term aerial survey segments are constructed as two types of independent evidence. Within the verification window, the point-to-plane iterative nearest point algorithm is used to obtain the local rigid verification solution. The normal median offset and the verification volume obtained by aggregating the vertex set of the verification window are calculated. The error source classification is completed based on the joint relationship between the two. The classification results and the verification window identifier are written into the registration log. Based on the source of error, reference domain update instructions and route fine-tuning suggestions are generated. The reference domain update instructions include grid addition and deletion, connected component redrawing, and collection rhythm adjustment. The results are structured into records at the granularity of the review window, archived as a prerequisite attachment to the next period's reference domain status record, and synchronized to the remarks column of the planning interface table.

[0013] According to a second aspect of the embodiments of this application, an electronic device is provided, comprising: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors perform the method as described in the first aspect.

[0014] According to a third aspect of the embodiments of this application, a computer-readable storage medium is provided that stores computer instructions thereon, which, when executed by a processor, implement the steps of the method as described in the first aspect.

[0015] The technical solutions provided by the embodiments of this application may include the following beneficial effects: As can be seen from the above embodiments, this application first uses dual-parameter mutual verification to provide the health of the reference domain, limiting registration to only within the trusted area to avoid incorporating real terrain changes into the transformation; then, elevation difference is performed on the trusted cells and aggregated according to construction zones to form a stable and comparable sequence of engineering quantities; deviation diagnosis triggers sampling verification and feedforward updates, and the reference domain index and flight path parameters are adaptively corrected according to changes in the field, so that errors do not accumulate in the time series; thus, the prediction is based on real changes, the planning schedule and resource allocation are quickly aligned, measurement disputes are reduced, the burden of re-measurement is reduced, and complex slopes and near-water areas still maintain stable geometric baselines, and the management end obtains structured inputs that can be directly called.

[0016] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] Figure 1 This is a flowchart illustrating a method for dynamically predicting the earthwork volume of a pumped storage power station according to an exemplary embodiment.

[0019] Figure 2 This is a schematic diagram of an electronic device according to an exemplary embodiment. Detailed Implementation

[0020] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0021] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0022] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0023] Figure 1 This is a flowchart illustrating a method for dynamically predicting the earthwork volume of a pumped storage power station according to an exemplary embodiment, such as... Figure 1 As shown, this method, when applied to a terminal, may include the following steps: S1: Obtain the reference domain boundary from the previous period and the UAV image point cloud from the current period, and perform coarse pose estimation based on airborne positioning and homonymous features to generate a reference domain state record; S2: Calculate the closure consistency and change compliance based on the reference domain status record and construction log, perform dual-parameter sequence synthesis, form the reference domain health label and output the available area index; S3: Perform restricted registration within the available area index to generate terrain alignment results, form a reliable coverage map based on residuals and sampling density, and implement an alternative benchmark strategy for areas of interest. S4: Calculate the difference in phased project quantities within the scope of the reliable coverage map, complete the spatial mapping according to the construction zone, generate the project quantity sequence, and output the deviation diagnosis and planning interface table; S5: Perform sampling review based on deviation diagnosis, record the source of error, generate reference domain update instructions and route fine-tuning suggestions for subsequent input.

[0024] In practice, the pumped storage area uses the previous period's reference domain boundary as the sole spatial anchor point for cross-period alignment, and the current period's UAV imagery point cloud as the sole observation source. Step S1 organizes the observations into a final visible range within the same coordinate frame, simultaneously establishing the initial registration relationship and the average reprojection residual. Geometric constraints are solidified using the reference domain grid and loop set, and connectivity is clarified using a grid cover map. A continuous link is formed from the data form to the geometric baseline, with the aim of enabling step S2 to directly determine the closure consistency and change conformity based on this, without relying on ad hoc inferences.

[0025] Specifically, step S1 may include the following sub-steps: (1) Data preparation and coordinate unification: Changes in construction phases lead to changes in observation coverage and occlusion patterns over time. Therefore, it is necessary to first normalize the original observations to a single coordinate frame. First, load the boundary of the previous reference domain as a spatial reference, and then import the original sequence of UAV image point clouds for the current phase. Based on the airborne positioning trajectory, remove out-of-bounds frames and obvious noise echoes to form a draft of the visible range. Use the airborne positioning extrinsic parameters as initial values ​​to unify the image exterior orientation elements and point cloud poses to the coordinate frame of the previous reference domain boundary, ensuring that all subsequent spatial objects are on the same reference. Lay out the reference domain grid with the density of the boundary point set and the draft coverage density as independent variables, so that the surface element scale adaptively converges with the density. Extract closed polylines along the main skeleton of the previous reference domain boundary, and then generate several closed lines along the grid skeleton, collectively referred to as the loop set. Each loop is checked for closure and continuity. Closure is used to verify the geometric consistency of the first and last points, and continuity is used to check the topological connectivity between the point series.

[0026] (2) Establishment of homonymous features and solution of initial registration relationship: The geometric baseline for inter-period comparability comes from reliable matching pairs. Scale-invariant features are extracted within the visible draft coverage area and geometric consistency matching between images is completed. Then, the matching results are projected onto the image plane of the previous period reference domain boundary to obtain a set of homonymous feature pairs. The three-dimensional coordinates of the image features are recovered based on the point cloud coordinates to form a three-dimensional homonymous set. Out-point suppression is performed using the random consistency method. Three pairs of three-dimensional points are taken in the sampling unit to establish candidate rigid solutions. In-point consistency is judged based on the adaptive threshold of the Euclidean distance residual. The threshold is dynamically updated with the observation noise model and baseline geometric conditions. Out-point suppression is stopped when the number of in-points is stable and no longer increases. The consistency set is retained as the source of matching pairs. A rigid transformation least squares model is established with the consistency set as input. The rotation matrix follows orthogonal constraints, and the translation vector is in three-dimensional Euclidean space. The solution is obtained using the orthogonal constraint least squares closed solution method. After obtaining the rotation matrix and translation vector, the average reprojection residual is calculated as a robustness measure. The rotation matrix is ​​denoted as The translation vector is denoted as The current three-dimensional feature points are denoted as The corresponding three-dimensional feature points in the previous period are denoted as The number of three-dimensional feature points is denoted as If the average reprojection residual exceeds the robust interval set based on the noise model, it reverts to the previous stable state of the uniform set and resolves until it enters the robust interval. This yields the initial registration relation and the average reprojection residual, which are considered the geometric baseline and quality signal read in steps S2 and S3, respectively.

[0027] (3) Visible Range Clipping and Reference Domain Status Record Solidification: Health determination requires clear visible boundaries and grid connectivity. The initial registration relationship is applied to the current observation, unifying the image point cloud to the previous reference domain boundary coordinate framework. The final visible range is updated from the draft based on the transformed coverage. A grid coverage map is generated using the reference domain grid as the carrier. Connectivity analysis is performed using the four-adjacency rule, isolated micro-patterns are removed, and connected components are retained to ensure the verifiability of subsequent neighborhood continuity determination. The reference domain grid and loop set, the initial registration relationship and the average reprojection residual, the final visible range and the grid coverage map are merged to form the reference domain status record.

[0028] Based on the above, step S1 can transform discrete observations into reference domain state records. The coordinate frame, geometric constraints, and quality signals within the records are consistent with each other. The final visible range defines the spatial boundary. The initial registration relationship and the average reprojection residual provide the geometric baseline. The reference domain grid and loop set serve subsequent path calculations. The grid cover map is used for connectivity verification.

[0029] Step S2 takes the final visible range, reference domain grid, loop set, initial registration relationship, and grid cover map from the reference domain state record, and reads the structured events from the construction log. The goal is to complete the health determination using two chains of evidence: geometric consistency and semantic fit. First, the closure consistency reflects the geometric closure of the loop path. Then, the change consistency verifies the spatial correspondence between log changes and observed changes. Subsequently, discrete cells are promoted to stable usable area indices through ordinal synthesis and connectivity verification, avoiding the misinclusion of actual terrain changes in the registration solution.

[0030] Specifically, this step may include the following sub-steps: (1) Data preparation and mask construction: To ensure that the judgment basis is authentic and consistent with the source, the final draft of the visible range and the reference domain grid in the reference domain status record are read first, and then the structured event stream of the construction log is read. Log change masks are generated according to the event location and the scope of influence. The observation side prepares the observation change mask within the final draft of the visible range, the image side uses the structural similarity algorithm to generate the difference map, and the point cloud side uses the voxel raster counting difference method to reveal the density change. Then, the two sources are aligned to the reference domain grid using the coordinate frame unification strategy.

[0031] (2) Closure Consistency Calculation and Grid Layout: To evaluate geometric self-consistency, the closure residual sequence of the loop set is calculated within the current coordinate frame, and the closure consistency signal is defined using Euclidean path error. Let the length of a certain loop be denoted as... The arc length position variable is denoted as The displacement residual is denoted as Its grid The internal closure consistency is expressed in the form of path average. The dimension of closure uniformity is length, and the integration interval is consistent with the path length. All traversed grids are considered. The loop segments are merged according to spatial coverage, and the average path is defined as the closure consistency of the grid to form a closure consistency graph.

[0032] (3) Change Consistency Calculation and Grid Layout: To verify semantic consistency, the spatial intersection-union ratio (IUU) of the log change mask and the observation change mask is calculated within each grid. Let the grid... The inner log mask region is denoted as The observation mask area is denoted as Changes in conformity are recorded as A higher degree of consistency indicates a stronger match between the construction records and observed changes. The results are then mapped onto a reference domain grid to form a consistency map.

[0033] (4) Health determination is completed by dual-parameter sequence composition and connectivity constraints: To obtain robust health labels, sequence mapping is performed on closure consistency and change conformity within each grid, and the quantile threshold method is used to set the inferior value zone and the pass zone. In the first stage, inferior value masking is performed. Grids whose closure consistency or change conformity falls into the inferior value zone are directly marked as avoidable. In the second stage, joint pass is performed. Grids with both signals in the pass zone are included in the candidate set. Then, four-adjacent connectivity verification is performed according to the grid coverage map. Only the candidate set that forms a connected component is retained as usable. The remaining grids that are not covered are marked as pending verification. In this way, the reference domain health label is obtained. The label set includes three categories: usable, avoidable, and pending verification. Then, the set of connected components with usable labels is summarized to form the usable area index, and the avoidable and pending verification are summarized as the area of ​​interest.

[0034] Step S2 generates reference domain health labels and available area indexes, and marks areas of interest. The available area index directly defines the registration scope of step S3, and the areas of interest provide a location basis for sampling verification and alternative benchmarks in step S5. In this stage, the initial registration relationship remains unchanged, and the average reprojection residual is only used as a quality reference indicator. Thus, the multi-temporal determination converges from both geometric and semantic dimensions into an operable spatial set, laying a stable foundation for the generation of reliable coverage maps and engineering quantity sequences.

[0035] In step S1, the previous spatial anchor point and the current observations form a reference domain status record, and the geometric baseline and visible range have been determined under a unified coordinate framework. Step S2 generates reference domain health labels and available area indexes based on this, while also marking areas of interest. Step S3 aims to strictly constrain inter-period registration within the available area index, construct a stable correspondence based on the initial registration relationship, output terrain alignment results, and form a reliable coverage map within a grid that satisfies both connectivity and coverage requirements. For areas of interest, an alternative benchmark strategy is needed to fill alignment gaps, ensuring that inter-period geometric constraints and observation coverage remain spatially consistent, thus providing a reliable entry point for subsequent engineering differentials and zonal mapping.

[0036] The specific implementation of step S3 may include the following sub-steps: (1) Anchor Set Selection and Correspondence Construction: To ensure that the solution falls within the credibility domain, the first step is to select a set of meshes with good closure consistency and usable labels in the reference domain mesh as the anchor set. The source of the set is the reference domain health label and usable area index from step S2. Using the initial registration relationship as the initial value of the extrinsic parameters, the current point cloud is projected onto the previous surface model, restricting the projection position to fall only within the coverage area of ​​the anchor set. A correspondence is established for each projection point. The current 3D point is denoted as point x, the corresponding point in the neighborhood of the previous surface is denoted as point y, and the unit normal vector of the previous surface at point y is denoted as the n-direction. The direction of the normal vector is taken from the normal field of the previous surface. The correspondence is constructed using the nearest point plus normal consistency criterion. The distance is defined by Euclidean distance, and the normal consistency is filtered by the included angle threshold. The threshold is given by the coarse-scale reference of the average reprojection residual in step S1 and then locally adaptively converges. Through the above screening, the corresponding set constrained within the usable area index is obtained, which serves as the sole data entry point for subsequent registration.

[0037] (2) Constrained Registration Solution and Multi-Level Refinement: To reduce the impact of attitude drift on the difference, a point-to-plane iterative nearest-point algorithm is adopted, and the rigid transformation is solved within the available region index. Let the rotation matrix be denoted as... The translation vector is denoted as The objective function is defined as follows: The solution employs an exponentially mapped parameterized Gaussian-Newton method, updating rotation increments with Lie groups and translation increments with 3D vectors. Each iteration reconstructs the corresponding region within the available region index; correspondences outside the boundary are directly discarded and not included in the objective function. To improve robustness, a three-layer voxel pyramid is constructed, optimizing layer by layer from coarse to fine, using the upper-layer solution as the initial value for the lower layer, until the residual decreases below a stopping criterion based on quantiles. After solving, a fine registration relationship is formed to unify the coordinate framework of the current image point cloud to the previous reference domain boundary, obtaining the terrain alignment result. Simultaneously, the iteration count and convergence status are recorded in the registration log for traceability in step S4.

[0038] (3) Credible Coverage Map Generation and Reference Area Substitution Processing: To characterize the spatial range available for differential analysis, two types of metrics are calculated on the reference domain grid based on the terrain alignment results. The first is the quantile magnitude of the residual distribution, defined as the quantile statistics of the corresponding residuals within a cell, metric named residual quantile magnitude, used to describe the quality of geometric fit. The second is sampling sufficiency, defined as the ratio of the number of valid points in the current period to the number of reference points in the previous period within a cell, used to describe the completeness of the observation coverage. The criteria for determining a credible coverage map are that the residual quantile magnitude is lower than the threshold and the sampling sufficiency is higher than the threshold, and the cell does not belong to the reference area. The threshold adopts the quantile threshold method. For the reference area, the reference substitution strategy is triggered. The nearest connected component is selected from the anchor set as a stable anchor segment first, and local point-to-plane solution is performed within the segment. The local solution is updated with constraints of small angle and small translation. If there are not enough stable anchor segments, the historical period slices that have not been affected by the operation are called as temporary reference segments. After local alignment is completed, they are incorporated into the credible coverage map according to the connectivity rules. All alternative operations are marked with the source fragment and the merged range in the registration log.

[0039] Step S3 completes the closed-loop process from corresponding construction to restricted registration within the available area index, outputting the terrain alignment result and generating a reliable coverage map. Simultaneously, an alternative benchmark strategy is implemented for areas of interest to compensate for gaps caused by local instability. The registration solution does not exceed the limits, and the reliable coverage map defines its available range using both quantile and coverage criteria. The fine registration relationship and registration log serve as the direct entry point for the difference calculation and partition mapping in step S4.

[0040] The previous stage yielded terrain alignment results and a reliable coverage map within a unified coordinate framework. The available area index has defined the effective space for inter-period registration, the reference domain grid has clearly defined sampling units, and the construction zoning set remains valid on the management side in the long term. Step S4 requires transforming the aforementioned geometric results into quantitative inputs that can be used for planning and scheduling. The goal is to generate phased project quantity differences and complete the construction zoning mapping, subsequently generating project quantity sequences, deviation diagnoses, and a planning interface table, enabling dynamic predictions to enter the management process with a stable baseline. Pumped storage areas have complex terrain such as steep slopes and steps. If sampling and integration are not performed within the scope defined by the reliable coverage map, the volume difference will treat the registration residuals as real changes, requiring careful handling.

[0041] Step S4 may specifically include the following sub-steps: (1) Surface Sampling and Height Function Construction: Based on the terrain alignment results, resampling is performed on the reference domain grid. The goal is to establish scalar functions of the two-stage surfaces within the cells marked on the reliable cover map. The elevation function of the upper-stage surface is defined along the unified coordinate frame and denoted as... The surface elevation function for the next period is denoted as Both are limited to the set of lattice cells in the trusted overlay graph that are marked as usable, and the name of the set is denoted as . The sampling method employs grid-by-grid bilinear interpolation. Interpolation nodes are taken from the triangular mesh and point cloud sample set intersecting with the grid cells in the terrain alignment result. The interpolation kernel is only effective within its own grid cell and does not cross grid cell boundaries. Each grid cell is denoted as... Extract the centroid coordinates of the lattice cells as and At the same time, the area of ​​the cell is recorded as . .

[0042] (2) Differential volume calculation and reliability constraints: The volume is solved using the digital elevation difference method, only when... Execute internally to avoid bringing the region of interest into the volumetric model. For any grid cell The elevation difference is defined as . The difference in size is recorded as For all belong Calculate each cell one by one At the same time, the reliable coverage determination of the source of each cell is recorded so as to retain traceability during subsequent partition aggregation.

[0043] (3) Construction zone mapping and quantity sequence generation: The management side has provided the set of construction zones and spatial boundaries. A mapping function denoted as π is established from the grid cell to the zone, with the function value being the zone identifier of the grid cell. For each zone, denoted as p, volume aggregation is performed. In the formula The difference in the phased work volume for partition p. All partitions... The current period's time label is denoted as τ and recorded together to form a new frame of the engineering quantity sequence in the time dimension. If a certain partition has no available cells in the current period, the partition is recorded as a missing measurement state. The missing measurement state is not filled by interpolation, and the empty space is reserved for sampling verification in step S5 and supplementary sampling in the next period.

[0044] (4) Deviation Diagnosis and Planning Interface Table Implementation: The management plan sets the current target volume for each partition, denoted as... The aggregated volume of each partition is compared with the target volume, and the deviation is denoted as . ,when When the value is greater than zero, it is considered that the amount filled is too much. When the amount is less than zero, it is considered that the excavation volume is too high. A value equal to zero is considered consistent. To provide reliable hints for the plan, within each partition, [the following is considered]: The median of the sufficiency of the lattice sampling is denoted as . .by The median statistics within each partition, along with the residual quantile amplitude from step S3, form the basis for determining the confidence level, following the quantile threshold method. Finally, a planning interface table is compiled, containing fields for partition identifier, quantity difference, deviation direction, confidence level, and time stamp.

[0045] Step S4 completes the lattice representation of the two phases of the surface within the space defined by the reliable coverage map. The volume difference of each cell is obtained using the digital elevation difference method, and then spatial aggregation is performed using the partition mapping function to form a new frame of the engineering quantity sequence. Subsequently, the deviation diagnosis is output with the partition target volume as a reference, and the reliability level is given according to the quantile rules of sampling sufficiency and residual quantile amplitude. Finally, the measurement results are stably transmitted to the management side through the planning interface table.

[0046] Step S4 has completed terrain alignment and reliable coverage maps within a unified coordinate framework. The phased project quantity differences have been mapped to a project quantity sequence through grid differencing and construction zone mapping. Deviation diagnosis reveals over-excavation and under-filling trends at the zone level. Step S5 needs to translate these measurement conclusions into actionable verification and feedforward actions. The aim is to verify the spatial reliability of the volume difference, locate the source of error, and generate reference domain update instructions and route fine-tuning suggestions, serving as prerequisites for the next steps S1 and S2. The pumped storage project area has areas of concern and residual hotspots, requiring precise spatial window selection and verification using independent anchored evidence; otherwise, errors will accumulate over time and propagate to the planning stage.

[0047] Step S5 may specifically include the following sub-steps: (1) Selection of verification windows and data preparation: The verification process begins with deviation diagnosis. At the zoning level, zoning zones with large absolute deviations are selected. Within these zones, reliable coverage maps and areas of interest are cross-referenced. Verification window sets are generated based on residual hotspots and missing holes. The verification window sets are defined as connected components on the reference domain grid, and the set name is denoted as W. Each verification window contains a set of cell indexes, time labels, and zoning identifiers. The coordinate reference is kept consistent with the terrain alignment results. Data preparation is carried out in two branches. One branch extracts stable segments that have not been overused by local solutions from the anchor set as independent geometric evidence. The other branch connects short-term aerial survey segments to the verification windows as independent observation evidence. Both are registered in the registration log to ensure traceability.

[0048] (2) Construction of Independent Anchored Sets and Local Alignment Solution: The local solution adopts the point-to-plane iterative nearest point algorithm, taking the correspondence of the verification window as input, and constructing a local rigid transformation within the window. The rotation matrix is ​​denoted as... The translation vector is denoted as The objective function is the sum of squared normal residuals, expressed as: The verification index set is denoted as The current three-dimensional points are denoted as The corresponding point in the previous period is recorded as The surface unit normal vector of the previous period is denoted as The solution process employs the exponential mapping Gauss-Newton method. Rotations are updated using Lie groups, and translations are updated using three-dimensional vectors. Iterations are performed only within the lattice set of the verification window to reconstruct the corresponding data and converge. The convergence status is written to the registration log.

[0049] (3) Calculation of verification indicators and interpretation of error sources: Verification indicators focus on two main lines: geometric consistency and volume consistency. Geometric consistency measures the degree of fit between the verification solution and the previous period's surface using the median offset of the normal direction, defined as follows: Consistency in volume is measured by the sum of the cell differences in volume within the verification solution recalculation window. This is first denoted as the cell set within the verification window. The internal solution updates the elevation difference between the two periods using verification, and then the volume aggregation is completed using the cell area. The expression is as follows: The verified elevation difference is recorded as "re-verification". The area of ​​a grid cell is denoted as Error sources are determined based on the combined relationship between geometric and volumetric indicators. If the median normal offset consistently leans to one side and the difference between the verified volume and the original volume meets the standard, it is classified as a registration drift. If the sampling sufficiency is generally low and the corresponding pairs are sparse within this window, it is classified as a coverage gap. If the verified volume and the original volume are similar and the geometric offset is within a robust range, it is classified as a true geometric change. The category label is written to the registration log and bound to the verification window identifier.

[0050] (4) Reference Domain Update Command and Flight Line Fine-Tuning Suggestion Generation: The reference domain update command generates spatial-level handling solutions based on window categories. Registration drift triggers the addition and deletion of grids and redrawing of connected components in the available area index. Coverage gap triggers grid densification and sampling rhythm adjustment. Real geometric change only records the status and keeps the existing index unchanged. Flight line fine-tuning suggestions extract the dominant direction from the geometric structure and normal statistics of the review window and provide suggestions for shooting angle adjustment, flight altitude fine-tuning, and operation time period.

[0051] To ensure the feasibility of the proposed flight path adjustments, the dominant direction, shooting angle, flight altitude, and operating time period are determined according to the following rules. First, within the grid cell set of the verification window, the surface unit normal vector from the previous period is horizontally projected to create an azimuth histogram, and the mode azimuth corresponding to the peak value is taken as the dominant direction. Simultaneously, the major axis direction of the verification window boundary is fitted with the grid cell centroid coordinates, and the consistency between the two is used to verify the dominant direction. If they are inconsistent, the mode azimuth of the normal is preferred to maintain the geometric stability of the observation facing the slope. Then, the flight path is determined based on the dominant direction, making the flight path nearly orthogonal to the dominant direction to reduce self-occlusion. The shooting angle is tilted inward along the dominant direction to maintain a stable incident relationship between the line of sight and the unit normal vector. Side-view compensation flight strips are set at the boundary between the verification window and the adjacent available area to cover the elevation transition. The flight altitude range is then set again based on the terrain undulation range of the verification window and the cell scale of the reference domain grid. The upper limit is given by the coverage width and overlap constraints, while the lower limit is given by the ground sampling resolution and minimum incident angle constraints. The flight altitude is taken as the median or slightly above the range. The roll angle is appropriately increased in near-water areas or high-reflectivity surfaces to reduce the mirror effect. Finally, the operation period is selected based on sunshine and wind field data. The sun azimuth is offset from the dominant direction to avoid the light path reflection direction being in the same direction as the camera's line of sight. The sun altitude should ensure that the slope shadow does not cross the main working zone. The wind speed and gust frequency meet the attitude margin required for a stable track. All parameters are structured into recommendations based on the verification window, recording the azimuth of the dominant direction, track direction, roll angle orientation, source of flight altitude range, and basis for operation period selection. The verification window identifier and zone identifier are associated in the registration log.

[0052] All actions are recorded in a structured record based on windows. Fields include window identifier, action category, action scope, associated partition, and time tag. The records are entered into the registration log and synchronized to the remarks column of the planning interface table for easy reception by the management side.

[0053] (5) Feedforward archiving and inter-period connection: Feedforward archiving organizes the review results into prerequisites for the next period, including a list of index additions and deletions for reference domain update instructions, a set of parameters for route fine-tuning suggestions, and local solutions. , Summary of review indicators. The above-mentioned archived documents are supplementary attachments to the reference domain status records. They will be directly imported during the loading process in the next step S1 and will be given priority for use as initial candidates for the available area index during the health determination process in the next step S2.

[0054] Step S5 begins with deviation diagnosis, precisely selecting a window in space. Local alignment is solved using independent anchor sets and short-term aerial survey segments, constructing two indicators: normal median offset and verification volume. The source of error is determined based on the joint relationship, and the processing results are solidified into reference domain update instructions and route fine-tuning suggestions, ultimately archived as prerequisites for the next phase. At this point, the inter-phase prediction link forms a verifiable closed loop between measurement and management, enabling the next phase to converge and operate on a stable reference domain health and available area index.

[0055] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0056] It should be noted that the system of the present invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting a variety of hardware environments and usage requirements.

[0057] Accordingly, this application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the dynamic prediction method for earthwork volume of a pumped storage power station as described above.

[0058] Accordingly, this application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; and when the one or more programs are executed by the one or more processors, causing the one or more processors to implement the dynamic prediction method for earthwork volume of pumped storage power stations as described above. Figure 2 The diagram shown is a hardware structure diagram of any data processing-capable device where a deep learning dataset access system is located, according to an embodiment of the present invention. Except for... Figure 2 In addition to the processor, memory, and network interface shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0059] Accordingly, this application also provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the dynamic prediction method for earthwork volume of a pumped storage power station as described above. The computer-readable storage medium can be an internal storage unit of any data processing device as described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0060] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

Claims

1. A method for dynamically predicting the earthwork volume of a pumped storage power station, characterized in that, include: S1: Obtain the reference domain boundary from the previous period and the UAV image point cloud from the current period, and perform coarse pose estimation based on airborne positioning and homonymous features to generate a reference domain state record; S2: Calculate the closure consistency and change compliance based on the reference domain status record and construction log, perform dual-parameter sequence synthesis, form the reference domain health label and output the available area index; S3: Perform restricted registration within the available area index to generate terrain alignment results, form a reliable coverage map based on residuals and sampling density, and implement an alternative benchmark strategy for areas of interest. S4: Calculate the difference in phased project quantities within the scope of the reliable coverage map, complete the spatial mapping according to the construction zone, generate the project quantity sequence, and output the deviation diagnosis and planning interface table; S5: Perform sampling review based on deviation diagnosis, record the source of error, generate reference domain update instructions and route fine-tuning suggestions for subsequent input.

2. The method for dynamically predicting the earthwork volume of a pumped storage power station according to claim 1, characterized in that, Step S1 includes: Obtain the reference domain boundary from the previous period and the point cloud of the UAV imagery from the current period; Using the previous reference domain boundary as a spatial reference, the original sequence of UAV image point clouds from this period is imported. A draft of the visible range is formed by recording the positioning, coordinate unification is performed, and a reference domain grid and loop set are constructed. Based on the aforementioned visible range draft, features with the same name are matched, and the initial registration relationship and average reprojection error are solved. Based on the initial registration relationship, the visible range draft is updated to the visible range final draft. A grid cover map is generated using the reference domain grid as the carrier. The reference domain grid, the loop set, the initial registration relationship, the average reprojection residual, the visible range final draft, and the grid cover map are merged to form a reference domain state record.

3. The method for dynamically predicting the earthwork volume of a pumped storage power station according to claim 2, characterized in that, Based on the aforementioned visible range draft, corresponding features are matched, and the initial registration relationship and average reprojection error are solved, including: Within the visible draft coverage area, scale-invariant features are extracted and geometrically consistent matching between images is completed. The matching results are projected onto the image plane of the previous reference domain boundary to obtain a set of identical feature pairs. The three-dimensional coordinates of the image features are recovered based on the point cloud coordinates to form a three-dimensional identical set. Out-point suppression is performed using the random consistency method. Three pairs of three-dimensional points are taken in the sampling unit to establish candidate rigid solutions. In-point consistency is judged based on the adaptive threshold of the Euclidean distance residual. The threshold is dynamically updated with the observation noise model and baseline geometry. Out-point suppression stops when the number of in-points is stable and no longer increases. The consistency set is retained as the source of the response. A rigid transformation least squares model is established using the consensus set as input. The rotation matrix follows orthogonal constraints, and the translation vector lies in three-dimensional Euclidean space. The solution is obtained using the orthogonal-constrained least squares closed-form solution method. After obtaining the rotation matrix and translation vector, the average reprojection residual is calculated as a robustness measure. The rotation matrix is ​​denoted as The translation vector is denoted as The current three-dimensional feature points are denoted as The corresponding three-dimensional feature points in the previous period are denoted as The number of three-dimensional feature points is denoted as If the average reprojection residual exceeds the robust interval set based on the noise model, it reverts to the previous stable state of the uniform set and solves again until it enters the robust interval, thereby obtaining the initial registration relation and the average reprojection residual.

4. The method for dynamically predicting the earthwork volume of a pumped storage power station according to claim 1, characterized in that, Step S2 includes: Obtain the structured event stream of the construction log, and combine it with the reference domain state record to generate a log change mask, i.e., an observation change mask; Based on the state records of the reference domain, the closure consistency of each grid is calculated to form a closure consistency graph; Within each grid, the spatial intersection-union ratio of the log change mask and the observation change mask is calculated to obtain the change consistency and form a consistency map; Within each grid, ordinal mapping is performed on the closure consistency and change compliance. The ordinal synthesis is completed using the quantile threshold method. Combined with the four-adjacent connectivity rule of the grid coverage map, three types of labels are marked: available, avoidable, and pending review. The available labels are summarized to form an available area index, and the avoidable and pending review labels are summarized to form an area of ​​interest.

5. The method for dynamically predicting the earthwork volume of a pumped storage power station according to claim 4, characterized in that, In step S2: Closedness consistency is expressed as path average. The length of the loop is The arc length position variable is The displacement residual is ; Grid The degree of consistency of the change is , where grid The inner log mask area is The observation mask area is .

6. The method for dynamically predicting the earthwork volume of a pumped storage power station according to claim 1, characterized in that, Step S3 includes: Within the range of available area index, select grids with better closure consistency as anchor sets, project the UAV image point cloud of this period based on the initial registration relationship, and construct the correspondence of each projection point; A point-to-plane iterative nearest-point algorithm is used for restricted registration. A voxel pyramid is constructed, and the algorithm is optimized layer by layer from coarse to fine. The solution of the upper layer is used as the initial value of the lower layer to obtain the fine registration relationship. The coordinate framework of the current image point cloud to the boundary of the previous reference domain is unified to obtain the terrain alignment result. The number of iterations and the convergence status are recorded as the registration log. Based on the terrain alignment results, the quantile magnitude and sampling sufficiency of the residual distribution are calculated on the reference domain grid. A reliable coverage map is generated according to the double threshold rule and excluding areas of interest. An alternative benchmark strategy is triggered for areas of interest, and stable anchored segments are used for local solutions first. If insufficient, historical slices from periods unaffected by the operation are introduced as temporary benchmark segments and incorporated into the reliable coverage map according to the connectivity rule.

7. The method for dynamically predicting the earthwork volume of a pumped storage power station according to claim 1, characterized in that, Step S4 includes: Establish surface elevation functions for the previous and current periods within the cells marked on the reliable coverage map; Based on the surface elevation functions of the previous and current periods, the volume difference is calculated according to the cell area. Based on the volume difference, the volume of each construction zone is aggregated and combined with the current time tag to form a sequence of engineering quantities. Among them, the zone with no available cells is marked as missing measurement status. Based on the engineering quantity sequence and the current target volume of each partition, deviation diagnosis is performed. Within each partition, the median sufficiency of available cell sampling is statistically analyzed. The confidence level is composed by combining the median statistics of the residual quantile amplitude within the partition, and a planning interface table is compiled.

8. The method for dynamically predicting the earthwork volume of a pumped storage power station according to claim 1, characterized in that, Step S5 includes: Taking deviation diagnosis as the entry point, a set of verification windows is generated within the partition by combining the credible coverage map and the region of interest. A set of verification window cells is extracted for each verification window, and an independent anchoring set and short-term aerial survey segments are constructed as two types of independent evidence. Within the verification window, the point-to-plane iterative nearest point algorithm is used to obtain the local rigid verification solution. The normal median offset and the verification volume obtained by aggregating the vertex set of the verification window are calculated. The error source classification is completed based on the joint relationship between the two. The classification results and the verification window identifier are written into the registration log. Based on the source of error, reference domain update instructions and route fine-tuning suggestions are generated. The reference domain update instructions include grid addition and deletion, connected component redrawing, and collection rhythm adjustment. The results are structured into records at the granularity of the review window, archived as a prerequisite attachment to the next period's reference domain status record, and synchronized to the remarks column of the planning interface table.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-8.

10. A computer-readable storage medium storing computer instructions thereon, characterized in that, When executed by the processor, this instruction implements the steps of the method as described in any one of claims 1-8.

Images