A method for compensating spatial deformation errors of 3D point clouds for engineering informatization

Abstract

This invention provides a method for compensating for spatial deformation errors in 3D point clouds for engineering informatization, comprising: reading the engineering informatization benchmark model of the target engineering object, multi-period on-site 3D point clouds, and external engineering records; generating a data package of the target engineering object, a set of component units, a record association table, and an initial spatial correspondence table; generating a benchmark credibility matrix based on the initial spatial residual, point cloud repetition rate, residual direction consistency rate, point cloud boundary continuity evaluation value, and model support features; identifying benchmark conflict units based on the benchmark credibility matrix, distinguishing between point cloud acquisition errors, model not-updated errors, and real spatial deformation, and generating a conflict handling strategy table by combining a reinforcement learning strategy model with hard constraint interception; and performing model reverse update, point cloud compensation, and real spatial deformation preservation respectively to generate a spatial consistency model and benchmark conflict handling records.

Description

Technical Field

[0001] This invention relates to the field of engineering informatization and 3D point cloud processing technology, and in particular to a method for compensating for spatial deformation errors of 3D point clouds for engineering informatization. Background Technology

[0002] With the application of engineering informatization and 3D modeling technology in scenarios such as buildings, bridges, and foundation pits, 3D point clouds are often used for as-built verification, component deviation analysis, and spatial deformation monitoring. Multi-stage on-site 3D point clouds obtained through ground laser scanning, kinematic surveying, or photogrammetry can reflect the spatial state of the target engineering object at different construction or operation and maintenance stages; while engineering informatization benchmark models are used to indicate the location of components, component boundaries, and relationships between adjacent components.

[0003] Existing 3D point cloud error compensation methods typically use an engineering information benchmark model as a fixed reference, registering the on-site 3D point cloud to this benchmark model and correcting residuals. This approach assumes that the benchmark model is accurate, complete, and has been synchronized with on-site changes, and mainly focuses on eliminating point cloud acquisition errors, coordinate transformation errors, and local registration errors.

[0004] However, in actual engineering projects, the engineering information benchmark model may differ from the actual site conditions due to unsynchronized design changes, outdated construction records, or unarchived component replacements. In such cases, the stable spatial deviations present in multiple phases of the site's 3D point cloud do not necessarily stem from point cloud errors; they may also originate from outdated models or actual spatial deformations. If the benchmark model is used as the sole reliable object for compensation, it is easy to incorrectly drag the actual site conditions back to the outdated model, resulting in a distorted 3D data foundation for the engineering information platform.

[0005] Therefore, this invention proposes a method for compensating for spatial deformation errors in three-dimensional point clouds for engineering informatization. The information disclosed in the background section is only for enhancing understanding of the background of this disclosure and may therefore contain prior art information that is not common knowledge to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by providing a three-dimensional point cloud spatial deformation error compensation method for engineering informatization, thereby solving the technical problems mentioned in the background section.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for compensating for spatial deformation errors in three-dimensional point clouds for engineering informatization includes the following steps: S1. Read the engineering information benchmark model, multi-phase on-site 3D point cloud and external engineering records of the target engineering object, perform coordinate format unification and validity verification on the multi-phase on-site 3D point cloud, generate target engineering object data package, and generate component unit set, record association table and initial spatial correspondence table. S2. Calculate the point cloud recurrence rate, residual amplitude fluctuation, residual direction consistency rate and point cloud boundary continuity evaluation value based on the initial spatial correspondence table, generate a point cloud fact feature table, generate a model support feature table based on the record association table, and fuse them to generate a benchmark confidence matrix. S3. Identify the set of benchmark conflict units and the set of undeterminable verification units based on the benchmark credibility matrix, make a preliminary judgment on the source of conflict for the set of benchmark conflict units, and generate a conflict handling strategy table by combining a reinforcement learning strategy model with hard constraint interception. S4. Based on the conflict resolution strategy table, perform model reverse update, point cloud compensation and deformation retention on the corresponding benchmark conflict units respectively to generate the updated engineering information benchmark model, the compensated on-site 3D point cloud and the real space deformation retention unit. S5. Based on the updated engineering information benchmark model, the compensated on-site 3D point cloud and the real spatial deformation preservation unit, perform 3D modeling and fusion to generate a spatial consistency model and benchmark conflict handling record, and output the engineering information compensation result.

[0008] S1 specifically includes: reading the engineering information benchmark model of the target engineering object, multi-phase on-site 3D point clouds, and external engineering records; unifying the coordinate format, time stamping, data acquisition source marking, and validity verification of the multi-phase on-site 3D point clouds; generating a target engineering object data package; dividing the target engineering object data package into component units, binding external engineering records and component adjacency relationship data, extracting point cloud regions, and generating a component unit set and record association table; performing 3D modeling processing based on the component unit set and record association table, establishing the initial spatial correspondence between component model regions, point cloud regions, and adjacent component relationships, calculating the initial spatial residual, and generating an initial spatial correspondence table.

[0009] S2 specifically includes: reading the initial spatial correspondence table, extracting the main point cloud region, initial spatial residual, residual direction, and point cloud boundary for each collection period according to the component identifier, calculating the point cloud recurrence rate, residual amplitude fluctuation, residual direction consistency rate, and point cloud boundary continuity evaluation value, and generating a point cloud fact feature table; reading the record association table and the initial spatial correspondence table, verifying the record integrity, record time consistency, component number consistency, geometric change support, and adjacent component relationship closure for each component unit, and generating a model support feature table; fusing the point cloud fact feature table and the model support feature table according to the component identifier, calculating the point cloud fact stability and model baseline reliability, and generating a baseline reliability matrix.

[0010] S3 specifically includes: reading the baseline confidence matrix, extracting point cloud factual stability, model baseline confidence, initial spatial residual mean, residual direction consistency rate and status according to component identifiers, generating component residual judgment thresholds based on component type, component scale, acquisition accuracy and component functional attributes, identifying the baseline conflict unit set and the undecidable verification unit set; generating preliminary conflict source judgment results based on the baseline conflict unit set and the baseline confidence matrix, combined with recorded conflict identifiers, point cloud boundary anomaly propagation identifiers and adjacent component relationship closure status; inputting the preliminary conflict source judgment results and the baseline conflict unit set into the reinforcement learning strategy model, and generating a conflict handling strategy table by combining hard constraint interception.

[0011] S4 specifically includes: reading the conflict resolution strategy table, performing update access verification on the benchmark conflict units of the assigned model reverse update strategy, forming a factual benchmark based on the stable main point cloud region, updating the component geometric boundaries, positioning parameters, and connection relationships of adjacent components of the engineering information benchmark model, and generating the updated engineering information benchmark model; performing compensation access verification on the benchmark conflict units of the assigned point cloud compensation strategy, using the component model region of the credible component unit as the compensation benchmark, generating point-level compensation vectors for the main point cloud region, and outputting the compensated on-site 3D point cloud; and generating real-space deformation preservation units for the benchmark conflict units of the assigned deformation preservation strategy based on the residual transfer consistency value and the closed state of the relationship between adjacent components.

[0012] S5 specifically includes: reading the updated engineering information benchmark model, the compensated on-site 3D point cloud, the real spatial deformation preservation unit, and the conflict handling strategy table; performing 3D modeling fusion according to the fusion priority; calculating the spatial consistency evaluation value; and generating a spatial consistency model; generating benchmark conflict handling records based on the spatial consistency model and the conflict handling strategy table, and generating a credibility identifier based on the credibility of the handling source, the spatial consistency evaluation value, the integrity of the strategy execution, and the hard constraint interception result; and encapsulating the spatial consistency model and the benchmark conflict handling records into engineering information compensation results, outputting a component index table, a credibility index table, a compensation displacement layer, and a review task list.

[0013] The beneficial effects of this invention are as follows: This invention simultaneously calculates the factual stability of point clouds and the benchmark reliability of the model, determining the source of spatial deviation before point cloud compensation. This distinguishes between point cloud acquisition errors, model update failure errors, and actual spatial deformation, preventing incorrect compensation of the actual on-site state to an outdated engineering information benchmark model. Based on the benchmark reliability matrix, it identifies benchmark conflict units and generates a conflict resolution strategy table by combining a reinforcement learning strategy model and hard constraint interception. This provides clear data support for model reverse updates, point cloud compensation, and deformation preservation, reducing misjudgments caused by relying solely on residual thresholds.

[0014] This invention improves the consistency of the 3D data base in the engineering information platform by performing reverse model updates on benchmark conflict units with high point cloud factual stability and low model benchmark reliability. This allows stable field facts from multiple phases of on-site 3D point clouds to be used to correct the engineering information benchmark model. Point cloud compensation is performed on benchmark conflict units with high model benchmark reliability and low point cloud factual stability. Compensation boundaries and prohibition markers limit the compensation range, correcting acquisition errors while preventing compensation actions from crossing real-space deformation regions.

[0015] This invention generates realistic spatial deformation-preserving units from benchmark conflict units with residual transfer consistency, ensuring that risk signals such as component displacement, connection node offset, and supporting structure deformation are preserved during the 3D modeling and fusion process. This improves the reliability of as-built review, operation and maintenance monitoring, and spatial deformation analysis. It outputs a spatial consistency model, benchmark conflict resolution records, credibility identifiers, and a review task list, making every model update, point cloud compensation, and deformation preservation traceable, verifiable, and rollbackable, thus enhancing the review value and engineering usability of the engineering information compensation results. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the three-dimensional point cloud spatial deformation error compensation method for engineering informatization according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example: Figure 1 As shown, this embodiment provides a method for compensating for spatial deformation errors in three-dimensional point clouds for engineering informatization, including the following steps: S1. Read the engineering information benchmark model, multi-phase on-site 3D point cloud and external engineering records of the target engineering object, perform coordinate format unification and validity verification on the multi-phase on-site 3D point cloud, generate target engineering object data package, and generate component unit set, record association table and initial spatial correspondence table. S2. Calculate the point cloud recurrence rate, residual amplitude fluctuation, residual direction consistency rate and point cloud boundary continuity evaluation value based on the initial spatial correspondence table, generate a point cloud fact feature table, generate a model support feature table based on the record association table, and fuse them to generate a benchmark confidence matrix. S3. Identify the set of benchmark conflict units and the set of undeterminable verification units based on the benchmark credibility matrix, make a preliminary judgment on the source of conflict for the set of benchmark conflict units, and generate a conflict handling strategy table by combining a reinforcement learning strategy model with hard constraint interception. S4. Based on the conflict resolution strategy table, perform model reverse update, point cloud compensation and deformation retention on the corresponding benchmark conflict units respectively to generate the updated engineering information benchmark model, the compensated on-site 3D point cloud and the real space deformation retention unit. S5. Based on the updated engineering information benchmark model, the compensated on-site 3D point cloud and the real spatial deformation preservation unit, perform 3D modeling and fusion to generate a spatial consistency model and benchmark conflict handling record, and output the engineering information compensation result.

[0019] S1 specifically includes the following sub-steps: S110. Generate target project object data package: Read the target project object's engineering information benchmark model, multi-period on-site 3D point cloud, design change record, construction batch record, acceptance image record, and component adjacent relationship data; among which, the engineering information benchmark model is the target project object's 3D model data stored in the engineering information platform, and the multi-period on-site 3D point cloud is the point cloud data obtained in different acquisition periods that reflects the on-site spatial state of the target project object.

[0020] The system performs time stamping, source marking, coordinate format standardization, and validity verification on multiple phases of on-site 3D point clouds. Time stamping is used to distinguish the point cloud acquisition phases, source marking is used to distinguish the sources of ground laser scanning, mobile measurement, UAV measurement, or photogrammetry, and coordinate format standardization is used to convert the point cloud coordinates from different acquisition sources to the project coordinate system.

[0021] When unifying coordinate formats, the external parameters of the acquisition equipment, the coordinates of engineering control points, and the project coordinate system of the engineering information benchmark model are read. When the multi-phase on-site 3D point cloud contains engineering control points, the engineering control points are used as coordinate transformation constraints. When the multi-phase on-site 3D point cloud does not contain engineering control points, the column bases, supports, wall corners, foundation edges, or permanent measurement marks that maintain stable positions in multiple acquisition phases in the engineering information benchmark model are used as auxiliary transformation constraints. The coordinate transformation parameters corresponding to each acquisition phase are obtained, and the transformed point cloud coordinates are written into the target engineering object data package.

[0022]

[0023] in, This represents the coordinates of the i-th point in the 3D point cloud of the k-th phase in the acquisition coordinate system. This indicates the coordinates of the point after transformation to the project coordinate system. This represents the rotation matrix corresponding to the 3D point cloud of the k-th phase. Let represent the translation vector corresponding to the 3D point cloud of the k-th phase.

[0024] If the control point residual of a certain acquisition period exceeds the preset coordinate verification threshold, or if there is an overall unidirectional offset between the transformed point cloud and the stable component, then the acquisition period will be marked as a coordinate verification pending state and will not participate in the subsequent point cloud factual stability calculation, but will only be retained as candidate point cloud data.

[0025] The resulting target engineering object data package includes the engineering information benchmark model, the converted multi-period on-site 3D point cloud, the collection period identifier, the collection source identifier, the coordinate verification status, and the external engineering records, which serve as the unified input for S120 component unit division and point cloud area extraction.

[0026] S120. Generate component unit sets and record association tables: Based on the target engineering object data package, divide the engineering information benchmark model into component units. Component unit division prioritizes component identifiers; when the engineering information benchmark model contains complete component identifiers, component units are directly generated according to the component identifier, component type, and component geometric boundary; when the engineering information benchmark model does not contain complete component identifiers, supplementary segmentation is performed according to component geometric continuity, connection nodes, material type, cross-sectional abrupt change location, and adjacent component relationships, and a unique component identifier is generated for the component units obtained from the supplementary segmentation.

[0027] A component unit includes at least a component model patch, a component geometric boundary, a component type, a component local coordinate axis, and adjacent component identifiers. The component model patch is used to establish a spatial correspondence with the point cloud patch later.

[0028] Based on the component identifier, each component unit is bound to design change records, construction batch records, acceptance image records, and component adjacency relationship data to form a record association table. When any of the design change records, construction batch records, or acceptance image records for the same component unit are missing, a record missing identifier is written into the record association table. When the component number, construction time, or adjacent component relationship in the record cannot be closed with the component identifier, a record conflict identifier is written. Record missing identifiers and record conflict identifiers are not directly used as model error conclusions, but rather as inputs for subsequent model baseline reliability calculations.

[0029] When extracting point cloud regions, a component-enclosed region is generated based on the geometric boundary of the component model region, and a candidate point cloud region is formed by expanding the boundary according to a preset distance. Points within the candidate region that are no more than a preset distance threshold from the surface of the component model region, have a normal angle no greater than a preset angle threshold, and a local point density no less than a preset density threshold are selected as the point cloud region corresponding to that component unit. If the number of valid points within a point cloud region is less than a preset number threshold, the component unit is marked as having insufficient point cloud regions, and this status is written to a record association table for later use in determining the stability of the point cloud.

[0030] in, This represents the point cloud region corresponding to the j-th component unit. This represents the point cloud points after transformation to the project coordinate system. This represents the component model patch of the j-th component unit. This represents the distance from a point in the point cloud to a patch in the component model. This indicates a preset distance threshold. This represents the angle between the point cloud point normal and the component model patch normal. This indicates the preset angle threshold.

[0031] This generates a set of component units and a record association table, which serve as input for S130 to establish the initial spatial correspondence table.

[0032] S130. Generate the initial spatial correspondence table: Based on the component unit set and record association table, perform 3D modeling processing on each component unit to form the initial spatial correspondence between component model areas, point cloud areas, and adjacent component relationships. When establishing the initial spatial correspondence, component identifier consistency is used as the first matching condition, and spatial proximity, boundary overlap rate, normal consistency, and adjacent component relationship closure are used as auxiliary matching conditions. When the same component model area corresponds to multiple point cloud areas, the point cloud area with the highest boundary overlap rate and adjacent component relationship closure is selected as the main point cloud area, and the remaining point cloud areas are retained as candidate point cloud areas. When multiple component model areas compete for the same point cloud area, a unique main correspondence is determined according to the adjacent component relationship closure, residual direction continuity, and component type consistency.

[0033] For each component unit, the initial spatial residual is calculated for the main point cloud region and the component model region. The initial spatial residual includes the residual amplitude and the residual direction. The residual direction is determined according to the outward normal of the component model region or the local coordinate axis of the component. The residual amplitude is determined according to the statistical value of the distance from the effective point in the main point cloud region to the component model region.

[0034] For component elements with insufficient point cloud regions, no effective initial spatial residual is generated; only the insufficient point cloud region state is retained to avoid misjudging occlusion or insufficient point count as model update error in the future.

[0035]

[0036] in, This represents the initial spatial residual of the j-th component element. This represents the number of valid points within the point cloud region corresponding to the j-th component unit. This represents the i-th point in the point cloud after transformation to the project coordinate system. This represents the component model patch of the j-th component unit. This represents the distance from a point in the point cloud to a region of the component model.

[0037] The component identifier, point cloud period, acquisition source identifier, model boundary, point cloud boundary, main point cloud region, candidate point cloud region, adjacent component identifier, record missing identifier, record conflict identifier, point cloud region insufficiency status, initial spatial residual, and residual direction are written into the initial spatial correspondence table. The initial spatial correspondence table serves as the common input for S210 to calculate the factual features of the point cloud, S220 to calculate the supporting features of the model, and S310 to identify the reference conflict unit.

[0038] S2 specifically includes the following sub-steps: S210. Generate a point cloud factual feature table: Read the initial spatial correspondence table generated in S130, and extract the main point cloud region, point cloud region deficiency status, initial spatial residual, residual direction, point cloud boundary, and adjacent component identifiers of the same component unit in each acquisition period according to the component identifier. Count the number of acquisition periods with valid main point cloud regions for each component unit, and calculate the point cloud recurrence rate; a valid main point cloud region refers to a point cloud region that is not marked as having a point cloud region deficiency status and whose number of valid points is not lower than a preset threshold.

[0039]

[0040] in, This represents the recurrence rate of the point cloud of the j-th component element. This indicates the number of collection periods in which the j-th component unit has a valid main point cloud region, and K represents the total number of collection periods involved in the calculation.

[0041] When the point cloud recurrence rate is not lower than the preset recurrence threshold, the component unit is marked as recurring; when the point cloud recurrence rate is lower than the preset recurrence threshold and there is a point cloud area insufficiency, the component unit is marked as incomplete acquisition and is not directly considered as having unstable point cloud facts.

[0042] Furthermore, the initial spatial residuals of the same component unit in each effective acquisition period are read, the residual amplitude fluctuation is calculated, and the residual direction consistency rate is statistically analyzed. The residual direction consistency rate is obtained by counting whether the angle between the residual direction and the main residual direction in each effective acquisition period is less than a preset direction angle threshold. When the residual amplitude fluctuation is not greater than the preset residual fluctuation threshold and the residual direction consistency rate is not lower than the preset direction consistency threshold, the component unit is marked as a point cloud residual stable state.

[0043]

[0044] in, This represents the residual amplitude fluctuation of the j-th component element. This indicates the number of acquisition periods for the j-th component unit to have a valid main point cloud region. This represents the initial spatial residual of the j-th component element in the k-th valid acquisition period. This represents the mean of the initial spatial residuals of the j-th component element during the effective acquisition periods.

[0045] The point cloud boundaries of the same component unit are compared with the point cloud boundaries of adjacent components to calculate the boundary spacing, boundary overlap rate, and boundary normal angle. When the boundary spacing is not greater than a preset boundary spacing threshold, the boundary overlap rate is not lower than a preset boundary overlap threshold, and the boundary normal angle is not greater than a preset boundary angle threshold, it is marked as a continuous point cloud boundary state. If abrupt changes in boundary spacing occur continuously in multiple acquisition periods and the residual direction is continuously propagated along adjacent components, it is not treated as acquisition noise but is written into the point cloud boundary abnormal propagation flag for S320 to call when judging the actual spatial deformation.

[0046] This generates a point cloud fact feature table, which includes component identification, point cloud recurrence rate, incomplete acquisition status, residual amplitude fluctuation, residual direction consistency rate, point cloud boundary continuity evaluation value, and point cloud boundary anomaly propagation identifier.

[0047] S220. Generate Model Support Feature Table: Read the record association table generated in S120 and the initial spatial correspondence table generated in S130, and perform model support verification for each component unit according to the component identifier. The model support verification includes at least record integrity verification, record time consistency verification, component number consistency verification, geometric change support verification, and adjacent component relationship closure verification.

[0048] Record integrity verification is used to determine whether design change records, construction batch records, and acceptance image records are complete; record time consistency verification is used to determine whether the record formation time is earlier than or equal to the corresponding collection period; component number consistency verification is used to determine whether the component number in the external engineering record is consistent with the component identifier; geometric change support verification is used to determine whether the record content can explain the initial spatial residual between the component model area and the main point cloud area; adjacent component relationship closure verification is used to determine whether the connection relationship between the component unit and adjacent components is consistent in the engineering information benchmark model, component adjacent relationship data, and external engineering records.

[0049] Adjacent component relationship closure includes connection object closure, connection type closure, and spatial position closure. Connection object closure means that adjacent components recorded in the engineering information benchmark model can be found in the component adjacency relationship data; connection type closure means that the connection relationship between two components is of the same type in both the model and the record; spatial position closure means that the position difference between the connection nodes of two components does not exceed a preset connection position threshold. If any verification item is not met, the corresponding verification item is written to the model support deficiency state, and the reason for the deficiency is recorded.

[0050] For unstructured external engineering records (such as design change texts and acceptance image records), the system introduces a semantic extraction module based on natural language processing (NLP) and optical character recognition (OCR).

[0051] Specifically, when performing geometric change support verification, the text in the engineering record is first identified by OCR, and a set of keywords containing spatial change semantics (such as "offset", "widening", "elevation adjustment") is extracted. Then, the triplet "component number-action-value" is extracted using dependency parsing. If the difference between the extracted numerical value and the initial spatial residual between the component model area and the main point cloud area is within the preset tolerance range, the geometric change support is determined to be valid, and the corresponding geometric change support evaluation value is assigned as 1; otherwise, it is 0. This automated analysis process avoids subjective human intervention in the engineering records.

[0052] This generates a model support feature table, which includes component identifiers, record integrity evaluation values, record time consistency evaluation values, component number consistency evaluation values, geometric change support evaluation values, adjacent component relationship closure evaluation values, record missing identifiers, record conflict identifiers, and reasons for insufficient model support.

[0053] S230. Generate the baseline credibility matrix: Based on the component identifier, fuse the point cloud fact feature table and the model support feature table, and calculate the point cloud fact stability and the model baseline credibility respectively. The point cloud fact stability is generated by weighting the point cloud recurrence rate, residual stability evaluation value, residual direction consistency rate, and point cloud boundary continuity evaluation value; among which, the residual stability evaluation value is determined based on the comparison result of the residual amplitude fluctuation amount and the preset residual fluctuation threshold, and the point cloud boundary continuity evaluation value is determined based on the comprehensive verification result of boundary spacing, boundary overlap rate, and boundary normal angle.

[0054]

[0055] in, This represents the point cloud factual stability of the j-th component element. This represents the recurrence rate of point clouds. This represents the residual stability evaluation value, and this represents the residual direction consistency rate. This represents the continuous evaluation value of the point cloud boundary. , , and These represent the weights of the corresponding evaluation items, and .

[0056] The benchmark credibility of the model is generated by weighting the evaluation values ​​of record integrity, record time consistency, component number consistency, geometric change support, and adjacent component relationship closure. When there are missing record indicators, conflicting record indicators, or non-closed adjacent component relationship indicators, the corresponding evaluation item is assigned a low evaluation value, and the reason for the deduction is written into the benchmark credibility matrix.

[0057]

[0058] in, This represents the model baseline confidence level of the j-th component element. This represents the record integrity evaluation value. This indicates the consistency evaluation value for the recorded time. This indicates the consistency evaluation value of the component number. This indicates the evaluation value supporting geometric changes. This indicates the evaluation value of the closure of the relationship between adjacent components. , , , and These represent the weights of the corresponding evaluation items, and .

[0059] A baseline reliability matrix is ​​generated using component identifiers as matrix row indices and evaluation fields as matrix column indices. Each matrix cell includes at least the point cloud factual stability, model baseline reliability, initial spatial residual mean, residual amplitude fluctuation, residual direction consistency rate, point cloud boundary continuity evaluation value, record missing identifier, record conflict identifier, adjacent component relationship closure status, point cloud region insufficiency status, and point cloud boundary anomaly propagation identifier. The baseline reliability matrix serves as direct input for S310 to screen baseline conflict cells, S320 to determine conflict sources, and S330 to generate a conflict resolution strategy table.

[0060] S3 specifically includes the following sub-steps: S310, Identify Benchmark Conflict Units: Read the benchmark credibility matrix generated in S230, and extract the point cloud factual stability, model benchmark credibility, initial spatial residual mean, residual amplitude fluctuation, residual direction consistency rate, point cloud area insufficiency status, coordinate verification pending status, incomplete acquisition status, record missing identifier, record conflict identifier, and adjacent component relationship closure status for each component unit according to the component identifier. For different component units, generate component residual judgment thresholds based on component type, component size, acquisition accuracy, and component functional attributes; among them, when flanges, bolt holes, supports, hanger connection points, and pipeline interfaces are used as functional interface components, the component residual judgment thresholds are lower than those for ordinary panel or wall components to avoid omission when small residuals affect assembly or connection relationships.

[0061] Calculate the normalized residual conflict value of the component element: in, This represents the normalized residual conflict value of the j-th component element. This represents the mean of the initial spatial residuals for the j-th component element. This represents the component residual judgment threshold corresponding to the j-th component unit.

[0062] When the normalized residual conflict value is not less than 1 and the residual direction consistency rate is not lower than the preset direction consistency threshold, the corresponding component unit is included in the candidate conflict unit. The candidate conflict unit is then subjected to a decisionability verification: if the baseline confidence matrix contains insufficient point cloud areas, coordinates pending verification, or incomplete data collection, and the number of valid data collection periods is lower than the preset minimum number of periods, it is not directly included in the baseline conflict unit set, but rather in the undecidable verification unit set; the undecidable verification unit set is not included in the conflict handling strategy generation in S330, but is only output as objects requiring supplementary data collection or coordinate verification in S530.

[0063] Through the above screening, a set of benchmark conflict units is generated. The set of benchmark conflict units includes component identification, conflict location, normalized residual conflict value, initial spatial residual mean, residual direction, point cloud factual stability, model benchmark confidence and closure state of adjacent component relationships, and serves as input for the initial judgment of S320 conflict source.

[0064] S320. Generate preliminary conflict source judgment results: Based on the benchmark conflict unit set and benchmark credibility matrix, perform preliminary conflict source judgment for each benchmark conflict unit. First, read the point cloud factual stability, model benchmark credibility, residual amplitude fluctuation, residual direction consistency rate, point cloud boundary anomaly propagation identifier, record missing identifier, record conflict identifier, and adjacent component relationship closure status to establish a conflict source judgment table.

[0065] If the point cloud has low factual stability, high residual amplitude fluctuation, low residual direction consistency rate, and high model baseline reliability, then the baseline conflict unit is initially judged as a point cloud acquisition error; if the point cloud has high factual stability, low model baseline reliability, and there are missing record identifiers, record conflict identifiers, or missing geometric change support, then the baseline conflict unit is initially judged as a model not updated error; if the point cloud has high factual stability, high model baseline reliability, high residual direction consistency rate, and there are point cloud boundary abnormal propagation identifiers, then the initial judgment of real space deformation continues.

[0066] To improve the reproducibility of the judgment of model not being updated, the propensity score for model not being updated is calculated:

[0067] in, This indicates that the model has not updated the tendency value for the j-th baseline conflict unit. Indicates the factual stability of the point cloud. Indicates the model's baseline confidence level. This indicates the record of conflict evaluation values. This indicates the missing evaluation value for geometric change support. , , and These represent the weights of the corresponding evaluation items. If the model not updated propensity value is higher than the preset model not updated threshold, the corresponding benchmark conflict unit is initially judged as a model not updated error.

[0068] Among them, when there are record conflict identifiers in the benchmark confidence matrix, The value is 1 if the geometric change support is missing, otherwise it is 0; The value is 1 if it is not 0 otherwise.

[0069] For the initial assessment of real-space deformation, calculate the residual propagation consistency value:

[0070] in, This represents the residual propagation consistency value of the j-th reference conflict cell. This represents the number of adjacent components that have an adjacent transmission relationship with the j-th reference conflict unit. This represents the angle between the residual direction of the j-th reference conflict element and the residual direction of the r-th adjacent component.

[0071] When the residual propagation consistency value is not lower than the preset propagation consistency threshold and the relationship between adjacent components is closed, a deformation causal chain identifier is generated, and the corresponding benchmark conflict unit is initially judged as a real spatial deformation. This generates an initial conflict source judgment result, which includes point cloud acquisition error, model not updated error, real spatial deformation, and undeterminable verification source, and serves as the input for S330 to generate the conflict handling strategy table.

[0072] S330. Generate a conflict resolution strategy table: Input the initial judgment result of the conflict source, the set of baseline conflict units, and the baseline credibility matrix into the reinforcement learning strategy model. The state variables of the reinforcement learning strategy model include point cloud factual stability, model baseline credibility, normalized residual conflict value, residual amplitude fluctuation, residual direction consistency rate, residual propagation consistency value, recorded conflict evaluation value, geometric change support missing evaluation value, point cloud region insufficiency state, and coordinate pending verification state; the action space includes point cloud compensation strategy, model reverse update strategy, deformation preservation strategy, and undecidable verification strategy; the reward function consists of spatial consistency improvement value, real space deformation preservation value, erroneous baseline correction value, and invalid compensation deduction value.

[0073] As a preferred implementation, the reinforcement learning policy model specifically employs a deep Q-network (DQN) architecture. Its network topology includes an input layer, multiple fully connected hidden layers, and an output layer.

[0074] Specifically, the input layer receives a set of state variables. The set is vectorized into a fixed-dimensional (e.g., 10-dimensional) state vector; the hidden layer contains at least two fully connected layers, each using the ReLU activation function for non-linear feature extraction; The output layer uses a linear activation function, and the number of nodes in the output layer is the same as the number of candidate actions A. It outputs the Q-value (i.e., the policy score) for each action. ).

[0075] The specific calculation logic of the reward function is as follows: ; in, This represents the improvement in spatial consistency (the difference between the mean residuals before and after fusion). Reserve a flag for real-world spatial deformation (1 if successful, 0 otherwise). Correct the penalty value for the error baseline. The penalty value for invalid compensation. to This corresponds to the reward weight.

[0076] The training process of the model adopts an experience replay mechanism and a dual-network structure of the target network. The mean squared error (MSE) is used as the loss function, and the gradient descent algorithm (such as the Adam optimizer) is used to backpropagate and update the weight parameters of the current network until the average reward value of the model on the validation set converges.

[0077] During training, historically confirmed engineering point cloud compensation samples are used to establish the correspondence between state variables and disposal actions; during execution, the reinforcement learning policy model outputs the policy score of each candidate disposal action based on the current state variable.

[0078]

[0079] in, This represents the conflict resolution strategy for the j-th reference conflict unit. Let 'a' represent the set of state variables for the j-th reference conflict unit, and 'a' represent the candidate action. Represents a set of candidate actions. This indicates that the reinforcement learning policy model applies to the set of state variables. The strategy score for executing candidate action a.

[0080] After the reinforcement learning strategy model outputs a strategy, hard constraint interception is executed: when the benchmark conflict unit has insufficient point cloud regions or coordinates pending verification, the output of the model reverse update strategy is prohibited; when the point cloud fact stability is lower than the preset fact threshold, the output of the deformation preservation strategy is prohibited; when the model benchmark credibility is lower than the preset benchmark threshold and there are record conflict markers, the engineering information benchmark model is prohibited from being directly used as the point cloud compensation benchmark. If the candidate action with the highest strategy score violates the hard constraints, it is rewritten as an undecidable verification strategy, or rewritten as a model reverse update strategy, point cloud compensation strategy, or deformation preservation strategy based on the initial judgment result of the conflict source.

[0081] Finally, a conflict resolution strategy table is generated, which includes component identifier, initial judgment result of conflict source, set of state variables, strategy score, hard constraint interception result and final resolution strategy, and serves as the control basis for S410, S420 and S430.

[0082] S4 specifically includes the following sub-steps: S410, Perform reverse update of engineering information benchmark model: Read the conflict handling strategy table generated by S330, and extract the component identifier, conflict location, point cloud factual stability, model benchmark credibility, initial spatial residual mean, residual direction, record conflict identifier, geometric change support missing evaluation value and adjacent component relationship closure status for the benchmark conflict unit assigned the model reverse update strategy.

[0083] For the benchmark conflict unit, an update access verification is first performed: if the point cloud fact stability is higher than the preset fact threshold, the model benchmark credibility is lower than the preset benchmark threshold, and there are no insufficient point cloud regions or coordinates pending verification states, then a reverse update of the engineering information benchmark model is allowed; if any of the above conditions are not met, a reverse update is not performed, and the benchmark conflict unit is written to the model update pending review state. The update access verification is used to prevent single-phase abnormal point clouds or coordinate transformation anomalies from contaminating the engineering information benchmark model.

[0084] For the benchmark conflict unit that has passed the updated access verification, the main point cloud region that exists stably in the multi-period field 3D point cloud is read, and the main point cloud region is used to form the factual benchmark. The model is not directly updated by using the point cloud region of a single collection period.

[0085] The reverse model update includes at least component geometric boundary update, component positioning parameter update, and adjacent component connection relationship update. Among them, component geometric boundary update is used to correct the boundary points, boundary lines, or component surfaces of component model areas; component positioning parameter update is used to correct the position offset and attitude deflection of component elements in the engineering project coordinate system; and adjacent component connection relationship update is used to correct the connection objects, connection node coordinates, and connection types between components.

[0086]

[0087] in, This represents the reverse update displacement of the j-th component element. This indicates the number of acquisition periods for the j-th component unit to have a valid main point cloud region. This represents the mean residual value of the j-th component unit during the k-th valid data acquisition period. This represents the residual direction unit vector of the j-th component unit in the k-th valid acquisition period.

[0088] Based on the reverse update displacement of the model, the component model area of ​​the corresponding component unit in the engineering information benchmark model is updated, and the model update source, update displacement, update direction, and reference collection period are recorded simultaneously. After the reverse update is completed, the connection objects, connection node positions, and connection types are re-verified according to the closed state of the relationship between adjacent components. If the difference in the updated connection node positions does not exceed the preset connection position threshold, the reverse update is confirmed to be valid, and the updated engineering information benchmark model is generated. If the update results in the relationship between adjacent components not being closed, the update calculation results are retained but written to the model update pending verification status, and the update results are not allowed to directly participate in the spatial consistency replacement of S510.

[0089] S420, Perform on-site 3D point cloud spatial deformation error compensation: Read the conflict handling strategy table, and for the benchmark conflict unit assigned the point cloud compensation strategy, extract the model benchmark credibility, point cloud factual stability, initial spatial residual mean, residual direction, component model area, main point cloud area, deformation retention strategy identifier, and prohibition compensation pre-identifier.

[0090] Compensation access verification is performed on the benchmark conflict unit: when the model benchmark credibility is not lower than the preset benchmark threshold, the point cloud fact stability is lower than the preset fact threshold, and the corresponding area is not marked as a deformation retention area by the conflict handling strategy table, on-site three-dimensional point cloud spatial deformation error compensation is allowed; when the corresponding area has a deformation retention strategy identifier, a prohibited compensation pre-identifier, or an undeterminable verification strategy, point cloud compensation is prohibited.

[0091] Point cloud spatial deformation error compensation uses the component model patch of the reliable component unit as the compensation benchmark. It reads the distance residual and residual direction from each valid point in the main point cloud patch to the component model patch and generates a point-level compensation vector. The compensation action only applies to the benchmark conflict unit assigned the point cloud compensation strategy and does not cross the deformation preservation strategy identification area, the undecidable verification unit, or the component unit corresponding to the model update pending verification state.

[0092]

[0093] in, This represents the point cloud points after conversion to the project coordinate system before compensation. This represents the point cloud points after compensation. This represents the compensation ratio coefficient of the j-th component element. This represents the distance from a point in the point cloud to the component model patch of the j-th component element. This represents the component model patch of the j-th component unit. This represents the unit vector of the compensation direction for the j-th component element.

[0094] To prevent overcompensation, a compensation boundary is established before point cloud compensation. This boundary is jointly defined by the geometric boundaries of the reference conflict elements, the relationships between adjacent elements, the deformation preservation strategy identifier, and the pre-marker for prohibiting compensation. Compensation is performed when a point cloud point is within the compensation boundary; otherwise, it is not performed when the point cloud point is outside the boundary, falls into the pre-marker area for prohibiting compensation, or is in a state of insufficient corresponding point cloud area. For point cloud points close to the compensation boundary, a compensation ratio coefficient is set based on their distance from the boundary, ensuring that the compensated displacement continuously attenuates at the boundary.

[0095]

[0096] in, This represents the compensation ratio coefficient of the j-th component element. This represents the distance from a point in the point cloud to the compensation boundary. This indicates the preset boundary attenuation distance; when the point cloud points are located within the pre-marked area where compensation is prohibited, The value is 0.

[0097] This generates a 3D point cloud of the site after compensation, and records the coordinates before compensation, the coordinates after compensation, the compensation direction, the compensation displacement, and the compensation boundary, which serve as inputs for the S510 to generate a spatial consistency model.

[0098] S430. Generate Real Spatial Deformation Preservation Units: Read the conflict handling strategy table. For the baseline conflict units assigned with deformation preservation strategies, extract the component identifier, deformation core location, residual direction, initial spatial residual mean, residual propagation consistency value, point cloud boundary anomaly propagation identifier, adjacent component relationship closure status, and preliminary conflict source judgment result. If the baseline conflict unit is determined to be a real spatial deformation, and the point cloud factual stability, model baseline credibility, and residual propagation consistency value all meet the corresponding threshold requirements, then the engineering information baseline model reverse update will not be performed on it, nor will on-site 3D point cloud spatial deformation error compensation be performed. Instead, a real spatial deformation preservation unit will be generated with the baseline conflict unit as the core.

[0099] The boundary of the real-space deformation preservation unit starts from the deformation core location and extends to adjacent components along a transmission path that is consistent with the residual direction and has a closed relationship between adjacent components. The expansion conditions include that the residual transmission consistency value is not lower than a preset transmission consistency threshold, the difference between the mean values ​​of the initial spatial residuals of adjacent components does not exceed a preset residual continuity threshold, and there is an abnormal transmission indicator at the point cloud boundary. Adjacent component units that meet the expansion conditions are written into the same real-space deformation preservation unit, while adjacent component units that do not meet the expansion conditions are not included in the preservation unit.

[0100]

[0101] in, This represents the j-th real-space deformation-preserving unit. This represents the j-th reference conflict unit. This represents an adjacent component element that has an adjacent transmission relationship with the j-th reference conflict element. Indicates the transitive consistency value of the residual. This indicates a preset transmission consistency threshold. This represents the mean of the initial spatial residuals for the j-th reference conflict cell. This represents the mean of the initial spatial residuals of adjacent component elements. This indicates the preset residual continuity threshold.

[0102] After generating realistic spatial deformation-preserving units, a "compensation-prohibited" flag is written to each unit to prevent subsequent spatial consistency fusion and engineering information platform calls from covering this area. A realistic spatial deformation-preserving unit includes at least the component identifier, core deformation location, deformation extension range, residual direction, residual amplitude, residual propagation consistency value, adjacent component propagation path, compensation-prohibited flag, and deformation preservation reliability. The deformation preservation reliability is determined comprehensively by the point cloud factual stability, model baseline reliability, residual direction consistency rate, and residual propagation consistency value. The realistic spatial deformation-preserving unit serves as input for the S510-generated spatial consistency model.

[0103] S5 specifically includes the following sub-steps: S510, Generate Spatial Consistency Model: Read the updated engineering information benchmark model generated by S410, the compensated on-site 3D point cloud generated by S420, the real spatial deformation preservation unit generated by S430, and the conflict handling strategy table generated by S330, and perform 3D modeling fusion processing under the same engineering project coordinate system.

[0104] During fusion, spatially overlapping areas are handled according to a fixed priority: real spatial deformation preservation units have the highest priority. Any component units or point cloud areas marked with a prohibition on compensation must not be covered by the compensated on-site 3D point cloud, nor can they be replaced by the updated engineering information benchmark model. The updated engineering information benchmark model that has been verified to be valid after connection is used as the model benchmark layer to carry the confirmed reverse update results of the model.

[0105] The compensated 3D point cloud serves as the point cloud observation layer, carrying the field observation results for which spatial deformation error compensation has been completed. Model updates awaiting verification, undeterminable verification units, and areas requiring additional data collection are written into the verification layer, but do not participate in spatial consistency replacement. This fusion priority avoids the same component unit being simultaneously subjected to model reverse updates, point cloud compensation, and deformation retention, thus preventing redundant processing.

[0106] The spatial consistency evaluation value is calculated for the fused component units. The spatial consistency evaluation value is determined by the component spatial residual, the connection node deviation, the closure state of the relationship between adjacent components, and the deformation preservation integrity. Among them, the component spatial residual is used to characterize the distance deviation between the fused component model area and the corresponding point cloud area, the connection node deviation is used to characterize the coordinate deviation of the connection nodes of adjacent components after fusion, the closure state of the relationship between adjacent components is used to characterize whether the connection object, connection type and connection node position remain closed, and the deformation preservation integrity is used to characterize whether the real space deformation preservation unit is completely preserved.

[0107]

[0108] in, This represents the spatial consistency evaluation value of the j-th component unit. This represents the normalized component spatial residual. This represents the normalized connection node bias. This represents the evaluation value for the closure of the relationship between adjacent components. This represents the evaluation value for the preservation of integrity in real-space deformation. , , and These represent the weights of the corresponding evaluation items, and .

[0109] When the spatial consistency evaluation value is lower than the preset consistency threshold, the corresponding component unit is written to the spatial consistency pending review status. This generates a spatial consistency model, which includes component identification, model baseline layer, point cloud observation layer, deformation preservation layer, and review layer, and serves as the input for S520 to generate baseline conflict handling records.

[0110] S520. Generate baseline conflict resolution records: Based on the spatial consistency model and conflict resolution strategy table, generate baseline conflict resolution records for each baseline conflict unit according to the component identifier. For component units implementing the model reverse update strategy, record the component identifier, model coordinates before update, model coordinates after update, model reverse update displacement, referenced acquisition period, conflict source, connection verification result, and model update pending review status; for component units implementing the point cloud compensation strategy, record the point cloud coordinates before compensation, point cloud coordinates after compensation, compensation direction, compensation displacement, compensation scale coefficient, and compensation boundary.

[0111] For component units implementing the deformation retention strategy, record the core deformation location, deformation extension range, residual propagation consistency value, adjacent component propagation path, prohibited compensation flag, and deformation retention reliability; for component units implementing the undeterminable verification strategy, record the undeterminable reason, the area requiring supplementary sampling, the required coordinate verification sampling period, and the verification triggering condition.

[0112] The above record fields correspond to the outputs of S410, S420 and S430 one by one, which is used to ensure that every model update, point cloud compensation and deformation preservation is traceable.

[0113] Furthermore, a credibility identifier is generated based on the credibility of the source of the conflict, the spatial consistency evaluation value, the integrity of the strategy execution, and the hard constraint interception result. The credibility of the source of the conflict is determined by the consistency between the initial judgment result of the conflict source and the conflict resolution strategy; the integrity of the strategy execution is determined by whether the corresponding field records are completed for model updates, point cloud compensation, or deformation preservation; the hard constraint interception result is used to reflect whether the component unit is in a state of strategy rewriting, model update pending verification, insufficient point cloud area, or spatial consistency pending verification.

[0114]

[0115] in, This represents the reliability evaluation value of the j-th component unit. Indicates the credibility of the source of the disposal. Indicates the spatial consistency evaluation value. This represents the evaluation value for the completeness of strategy execution. This indicates the deduction value for hard constraint interception or review status. , , and These represent the weights of the corresponding evaluation items.

[0116] When the credibility evaluation value reaches the preset high credibility threshold, a high credibility identifier is generated; when the credibility evaluation value is lower than the preset high credibility threshold but not lower than the preset usable threshold, a usable but requiring verification identifier is generated; when the credibility evaluation value is lower than the preset usable threshold, a non-directly callable identifier is generated. The resulting benchmark conflict handling record serves as the basis for the S530 to output the engineering information compensation results.

[0117] S530. Output Engineering Information Compensation Results: Encapsulate the spatial consistency model and benchmark conflict resolution records into engineering information compensation results and output them to the engineering information platform. The engineering information compensation results shall include at least the spatial consistency model file, component index table, benchmark conflict resolution records, reliability index table, deformation retention layer, compensation displacement layer, and review task list.

[0118] Among them, the spatial consistency model file is used to carry the fused 3D modeling results, the component index table is used to call the corresponding component unit according to the component identifier, the confidence index table is used to filter the component units that can be directly called, need to be reviewed, and cannot be directly called according to the confidence identifier, the deformation preservation layer is used to display the prohibited compensation area and the transfer path of adjacent components, the compensation displacement layer is used to display the point cloud compensation direction and compensation displacement, and the review task list is used to prompt the undetermined review unit, the model update pending review status, the spatial consistency pending review status, and the area that needs to be re-sampled.

[0119] For component units that execute the reverse update strategy of the model, the model state before the update and the model state after the update are retained at the same time. When the corresponding component unit in the review task list is confirmed to have data acquisition anomalies, coordinate anomalies, connection verification failures, or inconsistencies in external engineering records, the engineering information platform rolls the component unit back to the model state before the update based on the benchmark conflict handling record and re-marks it as an undeterminable review unit.

[0120] For component units implementing a point cloud compensation strategy, the platform retains the point cloud coordinates before and after compensation, enabling the compensated displacement to be verified. For component units implementing a deformation preservation strategy, the platform retains a compensation prohibition flag, ensuring that subsequent 3D modeling, as-built verification, operation and maintenance monitoring, and spatial deformation analysis do not cover the deformation area. This completes the output, retrieval, verification, and rollback of 3D point cloud spatial deformation error compensation results.

[0121] In a specific engineering application scenario (such as beam-column joint deformation compensation in a large steel structure factory building): The weight parameters and preset thresholds in each formula were fitted with historical engineering data to establish a set of optimal values. These values ​​were then used to calculate the actual stability of the point cloud. At that time, considering the decisive role of the recurrence rate of point clouds, we set... ; In calculating the baseline credibility of the model At that time, set .

[0122] The preferred engineering constraint threshold is: a preset distance threshold. Values Preset angle threshold Values ; In the normalized residual determination, when the component type is a steel column, the component residual determination threshold is... Preferred When the component type is a connecting flange, Preferred This solution aims to meet the high-precision assembly requirements of different functional components. By assigning specific numerical parameters, this solution can be directly deployed and effectively implemented to determine and compensate for the actual spatial deformation of plant nodes.

[0123] All the above formulas are performed using dimensionless numerical calculations; the relevant formulas are based on empirical models that approximate the real situation, obtained through extensive data collection and software simulation fitting. The preset parameters and thresholds involved in the formulas can be conventionally set and adjusted by those skilled in the art according to the physical constraints of the actual application scenario.

[0124] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0125] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0126] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for compensating for spatial deformation errors in three-dimensional point clouds for engineering informatization, characterized in that, Includes the following steps: S1. Read the engineering information benchmark model, multi-phase on-site 3D point cloud and external engineering records of the target engineering object, perform coordinate format unification and validity verification on the multi-phase on-site 3D point cloud, generate target engineering object data package, and generate component unit set, record association table and initial spatial correspondence table. S2. Calculate the point cloud recurrence rate, residual amplitude fluctuation, residual direction consistency rate and point cloud boundary continuity evaluation value based on the initial spatial correspondence table, generate a point cloud fact feature table, generate a model support feature table based on the record association table, and fuse them to generate a benchmark confidence matrix. S3. Identify the set of benchmark conflict units and the set of undeterminable verification units based on the benchmark credibility matrix, make a preliminary judgment on the source of conflict for the set of benchmark conflict units, and generate a conflict handling strategy table by combining a reinforcement learning strategy model with hard constraint interception. S4. Based on the conflict resolution strategy table, perform model reverse update, point cloud compensation and deformation preservation on the corresponding benchmark conflict units respectively, and generate the updated engineering information benchmark model, the compensated on-site 3D point cloud and the real space deformation preservation unit.

2. The method for compensating for spatial deformation errors of three-dimensional point clouds for engineering informatization according to claim 1, characterized in that, Also includes: S5. Based on the updated engineering information benchmark model, the compensated on-site 3D point cloud and the real spatial deformation preservation unit, perform 3D modeling and fusion to generate a spatial consistency model and benchmark conflict handling record, and output the engineering information compensation result.

3. The method for compensating for spatial deformation errors of three-dimensional point clouds for engineering informatization according to claim 1, characterized in that, S1 specifically includes: Read the engineering information benchmark model, multi-phase on-site 3D point cloud and external engineering records of the target project object, unify the coordinate format, time mark, collection source mark and validity verification of the multi-phase on-site 3D point cloud, and generate the target project object data package; Based on the target engineering object data package, component units are divided, external engineering records and component adjacency relationship data are bound, point cloud areas are extracted, and component unit sets and record association tables are generated. Based on the component unit set and record association table, perform 3D modeling processing, establish the initial spatial correspondence between component model areas, point cloud areas and adjacent components, calculate the initial spatial residual, and generate the initial spatial correspondence table.

4. The method for compensating for spatial deformation errors of three-dimensional point clouds for engineering informatization according to claim 1, characterized in that, S2 specifically includes: Read the initial spatial correspondence table, extract the main point cloud area, initial spatial residual, residual direction and point cloud boundary of each collection period according to the component identifier, calculate the point cloud recurrence rate, residual amplitude fluctuation, residual direction consistency rate and point cloud boundary continuity evaluation value, and generate a point cloud fact feature table. Read the record association table and the initial spatial correspondence table, and verify the integrity of the records, the consistency of the record time, the consistency of the component number, the support of geometric changes, and the closure of the relationship between adjacent components for each component unit, and generate the model support feature table.

5. The method for compensating for spatial deformation errors of three-dimensional point clouds for engineering informatization according to claim 4, characterized in that, Also includes: Based on the component identifier fusion point cloud fact feature table and model support feature table, the point cloud fact stability and model baseline confidence are calculated, and a baseline confidence matrix is ​​generated.

6. The method for compensating for spatial deformation errors of three-dimensional point clouds for engineering informatization according to claim 1, characterized in that, S3 specifically includes: Read the baseline confidence matrix, extract the point cloud fact stability, model baseline confidence, initial spatial residual mean, residual direction consistency rate and status according to component identifier, generate component residual judgment threshold based on component type, component scale, acquisition accuracy and component functional attributes, and identify the baseline conflict unit set and the undecidable verification unit set; Based on the benchmark conflict unit set and benchmark confidence matrix, combined with the recorded conflict identifier, point cloud boundary anomaly propagation identifier and adjacent component relationship closure status, a preliminary judgment result of the conflict source is generated. The initial judgment of the conflict source and the set of benchmark conflict units are input into the reinforcement learning strategy model, and a conflict handling strategy table is generated by combining hard constraint interception.

7. The method for compensating for spatial deformation errors of three-dimensional point clouds for engineering informatization according to claim 1, characterized in that, S4 specifically includes: Read the conflict handling strategy table, perform update access verification on the benchmark conflict unit of the assigned model reverse update strategy, form a fact benchmark based on the stable main point cloud area, update the component geometric boundary, positioning parameters and adjacent component connection relationship of the engineering information benchmark model, and generate the updated engineering information benchmark model. Compensation access verification is performed on the benchmark conflict unit of the assigned point cloud compensation strategy. The component model area of ​​the trusted component unit is used as the compensation benchmark. Point-level compensation vectors are generated for the main point cloud area, and the compensated on-site 3D point cloud is output.

8. The method for compensating for spatial deformation errors of three-dimensional point clouds for engineering informatization according to claim 7, characterized in that, Also includes: For the baseline conflicting elements assigned to the deformation preservation strategy, real spatial deformation preservation elements are generated based on the residual transfer consistency value and the closed state of the relationship between adjacent components.

9. The method for compensating for spatial deformation errors of three-dimensional point clouds for engineering informatization according to claim 2, characterized in that, S5 specifically includes: Read the updated engineering information benchmark model, the compensated on-site 3D point cloud, the real space deformation preservation unit and the conflict handling strategy table, perform 3D modeling fusion according to the fusion priority, calculate the spatial consistency evaluation value, and generate a spatial consistency model. Based on the spatial consistency model and conflict resolution strategy table, a baseline conflict resolution record is generated according to the component identifier, and a credibility identifier is generated based on the credibility of the resolution source, the spatial consistency evaluation value, the integrity of strategy execution, and the hard constraint interception result.

10. The method for compensating for spatial deformation errors of three-dimensional point clouds for engineering informatization according to claim 9, characterized in that, Also includes: The spatial consistency model and benchmark conflict handling records are encapsulated into engineering information compensation results, and the output includes a component index table, a credibility index table, a compensation displacement layer, and a review task list.

Images