Prediction method and prediction equipment for building deformation point cloud time sequence

Abstract

The invention provides a prediction method and prediction equipment for a building deformation point cloud time sequence, and relates to the technical field of deformation monitoring. The method comprises the following steps: analyzing an aligned point cloud set to obtain a point cloud set of an abnormal region; based on the multi-resolution features of the point cloud set of the abnormal region, extracting topological features of a preset connection structure of the abnormal region, and constructing a topological graph of the abnormal region; carrying out stress constraint and material characteristic constraint on the topological graph of the abnormal region, and determining submerged space characteristics of the topological graph; based on a latent diffusion model, performing interpolation and extrapolation processing on the latent space features, and determining a prediction vector field and a suspected damage area of the latent space features; performing multi-scale regression analysis on the suspected damage area to obtain local parameters; and carrying out fusion and consistency evaluation on the three-dimensional deformation prediction field corresponding to the prediction vector field of the submerged space features and the local parameters to obtain a fused global deformation prediction field, and obtaining high-risk deformation distribution. According to the method, deformation can be accurately predicted.

Description

Technical Field

[0001] The present invention relates to the technical field of building deformation monitoring, and particularly relates to a prediction method and prediction device for building deformation point cloud time series. Background Art

[0002] With the continuous acceleration of the urbanization process and the continuous increase of large and complex buildings, the safety and stability of buildings have received extensive attention. Due to its high precision, high efficiency and non-contact characteristics, three-dimensional laser scanning technology is applied to building deformation monitoring.

[0003] Three-dimensional laser scanning technology usually first collects the spatial information of the building surface through hardware devices such as lidar, structured light or photogrammetry, and then through feature extraction and coordinate calibration, realizes the alignment and data stitching of point clouds in different measurement time periods. Subsequently, registration methods based on the iterative closest point algorithm or graph optimization can be used to ensure that the multi-period scanning results maintain high precision and comparability under a unified benchmark.

[0004] However, although the current three-dimensional laser scanning technology has the capabilities of registration, noise reduction and structural crack identification, it is often limited to static or single-dimensional analysis of geometric information. When facing the large-scale data generated by multi-period acquisition, these methods are difficult to simultaneously characterize the material properties, stress field distribution and microscopic crack characteristics during the alignment process. In addition, in order to obtain the dynamic trend of deformation, most of the existing methods lack hierarchical modeling for stress concentration areas during regression or time series prediction, and cannot achieve flexible connection between local and global, resulting in inaccurate prediction. Summary of the Invention

[0005] Embodiments of the present invention provide a prediction method and prediction device for building deformation point cloud time series to solve the problem that the current prediction method for deformation point cloud cannot accurately predict.

[0006] In a first aspect, embodiments of the present invention provide a prediction method for building deformation point cloud time series, including: Performing coordinate system transformation and noise reduction processing on the acquired original point cloud data of the target building to obtain an aligned point cloud set; the original point cloud data includes point cloud data collected in multiple time periods and / or by multiple devices; Performing local difference degree analysis on the aligned point cloud set to obtain a point cloud set of the abnormal area; Based on the multi-resolution features of the point cloud set of the abnormal area, extracting the topological features of the preset connection structure of the abnormal area and constructing a topological graph of the abnormal area; performing stress constraint and material property constraint on the topological graph of the abnormal area to determine the latent space features of the topological graph of the abnormal area; Based on the latent diffusion model, perform interpolation and extrapolation processing on the latent space features to determine the prediction vector field of the latent space features and the suspected damage area; Perform multi-scale regression analysis on the suspected damage area to obtain the local deformation prediction value and the local uncertainty distribution; Fuse and perform consistency evaluation on the three-dimensional deformation prediction field corresponding to the prediction vector field of the latent space features, the local deformation prediction value, and the local uncertainty distribution to obtain the fused global deformation prediction field; Analyze the fused global deformation prediction field to obtain the high-risk deformation distribution.

[0007] In a possible implementation, based on the multi-resolution features of the point cloud set of the abnormal area, extract the topological features of the preset connection structure of the abnormal area and construct the topological graph of the abnormal area, including: Extract the multi-resolution features of the point cloud set of the abnormal area based on the multi-resolution encoder; among them, the multi-resolution encoder performs downsampling and feature extraction on the multi-resolution of the point cloud set of the abnormal area by introducing gradient vectors in the multi-resolution encoder; Extract the topological features of the preset connection structure of the abnormal area based on the topological feature extraction module in the multi-resolution encoder to obtain the topological features of the preset connection structure of the abnormal area; where the preset connection structure includes beams, columns, trusses, and facades of the target building; Construct the topological graph of the abnormal area based on the topological features of the preset connection structure of the abnormal area.

[0008] In a possible implementation, perform stress constraint and material property constraint on the topological graph of the abnormal area to determine the latent space features of the topological graph of the abnormal area, including: Perform stress constraint on the diffusion process of the latent space field corresponding to the topological graph of the abnormal area to obtain the first latent space feature; Based on the material property database, add material property constraints to each diffusion process of the latent space field corresponding to the topological graph of the abnormal area to obtain the second latent space feature; Based on the deformation amplitude and gradient, constrain the deformation area of the latent space field corresponding to the topological graph of the abnormal area to obtain the third latent space feature; Based on the first latent space feature, the second latent space feature, and the third latent space feature, determine the latent space features of the topological graph of the abnormal area.

[0009] In a possible implementation, based on the latent diffusion model, perform interpolation and extrapolation processing on the latent space features to determine the prediction vector field of the latent space features and the suspected damage area, including: Based on the trained latent diffusion model, as well as the stress constraint and material property constraint in the latent space, perform interpolation processing on adjacent time segments in the vector field of latent space features, and perform extrapolation processing on the vector field of latent space features to obtain the predicted deformation vector field of latent space features; Calculate the deformation intensity of the topological features of the preset connection structure, and determine the suspected damage area based on the preset deformation threshold and the predicted deformation vector field of latent space features.

[0010] In a possible implementation, perform multi-scale regression analysis on the suspected damage area to obtain local deformation prediction values and local uncertainty distributions, including: Based on the positioning information of the point set in the suspected damage area and the spatial filtering operator, screen high-risk suspected areas; the positioning information of the point set includes the coordinates of each point and the predicted deformation vector field of each point; Based on the pre-constructed composite kernel function, perform multi-scale regression analysis on each scale level in the high-risk suspected area to obtain local deformation prediction values and local uncertainty distributions; the composite kernel function includes geometric correlation, material property similarity kernel function, and topological coupling degree of the preset connection structure.

[0011] In a possible implementation, based on the pre-constructed composite kernel function, perform multi-scale regression analysis on each scale level in the high-risk suspected area to obtain local deformation prediction values and local uncertainty distributions, including: Based on the pre-constructed composite kernel function, perform multi-scale regression analysis on each scale level in the high-risk suspected area to obtain the first regression result; Dynamically update the preset weights in the composite kernel function to obtain the final weights of the composite kernel function; among them, a preset weight is set for each of the geometric correlation, material property similarity kernel function, and topological coupling degree of the preset connection structure in the composite kernel function; Based on the final weights of the composite kernel function, perform multi-scale regression analysis on each scale level in the high-risk suspected area to obtain the second regression result; Integrate the first regression result and the second regression result to obtain local deformation prediction values and local uncertainty distributions.

[0012] In a possible implementation, fuse and perform consistency evaluation on the three-dimensional deformation prediction field corresponding to the predicted vector field of latent space features, local deformation prediction values, and local uncertainty distributions to obtain the fused global prediction field, including: Based on the decoder, map the predicted vector field of latent space features to the physical coordinate system to obtain the three-dimensional deformation prediction field of latent space features; Based on the predicted vector field of latent space features and the suspected damage area, construct a damage map; Based on the three-dimensional deformation prediction field and the damage map, construct a global prediction field of latent space features; Determine the global skeleton of the global prediction field; wherein, the function of the global skeleton is determined by the double integral of the relevance weight of the target position and the first deformation difference, and the first deformation difference is the difference between the three-dimensional deformation prediction field and the preset average benchmark; Construct the weights of the local parameters and the global skeleton, and fuse the local parameters and the global skeleton to obtain the fused prediction field; wherein, the local parameters include the local deformation prediction value and the local uncertainty distribution; Based on the differences between the local parameters and the global skeleton in the high-risk suspected area, and the differences between the gradient of the fused prediction field and the gradient of the global skeleton, verify the fused prediction field to obtain the fused global deformation prediction field and the damage distribution.

[0013] In a possible implementation, analyze the fused global deformation prediction field to obtain the high-risk deformation distribution, including: Based on the fused global deformation prediction field and the high-risk suspected area, determine the deformation degree and uncertainty in the high-risk suspected area; Input the deformation degree and uncertainty in the high-risk suspected area, and the high-risk suspected area into the weight update module of the latent diffusion model for iterative processing to obtain the updated weights of the latent diffusion model; Based on the composite kernel function, perform multi-scale regression analysis on the high-risk suspected area to obtain the high-risk deformation distribution.

[0014] In a possible implementation, perform local difference degree analysis on the aligned point cloud set to obtain the point cloud set of the abnormal area, including: Based on the pre-constructed abnormal evaluation model, determine the intensity of each point cloud in the aligned point cloud set; Based on the intensities of all point clouds in the aligned point cloud set, determine the point cloud set of the abnormal area.

[0015] In a second aspect, an embodiment of the present invention provides a prediction device, including a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the method in the first aspect or any possible implementation manner of the first aspect above is implemented.

[0016] In the embodiments of the present invention, in order to comprehensively consider the influence of various factors on the deformation of a building, first, the original point cloud data of the target building obtained is subjected to coordinate system transformation and noise reduction processing to obtain an aligned point cloud set. After obtaining the aligned point cloud set, in order to narrow the research area, it is also necessary to perform a local difference degree analysis on the aligned point cloud set to obtain the point cloud set of the abnormal area, so that the point cloud set of the abnormal area can be studied more accurately. Then, based on the multi-resolution features of the point cloud set of the abnormal area, the topological features of the preset connection structure of the abnormal area are extracted, and a topological map of the abnormal area is constructed. Stress constraints and material property constraints are imposed on the topological map of the abnormal area to determine the latent space features of the topological map of the abnormal area, so that the local details of stress concentration in the preset connection structure can be captured. At the same time, by combining the stress constraints and the constraints of material properties, the constraints of materials and mechanics can be reflected in the diffusion process, and the deformation can be predicted more accurately. Next, based on the latent diffusion model, interpolation and extrapolation processing are performed on the latent space features to determine the prediction vector field and the suspected damage area of the latent space features, so that areas with potential damage possibilities can be captured by combining stress and material information at the latent space level. After determining the suspected damage area, multi-scale regression analysis is performed on the suspected damage area to obtain the local deformation prediction value and the local uncertainty distribution, so as to provide a refined compensation basis for the subsequent fusion with the global skeleton. In order to finally form a global deformation field and record the damage distribution in a unified framework, therefore, it is also necessary to fuse and perform consistency evaluation on the three-dimensional deformation prediction field corresponding to the prediction vector field of the latent space features, the local deformation prediction value, and the local uncertainty distribution to obtain the fused global deformation prediction field. Finally, the fused global deformation prediction field is analyzed to obtain the high-risk deformation distribution. Thus, a more accurate high-risk deformation distribution can be formed in the time series continuity. During the prediction process, the linked analysis of stress distribution, material properties, and local structural differences is realized, especially the comprehensive influence of fine cracks, complex component connections, and multiple materials, so as to improve the connection between global and local information for more accurate deformation prediction in the time series dimension. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is the implementation flowchart of the prediction method for the time series of building deformation point clouds provided by the embodiments of the present invention; Figure 2 is the structural schematic diagram of the prediction device for the time series of building deformation point clouds provided by the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The embodiments of the present invention will be described in detail below with reference to the drawings.

[0019] The inventors have found that although the current methods for three-dimensional point cloud monitoring and data fusion have the capabilities of registration, noise reduction, and structural crack identification, they are often limited to static or single-dimensional analysis of geometric information. When faced with large-scale data generated from multi-period acquisitions, these methods are difficult to simultaneously represent material properties, stress field distribution, and microscopic crack characteristics during the alignment process. Existing research usually focuses on data cleaning or macroscopic deformation detection using traditional filtering and matching algorithms, and does not achieve dynamic tracking of local refined structures and material differences at a deeper level. For complex building components, such as beam-column connections and special-shaped facades, different materials may exhibit inconsistent deformation behaviors. However, existing methods usually rely on additional offline simulations or empirical parameters to determine these local differences and cannot perform timely and unified processing during multi-period monitoring and prediction. At the same time, in order to obtain the dynamic trend of deformation, most existing methods lack hierarchical modeling for stress concentration regions during regression or time series prediction and are also difficult to achieve flexible connection between local and global levels. This results in insufficient insight into potential crack evolution and stress distribution under heterogeneous materials.

[0020] To solve the above problems, the present invention provides a method and device for predicting the time series of building deformation point clouds.

[0021] See Figure 1 , which shows the implementation flowchart of the method for predicting the time series of building deformation point clouds provided by the embodiments of the present invention, and is described in detail as follows: S110. Perform coordinate system transformation and noise reduction processing on the acquired original point cloud data of the target building to obtain an aligned point cloud set.

[0022] Among them, the original point cloud data includes point cloud data collected in multiple periods and by multiple devices.

[0023] In some embodiments, coordinate system transformation and noise reduction processing can be performed based on environmental parameters and the hardware attributes of each device.

[0024] S1110. Perform quantitative analysis on noise and density distribution based on the recorded environmental parameters and the hardware attributes of each device.

[0025] Define the sampling data matrix , represents the original scan value corresponding to the time index and the device index . Construct the function , which is used to characterize the cumulative noise difference between different time periods and and under the same device ; Where, Indicates at the time index under the device index k, the corresponding coordinates at the scanning intensity value, and respectively represent the index ranges of the point cloud in the horizontal and vertical directions. According to the peak distribution law of and the environmental parameters in the measurement scene, the noise characteristics of the target area are predicted.

[0026] S1120. Perform coordinate system transformation For the original point cloud coordinate set corresponding to the device index , let the rotation matrix and the translation vector and . On the basis of the initial external parameter calibration, perform the conversion between the hybrid coordinate system and the standard coordinate system: ; where represents the corrected coordinate vector in the standard coordinate system. By retaining the internal and external parameters of the sensor, the subsequent batch correction link can achieve multi-source data synchronization based on the same benchmark.

[0027] S1130. Based on the building structure layout of the target building, select several stable and easily recognizable reference points as key reference points for precise positioning during subsequent time series alignment. Store each key reference point as , where r is the reference point number. Map the coordinate information of these key reference points to the unified coordinate system obtained in step S1120.

[0028] S1140. Combine the result obtained in step S1110 distribution, and record the result of this distribution as the noise threshold , define the judgment function , and monitor the distance value between each three-dimensional point and the center of its neighborhood. If it exceeds the noise threshold , then eliminate this point, and count the finally remaining point cloud: ; where represents the centroid vector of the points in the neighborhood.

[0029] By executing the operation batch by batch, this step continuously reduces the abnormal range and reduces the misjudgment of normal structure points, obtaining a point cloud with less noise.

[0030] S1150. Downsample the finally remaining point cloud through a unified step size and scale factor . Let the finally remaining point cloud be , through the function to obtain , enabling multi-temporal point clouds to be characterized within the same resolution scale, reducing the mismatch between different densities and taking into account the computational efficiency of the subsequent latent space model.

[0031] S1160. Perform registration for each time period.

[0032] Given adjacent time periods and the corresponding point cloud sets and , find the rotation matrix and the translation vector , minimizing the sum of the squares of the Euclidean distances between corresponding points: ; where and respectively represent the corresponding point coordinates in time period and time period .

[0033] Thus, based on the results output from the above steps, an aligned multi-temporal point cloud set can be formed.

[0034] S120. Perform local difference analysis on the aligned multi-temporal point cloud set to obtain the point cloud set of the abnormal region.

[0035] In order to provide a candidate region for subsequent processing and correction and be able to process abnormal point clouds more targeted, the present invention performs local difference analysis on the point cloud sequence, so as to quickly lock the abnormal region.

[0036] In some embodiments, first, based on a pre-constructed abnormal evaluation model, determine the intensity of each point cloud in the aligned point cloud set. Among them, the abnormal evaluation model is constructed based on the difference between the gray scale or intensity information of each point cloud and a preset average intensity value. Then, based on the intensities of all point clouds in the aligned point cloud set, determine the point cloud set of the abnormal region.

[0037] In this embodiment, the aligned multi-temporal point cloud set is denoted as , and perform local difference analysis on each . Suppose at time index , the coordinate index of is , and its gray scale or intensity information is .

[0038] The constructed abnormal evaluation model can quantify the noise or deformation abnormal distribution in each point cloud frame: ; Among them, represents the integration domain of the visible region in the two-dimensional index and represents the average intensity value of Xᵗ in this region. By calculating under different , the noise concentration area and the candidate area with a large displacement gradient can be located. The larger is, the more significant the noise or geometric change is under this time slice. By comparing corresponding to different t, the "abnormal" area can be quickly locked.

[0039] S130. Based on the multi-resolution features of the point cloud set in the abnormal area, extract the topological features of the preset connection structure in the abnormal area, and construct a topological graph of the abnormal area.

[0040] In order to be able to retain both the overall outline of the target building and extract the local details of the preset connection structure of key parts such as beams and columns, an encoder is used for separate multi-resolution feature extraction.

[0041] In some embodiments, first, based on the multi-resolution encoder, the multi-resolution of the point cloud set in the abnormal area is extracted to obtain the multi-resolution features of the point cloud set in the abnormal area. Among them, the multi-resolution encoder is based on introducing a gradient vector in the multi-resolution encoder to perform downsampling and feature extraction on the multi-resolution of the point cloud set in the abnormal area.

[0042] Then, based on the topological feature extraction module in the multi-resolution encoder, the topological features of the preset connection structure in the abnormal area are extracted to obtain the topological features of the preset connection structure in the abnormal area. Among them, the preset connection structure includes beams, columns, trusses, and facades of the target building.

[0043] Finally, based on the topological features of the preset connection structure in the abnormal area, a topological graph of the abnormal area is constructed.

[0044] In this embodiment, after determining the abnormal area, it is necessary to characterize the point cloud at multiple scales to take into account both the overall structural form and local details.

[0045] Let the encoder be denoted as , its input is , and the output is , where represents the resolution level. To ensure geometric consistency at multiple scales, the vector is introduced to represent the feature representation at the corresponding position at the resolution , and adaptive downsampling and feature extraction are performed: ; Among them, represents the channel dimension at the resolution , represents the local geometric gradient vector for at the position . is the set of weights, is the bias term at the resolution .

[0046] In addition, in order to strengthen the topological information of the preset connection structure and the topological information of spatially hierarchical complex parts such as reinforced beams, columns, trusses, and facades, a topological feature extraction module is added, and its output is denoted as .

[0047] ; wherein, it is assumed that each local connection structure contains construction units, and the topological connection matrix of each construction unit is , corresponding to the node set , and cc is the multi-scale feature channel matrix of each node in the corresponding node set . Thus, accurate capture of local details and at the preset connection structure can be achieved.

[0048] Finally, construct the topological graph of the abnormal area .

[0049] S140. Apply stress constraints and material property constraints to the topological graph of the abnormal area to determine the latent space features of the topological graph of the abnormal area.

[0050] Since the structure of the target building is also subject to stress constraints and material property constraints, simply relying on geometric denoising may disrupt the originally satisfied mechanical balance. Therefore, this application explicitly incorporates the "stress field" into the iterative process of latent space denoising, so that the denoised deformation result is not only smooth but also consistent with the building mechanics.

[0051] In some embodiments, first, stress constraints can be applied to the diffusion process of the latent space field corresponding to the topological graph of the abnormal area to obtain the first latent space feature.

[0052] Then, based on the material property database, material property constraints are added to each diffusion process of the latent space field corresponding to the topological graph of the abnormal area to obtain the second latent space feature.

[0053] Next, based on the deformation amplitude and gradient, constraints are applied to the deformation area of the latent space field corresponding to the topological graph of the abnormal area to obtain the third latent space feature.

[0054] Finally, based on the first latent space feature, the second latent space feature, and the third latent space feature, determine the latent space feature of the topological map of the abnormal area.

[0055] In this embodiment, an energy functional can be constructed based on adding a gravitational field to the topological map of the abnormal area to perform stress constraint on the diffusion process. Define the stress field to represent the estimated stress magnitude at time , coordinate . To incorporate into the latent space diffusion, construct the energy functional to guide the latent space diffusion process, ; where is the volume region of the building target structure, represents the stress estimate value at the corresponding position after latent space mapping, is the weighting coefficient for adjusting the smoothness of the latent space and the matching degree of the stress field. By minimizing , can be made to be consistent with the true structural mechanics characteristics in the denoising iteration. On the basis of conventional geometric denoising, explicitly introduce the stress field as a physical prior, and guide the diffusion / denoising process of the latent space field by constructing and minimizing the energy functional. This variational framework of mechanical constraint + smooth regularization can make the final deformation field not only get rid of noise interference but also not violate the building mechanics characteristics, thus providing a more reliable basis for subsequent material, crack, and damage assessment.

[0056] In this embodiment, utilize the material property database to further distinguish the mechanical properties of different building materials in the latent space. Let the lookup result of the material type at coordinate be , and its parameter vector in is denoted as . The diffusion process introduces the material influence factor at each iteration step: ; where, represents the elastic modulus, represents the Poisson's ratio, and the remaining are other material mechanics parameters, is the adjustable coefficient.

[0057] By dynamically regulating the noise injection amount and the diffusion stable range during the latent space update process, reduce the interference of invalid white noise and highlight the deformation characteristics caused by material differences.

[0058] In this embodiment, the deformable or cracked regions are centrally processed to enhance the fusion of damage information during the diffusion denoising process. Let the set of deformable regions be denoted as , and its corresponding coordinates in space are denoted as ). To directionally amplify the deformation characteristics of these regions, a directional enhancement damage function

[0059] is constructed based on the deformation amplitude and gradient; where and respectively control the enhancement intensity of the deformation amplitude and local gradient. By incorporating into the optimization objective during the denoising iteration, the potential damage signs in the micro-cracks and stress concentration regions are revealed.

[0060] Finally, the latent space features of the topological map of the abnormal region not only contain the geometric multi-resolution features but also fuse the material and stress features. However, they are only distributed at discrete times t = 1,..., T, and there are still certain noises or discontinuities. Therefore, temporal interpolation is required for the latent space representation to further smooth, diffuse, or infer the deformation evolution trend.

[0061] S150. Based on the latent diffusion model, perform interpolation and extrapolation processing on the latent space features to determine the predicted vector field of the latent space features and the suspected damage regions.

[0062] In some embodiments, first, based on the trained latent diffusion model, as well as the stress constraint and material property constraint in the latent space, perform interpolation processing on adjacent time segments in the vector field of the latent space features, and perform extrapolation processing on the vector field of the latent space features to obtain the predicted deformation vector field of the latent space features.

[0063] Finally, calculate the deformation intensity of the topological features of the preset connection structure, and determine the suspected damage regions based on the preset deformation threshold and the predicted deformation vector field of the latent space features.

[0064] In this embodiment, the latent space features of the topological map of the abnormal region are denoted as , where represents the latent space vector field at the -th time slice. To accurately retain the key structural information during the temporal interpolation and subsequent extrapolation operations, a latent diffusion model network with a guiding mechanism is constructed, and the physical prior and material stratification information are embedded in the network weight initialization process. Denote the core parameter set of as , and define the initial guiding function as the network guiding term, which is combined with the above latent space features Obtained after superposition The initial parameters are determined by minimizing the following objective function : ; where represents the latent space coordinate domain, is the weighting coefficient for adjusting the smoothness of the model parameters.

[0065] In the initial training, the constraints of stress and material properties are incorporated, enabling the network to be more sensitive to vulnerable parts during subsequent interpolation and denoising. By introducing LDM as a "neural network model with physical / material priors", the dynamic evolution law of latent space representation is learned, making it particularly sensitive to structurally weak / vulnerable parts during interpolation and denoising.

[0066] After the training of LDM is completed, the network is applied to the interpolation of adjacent time slices. Given adjacent time slices , between and insert intermediate deformation frames to generate the interpolation sequence . Define the dynamic diffusion operator such that is combined with the guiding mechanism during the diffusion process: ; where is the latent space vector of the previous interpolation frame, represents the latent space vector of the next time slice, represents the physical field and noise suppression factor input at the th step of interpolation.

[0067] By gradually reducing the distortion of temporal unevenness through diffusion composite interpolation, the overall deformation trajectory of the target building is made continuous and smooth in the time dimension.

[0068] Next, at the current known time , perform deformation extrapolation for subsequent future times to support the early detection of potential anomalies. Define the extrapolation operation and estimate the latent space representation of time with reference to the generated interpolation sequence and mechanical constraint terms.

[0069] Let the extrapolation order be , and combine the historical multi-frame interpolation results , …, , with the structural material prior to obtain the following recurrence formula:

[0070] wherein represents the extrapolation coefficient, and respectively represent the stress constraint field and the material property field in the latent space, is the boundary of the building target area.

[0071] By jointly considering multi-frame interpolation and material mechanics hierarchical information, predict the potential deformation trend at time .

[0072] Thus, the predicted deformation vector field of the latent space features can be obtained based on the results of the above steps.

[0073] Next, it is also necessary to determine the suspected damage area based on the predicted deformation vector field of the latent space features, the deformation strength of the topological features of the preset connection structure, and the preset deformation threshold.

[0074] Based on the topological map of the abnormal area for deformation evaluation, representative component units can be selected, and the deformation strength is calculated by combining the topological features extracted in step S2.

[0075] Define the deformation strength measurement system : ; wherein represents the gradient vector at the position in the latent space under , represents the structural stress value, represents the material property factor is the weighting coefficient. By judging which components have a high degree of deformation, it is possible to focus on screening for damage at these high-strength parts.

[0076] Based on results and the preset deformation threshold, determine the suspected damage area exceeding the threshold .

[0077] Select the judgment function : ; wherein, (x) represents the deformation strength of the topological segment where it is located.

[0078] It represents that abnormal deformation has occurred at this location. Store such points in the suspected damage list, retaining the coordinate information and adjacent reference marks. In addition, an additional diffusion denoising cycle needs to be performed on the suspected damage area to suppress sudden noise errors.

[0079] Summarize all suspected damage areas and match them with the continuous time series deformation trajectories generated by the interpolation and extrapolation sequences to form a preliminary damage map. 。 At the time index Record the spatial coordinates and deformation amplitude of the suspected damage points. By comparing the position changes of the suspected area after multiple iterations across time slices, combine local anomalies with the overall deformation trend.

[0080] Finally, output the final interpolation and extrapolation sequences and the damage map. Maintain consistency in adjacent time periods and future time periods, and solidify the positions of suspected cracks or beam-column connections into the damage calibration list.

[0081] S160. Perform multi-scale regression analysis on the suspected damage area to obtain local deformation prediction values and local uncertainty distributions.

[0082] In some embodiments, first, based on the positioning information of the point set in the suspected damage area and the spatial filtering operator, screen high-risk suspected areas. Among them, the positioning information of the point set includes the coordinates of each point and the predicted deformation vector field of each point.

[0083] Then, based on the pre-constructed composite kernel function, perform multi-scale regression analysis on each scale level in the high-risk suspected area to obtain local deformation prediction values and local uncertainty distributions. The composite kernel function includes geometric correlation, material property similarity kernel function, and topological coupling degree of the preset connection structure.

[0084] In this embodiment, first, based on the pre-constructed composite kernel function, perform multi-scale regression analysis on each scale level in the high-risk suspected area to obtain the first regression result.

[0085] Then, dynamically update the preset weights in the composite kernel function to obtain the final weights of the composite kernel function. Among them, a preset weight is set for each of the geometric correlation, material property similarity kernel function, and topological coupling degree of the preset connection structure in the composite kernel function.

[0086] Next, based on the final weights of the composite kernel function, perform multi-scale regression analysis on each scale level in the high-risk suspected area to obtain the second regression result.

[0087] Finally, integrate the first regression result and the second regression result to obtain local deformation prediction values and local uncertainty distributions.

[0088] Specifically, the positioning information of the midpoint set in the suspected damage area forms a target data set , including coordinates and its predicted deformation vector field .

[0089] Define the target domain to represent the local area containing potential cracks or component defects, and filter out the high-risk suspected areas of the building structure through the following spatial filtering operator (·) : ; where judge whether the point belongs to the high-risk suspected area through the threshold . If it meets , then determine that the point is an input that needs to be focused on.

[0090] In addition, based on the traditional Gaussian process regression kernel function, combined with the geometric characteristics of the building and material differences, the present invention proposes a composite kernel function . Construct the spatial proximity, topological coupling degree, and material properties, namely geometric correlation degree, material characteristic similarity kernel function, and topological coupling degree of the preset connection structure, into a comprehensive metric: ; where represents the geometric correlation degree based on the Euclidean distance, represents the material characteristic similarity kernel function between material parameters, is the material characteristic vector at the position, represents the topological coupling degree of local preset connection structures such as beam-column joints or facade connections, is an adjustable weighting coefficient.

[0091] Perform multi-scale regression on the high-risk suspected areas, and gradually fit the regional deformation from the macroscopic scale to the crack details. Assume that each area has scale levels , and call the inference operator for each layer, and output the prediction result : ; where represents the prior mean function at scale , represents the kernel vector, is the kernel matrix, Denote the observation data vector processed by this scale layer, which is the prior mean in the corresponding dimension. This process performs step-by-step refined fitting from the macro layer to the micro crack layer and shrinks the prediction uncertainty interval according to the error metric function after the regression of each layer until a hierarchical deformation prediction meeting the accuracy requirements is obtained.

[0092] In addition, in order to perform regression analysis more accurately, a weighted update mechanism (·) can also be defined to dynamically optimize in the composite kernel function: ; wherein, is the number of iterations, represents the mean square error between the predicted value and the observed value after the th regression, and

[0093] is the learning rate. After determining the final weights based on the composite kernel function, multi-scale regression analysis is performed again on each scale level in the high-risk suspected area to obtain the second regression result.

[0094] Integrate the results of the two regression analyses to generate accurate deformation prediction values and local uncertainty distributions for the local area.

[0095] Define the result set containing the predicted values and variance information at each position in all regions :

[0096] wherein represents the final deformation amount prediction based on the improved , represents the uncertainty measure, which is estimated through the variance of the regression posterior distribution.

[0097] Thus, high-precision local supplementation can be provided for the deformation evaluation of the target building, which is applicable to high-risk areas such as beam-column joints or crack ends.

[0098] S170. Fusion and consistency evaluation are performed on the three-dimensional deformation prediction field corresponding to the predicted vector field of the latent space features, the local deformation prediction value, and the local uncertainty distribution to obtain the fused global deformation prediction field.

[0099] In some embodiments, first, based on the decoder, the predicted vector field of the latent space features is mapped to the physical coordinate system to obtain the three-dimensional deformation prediction field of the latent space features.

[0100] Next, based on the predicted vector field of the latent space features and the suspected damage areas, a damage map is constructed.

[0101] Secondly, based on the three-dimensional deformation prediction field and the damage map, a global prediction field of the latent space features is constructed.

[0102] Then, the global skeleton of the global prediction field is determined. Among them, the function of the global skeleton is determined by the double integral of the relevance weight of the target position and the first deformation difference, and the first deformation difference is the difference between the three-dimensional deformation prediction field and the preset average benchmark.

[0103] Again, the weights of the local parameters and the global skeleton are constructed, and the local parameters and the global skeleton are fused to obtain the fused prediction field. Among them, the local parameters include the local deformation prediction value and the local uncertainty distribution.

[0104] Finally, based on the differences between the local parameters and the global skeleton in the high-risk suspected areas, and the differences between the gradients of the fused prediction field and the global skeleton, the fused prediction field is verified to obtain the fused global deformation prediction field and the damage distribution.

[0105] In this embodiment, the predicted vector field of the latent space features is: ; Among them,

[0106] For the convenience of global-local fusion, the decoder of the same network is called , and the predicted vector field of the latent space features is mapped back to the physical coordinate system to form a three-dimensional deformation prediction field ; Among them, represents the displacement vector at the spatial position at time .

[0107] Based on the three-dimensional deformation prediction field and the damage map, the constructed global prediction field of the latent space features is: ; Among them, represents the time index, represents the spatial coordinates of the overall building range, and is used as the reference zero-benchmark for the subsequent main ridge line extraction to maintain the macroscopic contour and trend.

[0108] Furthermore, a global field average benchmark ; characterizes the average deformation of the entire building at time , and is used as the reference zero-benchmark for the subsequent main ridge line extraction.

[0109] The relevance weight with the target position can be expressed as: , the time-varying correlation weight is: ; in Controls the falloff scale of spatial and deformation amplitudes.

[0110] In this embodiment, the process of determining the global skeleton of the global prediction field is as follows: Global Skeleton Functions Represents the overall deformation reference skeleton.

[0111] Select the main ridge line and important topological node information of the deformation, and filter out local high-frequency noise: ; in, Representation and target position The relevance weight of is the average benchmark of the global field, For the overall building space area, is the three-dimensional deformation prediction field.

[0112] The function of the global skeleton provides a global coordinate frame in subsequent steps to ensure that small-scale correction actions do not destroy the overall deformation trend.

[0113] In this embodiment, after determining the global skeleton and the local deformation prediction value and the local uncertainty distribution, fusion processing can be performed to fuse local details and global information. Multiple evaluations and cyclic verifications are required before and after fusion to improve the overall accuracy and stability of the deformation field. The specific fusion and verification process is as follows: Before fusion, evaluate the local and global error range, and set the evaluation function Measuring the difference between the local prediction and the global skeleton: ; in, is the local uncertainty, A set of coordinates representing high-risk areas.

[0114] For high-risk areas, a weight mapping function is constructed to prevent small-scale damage from being ignored in the macro skeleton. , in high-risk areas Take a larger value near the edge and gradually decay in other areas: ; in, is the prediction field obtained after fusion, Adjust the ratio of the global skeleton and the local high-precision regression result. When the local strain is large or the uncertainty is high, tend to 1, emphasizing the dependence on local correction to ensure the balance of low-frequency and high-frequency features in the global field.

[0115] After the preliminary fusion is completed, it is also possible to perform a secondary comparison on the crack endpoints or positions with sudden increases in strain in

[0116] to ensure that there will be no local distortion after correction and it can also be consistent with the global skeleton trend. Set the local verification function ; If exceeds the threshold , it means that the local gradient change after fusion is too large, and is limited to avoid overcorrection.

[0117] Then, cyclic verification can be performed, and and are both incorporated into the fusion result evaluation criteria, and the following fusion scoring function is set: ; Among them, and are adjustment coefficients. By minimizing , the fused prediction field is verified to exclude local overfitting and global deviation, and the differences between sub-scales are continuously flattened, ultimately improving the overall accuracy and stability of the deformation field.

[0118] After all cyclic verifications are completed, the finally fused prediction field and the damage distribution information are output together, covering the building detail area.

[0119] Define the output carrier to record the deformation vector and risk marker at each moment t: ; The output can truly reflect the coexistence of macroscopic structural deformation and local damage, providing an intuitive and visual basis for building maintenance and monitoring strategies.

[0120] S180. Analyze the fused global deformation prediction field to obtain the high-risk deformation distribution.

[0121] In some embodiments, first, based on the fused global deformation prediction field and the high-risk suspected region, the deformation degree and uncertainty within the high-risk suspected region are determined.

[0122] Then, the deformation degree and uncertainty within the high-risk suspected region, as well as the high-risk suspected region, are input into the weight update module of the latent diffusion model for iterative processing to obtain the updated weights of the latent diffusion model.

[0123] Finally, based on the composite kernel function, multi-scale regression analysis is performed on the high-risk suspected region to obtain the high-risk deformation distribution.

[0124] In this embodiment, the fused global deformation prediction field and the high-risk suspected region are centrally stored to form a data set , where, is the high-risk suspected region, v i represents the deformation prediction value at that location, represents the corresponding uncertainty.

[0125] Construct a cumulative risk assessment function for statistically analyzing the comprehensive level of deformation degree and uncertainty within the high-risk region: ; where is the coefficient for adjusting the relative weights of the deformation amplitude and uncertainty.

[0126] By quantifying the deformation evolution trend of the high-risk region, it provides a reference for latent space guidance update.

[0127] In this embodiment, the extracted and the local high-risk distribution are input into the guidance weight update module of the latent space diffusion network.

[0128] Define the key parameter set of the latent space diffusion network , and dynamically adjust the guidance weight in each iteration to pay more attention to the structural details of the high-risk region during latent space mapping. Denote the iteration step as : ; where, is the learning rate, represents the loss function of the diffusion network, including the weight gain for the high-risk suspected region to make the gradient descent pay more attention to the latent space representation at the crack or strain concentration point.

[0129] After completing the latent space guidance weight update, for the high-risk suspected region Perform a new round of improved Gaussian process regression. Based on the composite kernel function , and bring the updated weight factor into the kernel function parameters during this iteration. The regression output is denoted as , the new predicted value corresponding to the high-risk area: ; where is the prior mean function of this iteration, represents the kernel vector, is the calculation result of the improved kernel function after introducing the new weight, is the observed value vector of the previous round of prediction.

[0130] Through re-regression, incorporate the latent space correction information, material differences, and topological connection elements to further refine the accuracy of predicting cracks or strain concentration points.

[0131] In addition, to avoid local overfitting or noise superposition caused by multiple corrections, perform statistical analysis on the difference between the global skeleton and , and define the matching degree metric : ; where is the weighting coefficient for high-risk suspected areas or high-noise areas to highlight the difference from the global skeleton. If is too large, it indicates a risk of noise accumulation, and it is necessary to appropriately weaken the latent space update intensity or add a smoothing term in the regression. Thus, ensure that after multiple rounds of correction, the local correction still has a robust fit with the overall trend.

[0132] Fuse the local update with the initial deformation field again, and denote the fused field as to gradually correct the overall deformation distribution.

[0133] Let the fusion function map the local and global deformations:

[0134] where is the function that maps to the interval [0,1], which is used to allocate the fusion ratio of the local and global deformations. Maintain the updated time series curve | in the time dimension to improve the continuity and accuracy of the deformation evolution.

[0135] Thus, by performing multi-scale regression analysis on high-risk suspected areas, a high-risk deformation distribution can be obtained. This result includes both the global and local fusion deformation fields and also records the local regression predictions of local details.

[0136] In addition, after obtaining the high-risk deformation distribution, it is also possible to perform visual output and risk identification on the high-risk deformation distribution. After reorganizing the deformation data for each time period, 3D rendering is performed to highlight the deformation amounts and crack propagation of key components such as beams, columns, facades, and high-risk cracks, and a visual report and classification list are recorded for reference in subsequent structural assessment and maintenance strategies. The visualization process is as follows: First, load the high-risk deformation distribution.

[0137] To perform data reorganization before visualization and risk analysis, a merging function is defined , and the global and local comprehensive deformations are aggregated within the time index . The merging for all moments is based on the following expression convergence operation: ; where allocates the convergence ratio of the global and local predictions.

[0138] The full-time series deformation set is obtained, laying a consistent data foundation for visual rendering and damage analysis.

[0139] Next, for the merged deformation set , 3D visual rendering is performed, and the key areas of beams, columns, facades, and high-risk cracks, i.e., the preset connection structures, are highlighted with different colors and line widths.

[0140] The view transformation matrix is introduced to project the deformation vector into an interactive visualization space, and the mapping operation is defined as: ; where represents the planar coordinates, represents the vertical displacement or deformation value at time , and and are the rendered screen coordinates and depth information, respectively.

[0141] The Gaussian process uncertainty distribution and the latent space prediction field are combined to quantify the potential risk level at the component or crack location, and the damage scoring function is defined as: ; Among them, represents the surrounding local neighborhood, represents the deformation increment field in the latent space, represents the uncertainty at the same position, and is the coefficient for adjusting the weights of different elements.

[0142] Based on the value, divide the potential damage levels of the components and crack locations, and initially form high, medium, and low risk intervals.

[0143] For the locations exceeding a certain risk threshold in this step, incorporate the coordinate positions and their time variation information into the classification list. Establish a record matrix whose row index corresponds to the high-risk coordinate points, and the column index covers the deformation increment, crack expansion amount, and historical trend: ; If , then create a new entry in the list to record the abnormal stress and crack expansion rate of x to support subsequent structural maintenance and reinforcement strategies.

[0144] Then, based on the results generated in the above steps, generate a visualization report and a data interface. The report includes the global building deformation time series diagram, the local crack expansion diagram, and the high-risk list details, and construct a data interface to output each result in a standardized manner: ; Among them, , , respectively represent the visual encapsulation of the chart, risk data, and time series information for subsequent calls by external safety assessment systems or maintenance strategy modules.

[0145] Finally, output the final set of building deformation and damage information which contains the visualization chart, risk level markings, and the high-risk area classification list: ; Through this set, the building deformation trajectory can be accurately displayed in the time dimension and the degree of structural damage can be marked with quantitative indicators, ultimately providing an intuitive reference for subsequent safety assessment and maintenance decisions.

[0146] In the embodiments of the present invention, in order to comprehensively consider the influence of various factors on the deformation of a building, first, the original point cloud data of the target building obtained is subjected to coordinate system transformation and noise reduction processing to obtain an aligned point cloud set. After obtaining the aligned point cloud set, in order to narrow the research area, local difference analysis is also required for the aligned point cloud set to obtain the point cloud set of the abnormal area, so that the point cloud set of the abnormal area can be studied more accurately. Then, based on the multi-resolution features of the point cloud set of the abnormal area, the topological features of the preset connection structure of the abnormal area are extracted, and a topological map of the abnormal area is constructed. Stress constraints and material property constraints are imposed on the topological map of the abnormal area to determine the latent space features of the topological map of the abnormal area, so that the local details of stress concentration in the preset connection structure can be captured. At the same time, by combining the stress constraints and the constraints of material properties, the constraints of materials and mechanics can be reflected in the diffusion process, and the deformation can be predicted more accurately. Next, based on the latent diffusion model, interpolation and extrapolation processing are performed on the latent space features to determine the prediction vector field and the suspected damage area of the latent space features, so that areas with potential damage can be captured by combining stress and material information at the latent space level. After determining the suspected damage area, multi-scale regression analysis is performed on the suspected damage area to obtain the local deformation prediction value and the local uncertainty distribution, so as to provide a refined compensation basis for the subsequent fusion with the global skeleton. In order to finally form a global deformation field and record the damage distribution in a unified framework, therefore, it is also necessary to fuse and perform consistency evaluation on the three-dimensional deformation prediction field corresponding to the prediction vector field of the latent space features, the local deformation prediction value, and the local uncertainty distribution to obtain the fused global deformation prediction field. Finally, the fused global deformation prediction field is analyzed to obtain the high-risk deformation distribution. Thus, a more accurate high-risk deformation distribution can be formed in the time series continuity. During the prediction process, the linkage analysis of stress distribution, material properties, and local structural differences is realized, especially the comprehensive influence of fine cracks, complex component connections, and multiple materials, so as to improve the connection between global and local information for more accurate deformation prediction in the time series dimension.

[0147] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The order of execution of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0148] The following are the device embodiments of the present invention. For the details not described in detail therein, reference may be made to the corresponding method embodiments above.

[0149] Figure 2 The structural schematic diagram of the prediction device for the building deformation point cloud time series provided by the embodiments of the present invention is shown. For the sake of convenience of description, only the parts related to the embodiments of the present invention are shown and are described in detail as follows: As Figure 2As shown in the figure, the prediction device 200 for the time series of building deformation points includes: An acquisition alignment module 210, configured to perform coordinate system transformation and noise reduction processing on the acquired original point cloud data of the target building to obtain an aligned point cloud set; the original point cloud data includes point cloud data collected in multiple time periods and / or by multiple devices; A difference analysis module 220, configured to perform local difference degree analysis on the aligned point cloud set to obtain a point cloud set of the abnormal area; A construction module 230, configured to extract topological features of a preset connection structure of the abnormal area based on multi-resolution features of the point cloud set of the abnormal area, and construct a topological graph of the abnormal area; A feature extraction module 240, configured to perform stress constraint and material property constraint on the topological graph of the abnormal area to determine the latent space features of the topological graph of the abnormal area; An interpolation inference module 250, configured to perform interpolation and extrapolation processing on the latent space features based on the latent diffusion model to determine the predicted vector field and the suspected damage area of the latent space features; An analysis module 260, configured to perform multi-scale regression analysis on the suspected damage area to obtain local deformation prediction values and local uncertainty distributions; A fusion module 270, configured to fuse and perform consistency evaluation on the three-dimensional deformation prediction field corresponding to the predicted vector field of the latent space features, the local deformation prediction values, and the local uncertainty distributions to obtain a fused global deformation prediction field; A fusion analysis module 280, configured to analyze the fused global deformation prediction field to obtain a high-risk deformation distribution.

[0150] In a possible implementation manner, the construction module 230 is configured to extract the multi-resolution of the point cloud set of the abnormal area based on a multi-resolution encoder to obtain multi-resolution features of the point cloud set of the abnormal area; wherein, the multi-resolution encoder is based on introducing a gradient vector in the multi-resolution encoder to perform downsampling and feature extraction on the multi-resolution of the point cloud set of the abnormal area; Based on the topological feature extraction module in the multi-resolution encoder, extract the topological features of the preset connection structure of the abnormal area to obtain the topological features of the preset connection structure of the abnormal area; wherein, the preset connection structure includes beams, columns, trusses, and facades of the target building; Based on the topological features of the preset connection structure of the abnormal area, construct a topological graph of the abnormal area.

[0151] In a possible implementation manner, the construction module 230 is configured to perform stress constraint on the diffusion process of the latent space field corresponding to the topological graph of the abnormal area to obtain first latent space features; Based on the material property database, material property constraints are added during each diffusion process of the latent space field corresponding to the topological map of the abnormal region to obtain the second latent space feature; Based on the deformation amplitude and gradient, the deformation region of the latent space field corresponding to the topological map of the abnormal region is constrained to obtain the third latent space feature; Based on the first latent space feature, the second latent space feature, and the third latent space feature, the latent space feature of the topological map of the abnormal region is determined.

[0152] In a possible implementation, the interpolation inference module 250 is used to perform interpolation processing on adjacent time segments in the vector field of the latent space feature based on the trained latent diffusion model, as well as stress constraints and material property constraints in the latent space, and perform extrapolation processing on the vector field of the latent space feature to obtain the predicted deformation vector field of the latent space feature; Calculate the deformation intensity of the topological features of the preset connection structure, and determine the suspected damage region based on the preset deformation threshold and the predicted deformation vector field of the latent space feature.

[0153] In a possible implementation, the analysis module 260 is used to screen high-risk suspected regions based on the positioning information of the point set in the suspected damage region and the spatial filtering operator; the positioning information of the point set includes the coordinates of each point and the predicted deformation vector field of each point; Based on the pre-constructed composite kernel function, multi-scale regression analysis is performed on each scale level in the high-risk suspected region to obtain the local deformation prediction value and the local uncertainty distribution; the composite kernel function includes geometric correlation, material property similarity kernel function, and topological coupling degree of the preset connection structure.

[0154] In a possible implementation, the analysis module 260 is used to perform multi-scale regression analysis on each scale level in the high-risk suspected region based on the pre-constructed composite kernel function to obtain the first regression result; Dynamically update the preset weights in the composite kernel function to obtain the final weights of the composite kernel function; among them, a preset weight is set for each of the geometric correlation, material property similarity kernel function, and topological coupling degree of the preset connection structure in the composite kernel function; Based on the final weights of the composite kernel function, multi-scale regression analysis is performed on each scale level in the high-risk suspected region to obtain the second regression result; Integrate the first regression result and the second regression result to obtain the local deformation prediction value and the local uncertainty distribution.

[0155] In a possible implementation, the fusion module 270 is used to map the predicted vector field of the latent space feature to the physical coordinate system based on the decoder to obtain the three-dimensional deformation prediction field of the latent space feature; Construct a damage map based on the predicted vector field and suspected damage areas in the latent space feature; Construct a global prediction field of the latent space feature based on the three-dimensional deformation prediction field and the damage map; Determine the global skeleton of the global prediction field; wherein, the function of the global skeleton is determined by the relevance weight of the target position and the double integral of the first deformation difference, and the first deformation difference is the difference between the three-dimensional deformation prediction field and the preset average benchmark; Construct the weights of the local parameters and the global skeleton, and fuse the local parameters and the global skeleton to obtain the fused prediction field; wherein, the local parameters include the local deformation prediction value and the local uncertainty distribution; Based on the differences between the local parameters and the global skeleton in the high-risk suspected area, and the differences between the gradient of the fused prediction field and the gradient of the global skeleton, verify the fused prediction field to obtain the fused global deformation prediction field and the damage distribution.

[0156] In a possible implementation manner, the fusion analysis module 280 is configured to determine the deformation degree and uncertainty in the high-risk suspected area based on the fused global deformation prediction field and the high-risk suspected area; Input the deformation degree and uncertainty in the high-risk suspected area, and the high-risk suspected area into the weight update module of the latent diffusion model for iterative processing to obtain the updated weights of the latent diffusion model; Perform multi-scale regression analysis on the high-risk suspected area based on the composite kernel function to obtain the high-risk deformation distribution.

[0157] In a possible implementation manner, the difference analysis module 220 is configured to determine the intensity of each point cloud in the aligned point cloud set based on the pre-constructed anomaly evaluation model; Determine the point cloud set of the abnormal area based on the intensities of all point clouds in the aligned point cloud set.

[0158] An embodiment of the present invention further provides a prediction device, including a memory and a processor, where the memory stores a computer program, and the processor implements the method in the above method embodiment when executing the computer program.

[0159] In the above embodiments, the descriptions of the respective embodiments have their own emphases. For parts not detailed or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments. If there is no special description and logical conflict, the terms and / or descriptions between different embodiments are consistent and can be referenced to each other, and the technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationships.

[0160] The above-described embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for predicting building deformation point cloud time series, characterized in that: include: Perform coordinate system transformation and noise reduction on the original point cloud data of the target building to obtain an aligned point cloud set; The original point cloud data includes point cloud data collected in multiple time periods and / or by multiple devices; Performing local difference analysis on the aligned point cloud set to obtain a point cloud set in an abnormal area; Based on the multi-resolution features of the point cloud set of the abnormal area, extract the topological features of the preset connection structure of the abnormal area, and construct a topological map of the abnormal area; Performing stress constraints and material property constraints on the topological map of the abnormal area to determine the latent space characteristics of the topological map of the abnormal area; Based on the latent diffusion model, the latent space features are interpolated and extrapolated to determine the predicted vector field of the latent space features and the suspected damage area; Perform multi-scale regression analysis on the suspected damaged area to obtain the local deformation prediction value and local uncertainty distribution; The three-dimensional deformation prediction field corresponding to the prediction vector field of the latent space feature, the local deformation prediction value and the local uncertainty distribution are fused and consistency evaluated to obtain a fused global deformation prediction field; The fused global deformation prediction field is analyzed to obtain a high-risk deformation distribution.

2. The method for predicting building deformation point cloud time series according to claim 1, characterized in that: The method of extracting the topological features of the preset connection structure of the abnormal area based on the multi-resolution features of the point cloud set of the abnormal area and constructing the topological map of the abnormal area includes: Extracting the multi-resolution of the point cloud set of the abnormal area based on a multi-resolution encoder to obtain the multi-resolution features of the point cloud set of the abnormal area; wherein the multi-resolution encoder is based on introducing a gradient vector in the multi-resolution encoder to downsample and extract features of the multi-resolution of the point cloud set of the abnormal area; Extracting the topological features of the preset connection structure of the abnormal area based on the topological feature extraction module in the multi-resolution encoder to obtain the topological features of the preset connection structure of the abnormal area; wherein the preset connection structure includes beams, columns, trusses and facades of the target building; Based on the topological features of the preset connection structure of the abnormal area, a topological map of the abnormal area is constructed.

3. The method for predicting building deformation point cloud time series according to claim 1, characterized in that: The step of performing stress constraints and material property constraints on the topological map of the abnormal region to determine the latent space characteristics of the topological map of the abnormal region includes: Performing stress constraints on the diffusion process of the latent space field corresponding to the topological map of the abnormal area to obtain a first latent space feature; Based on the material property database, material property constraints are added to each diffusion process of the latent space field corresponding to the topological map of the abnormal area to obtain a second latent space feature; Based on the deformation amplitude and gradient, constraining the deformation area of ​​the latent space field corresponding to the topological map of the abnormal area to obtain a third latent space feature; Based on the first latent space feature, the second latent space feature and the third latent space feature, a latent space feature of the topological map of the abnormal area is determined.

4. The method for predicting building deformation point cloud time series according to claim 1, characterized in that: The method of interpolating and extrapolating the latent space features based on the latent diffusion model to determine the predicted vector field and suspected damage area of ​​the latent space features includes: Based on the trained latent diffusion model, as well as the stress constraints and material property constraints in the latent space, adjacent time segments in the vector field of the latent space feature are interpolated, and the vector field of the latent space feature is extrapolated to obtain a predicted deformation vector field of the latent space feature; The deformation intensity of the topological feature of the preset connection structure is calculated, and the suspected damage area is determined based on a preset deformation threshold and the predicted deformation vector field of the latent space feature.

5. The method for predicting building deformation point cloud time series according to claim 1, characterized in that: The multi-scale regression analysis is performed on the suspected damage area to obtain the local deformation prediction value and the local uncertainty distribution, including: Based on the positioning information of the point set in the suspected damage area and the spatial filtering operator, the high-risk suspected area is screened; the positioning information of the point set includes the coordinates of each point and the predicted deformation vector field of each point; Based on a pre-constructed composite kernel function, a multi-scale regression analysis is performed on each scale level in the high-risk suspected area to obtain a local deformation prediction value and a local uncertainty distribution; the composite kernel function includes geometric correlation, material property similarity kernel function and topological coupling of the preset connection structure.

6. The method for predicting building deformation point cloud time series according to claim 5, characterized in that: Based on the pre-constructed composite kernel function, a multi-scale regression analysis is performed on each scale level in the high-risk suspected area to obtain a local deformation prediction value and a local uncertainty distribution, including: Based on the pre-constructed composite kernel function, a multi-scale regression analysis is performed on each scale level in the high-risk suspected area to obtain a first regression result; Dynamically updating the preset weights in the composite kernel function to obtain the final weight of the composite kernel function; wherein the geometric correlation in the composite kernel function, the material property similarity kernel function and the topological coupling degree of the preset connection structure are each provided with a corresponding preset weight; Based on the final weight of the composite kernel function, a multi-scale regression analysis is performed on each scale level in the high-risk suspected area to obtain a second regression result; The first regression result and the second regression result are integrated to obtain a local deformation prediction value and a local uncertainty distribution.

7. The method for predicting building deformation point cloud time series according to claim 5 or 6, characterized in that: The three-dimensional deformation prediction field corresponding to the prediction vector field of the latent space feature, the local deformation prediction value and the local uncertainty distribution are fused and consistency evaluated to obtain a fused global prediction field, including: Mapping the prediction vector field of the latent space feature to a physical coordinate system based on a decoder to obtain a three-dimensional deformation prediction field of the latent space feature; constructing a damage map based on the predicted vector field and suspected damage areas of the latent space features; Based on the three-dimensional deformation prediction field and the damage map, construct a global prediction field of the latent space feature; Determining a global skeleton of the global prediction field; wherein a function of the global skeleton is determined by a double integral of a correlation weight of a target position and a first deformation difference, wherein the first deformation difference is a difference between the three-dimensional deformation prediction field and a preset average reference; Constructing weights of local parameters and the global skeleton, and fusing the local parameters and the global skeleton to obtain a fused prediction field; wherein the local parameters include a local deformation prediction value and a local uncertainty distribution; Based on the difference between the local parameters in the high-risk suspected area and the global skeleton, and the difference between the gradient of the fused prediction field and the gradient of the global skeleton, the fused prediction field is verified to obtain the fused global deformation prediction field and damage distribution.

8. The method for predicting building deformation point cloud time series according to claim 5 or 6, characterized in that: The analyzing the fused global deformation prediction field to obtain a high-risk deformation distribution includes: Determining the deformation degree and uncertainty in the high-risk suspected area based on the fused global deformation prediction field and the high-risk suspected area; Inputting the deformation degree and uncertainty in the high-risk suspected area and the high-risk suspected area into the weight updating module of the latent diffusion model, performing iterative processing, and obtaining the updated weight of the latent diffusion model; Based on the composite kernel function, a multi-scale regression analysis is performed on the high-risk suspected area to obtain the high-risk deformation distribution.

9. The method for predicting building deformation point cloud time series according to claim 1, characterized in that: The performing local difference analysis on the aligned point cloud set to obtain the point cloud set of the abnormal area includes: Based on a pre-constructed anomaly evaluation model, determining the intensity of each point cloud in the aligned point cloud set; wherein the anomaly evaluation model is constructed based on the difference between the grayscale or intensity information of each point cloud and a preset average intensity value; Based on the intensities of all point clouds in the aligned point cloud set, a point cloud set of the abnormal area is determined.

10. A prediction device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 9 when executing the computer program.

Images