Precise Registration Method for Building Monitoring Data Based on Spatiotemporal Dynamic Weights

Abstract

The present invention discloses a precise registration method for building monitoring data based on spatio-temporal dynamic weights, including: establishing a response model considering the nonlinear behavior of materials by analyzing the deformation characteristics in multi-source monitoring data; constructing a dynamic weight function based on the structural stress state to achieve the adaptive allocation of the reliability weights of sensor data; using the material nonlinear response model for geometric deformation correction to correct the geometric distortion caused by the nonlinear behavior of materials; adopting a weight fusion and deformation gradient constraint method to achieve the spatio-temporal registration of multi-source data; and performing parameter adaptive optimization through an online learning mechanism. The present invention solves technical problems such as low geometric correction accuracy in high-stress areas and dynamic changes in sensor reliability, and realizes the registration of building monitoring data with sub-millimeter-level accuracy.

Description

Technical Field

[0001] The present invention belongs to structural health monitoring, and in particular, to a precise registration method for building monitoring data based on spatio-temporal dynamic weights. Background Art

[0002] The precise registration of building monitoring data is a key link in structural health monitoring and safety assessment. With the development of multi-source monitoring technologies such as satellite remote sensing, unmanned aerial vehicles, and ground monitoring equipment, the building deformation monitoring ability has been significantly improved. However, it also brings challenges of data spatio-temporal inconsistency. Due to differences in acquisition cycles, positions, and observation angles of multi-source monitoring data, registration errors that are difficult to eliminate are generated. Especially in building monitoring scenarios with sub-millimeter accuracy requirements, these errors will seriously affect the reliability of analysis results. Precise registration not only affects the accuracy of structural safety assessment but also directly relates to post-disaster emergency decision-making, building life prediction, and the formulation of repair and reinforcement plans, and has important economic and social values.

[0003] Current building monitoring data registration methods mainly rely on three types of technical routes: feature-based registration methods, region-based registration methods, and model-based registration methods. Feature-based methods achieve registration by extracting and matching key points, but the matching success rate is often low on the surface of modern buildings with sparse texture features; region-based methods achieve registration by maximizing similarity measures such as mutual information, but are sensitive to noise and have high computational complexity; model-based methods establish a geometric transformation model to describe the corresponding relationship between data, but mostly use simplified linear models and are difficult to handle complex non-linear deformations. Existing methods generally adopt a static weight strategy, presetting weights based on the inherent accuracy of sensors and environmental factors, and cannot adapt to changing monitoring conditions. At the same time, traditional geometric correction methods are mostly based on rigid body or linear elastic assumptions, and the accuracy significantly decreases when the building structure is under high stress or has undergone plastic deformation.

[0004] These technical routes face multiple technical problems in practical applications: First, existing registration methods ignore the non-linear behavior of building materials under high stress conditions, resulting in a significant increase in errors in geometric correction in the region where the stress is close to the yield point; second, the weight allocation of sensor data does not consider the influence of the structural stress state, and the change in the reliability of the same sensor under different stress conditions is not fully reflected. These technical problems severely restrict the further improvement of the registration accuracy of building monitoring data, and innovative solutions are urgently needed. Summary of the Invention

[0005] The object of the invention is to provide a precise registration method for building monitoring data based on spatio-temporal dynamic weights to solve the above problems existing in the prior art.

[0006] Technical Solution: A precise registration method for building monitoring data based on spatio-temporal dynamic weights includes the following steps:

[0007] Read satellite data, UAV data and ground monitoring data, preprocess and standardize them to generate a standardized monitoring dataset;

[0008] Analyze the deformation characteristics in the standardized monitoring dataset, combine with the characteristics of building materials, establish a material model considering non - linear behavior, and generate a material non - linear response model and a structural stress distribution map;

[0009] According to the structural stress distribution map and the pre - stored sensor characteristics, construct a dynamic weight function, calculate the reliability weights of each sensor data in different stress regions, and output a dynamic sensor weight matrix;

[0010] Use the material non - linear response model to perform geometric deformation correction on the standardized monitoring dataset to obtain a non - linear correction dataset;

[0011] Fuse the non - linear correction dataset and the dynamic sensor weight matrix to generate an accurate registration result;

[0012] Based on the accurate registration result, perform parameter adaptive optimization, and at the same time identify possible data anomalies or structural anomalies, generate optimized configuration parameters and feedback for iterative optimization.

[0013] Advantageous effects: The present invention introduces two core factors, namely the material non - linear behavior and the structural stress state, and establishes a deep integration of data registration and physical models; through non - linear material modeling of the standardized monitoring dataset, the geometric correction can accurately reflect the complex deformation behavior of materials under high - stress states; through the dynamic weight allocation based on the perception of the structural stress state, the problem of reliability evaluation of different monitoring devices under changing conditions is solved; the registration accuracy is improved, the adaptability and robustness of the method are enhanced, and more reliable data support is provided for the safety monitoring of building structures. Description of the Drawings

[0014] Figure 1 It is a step - flow chart of the accurate registration method for building monitoring data based on spatio - temporal dynamic weights provided by an embodiment of the present application.

[0015] Figure 2 It is a step - flow chart of generating a material non - linear response model provided by an embodiment of the present application.

[0016] Figure 3 It is a step - flow chart of generating a structural stress distribution map provided by an embodiment of the present application.

[0017] Figure 4 It is a step - flow chart of constructing a dynamic weight function and outputting a dynamic sensor weight matrix provided by an embodiment of the present application. Detailed Embodiments

[0018] To enable those skilled in the art to better understand the solution of the present invention, the following will clearly and completely describe the technical solution in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

[0019] It should be particularly noted that, for clearly showing the step flow of the present application, serial numbers are marked for each step in the specification. These serial numbers are only for the convenience of description and do not limit the execution order of the steps. In actual operation, according to the technical requirements of the specific implementation scenario, each step can be executed in a different order from that shown in the specification, and in some cases, parallel processing between steps can also be achieved.

[0020] As Figure 1 shown, the accurate registration method for building monitoring data based on spatio-temporal dynamic weights includes the following steps:

[0021] S1. Read satellite data, UAV data, and ground monitoring data and perform preprocessing and standardization, including coordinate system unification, noise filtering, and data format standardization, to generate a standardized monitoring data set;

[0022] S2. Analyze the deformation characteristics in the standardized monitoring data set, combine with the characteristics of building materials, establish a material model considering non-linear behavior, and generate a material non-linear response model and a structural stress distribution map;

[0023] S3. According to the structural stress distribution map and the pre-stored sensor characteristics, construct a dynamic weight function, calculate the reliability weights of each sensor data in different stress regions, and output a dynamic sensor weight matrix;

[0024] S4. Use the material non-linear response model to perform geometric deformation correction on the standardized monitoring data set, correct the geometric distortion caused by the non-linear behavior of the material, and obtain a non-linear correction data set;

[0025] S5. Integrate the non-linear correction data set and the dynamic sensor weight matrix to achieve spatio-temporal registration of multi-source data, generate an accurate registration result; and perform accuracy evaluation by comparing with reference points, and output a registration accuracy report;

[0026] S6. Based on the accurate registration result, perform parameter adaptive optimization, and at the same time identify possible data anomalies or structural anomalies, generate optimized configuration parameters and an anomaly detection report, and feedback them to the foregoing steps for iterative optimization.

[0027] According to one aspect of the present application, step S1 is further as follows:

[0028] S11. Conduct multi-source data collection and metadata extraction: Read satellite monitoring data, UAV monitoring data, and ground monitoring equipment data, extract monitoring metadata such as timestamps, collection locations, observation angles, and sensor accuracies, and at the same time conduct a preliminary verification of data integrity to generate an initial data quality report.

[0029] S12. Unify coordinate systems and align time bases: Convert different coordinate systems (such as WGS84, local coordinate systems, etc.) of satellite monitoring data, UAV monitoring data, and ground monitoring equipment data to a unified engineering coordinate system, and at the same time align different time bases (such as UTC, local time) to a unified time base, and output a coordinate-time unified dataset.

[0030] S13. Harmonize multi-scale data resolutions: Conduct a harmonization process on data with different resolutions in the coordinate-time unified dataset, and convert satellite data (meter level), UAV data (centimeter level), and ground monitoring data (millimeter level) into a dataset with consistent resolution through an adaptive interpolation algorithm to ensure comparability in the spatial dimension.

[0031] S14. Conduct noise identification and signal enhancement: Analyze the noise patterns in the dataset with consistent resolution, and through a combination of wavelet transform and adaptive filters, identify and remove environmental noise, sensor noise, and random errors, and enhance the effective signals to obtain a noise-suppressed dataset.

[0032] S15. Standardize data formats and conduct quality assessment: Convert the noise-suppressed dataset into a unified data structure and format, append quality indicators (signal-to-noise ratio, integrity, reliability), and generate a final standardized monitoring dataset and a data quality assessment report.

[0033] According to one aspect of the present application, step S2 is further as follows:

[0034] S21. Initialize structural material parameters: Read building structure design parameters and material property databases, and in combination with the initial deformation characteristics in the standardized monitoring dataset, initialize the constitutive parameters of building materials such as concrete and steel to establish an initial material parameter set.

[0035] S22. Analyze multi-temporal deformation data: Extract structural deformation data at multiple time points from the standardized monitoring dataset, calculate the deformation rate, deformation gradient, and cumulative deformation amount, identify deformation significant regions and deformation patterns, and output a deformation feature map.

[0036] S23. Identify the non-linear stress-strain relationship: Based on the deformation feature map and the initial material parameter set, use the reverse derivation method to calculate the stress states of key points of the structure, and in combination with the measured deformation data, identify the actual stress-strain curve, especially the non-linear characteristics in the elastic-plastic transition region, and generate a non-linear stress-strain model.

[0037] S24. Evaluate the material state in sub - regions: Divide the building structure into multiple regions, and evaluate the current state (elastic, elastic - plastic transition, plastic) of the materials in each region according to the non - linear stress - strain model and the deformation characteristic atlas, and establish a material state partition map.

[0038] S25. Construct a non - linear material response model: Integrate the non - linear stress - strain model and the material state partition map to construct a mathematical model considering the non - linear characteristics of materials. This model can describe the deformation response of materials under different stress levels, and output the non - linear material response model and a detailed structural stress distribution map.

[0039] As Figure 2 shown, according to one aspect of the present application, step S23 is further as follows:

[0040] S231. Based on the standardized monitoring data set and the building material properties, extract the deformation data and the initial material parameter set, and calculate the initial stress distribution field through the finite - element inversion algorithm.

[0041] S232. Based on the initial stress distribution field and the initial material parameter set, extract the multi - time - step stress - strain data pairs to form a regional stress - strain data set.

[0042] S233. Adopt a piece - wise non - linear fitting method for the regional stress - strain data set to construct a non - linear constitutive model, and calculate the strain rate influence factor at the same time.

[0043] S234. Construct an environmental factor correction function in combination with environmental parameters, and integrate the non - linear constitutive model, the strain rate influence factor and the environmental factor correction function to generate a non - linear stress - strain model.

[0044] S235. Verify the prediction accuracy of the non - linear stress - strain model, quantify the model uncertainty, and output a model reliability evaluation report and a non - linear material response model.

[0045] In an embodiment of the present application, the initial stress distribution is calculated in reverse. Read the deformation characteristic atlas and the initial material parameter set, apply the finite - element inversion algorithm, and solve the unknown stress field through the known deformation field. The specific process is as follows: Establish a discrete finite - element model of the structure, use the displacement in the deformation characteristic atlas as the boundary condition, and use the conjugate gradient method to iteratively solve the stress distribution equation to obtain the initial stress distribution field. This calculation considers the structural continuity constraint to ensure that the stress distribution satisfies the static equilibrium equation.

[0046] Separate the linear and nonlinear regions of the material. Analyze the initial stress distribution field and, in combination with the material yield strength in the initial material parameter set, identify the linear deformation regions (regions where the stress is below the yield strength) and potential nonlinear deformation regions (regions where the stress is close to or exceeds the yield strength) in the structure. Use an adaptive threshold segmentation algorithm, considering stress gradient and stress concentration factors, to generate a material behavior partition map that divides the structure into multiple regions and labels the deformation behavior characteristics (linear, critical, or nonlinear) of each region.

[0047] Extract multi-temporal stress-strain data. From the standardized monitoring data set and the initial stress distribution field, extract stress-strain data pairs at multiple time points. For each region (especially the nonlinear regions marked in the material behavior partition map), establish a stress-strain data matrix where each row represents a time point, each column represents a spatial point, and the matrix elements are the corresponding stress-strain value pairs. Apply a noise-robust data cleaning algorithm to remove outliers and obtain the regional stress-strain data set.

[0048] Perform iterative fitting of the nonlinear constitutive relationship. For each region in the regional stress-strain data set, use a piecewise nonlinear fitting method to construct a stress-strain relationship model. First, use piecewise linear fitting to determine the elastic modulus and yield point of the material, and then for the data beyond the yield point, use a nonlinear optimization algorithm (such as the Levenberg-Marquardt algorithm) to fit a multi-parameter nonlinear model (such as the Ramberg-Osgood model or the modified Johnson-Cook model). During the iterative optimization process, evaluate the prediction ability of the model under different load conditions through cross-validation, dynamically adjust the model parameters until the residual reaches the minimum, and output the regional nonlinear constitutive model.

[0049] Calculate the strain rate influence factor. Analyze the dynamic deformation data in the regional stress-strain data set, calculate the strain rates in different regions, and evaluate the influence of the strain rate on the nonlinear behavior of the material. The specific method is: calculate the ratio of the strain increment between consecutive time points to the time interval to obtain the strain rate field; divide the strain rate into multiple levels, fit the constitutive model separately for the stress-strain data at each level, compare the changes in model parameters at different strain rates, establish a relationship function between the parameters and the strain rate, and output the strain rate influence factor.

[0050] Perform temperature and environmental factor correction. Read the environmental parameters (temperature, humidity, etc.) in the monitoring metadata and analyze the influence of these factors on the nonlinear behavior of the material. Use the multivariate regression analysis method to establish the mapping relationship between the environmental parameters and the constitutive model parameters. For the temperature effect, particularly consider the coupling of the thermal expansion effect and the material softening effect; for concrete materials, also consider the influence of humidity on the Young's modulus. Based on the analysis results, generate the environmental factor correction function for adjusting the constitutive model parameters under different environmental conditions.

[0051] Construct a comprehensive nonlinear stress-strain model. Integrate the regional nonlinear constitutive model, the strain rate influence factor, and the environmental factor correction function to construct a comprehensive nonlinear stress-strain model considering multiple factors. This model adopts a hierarchical structure: the basic layer is the static nonlinear constitutive relationship, the middle layer considers the strain rate influence, and the top layer applies environmental correction. The model is expressed by tensors and can describe the nonlinear behavior of anisotropic materials under three-dimensional stress states. The model parameters are stored in the form of spatial distribution, covering the entire monitoring area, and finally output the nonlinear stress-strain model, which will be used in the subsequent steps for accurate prediction of deformation behavior.

[0052] Perform model verification and uncertainty quantification. By using a part of the regional stress-strain dataset (about 20%) as the validation set, evaluate the prediction accuracy of the nonlinear stress-strain model. Calculate the relative error between the predicted stress and the measured stress and generate the error distribution statistical information. At the same time, use the Monte Carlo method to perform random sampling in the model parameter space, evaluate the influence of parameter uncertainty on the model output, and quantify the reliability interval of the model. Based on the verification results and uncertainty analysis, generate a model reliability assessment report, which contains the reliability indicators of the model at different stress levels and different regions and will be used as an important reference for subsequent weight allocation.

[0053] This embodiment constructs a comprehensive nonlinear stress-strain model, overcoming the limitations of traditional linear material models. This model can accurately describe the nonlinear behavior of materials under different stress states, especially the deformation characteristics in the vicinity of the yield point and the plastic region. Compared with traditional linear material models, the prediction accuracy of this nonlinear model in the high-stress region is increased by more than 3 times, reducing the material behavior prediction error from more than 30% to less than 10%. This high-precision prediction of material behavior directly improves the accuracy of geometric correction, enables the registration method to handle complex nonlinear deformation fields, and is particularly suitable for the monitoring of building structures undergoing large deformations or partial plastic deformations, improving the reliability of monitoring data under extreme conditions (such as after earthquakes and during construction).

[0054] As Figure 3 shown, according to one aspect of the present application, step S24 is further:

[0055] S241. Read the building structure design parameters and the standardized monitoring data set, and divide the structural function partition map based on the structural function characteristics and geometric morphology;

[0056] S242. Analyze the deformation characteristics in the standardized monitoring data set, identify the regions with similar deformation characteristics, and generate the deformation behavior partition map;

[0057] S243. Read the initial stress distribution field and the non-linear stress-strain model, calculate the stress state index, and generate the stress state grading map;

[0058] S244. Determine the elastic-plastic state of the materials in each region based on the material yield criterion, and consider the material property differences to output the material elastic-plastic state map;

[0059] S245. Analyze the historical load data in the standardized monitoring data set, calculate the cumulative plastic strain and fatigue damage, and generate the cumulative damage distribution map;

[0060] S246. Integrate the structural function partition map, the deformation behavior partition map, the stress state grading map, the material elastic-plastic state map and the cumulative damage distribution map, and generate the material state partition map and the structural stress distribution map through multi-index evaluation.

[0061] In an embodiment of the present application, the structural function areas are divided. Read the building structure design parameters and the standardized monitoring data set, and based on the functional characteristics and geometric morphology of the structure, use the hierarchical clustering algorithm to divide the entire structure into regions with similar functions. The algorithm first performs an initial classification based on the material type and component function, and then considers the geometric continuity and stress transfer path for region merging or subdivision, and finally outputs the structural function partition map. The partition map includes region boundaries, region numbers and region function attributes, providing a basic spatial reference framework for subsequent material state evaluation.

[0062] Analyze the deformation feature clustering. Read the deformation feature map, and for the displacement, velocity and acceleration data of each spatial point, apply the K-means++ clustering algorithm to identify the regions with similar deformation characteristics. Before clustering, use the principal component analysis (PCA) for dimensionality reduction, and retain the principal components with an explained variance of more than 90%; during the clustering process, automatically determine the optimal number of clusters through the Silhouette Coefficient. Apply the spatial continuity constraint to the clustering result to ensure that the same clustering region is relatively continuous in space, and output the deformation behavior partition map.

[0063] Conduct a refined analysis of the stress state. Read the initial stress distribution field and the nonlinear stress-strain model, and calculate the stress state indicators within each region, including: principal stress distribution, shear stress distribution, stress triaxiality, and stress intensity. Using the stress invariant theory, calculate the von Mises equivalent stress and the maximum principal stress at each point, and compare them with the material yield strength to determine the stress utilization rate at each point. Based on these indicators, use the fuzzy logic algorithm to classify the stress state and output the stress state classification map.

[0064] Determine the elastic-plastic state of the material. Read the stress state classification map and the nonlinear stress-strain model, and determine the elastic-plastic state of the material in each region based on the material yield criterion (such as the von Mises criterion for metal materials and the Drucker-Prager criterion for concrete). For each region, calculate the probability distributions of the elastic, elastic-plastic transition, and plastic states, and use the Bayesian inference method to determine the final state classification considering the uncertainty information in the model reliability assessment report. For concrete materials, particularly consider the tensile-compressive asymmetry and cracking behavior, and output the material elastic-plastic state map.

[0065] Evaluate the cumulative damage. Analyze the historical load data in the standardized monitoring dataset, and calculate the cumulative plastic strain and fatigue damage experienced by the material in each region. For metal materials, use the Palmgren-Miner linear cumulative damage theory and combine it with the Rainflow counting method to process variable amplitude loads; for concrete materials, consider the cracking damage model and evaluate the development state of microcracks. Calculate the regional damage index through a method that combines low-cycle fatigue and high-cycle fatigue models, and output the cumulative damage distribution map.

[0066] Conduct a comprehensive assessment of the material state. Integrate the structural function partition map, deformation behavior partition map, stress state classification map, material elastic-plastic state map, and cumulative damage distribution map to establish a multi-index evaluation system. Use the Analytic Hierarchy Process (AHP) to determine the weights of each index, and calculate the comprehensive state index through weighted summation. According to the index value, divide the material state into five grades: fully elastic, elastic-dominated, elastic-plastic transition, plastic-dominated, and near failure, and generate the material comprehensive state assessment map.

[0067] Refine the state boundary. Read the material comprehensive state assessment map, detect the boundary positions of different state regions, and use the adaptive grid refinement algorithm to improve the evaluation accuracy of the boundary region. In the region near the boundary, increase the density of calculation points and re-perform the state evaluation to ensure an accurate description of the state change region. Apply the boundary smoothing algorithm to reduce unnecessary boundary complexity while retaining the true state change characteristics, and output the refined material state boundary map.

[0068] Conduct a time-evolution analysis of the material state. Compare the comprehensive material state assessment diagrams at different time points to analyze the time-evolution trend of the material state. Use a state transition matrix to describe the transition probabilities between states, and identify the regions with rapid state changes and stable regions. Based on historical evolution laws, use a time series prediction model (such as ARIMA or LSTM network) to predict the state change trend in the short term, and output a state evolution prediction report and the final material state partition diagram, which will be used as an important basis for subsequent non-linear correction and weight assignment.

[0069] This embodiment realizes a refined assessment of the state of building structure materials. Compared with the traditional single-standard zoning method, this embodiment enhances the adaptability to complex structures, and the zoning accuracy rate is increased by more than 35%. The refined material state assessment enables the system to adopt differential processing strategies for different state regions, especially the refined identification of the elastoplastic transition region and the damage region, providing the possibility for subsequent high-precision registration. In addition, by considering the cumulative damage factor, this embodiment also improves the monitoring ability of long-term service structures, providing more reliable data support for structural life assessment.

[0070] Existing weight assignment methods are mainly based on the inherent accuracy of sensors and environmental factors, ignoring the stress state of the building structure itself. Under different stress distribution conditions, the reliability of the same sensor will vary significantly, and the structural stress state is not considered. Existing static weight schemes and dynamic calibration techniques use predefined weights or simple environmental parameter adjustments and cannot respond to changes in stress distribution in the structure. This results in a significant reduction in the registration accuracy under complex loading conditions, especially in areas with non-uniform deformation. Therefore, as Figure 4 shown, according to one aspect of the present application, step S3 is further as follows:

[0071] S31. Analyze the accuracy characteristics of various sensors (satellite SAR, UAV optical / laser, ground displacement / tilt / strain sensors, etc.) under different working conditions, establish a sensor accuracy characteristic model, and quantify the relationship between influencing factors and measurement errors;

[0072] S32. Read the structural stress distribution diagram and the sensor accuracy characteristic model, analyze the variation law of the reliability of various sensors under different stress levels, establish a quantitative correlation function between stress level and sensor reliability, and output a stress-reliability mapping table;

[0073] S33. Based on the structural stress distribution diagram and engineering safety assessment criteria, identify high-stress regions, stress concentration regions, regions with large stress gradients, and key load-bearing parts in the structure, set the monitoring priorities for different regions, and generate a structural region priority diagram;

[0074] S34. Integrate the stress-reliability mapping table and the structural area priority map to construct a dynamic weight calculation function. This function can adaptively adjust the weight coefficient according to the regional stress state, sensor type, and regional priority, and generate a weight calculation function (hereinafter referred to as the dynamic weight function).

[0075] S35. Apply the weight calculation function to calculate the weight of each data point in the standardized monitoring dataset, considering the spatial position, acquisition time, stress state of the area where the data point is located, and sensor characteristics, and generate a dynamic sensor weight matrix covering the entire monitoring area and time period.

[0076] This embodiment overcomes the limitations of traditional static weight methods. Compared with traditional static weight methods, this dynamic weight mechanism improves the rationality of weight allocation under changing load conditions, and the data fusion accuracy is increased by 25 - 40%. Especially in complex structures with uneven stress distribution, it can accurately reflect the monitoring requirements of local areas and improve the ability to identify potential risk areas. In addition, this embodiment also enhances the system's adaptability to sensor anomalies. When the performance of some sensors deteriorates, the system can automatically reduce their weights to maintain the overall monitoring quality, enhancing the robustness and fault tolerance of the monitoring system.

[0077] According to one aspect of the present application, step S34 is further as follows:

[0078] S341. Read the stress-reliability mapping table, the structural area priority map, and the sensor accuracy characteristic model, and establish a multi-dimensional factor evaluation system for influencing weight allocation, including stress state indicators, regional importance indicators, and sensor characteristic indicators, to obtain a weight influence factor evaluation system.

[0079] S342. Based on the weight influence factor evaluation system, construct an objective function for weight optimization, and at the same time consider maximizing the deformation monitoring accuracy, minimizing the spatial discontinuity of weight allocation, and maximizing the monitoring reliability of key areas, to obtain a multi-objective weight optimization function.

[0080] S343. Read the structural stress distribution map and the stress-reliability mapping table, design stress response functions for each sensor type, and output a set of stress response functions.

[0081] S344. Construct a constraint condition to ensure the spatial continuity of weight allocation, construct a gradient energy functional of the weight field, and output a spatial continuity constraint function.

[0082] S345. Analyze the deformation rate data in the standardized monitoring dataset, construct a time response mechanism for dynamic weight adjustment, including a fast response part and a slow adaptation part, and output a time response adjustment function.

[0083] S346. Construct a robust processing mechanism for abnormal data using a weight function to reduce the impact of outliers on weight calculation and output an abnormal data processing function;

[0084] S347. Integrate the multi-objective weight optimization function, stress response function set, spatial continuity constraint function, time response adjustment function, and abnormal data processing function to construct a complete dynamic weight calculation framework and generate a weight calculation function.

[0085] This embodiment realizes the optimal weight allocation under complex conditions. Compared with the traditional single-factor weight method, this embodiment improves the rationality and adaptability of weight allocation, and the fusion accuracy is increased by more than 30% in complex monitoring scenarios. In particular, the introduction of the time response mechanism enables the system to quickly respond to emergencies, such as instantaneous deformations caused by seismic or explosion loads, and gradually adapt to slow processes such as creep or seasonal changes of buildings through long-term learning. The robust processing mechanism for abnormal data improves the reliability of the system in harsh environments or partial sensor failures, extending the effective working time of the monitoring system by 35%.

[0086] According to one aspect of the present application, the steps of constructing a dynamic weight function and outputting a dynamic sensor weight matrix include:

[0087] Read the structural stress distribution map and the stress-reliability mapping table, and design stress response functions for each type of sensor in the form of a piecewise function: in the elastic region, use a slow-changing S-shaped curve to describe the change in sensor reliability; in the elastic-plastic transition region, increase the function slope; in the plastic region, introduce an inflection point or saturation characteristic;

[0088] Determine the stress response function parameters of various sensors by fitting the pre-stored measured data using the non-linear least squares method, and output the stress response function set.

[0089] This embodiment realizes the accurate mapping between the structural stress state and sensor reliability. Compared with simple linear mapping, the accuracy of reliability estimation in this embodiment is increased by 45%. The accurate stress-reliability mapping enables the system to preferentially select the most reliable data sources for fusion under complex stress distributions, avoiding the problem of error amplification caused by the decrease in sensor reliability in high-stress regions. In addition, this embodiment can also adapt to the characteristic differences of different sensors, such as the decrease in accuracy of SAR interferometric sensors in large deformation regions induced by stress, and the insufficient sensitivity of optical correlation techniques in small deformation regions, thereby maximizing the advantages of various sensors and improving the overall monitoring efficiency.

[0090] In an embodiment of the present application, the influencing factors of the analysis weight are quantified. Read the stress-reliability mapping table, the structural area priority map, and the sensor accuracy characteristic model to establish a multi-dimensional factor space that affects weight allocation. Use the principal component analysis method to identify the main influencing factors and their relative importance. For each factor, establish a quantitative evaluation index, such as stress state index (stress level, stress gradient), regional importance index (structural safety factor, functional criticality), and sensor characteristic index (baseline accuracy, random error, systematic error), and output the evaluation system of the influencing factors of the weight.

[0091] Construct a multi-objective weight optimization function. Based on the evaluation system of the influencing factors of the weight, construct an objective function for weight optimization. This objective function contains three sub-objectives: maximizing the deformation monitoring accuracy, minimizing the spatial discontinuity of the weight allocation, and maximizing the monitoring reliability of the key area. Use the Pareto optimization method to handle multi-objective conflicts and construct a comprehensive objective function through convex optimization techniques. For different application scenarios (such as conventional monitoring, post-disaster assessment, construction monitoring), design different combinations of objective weights and output the multi-objective weight optimization function.

[0092] Construct a stress state response function. Read the structural stress distribution map and the stress-reliability mapping table, and design a stress response function for each type of sensor. This function describes the law of the sensor reliability changing with the stress level and adopts a piecewise function form: in the low stress area (within the elastic range), the function is a slowly changing S-shaped curve; in the transition area (elastic-plastic transition zone), the slope of the function increases; in the high stress area (plastic zone), the function may have an inflection point or saturation phenomenon. The function parameters are determined by fitting the measured data through the nonlinear least squares method, and a set of stress response functions for various types of sensors is output.

[0093] Construct a spatial continuity constraint. Design a constraint condition to ensure the spatial continuity of the weight allocation. Use the principle of variational method to construct the gradient energy functional of the weight field and quantify the spatial discontinuity of the weight space as an energy value. In the areas with similar material states, set stronger continuity constraints; at the boundaries of material states, allow appropriate discontinuities. The constraint strength parameter is determined through numerical experiments on typical scenarios, so that the weight change is both smooth and can reflect the real change of sensor reliability, and output the spatial continuity constraint function.

[0094] Construct a time responsiveness mechanism. Analyze the deformation rate data in the deformation characteristic spectrum and design a time response mechanism for dynamic weight adjustment. This mechanism consists of two parts: a fast response part and a slow adaptation part. The fast response part is realized through an exponential filter to quickly adjust the weight for sudden deformation events (such as impact loads); the slow adaptation part uses a long short-term memory (LSTM) network to learn the long-term deformation trend and predict the weight adjustment requirements. The results of the two parts are combined through an adaptive mixer, and the time response adjustment function is output.

[0095] Build an outlier robust processing mechanism. Design a weight function for the robust processing mechanism of abnormal data. Use the Huber loss function to replace the traditional mean square error to reduce the impact of outliers on weight calculation. Combine the Random Sample Consensus (RANSAC) algorithm to detect and process data outliers and prevent unreasonable weight allocation caused by abnormal data. For the situation of temporary failure or sudden change in accuracy of sensors, design a fast weight reduction strategy and a backup sensor activation mechanism, and output an abnormal data processing function.

[0096] Carry out weight normalization and smoothing strategies. Design a weight normalization strategy to ensure the comparability of weights of different types of sensors and the rationality of the overall weight distribution. Use a non-linear normalization method (such as a variant of the softmax function) to control the ratio of the maximum and minimum weights while maintaining the relative differences in weights. To avoid sudden changes in weights over time, apply a time smoothing filter to limit the maximum amplitude of weight changes per unit time, and output a weight normalization and smoothing function.

[0097] Integrate a comprehensive dynamic weight function. Integrate multi-objective weight optimization functions, stress response function sets, spatial continuity constraint functions, time response adjustment functions, abnormal data processing functions, and weight normalization and smoothing functions to build a complete dynamic weight calculation framework. Adopt a hierarchical calculation strategy: first calculate the basic weights based on the stress state and sensor characteristics, then apply spatial continuity constraints, followed by time adjustment, and finally handle outliers and normalize. The framework is implemented in the form of a function library and can be flexibly configured according to different monitoring requirements. Finally, output a weight calculation function, which will be used to generate a dynamic weight matrix covering the entire monitoring area.

[0098] According to one aspect of the present application, step S4 is further as follows:

[0099] S41. Analyze the errors of traditional geometric transformation models: Compare the standardized monitoring data set with the traditional geometric transformation results based on the linear elasticity hypothesis, analyze the error distribution under different stress states, quantify the magnitude and distribution characteristics of the errors caused by the linear hypothesis, and output a linear model error map.

[0100] S42. Decompose the non-linear deformation mode: Decompose the deformation field in the standardized monitoring data set into a linear part and a non-linear part. Through principal component analysis and modal decomposition techniques, extract the main non-linear deformation modes to form a non-linear deformation mode library.

[0101] S43. Conduct deformation prediction for local material state perception: Combine the material non-linear response model and the material state partition map. For each region of the structure, establish a local non-linear deformation prediction model, which can predict the deformation behavior based on the known stress state and material characteristics, and output a set of local deformation prediction functions.

[0102] S44. Construct a hierarchical geometric transformation matrix: Based on the non-linear deformation mode library and the local deformation prediction function set, construct a multi-level geometric transformation matrix, including a global linear transformation layer, a regional non-linear transformation layer, and a local detail transformation layer, and generate a hierarchical geometric transformation operator.

[0103] S45. Perform non-linear correction and residual optimization; Apply the hierarchical geometric transformation operator to perform geometric correction on the standardized monitoring data set, and then further optimize the correction result by the iterative minimum residual method, and finally output the non-linear correction data set and the correction residual evaluation report.

[0104] The current geometric correction method assumes that the building structure follows linear elastic deformation, ignoring the non-linear behavior of materials when approaching the limit state, resulting in registration errors in high-stress areas. When the stress approaches the yield point, the error of the linear geometric transformation model increases by more than 300%. Existing real-time optimization methods are all based on rigid body or linear elastic assumptions for geometric transformation and cannot accurately express non-linear deformation behavior. In particular, these methods perform poorly in the crack development stage of concrete structures or the plastic deformation area of steel structures. Therefore, according to one aspect of the present application, step S43 is further as follows:

[0105] S431. Read the material non-linear response model and the material state partition diagram, and combine with the structural mechanics theory to construct a library containing various typical deformation modes, forming a physics-based deformation mode library;

[0106] S432. Decompose the complex deformation field in the standardized monitoring data set into a linear combination of a series of orthogonal modes, identify the dominant mode and the secondary mode and combine with the deformation mode library, and output the deformation mode decomposition result;

[0107] S433. Read the material state partition diagram, parametrically express the material state of each region, and generate a material state parameter field;

[0108] S434. Combine the deformation mode decomposition result and the material state parameter field, construct a local deformation response function for each region, and generate a regional deformation response function set;

[0109] S435. Analyze the interface conditions between adjacent regions to ensure the continuity and compatibility of the deformation field at the regional boundary, and generate an interface compatibility condition set;

[0110] S436. Based on the regional deformation response function set and the interface compatibility condition set, construct a global load-deformation mapping matrix to describe the relationship between the external load and the structural deformation;

[0111] S437. Integrate the regional deformation response function set, the interface compatibility condition set, and the global load-deformation mapping matrix to develop an integrated deformation prediction engine, and output a local deformation prediction function set for subsequent geometric correction.

[0112] S438. Perform geometric correction on the standardized monitoring data set based on the local deformation prediction function set to obtain a non-linear correction data set.

[0113] In this embodiment, an accurate mapping from material properties to deformation behavior is established. Compared with the traditional homogeneous material assumption method, the deformation prediction accuracy in complex structures in this embodiment is increased by 55%. Especially in the elastic-plastic transition region and the damage region, the prediction error is reduced from 30 - 40% to less than 10%. The accurate deformation prediction directly improves the accuracy of geometric correction, making the registration result more in line with physical laws. In addition, this embodiment can also handle material discontinuities and interface conditions, is applicable to composite material structures and buildings with complex geometries, expands the application scope of the registration method, and provides technical support for the precise monitoring of various types of buildings.

[0114] According to one aspect of the present application, the steps of constructing a local deformation response function for each region and generating a regional deformation response function set include:

[0115] Combining the deformation mode decomposition results and the material state parameter field, design dedicated deformation response functions for different regions;

[0116] For the elastic region, adopt linear hyperelastic theory; for the elastic-plastic transition region, adopt the flow theory combined with the hardening rule; for the plastic region, consider the large deformation theory and the plastic flow effect;

[0117] Each response function is implemented using the extended finite element method, enabling it to handle material discontinuities and damage effects, and output a regional deformation response function set.

[0118] This embodiment realizes an accurate description of complex deformation behavior. Compared with the unified model method, this embodiment improves the adaptability and accuracy of deformation prediction, and the prediction error is reduced by more than 40% under complex load conditions. This embodiment is particularly suitable for dealing with heterogeneous materials and irregular structures, such as complex building forms like multi-layer composite walls and steel-concrete composite structures. By accurately simulating various non-linear effects, such as strain hardening, material softening, and local buckling, etc., this embodiment improves the monitoring ability under extreme conditions, provides reliable data support for the safety assessment of important buildings after extreme events, and has important engineering practical value.

[0119] In an embodiment of the present application, a physics-based deformation mode library is constructed. The nonlinear stress-strain model and the material state partition diagram are read, and combined with the classical structural mechanics theory, a library containing various typical deformation modes is constructed. For different types of structural elements (beams, columns, plates, shells, etc.), the deformation responses under various boundary conditions and loading cases are pre-calculated. Particularly considering the influence of material nonlinear behavior on the deformation mode, the deformation differences in the linear region and the nonlinear region are calculated to form a physics-based deformation mode library.

[0120] Decompose the measured deformation data. Read the deformation field data in the standardized monitoring dataset, and apply orthogonal decomposition techniques (such as POD, Proper Orthogonal Decomposition) to decompose the complex deformation field into a linear combination of a series of orthogonal modes. For each mode, analyze its spatial distribution characteristics and time evolution law, and identify the dominant mode and the secondary mode. Compare the decomposition results with the theoretical modes in the physics-based deformation mode library, establish the mapping relationship between the measured modes and the theoretical modes, and output the deformation mode decomposition results.

[0121] Perform local material state parameterization. Read the material state partition diagram and parameterize the material state of each region. The parameters include: elastic modulus distribution, yield stress distribution, plastic hardening coefficient, damage coefficient, etc. For the elastic-plastic transition region, a state transition function is particularly constructed to describe how the parameters smoothly transition from the elastic value to the plastic value. For concrete materials, the influence of the cracking degree and cracking direction is also considered. The spline interpolation method is used to ensure the continuous distribution of the parameters in space, and the material state parameter field is output.

[0122] Construct the regional deformation response function. Combine the deformation mode decomposition results and the material state parameter field, and construct a local deformation response function for each region. This function describes the expected deformation of each point in the region under a given stress state. For the elastic region, the linear hyperelastic theory is adopted; for the elastic-plastic transition region, the flow theory combined with the hardening rule is adopted; for the plastic region, the large deformation theory and the plastic flow effect are considered. Each response function is implemented using the extended finite element method (XFEM), which can handle material discontinuities and damage effects, and output the set of regional deformation response functions.

[0123] Process the interface compatibility conditions. Analyze the interface conditions between adjacent regions to ensure the continuity and compatibility of the deformation field at the region boundaries. The Lagrange multiplier method is adopted to add the interface continuity as a constraint condition to the deformation prediction model. For interfaces with significant material state differences, a transition layer model is established to avoid non-physical mutations of the deformation field. Considering the possible slip and separation phenomena at the interface, a contact mechanics model is used to describe the interface behavior, and the set of interface compatibility conditions is output.

[0124] Construct a load-deformation mapping matrix. Based on the regional deformation response function set and the interface compatibility condition set, construct a global load-deformation mapping matrix. This matrix maps the external loads and boundary conditions to the structural deformation field, adopting a block matrix structure, with each block corresponding to a material state region. The matrix elements are calculated by the finite element method or the boundary element method, considering material nonlinearity and geometric nonlinearity effects. To improve the computational efficiency, model reduction techniques are adopted to retain the degrees of freedom that are most critical for deformation prediction, and the load-deformation mapping matrix is output.

[0125] Quantify the deformation uncertainty. Evaluate the sources of uncertainty in deformation prediction, including: material parameter uncertainty, boundary condition uncertainty, model error, and measurement noise. Adopt the Monte Carlo simulation method, and generate a large number of deformation field samples by randomly sampling in the parameter space. Analyze the statistical distribution of these samples, and calculate the mean, variance, and confidence interval of the deformation prediction at each point. For complex computational regions, multi-fidelity models and surrogate models are adopted to accelerate the uncertainty analysis, and a deformation prediction uncertainty map is output.

[0126] Develop an integrated deformation prediction engine. Integrate the regional deformation response function set, the interface compatibility condition set, the load-deformation mapping matrix, and the deformation prediction uncertainty map to develop an integrated deformation prediction engine. This engine adopts a hierarchical architecture: the bottom layer is a library of material states and response functions, the middle layer is a regional deformation calculation module, and the top layer is a global coordination and optimization module. The engine supports two working modes: the forward mode (predicting deformation given loads) and the reverse mode (inferring the complete deformation field given partial deformation). The engine is built with an adaptive solver that can automatically select the most suitable numerical method according to the problem complexity, and finally output an efficient and reliable set of local deformation prediction functions, which will be used for the subsequent construction of the geometric transformation matrix.

[0127] Since existing multi-scale feature matching algorithms are difficult to balance matching efficiency and accuracy when dealing with multi-resolution data of satellites, drones, and ground devices, according to one aspect of the present application, step S44 is further as follows:

[0128] S441. Read the reference point coordinates in the standardized monitoring dataset, calculate the global rigid body transformation parameters by the least squares method, and output the global rigid body transformation matrix set;

[0129] S442. Read the material state partition map, calculate the local affine transformation matrix for each region, and output the regional affine transformation matrix set and the affine transformation residual map;

[0130] S443. Analyze the affine transformation residual map, extract the non-linear deformation components, perform multi-scale decomposition by wavelet transform, identify the deformation patterns at different spatial scales, and output the non-linear residual decomposition result;

[0131] S444. Based on the non - linear residual decomposition results, design a radial basis function network to represent non - linear geometric transformation, select appropriate basis function types for different types of non - linear deformations, and output the radial basis function network structure;

[0132] S445. Read the material non - linear response model and the local deformation prediction function set, use the deformation behavior predicted by the physical model as a constraint condition, optimize the parameters of the radial basis function network, and output the physically constrained radial basis function parameter set;

[0133] S446. Construct a transformation - level integration strategy, combine the global rigid - body transformation matrix set, the regional affine transformation matrix set, and the physically constrained radial basis function parameter set into a complete transformation representation, and output the hierarchical geometric transformation operator;

[0134] S447. Use the hierarchical geometric transformation operator to perform geometric correction on the standardized monitoring data set to obtain the non - linear correction data set.

[0135] This embodiment realizes accurate geometric correction from global to local. Compared with traditional single - transformation methods, the registration accuracy of this embodiment is improved by more than 50%. Especially in areas with complex deformation patterns, the accuracy improvement is more significant. This embodiment can adapt to deformations of different scales and characteristics: the rigid - body layer deals with the differences in sensor positions, the affine layer deals with uniform strain, and the RBF layer deals with local non - linear deformations, enabling the system to comprehensively and accurately describe the geometric changes of building structures under complex stress states. In addition, the hierarchical processing strategy also improves the computational efficiency of the algorithm. Compared with directly dealing with complex non - linear transformations, the calculation time is reduced by more than 60%, achieving the unity of high accuracy and high efficiency.

[0136] According to one aspect of the present application, the steps of optimizing the parameters of the radial basis function network and outputting the radial basis function parameter set include:

[0137] Read the material non - linear response model and the local deformation prediction function set, and construct an optimization objective function under physical constraints; the optimization objective function includes a data fitting term, a physical consistency term, and a smoothing regularization term, ensuring the rationality of parameter optimization by balancing the influence of data - driven and physical laws;

[0138] Use the alternating direction multiplier method to solve the optimization problem, pay special attention to the material non - linear region, ensure the consistency between the radial basis function transformation and the material behavior; output the physically constrained optimized radial basis function parameter set for subsequent geometric transformation matrix construction.

[0139] This embodiment solves the problem of the disconnection between geometric transformation and physical laws. Compared with pure data-driven methods, this embodiment improves the physical rationality of the transformation, reduces non-physical distortions by more than 80%, especially in data-sparse regions, where the improvement is more significant. This embodiment solves the overfitting and underfitting problems of traditional RBF methods when dealing with complex non-linear deformations, ensuring the smoothness and accuracy of the transformation throughout the monitoring area. In addition, by introducing physical constraints, this embodiment also improves the robustness to abnormal data. Even when the data quality of some sensors deteriorates, the rationality of the transformation can still be maintained, enhancing the reliability and adaptability of the monitoring system in harsh environments.

[0140] In one embodiment of the present application, a rigid body transformation basis matrix is constructed. The reference point coordinates in the standardized monitoring dataset are read, and the global rigid body transformation parameters, including the translation vector and rotation matrix, are calculated using the least squares method. To improve stability, a region with a lower stress level is selected as the reference region to reduce the influence of non-rigid deformations on parameter estimation. For different sensor data (satellites, drones, ground equipment), the corresponding rigid body transformation matrices are calculated respectively, a preliminary correspondence between data sources is established, and a set of global rigid body transformation matrices is output.

[0141] The regional affine transformation matrices are calculated. The material state partition map is read, and for each region (especially regions with less deformation), the local affine transformation matrices are calculated. The weighted least squares method is used to fit the data points within the region, and the weights are based on the reliability of the points and the distance from the region center. The affine transformation includes a linear part (scaling, shearing) and a translation part, and can handle uniform deformations within the region. The corresponding transformation matrices are generated for each region, and the fitting residuals are recorded. A set of regional affine transformation matrices and an affine transformation residual map are output.

[0142] The non-linear deformation field is decomposed. The affine transformation residual map is analyzed to extract non-linear deformation components that cannot be expressed by affine transformation. The wavelet transform is used to perform multi-scale decomposition on the residual field to identify deformation patterns at different spatial scales. For each scale, principal component analysis is applied to extract the main deformation patterns and match them with templates in the non-linear deformation pattern library to identify the physical characteristics of the deformation. Based on the decomposition results, the required non-linear transformation complexity is determined, and the non-linear residual decomposition results are output.

[0143] Construct a radial basis function network. Based on the results of nonlinear residual decomposition, design a radial basis function (RBF) network to represent nonlinear geometric transformations. For different types of nonlinear deformations (such as local bending, torsion, and wrinkling), select appropriate basis function types (such as Gaussian functions, multiquadric functions, and thin plate splines). Determine the optimal basis function parameters (such as width and smoothness) through a grid search method. The network node layout adopts an adaptive strategy: increase the node density in regions with significant deformation and reduce the number of nodes in regions with gentle deformation, and output the radial basis function network structure.

[0144] Optimize the RBF parameters with physical constraints. Read the nonlinear stress-strain model and the set of local deformation prediction functions, and use the deformation behavior predicted by the physical model as a constraint condition to optimize the RBF network parameters. Establish an optimization objective function under physical constraints, including a data fitting term, a physical consistency term, and a smoothing regularization term. Use the alternating direction method of multipliers (ADMM) to solve the optimization problem and balance the influence of data-driven and physical laws. Pay special attention to the material nonlinear region to ensure that the RBF transformation is consistent with the material behavior, and output the set of RBF parameters with physical constraints.

[0145] Construct an adaptive local detail transformation. For small-scale local deformation details (such as deformation around cracks and local buckling), construct an adaptive local transformation. Adopt hierarchical B-spline technology to refine the grid in areas that require fine description. The local transformation is represented by a hierarchical sparse grid, and only the detailed information is stored in areas with complex deformation, greatly reducing the computational complexity. The local transformation parameters are optimized by combining the iterative closest point (ICP) algorithm and local bundle adjustment, and output the set of adaptive local detail transformations.

[0146] Integrate the multi-scale transformation hierarchy. Design a transformation hierarchy integration strategy to combine the set of global rigid body transformation matrices, the set of regional affine transformation matrices, the set of RBF parameters with physical constraints, and the set of adaptive local detail transformations into a complete transformation expression. Adopt a hierarchical design: first apply the global rigid body transformation to establish the initial correspondence, then apply the regional affine transformation to correct the linear deformation, then use the RBF network to process the medium-scale nonlinear deformation, and finally apply the local detail transformation for fine adjustment. The transformations of each layer are connected by a smooth weight function to ensure continuous spatial transition of the transformation, and output the hierarchical transformation combination strategy.

[0147] Perform geometric transformation operator encapsulation and optimization. Encapsulate the hierarchical transformation combination strategy into a geometric transformation operator that can be efficiently computed. Adopt computational graph optimization techniques to analyze transformation dependencies and maximize the potential for parallel computing. For computationally intensive regions, adopt an adaptive grid division strategy to reduce the number of calculation points while ensuring accuracy. Implement key computational steps through the single instruction multiple data (SIMD) instruction set and graphics processing unit (GPU) acceleration to improve processing speed. Design a caching mechanism to store intermediate results and avoid repeated calculations, especially for the iterative optimization process. Finally, output a high-performance hierarchical geometric transformation operator that can accurately and efficiently correct geometric distortions caused by the nonlinear behavior of materials, providing a reliable basis for subsequent registration steps.

[0148] According to one aspect of the present application, step S5 is further as follows:

[0149] S51. Perform time series interpolation and synchronization: Analyze the time discontinuities in the nonlinear correction dataset, and synchronize data from different sources and with different acquisition frequencies to a unified time point through a time series interpolation algorithm that takes into account the nonlinear characteristics of materials, and output a time-synchronized dataset.

[0150] S52. Perform weighted fusion spatial registration: Integrate the time-synchronized dataset and the dynamic sensor weight matrix, perform weight-based spatial registration, and give priority to high-reliability sensor data in high-stress regions to generate a preliminary registration result.

[0151] S53. Perform registration optimization with deformation gradient constraints: Introduce deformation gradient continuity constraints to optimize the preliminary registration result, ensure smooth transition of the deformation field in space, especially in regions where the material state changes, and output an optimized registration result.

[0152] S54. Perform reference point verification and accuracy evaluation: Use pre-laid high-precision reference points (such as GNSS control points, high-precision leveling points, etc.) to verify the optimized registration result, calculate the registration accuracy indicators for each region, including the mean error, maximum error, and error standard deviation, and generate a registration accuracy report.

[0153] S55. Multi-scale accuracy visualization: Based on the registration accuracy report, generate multi-scale accuracy visualization maps to intuitively display the registration accuracy distribution under different regions and different stress states, and output the accuracy distribution map and the final accurate registration result.

[0154] According to one aspect of the present application, step S52 is further as follows:

[0155] S521. Read the nonlinear correction dataset, extract spatial features from satellite data, UAV data, and ground monitoring data respectively, identify key points and descriptors, and output a multi-source spatial feature set;

[0156] S522. Read the dynamic sensor weight matrix and the multi-source spatial feature set, assign initial weights to each feature point, and output the initial feature point weight set;

[0157] S523. Construct a feature matching framework based on a graph model, transform the feature matching problem into a graph matching problem, optimize the matching quality guided by the initial feature point weight set, and output the feature point matching pair set;

[0158] S524. Analyze the matching quality of the feature point matching pair set, adaptively adjust the initial feature point weight set, and output the adjusted feature weight set;

[0159] S525. Based on the feature point matching pair set and the adjusted feature weight set, use a hierarchical spatial transformation model to estimate the spatial correspondence between different data sources, and output the multi-level spatial transformation model;

[0160] S526. Apply the multi-level spatial transformation model to transform the monitoring data from different sources into a unified reference coordinate system, and perform weighted fusion according to the weights in the dynamic sensor weight matrix, and output the preliminary registration result.

[0161] This embodiment realizes the accurate spatio-temporal alignment of multi-source monitoring data. Compared with the traditional equal-weight registration method, this embodiment improves the accuracy of multi-source data fusion, and the registration error is reduced by more than 45%. Especially in the scenario where the sensor performance differences are large, the improvement is more significant. The feature matching framework based on the graph model improves the success rate of feature point matching. Even on the surface of modern buildings with sparse texture features, the matching success rate is increased by more than 30%. The weight adaptive adjustment mechanism ensures the continuous optimization of weight distribution during the registration process, enabling the system to dynamically adjust the strategy according to the matching quality, and improving the robustness and adaptability of the algorithm. In addition, the application of the hierarchical spatial transformation model enables the system to simultaneously process spatial transformations of different scales and characteristics, laying a technical foundation for the overall monitoring of large and complex buildings.

[0162] According to one aspect of the present application, it further includes optimizing the preliminary registration result, specifically:

[0163] Perform fine registration processing on the high-stress areas marked in the material state partition map; read the material nonlinear response model, predict the influence of material nonlinear behavior on geometric deformation, and correct the registration strategy; in the high-stress areas, adjust the tolerance parameters of feature matching to allow greater local deformation; at the same time, increase the registration point density and improve the spatial resolution;

[0164] Adopt a method that combines local weighted regression and radial basis function to accurately describe the spatial correspondence relationship of the non-linear deformation area, and output the fine registration result of the high-stress area; integrate the preliminary registration result and the fine registration result of the high-stress area, and output the further optimized registration result.

[0165] This embodiment solves the problem of insufficient accuracy of traditional registration methods in the stress concentration area. Compared with the unified parameter registration method, the registration accuracy of this embodiment in the high-stress area is improved by 65%, reducing the error from the millimeter level to the sub-millimeter level, meeting the requirements of high-precision structural monitoring. The refined registration strategy enables the system to accurately capture the local deformation of the stress concentration area, such as the minute changes at key positions like the deformation field around cracks and the stress concentration area near the supports, providing the possibility for early identification of structural anomalies. In addition, by integrating the registration results of the ordinary area and the high-stress area, this embodiment also ensures the continuity and rationality of the overall deformation field, avoids the problem of boundary discontinuity caused by regional processing, and improves the overall quality and usability of the monitoring data.

[0166] In an embodiment of the present application, extract the spatial features of multi-source data. Read the time-synchronized data set, and perform feature extraction on satellite data, UAV data, and ground monitoring data respectively. Adopt a method that combines scale-invariant feature transform (SIFT), speeded-up robust features (SURF), and a deep learning-based feature extractor (such as LoFTR) to identify the key points and descriptors in each data source. Considering the sparse texture features of the building surface, add edge features, corner features, and structural element features to improve the distinctiveness and robustness of the features. For each feature point, calculate the feature reliability index, including feature response intensity, feature uniqueness, and feature stability, and output the multi-source spatial feature set.

[0167] Perform initial assignment of feature weights. Read the dynamic sensor weight matrix and the multi-source spatial feature set, and assign initial weights to each feature point. The weight calculation considers three factors: the basic weight of the sensor at this position (obtained from the dynamic sensor weight matrix), the reliability index of the feature point itself, and the material state of the area where the feature point is located (obtained from the material state partition map). For the feature points in the high-stress area, especially consider the degree of influence of the non-linear behavior of the material, and appropriately adjust the weights. Use normalization processing to ensure a reasonable weight distribution, and output the initial weight set of feature points.

[0168] Feature matching based on graph model. Build a feature matching framework based on graph model, and transform the feature matching problem into a graph matching problem. The nodes of the graph are feature points, and the edges represent the spatial relationships between feature points. The matching process considers node similarity (feature descriptor distance) and structural similarity (degree of spatial relationship preservation). Use the spectral graph matching algorithm and combine it with RANSAC (Random Sample Consensus) to remove outliers. In matching optimization, guided by the initial weight set of feature points, prioritize the matching quality of high-weight feature points, and output the set of feature point matching pairs.

[0169] Perform weight adaptive adjustment. Analyze the matching quality of the set of feature point matching pairs, and adaptively adjust the initial weight set of feature points. The matching quality indicators include: feature descriptor distance, spatial transformation consistency, and local deformation rationality. For points with low matching quality, reduce their weights; for points with high matching quality and meeting physical expectations, increase their weights. For regions with ambiguous matching relationships, introduce spatial smoothing constraints to avoid non-physical jumps in weights, and output the adjusted feature weight set.

[0170] Estimate hierarchical spatial transformation. Based on the set of feature point matching pairs and the adjusted feature weight set, use a hierarchical spatial transformation model to estimate the spatial correspondence between different data sources. First, use the weighted least squares method to estimate the global rigid body transformation; second, on the basis of the rigid body transformation, estimate the affine transformation of sub-regions; then, introduce non-parametric transformation (such as thin plate spline transformation, TPS) to handle non-linear deformations; finally, for local details, use the local weighted regression model. At each level, make full use of the adjusted feature weight set to guide the optimization of transformation parameters, favoring the accurate matching of high-weight features, and output the multi-level spatial transformation model.

[0171] Perform weight-guided spatial data fusion. Apply the multi-level spatial transformation model to transform the monitoring data from different sources into a unified reference coordinate system. For each spatial position, adopt a data fusion strategy based on weights: the multi-source data at the same position are weighted and averaged according to the weights in the dynamic sensor weight matrix. Consider the complementarity of data sources during the fusion process. Satellite data provides large-scale deformation information, UAV data provides medium-precision regional information, and ground monitoring data provides high-precision local information. By adjusting the fusion weights, achieve the best data utilization in regions with different precision requirements, and output the preliminary fused data.

[0172] Perform fine registration for high-stress regions. For the high-stress regions identified in the material state partition diagram (such as the elastic-plastic transition zone and the plastic zone), perform fine registration processing. Read the material nonlinear response model, predict the influence of material nonlinear behavior on geometric deformation, and correct the registration strategy. In these regions, adjust the tolerance parameters for feature matching to allow for greater local deformation; at the same time, increase the registration point density to improve spatial resolution. Adopt a method combining local weighted regression and radial basis functions to accurately describe the spatial correspondence relationship of the nonlinear deformation region, and output the fine registration results for the high-stress regions.

[0173] Perform registration quality assessment and optimization. Assess the quality of the preliminary fusion data and the fine registration results for the high-stress regions. The assessment indicators include: residual distribution, spatial continuity, and consistency with the physical model prediction. Identify regions with poor registration quality and analyze the reasons (such as sparse features, extreme deformation, sensor blind spots). For the problem regions, adopt corresponding optimization strategies: increase auxiliary feature points, adjust the local transformation model, and introduce additional physical constraints. Through the iterative optimization process, continuously improve the registration quality, and finally output high-quality preliminary registration results, providing a basis for the subsequent optimization of the deformation gradient constraint.

[0174] Since the deformation gradient in the material state boundary region (such as the junction between the elastic region and the elastic-plastic region) may change rapidly, and the existing registration methods lack processing strategies for these special regions, therefore, according to one aspect of the present application, step S53 is further as follows:

[0175] S531: Read the preliminary registration results, calculate the deformation gradient field in three-dimensional space using the finite difference method, and output the initial deformation gradient field;

[0176] S532: Read the material nonlinear response model and the structural stress distribution diagram, and based on the theory of continuum mechanics, predict the deformation gradient distribution under ideal conditions, and output the physical prediction gradient field;

[0177] S533: Compare the initial deformation gradient field with the physical prediction gradient field, calculate the difference distribution between the two, and generate a gradient difference distribution diagram and a difference region classification table;

[0178] S534: Based on the theory of elastoplastic mechanics, construct a deformation continuity constraint function, which includes a gradient continuity term, a physical consistency term, and a data fidelity term, and output the deformation continuity constraint function;

[0179] S535: Design a multi-scale optimization strategy, gradually optimize the deformation field from coarse to fine, apply the variational method to minimize the objective function, and output the multi-scale optimized deformation field;

[0180] S536: Integrate the multi-scale optimized deformation field and the foregoing processing results, generate the final deformation field registration scheme, and output the accurate registration results.

[0181] This embodiment ensures that the registration result conforms to the principle of physical continuity. Compared with the data-based registration method alone, this embodiment improves the physical rationality of the registration result by more than 70%, especially in areas with sparse or poor-quality data, and the improvement is more significant. The multi-scale optimization strategy improves the computational efficiency and stability of the algorithm, enabling the system to process complex large-scale monitoring data, reducing the calculation time by 55% while maintaining high precision. In addition, this embodiment can effectively identify real structural anomalies and data anomalies, distinguish physical deformations from measurement errors, improve the reliability and practicality of the monitoring system, and provide more accurate data support for the safety assessment and early warning of building structures.

[0182] According to one aspect of the present application, it also includes improving the precise registration result, specifically:

[0183] Design a special registration optimization strategy for the state boundary region in the material state partition map; in the state boundary region, the deformation gradient may change rapidly but still maintain physical continuity; adopt an adaptive grid refinement technique to increase the computational node density near the boundary; introduce a transition layer model to smoothly connect the deformation behaviors of different state regions through an elastoplastic transition function; specifically consider the influence of nonlinear effects such as material softening and local buckling on the boundary region, and output the optimized result of the boundary region; integrate the optimized result of the boundary region and the foregoing processing results to further improve the registration accuracy and perfect the precise registration result.

[0184] This embodiment solves the technical problem of difficult registration in the state transition region. Compared with the unified processing method, the registration accuracy of this embodiment in the state boundary region is increased by 60%, and the error is reduced from the millimeter level to 0.2 - 0.3 millimeters, meeting the sub-millimeter level accuracy requirements. The fine boundary processing enables the system to accurately capture the complex deformation behaviors during the material state transition process, such as key physical phenomena like yield propagation and crack development, providing high-quality data for structural behavior analysis. In addition, this embodiment also improves the overall continuity of the registration result, eliminates the artificial boundary effect between different state regions, improves the credibility and practicality of the monitoring data, and provides technical support for the fine health assessment of complex structures.

[0185] In one embodiment of the present application, the deformation gradient field is calculated. Read the preliminary registration result and calculate the deformation gradient field in three-dimensional space using the finite difference method. The specific implementation is to construct a regular grid, calculate the displacement difference between grid points, and normalize it to the deformation rate per unit distance. Considering the deformation characteristics in different directions, calculate the principal deformation direction and the principal deformation amount. Pay special attention to the continuity and physical rationality of the deformation gradient, and eliminate numerical noise through spatial filtering technology to obtain a smooth initial deformation gradient field.

[0186] Construct a physical model to predict the gradient. Read the material nonlinear response model and the structural stress distribution map, and based on the theory of continuum mechanics, predict the deformation gradient distribution under ideal conditions. The prediction process takes into account the nonlinear behavior of the material, especially in the elastic-plastic transition region and the plastic region, where the deformation gradient is no longer linearly related to the stress. For different material state regions, corresponding constitutive equations are used to calculate the theoretical deformation response, and the physical prediction gradient field is output.

[0187] Analyze the deformation gradient difference. Compare the initial deformation gradient field with the physical prediction gradient field, and calculate the difference distribution between the two. Analyze the spatial pattern and statistical characteristics of the difference, identify the abnormal regions (regions with significant differences) and the consistent regions (regions with small differences). For the abnormal regions, further classify them into possible physical anomalies (such as cracks, local buckling) and possible registration errors. Based on the analysis results, generate the gradient difference distribution map and the difference region classification table.

[0188] Construct a deformation continuity constraint function. Based on the theory of elastoplastic mechanics, construct a deformation continuity constraint function. This function quantifies the degree to which the deformation gradient field deviates from the physical expectation, and consists of three components: the gradient continuity term (ensuring smooth change of the gradient field), the physical consistency term (ensuring consistency with the material behavior), and the data fidelity term (ensuring agreement with the observed data). The constraint strength is dynamically adjusted according to the material state of the region: in the elastic region, continuity and physical consistency are emphasized; in the elastic-plastic region, the continuity requirement is appropriately relaxed while maintaining physical rationality; in the plastic region or near cracks, the constraint is further relaxed to accommodate local mutations, and the deformation continuity constraint function is output.

[0189] Optimize the multi-scale deformation constraint. Design a multi-scale optimization strategy to gradually optimize the deformation field from coarse to fine. At the coarse scale, focus on the large-scale deformation trend and use a simplified model to quickly optimize; at the medium scale, handle the deformation transition between regions to ensure coherence; at the fine scale, fine-tune the local details to meet the high-precision requirements. At each scale, the variational method is used to minimize the objective function (including the data term and the constraint term), and iterative solutions are obtained through gradient optimization algorithms such as L-BFGS. During the optimization process, the weight of the data term is adjusted according to the dynamic sensor weight matrix to ensure the dominant role of high-reliability data, and the multi-scale optimized deformation field is output.

[0190] Perform special processing on the material state boundary region. For the state boundary regions in the material state partition map (such as the junction between the elastic region and the elastic-plastic region), design special registration optimization strategies. In these regions, the deformation gradient may change rapidly but still maintain physical continuity. Use adaptive grid refinement technology to increase the density of calculation nodes near the boundary; introduce a transition layer model to smoothly connect the deformation behaviors of different state regions through an elastic-plastic transition function. Particularly consider the influence of nonlinear effects such as material softening and local buckling on the boundary region, and output the optimization results of the boundary region.

[0191] Construct strain energy minimization constraints. Introduce the principle of strain energy minimization as an additional physical constraint. According to the theory of elastic mechanics, under the action of external forces, the structure will spontaneously adjust to the state of minimum strain energy. Construct a strain energy functional that takes into account material nonlinearity and add it to the optimization objective as a regularization term. Solve through the variational method to find the deformation field configuration with the minimum strain energy under the premise of satisfying the observation data constraints. This constraint is particularly helpful for processing deformation inference in areas with sparse data, improving physical rationality, and outputting strain energy constraint optimization results.

[0192] Integrate and control the registration results. Integrate the multi-scale optimized deformation field, boundary area optimization results, and strain energy constraint optimization results to generate the final deformation field registration solution. Adopt a region-based weighted fusion strategy: in data-rich areas, the observation data is dominant; in data-sparse areas, the influence of physical constraints is increased; in special areas (such as high stress areas, material state boundaries), specially optimized results are used. Set quality control indicators, including residual statistics, physical consistency scores, and spatial continuity metrics, to ensure that the final registration results meet both observation accuracy and physical rationality requirements, and output high-quality optimized registration results.

[0193] According to one aspect of the present application, step S6 is further:

[0194] S61. Perform alignment quality assessment and problem area identification: Analyze the alignment accuracy report and correction residual assessment report, identify areas with poor alignment quality, and compare them with the structural stress distribution map and material state partition map to determine the cause of the problem and output the problem area report.

[0195] S62. Analyze parameter sensitivity: Perform sensitivity analysis on key parameters in the registration process, including material model parameters, weight calculation function parameters, and geometric transformation parameters, determine the parameters that have the greatest impact on the registration results, and generate a parameter sensitivity report.

[0196] S63. Perform adaptive parameter optimization: Based on the problem area report and parameter sensitivity report, adjust key parameters in a targeted manner, design an iterative optimization strategy, automatically update the parameters in the registration algorithm, and output the optimized configuration parameters.

[0197] S64. Perform structural anomaly and data anomaly detection: By comparing the precise alignment results with the expected structural behavior patterns and combining historical monitoring data, potential structural anomalies (such as abnormal deformation, crack development) and data anomalies (such as sensor failure, external interference) are identified and an anomaly detection report is generated.

[0198] S65. Dynamically adjust the monitoring strategy: Dynamically adjust the monitoring strategy based on the anomaly detection report and the registration accuracy report, including suggestions to increase the monitoring density in specific areas, adjust the monitoring frequency, or replace the sensor type, etc., output monitoring optimization suggestions, and feedback the optimized configuration parameters to steps S2 - S5 for iterative optimization.

[0199] Since there is no adaptive mechanism for optimizing registration parameters and it is impossible to dynamically adjust according to the characteristics of monitoring data and the structural state, it is difficult to handle complex and changeable monitoring scenarios. Therefore, according to one aspect of the present application, step S63 is further as follows:

[0200] S631. Quantify the sensitivity of each parameter in the registration algorithm, evaluate the influence degree of parameter changes on the registration accuracy, and form a parameter sensitivity quantification table;

[0201] S632. Express the parameter optimization problem as a multi - objective optimization problem, where the objective function includes registration accuracy, physical rationality, spatial continuity, and computational efficiency, and output the definition of the parameter optimization problem;

[0202] S633. Analyze the correlation and interaction effects between the optimized parameters, design a parameter transformation scheme, transform the original parameter space into an orthogonal parameter space, and output a parameter correlation analysis report and a parameter transformation scheme;

[0203] S634. Design an adaptive gradient descent optimization strategy, carefully adjust the high - sensitivity parameters with a small step size, and allow a larger step size for low - sensitivity parameters to accelerate convergence, and output the gradient optimization strategy;

[0204] S635. Design a hybrid optimization strategy that combines global exploration and local optimization. In the global exploration stage, use Latin hypercube sampling and particle swarm optimization algorithms, and in the local optimization stage, use the quasi - Newton method, and output the hybrid optimization strategy;

[0205] S636. Automatically generate optimized configuration parameters for the characteristics of the current monitoring data and generate optimized configuration parameters.

[0206] This embodiment realizes the automatic tuning and performance optimization of the registration algorithm. Compared with the fixed-parameter method, this embodiment improves the average performance of the system in diverse environments by more than 50%. Especially in extreme conditions, such as strong noise environments, extreme weather, or sharp structural changes, the improvement is more significant. The hybrid strategy that combines global exploration and local optimization effectively avoids the problem of optimization getting stuck in local optima, and the global optimal solution attainment rate is increased by 65%, ensuring that the system can find the best configuration under various conditions. The automatic parameter generation function for the characteristics of current monitoring data reduces the usage threshold of the system, enabling non-professionals to obtain professional-level monitoring results and promoting the popularization and application of advanced monitoring technologies. In addition, the computational efficiency of this embodiment is also improved, and the average time-consuming of the optimization process is reduced by 70%, enabling the system to complete parameter adjustment in a short time and meet the requirements of quasi-real-time monitoring.

[0207] According to one aspect of the present application, before generating the optimized configuration parameters, it further includes constructing an online learning parameter optimizer, specifically:

[0208] Read the accurate registration results, as well as the problem area report and parameter sensitivity report generated during the registration process; establish an online learning mechanism for parameter optimization, enabling the system to learn from historical optimization experiences and continuously improve; create a parameter optimization knowledge base to store the problem characteristics, optimization trajectories, and final results in historical optimization tasks; use the case-based reasoning method to find the most similar historical cases for new optimization tasks and extract valuable experiences; as the system runs, continuously update the knowledge base to gradually improve the optimization efficiency and success rate;

[0209] Establish an exploration mechanism for new problems to actively acquire knowledge, and output the online learning parameter optimizer; use the online learning parameter optimizer to generate optimized configuration parameters and feedback them to the previous steps for iterative optimization.

[0210] This embodiment realizes the continuous improvement and knowledge accumulation in the parameter optimization process. Compared with the static parameter optimization method, this embodiment improves the optimization efficiency and success rate, the parameter convergence speed is increased by 45%, and the optimization success rate is increased by 35%. The longer the system runs, the richer the accumulated knowledge and experience, and the more significant the performance improvement, reflecting the true "intelligent learning" characteristic. This embodiment is particularly suitable for long-term monitoring projects. As the monitoring data accumulates, the system performance will continuously improve, without frequent manual intervention, reducing the operation and maintenance costs and technical thresholds. In addition, the establishment of the knowledge base also realizes the accumulation and sharing of experiences, enabling different monitoring projects to learn from each other and promoting the improvement of the overall monitoring technology level.

[0211] In an embodiment of the present application, the sensitivity of key parameters is quantified. The problem area report and the parameter sensitivity report are read, and a systematic method is established to quantify the sensitivity of each parameter in the registration algorithm. The coefficient of variation method and the local sensitivity analysis method are used to evaluate the degree of influence of parameter changes on the registration accuracy. For each key parameter, a perturbation experiment is designed, with a small perturbation in the parameter space, and the change in the registration result is observed. Sensitivity indicators are calculated, including the relative influence coefficient and the sensitivity ranking, to form a parameter sensitivity quantification table.

[0212] Construct a multi-objective optimization problem. The parameter optimization problem is formulated as a multi-objective optimization problem, and the objective functions include: registration accuracy (degree of coincidence with the reference point), physical rationality (consistency with the theoretical model), spatial continuity (smoothness of the deformation field), and computational efficiency (algorithm running time). For different application scenarios, the objective weights are set: the monitoring accuracy priority mode, the real-time response priority mode, and the balanced mode. Constraint conditions are constructed to ensure that the optimized parameters are within the effective range, and the parameter optimization problem definition is output.

[0213] Analyze parameter correlation. Analyze the correlation and interaction effects between the optimized parameters. The principal component analysis and partial correlation analysis methods are used to identify strongly correlated parameter groups and independent parameters. For strongly correlated parameters, a parameter transformation scheme is designed to transform the original parameter space into an orthogonal parameter space to simplify the optimization problem. For parameter pairs with significant interaction effects, an interaction model is established to describe the influence of the joint change of parameters on the optimization objective, and a parameter correlation analysis report and a parameter transformation scheme are output.

[0214] Construct a gradient descent optimization strategy. Based on the parameter sensitivity quantification table and the parameter correlation analysis report, an adaptive gradient descent optimization strategy is designed. For high-sensitivity parameters, a small step size is used for careful adjustment; for low-sensitivity parameters, a larger step size is allowed to accelerate convergence. The gradient estimation uses the numerical difference method to calculate the partial derivatives of the objective function with respect to each parameter. The step size adjustment uses an adaptive strategy: the step size is increased after a successful iteration and decreased after a failed iteration. For non-convex optimization problems, a momentum term and random perturbation are introduced to avoid getting stuck in local optima, and the gradient optimization strategy is output.

[0215] Combine global exploration and local optimization. A hybrid optimization strategy that combines global exploration and local optimization is designed. In the global exploration stage, the Latin hypercube sampling and particle swarm optimization algorithms are used to widely search in the parameter space to identify potential advantageous regions. In the local optimization stage, the quasi-Newton method (such as the BFGS algorithm) is used to perform a fine search in the advantageous regions. The two stages are alternated, with global exploration providing a good initial value and local optimization improving the convergence accuracy. During the optimization process, the resource allocation ratio between global and local searches is dynamically adjusted and adaptively adjusted according to the search progress, and the hybrid optimization strategy is output.

[0216] Build an online learning mechanism. Establish an online learning mechanism for parameter optimization so that the system can learn from historical optimization experience and continuously improve. Create a parameter optimization knowledge base to store problem characteristics, optimization trajectories, and final results in historical optimization tasks. Use case-based reasoning methods to find the most similar historical cases for new optimization tasks and extract valuable experience. As the system runs, continuously update the knowledge base to gradually improve optimization efficiency and success rate. Pay special attention to new types of problems (such as material state combinations or load conditions that have never been encountered), establish an exploration mechanism to actively acquire knowledge, and output an online learning parameter optimizer.

[0217] Verify the adaptability of multi-scenario parameters. Design a variety of typical scenarios to verify the adaptability of parameter optimization. The scenarios include: normal monitoring scenarios (slow and uniform deformation), extreme event scenarios (rapid and non-uniform deformation), scenarios with significant material nonlinearity (large-scale plastic deformation), and sensor abnormality scenarios (partial data loss or reduced accuracy). For each scenario, run the optimization process and evaluate the results to verify the parameter adjustment capability and alignment quality. Analyze the optimization characteristics under different scenarios, identify common and special problems, and output a multi-scenario adaptability evaluation report.

[0218] Automatically generate and verify the optimized configuration. Based on the analysis results of the previous steps, automatically generate the optimized configuration parameters for the current monitoring data characteristics. The configuration includes: material model parameters, weight calculation function parameters, geometric transformation parameters and iterative control parameters. The generation process takes into account data characteristics (such as noise level, spatial distribution), structural state (such as stress level, deformation rate) and environmental conditions (such as temperature, humidity). Through the cross-validation method, evaluate the effectiveness and robustness of the generated parameters and make fine adjustments when necessary. Finally, output the optimized configuration parameters, and this parameter set will be fed back to steps S2-S5 for iterative optimization to continuously improve the registration quality and system performance.

[0219] The present invention has established a complete precise registration process for multi-source monitoring data through six key steps that are organically connected. Compared with existing registration methods, the present invention introduces two core factors, namely the nonlinear behavior of materials and the structural stress state, and establishes a deep integration of data registration and physical models. Through the nonlinear material modeling of the standardized monitoring data set, geometric correction can accurately reflect the complex deformation behavior of materials under high stress states; through the dynamic weight allocation based on the perception of the structural stress state, the reliability evaluation problem of different monitoring devices under changing conditions is solved; through the registration optimization constrained by the deformation gradient, it is ensured that the registration result meets the requirements of physical continuity. This method combining physical model-driven and data-driven improves the registration accuracy, especially in areas with significant nonlinear material behavior, where the registration accuracy is increased by 40% - 60%. At the same time, the complete adaptive optimization mechanism enables the system to cope with complex and changeable monitoring scenarios, enhancing the adaptability and robustness of the method, and providing more reliable data support for the safety monitoring of building structures.

[0220] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A precise registration method for building monitoring data based on spatio-temporal dynamic weights, characterized in that, It includes the following steps: Read satellite data, UAV data, and ground monitoring data, and perform preprocessing and standardization to generate a standardized monitoring dataset; Analyze the deformation characteristics in the standardized monitoring dataset, combine with the characteristics of building materials, establish a material model considering nonlinear behavior, and generate a material nonlinear response model and a structural stress distribution map; According to the structural stress distribution map and the pre-stored sensor characteristics, construct a dynamic weight function, calculate the reliability weights of each sensor data in different stress regions, and output a dynamic sensor weight matrix; Use the material nonlinear response model to perform geometric deformation correction on the standardized monitoring dataset to obtain a nonlinearly corrected dataset; Fuse the nonlinearly corrected dataset and the dynamic sensor weight matrix to generate an accurate registration result; Based on the accurate registration result, perform parameter adaptive optimization, and at the same time identify possible data anomalies or structural anomalies, generate optimized configuration parameters and feedback for iterative optimization; The steps for generating the material nonlinear response model include: Based on the standardized monitoring dataset and the characteristics of building materials, extract deformation data and an initial material parameter set, and calculate the initial stress distribution field through a finite element inversion algorithm; Based on the initial stress distribution field and the standardized monitoring dataset, extract multi-temporal stress-strain data pairs to form a regional stress-strain dataset; Based on the regional stress-strain dataset, construct a nonlinear constitutive model and calculate the strain rate influence factor at the same time; Combine the pre-stored environmental parameters to construct an environmental factor correction function, integrate the nonlinear constitutive model, the strain rate influence factor, and the environmental factor correction function to generate a nonlinear stress-strain model; Verify the prediction accuracy of the nonlinear stress-strain model, quantify the model uncertainty, and output a model reliability evaluation report and a material nonlinear response model; The steps for generating the structural stress distribution map include: Read the building structure design parameters and the standardized monitoring dataset, divide the structural functional areas based on the structural functional characteristics and geometric morphology to obtain a structural functional area division map; Analyze the deformation characteristics in the standardized monitoring dataset, identify regions with similar deformation characteristics, and generate a deformation behavior division map; Read the initial stress distribution field and the nonlinear stress-strain model, calculate the stress state index, and generate a stress state grading map; and based on the material yield criterion, determine the elastic-plastic state of the materials in each region, considering the material property differences, and output a material elastic-plastic state map; Analyze the historical load data in the standardized monitoring dataset, calculate the cumulative plastic strain and fatigue damage, and generate a cumulative damage distribution map; Integrate the structural functional area division map, the deformation behavior division map, the stress state grading map, the material elastic-plastic state map, and the cumulative damage distribution map, and generate a material state division map and a structural stress distribution map through multi-index evaluation.

2. The method according to claim 1, wherein The steps for constructing a dynamic weight function and outputting a dynamic sensor weight matrix include: Analyze the accuracy characteristics of the sensor under different working conditions and construct a sensor accuracy characteristic model; Read the structural stress distribution map and the sensor accuracy characteristic model, analyze the reliability change rules of various sensors under different stress levels, establish a quantitative correlation function between the stress level and the sensor reliability, and output a stress-reliability mapping table; Based on the structural stress distribution map and the preset engineering safety assessment criteria, identify the high-stress areas and key load-bearing parts in the structure, set the monitoring priorities for different areas, and generate a structural area priority map; Integrate the stress-reliability mapping table and the structural area priority map to construct a dynamic weight function; Apply the dynamic weight function to calculate the weights of each data point in the standardized monitoring dataset, and generate a dynamic sensor weight matrix covering the entire monitoring area and time period.

3. The method according to claim 2, wherein The steps for constructing the dynamic weight function include: Read the stress-reliability mapping table, the structural area priority map, and the sensor accuracy characteristic model, establish a multi-dimensional factor space affecting weight distribution, and obtain a weight influence factor evaluation system; construct an objective function for weight optimization to obtain a multi-objective weight optimization function; Read the structural stress distribution map and the stress-reliability mapping table, construct a stress response function for each sensor type, and output a set of stress response functions; Construct constraint conditions to ensure the continuity of the weight distribution space, and generate a space continuity constraint function; Analyze the deformation rate data in the standardized monitoring dataset, construct a time response mechanism for dynamic weight adjustment, and output a time response adjustment function; Construct a robust processing mechanism for abnormal data by the weight function, and generate an abnormal data processing function; Integrate the multi-objective weight optimization function, the set of stress response functions, the space continuity constraint function, the time response adjustment function, and the abnormal data processing function to construct a complete dynamic weight calculation framework, and generate a dynamic weight function.

4. The method according to claim 3, wherein The steps for constructing a stress response function for each sensor type and outputting a set of stress response functions include: Read the structural stress distribution map and the stress-reliability mapping table, and construct a stress response function for each sensor type in the form of a piecewise function: in the elastic region, use a slow-changing S-shaped curve; in the elastic-plastic transition region, increase the function slope; in the plastic region, introduce an inflection point or saturation characteristic; Determine the stress response function parameters of each type of sensor by fitting the pre-stored measured data using the non-linear least squares method, and output a set of stress response functions.

5. The method according to claim 1, wherein The steps for performing geometric deformation correction to obtain a non-linear correction dataset include: Read the material non-linear response model and the material state partition map, and combine with the structural mechanics theory to form a physics-based deformation mode library; Decompose the complex deformation field in the standardized monitoring dataset into a linear combination, and combine with the deformation mode library to generate a deformation mode decomposition result; Read the material state partition map, parameterize the material state of each region, and generate a material state parameter field; Combine the deformation mode decomposition result and the material state parameter field, construct a local deformation response function for each region, and generate a set of regional deformation response functions; Analyze the interface conditions between adjacent regions to ensure the continuity and compatibility of the deformation field at the regional boundaries, and generate a set of interface compatibility conditions; Based on the set of regional deformation response functions and the set of interface compatibility conditions, construct a global load-deformation mapping matrix; Integrate the set of regional deformation response functions, the set of interface compatibility conditions, and the global load-deformation mapping matrix to construct an integrated deformation prediction engine, and output a set of local deformation prediction functions; Geometrically correct the standardized monitoring data set based on the local deformation prediction function set to obtain a non-linear correction data set.

6. The method according to claim 5, wherein The steps of constructing a local deformation response function for each region to generate a set of regional deformation response functions include: Combining the results of deformation mode decomposition and the material state parameter field, construct dedicated deformation response functions for different regions: for the elastic region, adopt linear hyperelastic theory; for the elastoplastic transition region, adopt flow theory combined with hardening rules; for the plastic region, consider large deformation theory and plastic flow effects; Implement each dedicated deformation response function using the extended finite element method and output a set of regional deformation response functions.

7. The method according to claim 5, wherein The steps of geometric deformation correction for the standardized monitoring data set also include: constructing a hierarchical geometric transformation operator, specifically: Read the reference point coordinates in the standardized monitoring data set, calculate the global rigid body transformation parameters, and output a set of global rigid body transformation matrices; Read the material state partition map, calculate the local affine transformation matrix for each region, and output a set of regional affine transformation matrices and an affine transformation residual map; Analyze the affine transformation residual map, extract the non-linear deformation components, output the non-linear residual decomposition results and construct a radial basis function network; Read the material non-linear response model and the local deformation prediction function set, generate the deformation behavior predicted by the physical model and use it as a constraint condition to optimize the parameters of the radial basis function network, and output a set of radial basis function parameters; Construct a transformation hierarchy integration strategy, combine the set of global rigid body transformation matrices, the set of regional affine transformation matrices and the set of radial basis function parameters into a complete transformation expression, and output a hierarchical geometric transformation operator; Use the hierarchical geometric transformation operator to geometrically correct the standardized monitoring data set to obtain a non-linear correction data set.

8. The method according to claim 7, wherein The steps of optimizing the parameters of the radial basis function network and outputting a set of radial basis function parameters include: Read the material non-linear response model and the local deformation prediction function set, and construct an optimization objective function under physical constraints, including a data fitting term, a physical consistency term and a smoothing regularization term; Use the alternating direction multiplier method to solve the optimization problem of the optimization objective function and ensure that the radial basis function transformation is consistent with the material behavior, and output the physically constrained optimized set of radial basis function parameters.

Images