Ground surveying and mapping driven space-time evolution modeling method and system

Abstract

The embodiment of the invention relates to the technical field of ground surveying and mapping data modeling, in particular to a ground surveying and mapping driven space-time evolution modeling method and system.The method comprises the steps that firstly, multi-dimensional correlation features of elevations, coverage types and deformation data in at least three continuous periods are integrated; the limitation of single time point or single parameter analysis is broken through, and the nonlinear space-time coupling rule in earth surface evolution can be accurately captured; secondly, a spatial-temporal model construction method based on dynamic parameter optimization effectively fuses geographic spatial correlation features and coverage evolution trends, so that the model has adaptive characterization capability for a complex geographic process; in addition, by establishing a closed-loop feedback mechanism of prediction results and model parameters, real-time iterative optimization of the model parameters is realized, and long-term prediction stability in different geographical scenes is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of ground surveying and mapping data modeling, and in particular, to a spatio-temporal evolution modeling method and system driven by ground surveying and mapping. Background Art

[0002] The geographical evolution analysis technology is the core means to study the dynamic change law of the earth's surface, and has important application value in the fields of disaster warning, ecological governance and geographical planning. By analyzing the spatio-temporal evolution characteristics of key elements such as elevation change, surface cover migration and deformation, this technology provides a scientific basis for identifying potential geological disaster hazards and predicting the trend of ecological degradation.

[0003] However, traditional geographical evolution analysis mainly relies on single-period remote sensing data or manual surveys, combined with static geographical information systems for overlay analysis. Although it can achieve the assessment of the basic surface state, its static modeling framework is difficult to adapt to the complexity of dynamic geographical processes, restricting the prediction accuracy and decision timeliness. In summary, the existing technology is difficult to adapt to the changes in the dynamic geographical environment, and thus difficult to meet the long-term prediction stability requirements for different geographical scenarios at the present stage. Summary of the Invention

[0004] In order to at least overcome the above deficiencies in the existing technology, one of the objectives of the present invention is to provide a spatio-temporal evolution modeling method and system driven by ground surveying and mapping.

[0005] An embodiment of the present invention provides a spatio-temporal evolution modeling method applied to a spatio-temporal evolution modeling system. The method includes: obtaining a multi-period ground surveying and mapping data set of a target geographical area, where the multi-period ground surveying and mapping data set includes surface surveying and mapping data sets of at least three consecutive surveying periods, and each surface surveying and mapping data set is composed of surface elevation data, surface cover type data and surface deformation monitoring data under the same geographical coordinate system; performing spatio-temporal feature coupling processing on the multi-period ground surveying and mapping data set to generate a spatio-temporal evolution feature set, where the spatio-temporal evolution feature set includes surface deformation features, geographical space association features and surface cover evolution features corresponding to each surveying period; performing dynamic parameter optimization processing on the spatio-temporal evolution feature set based on a preset basic spatio-temporal evolution model to generate a target spatio-temporal evolution model, where the target spatio-temporal evolution model is used to represent the mapping relationship between surface surveying and mapping data and geographical space evolution patterns; calling the target spatio-temporal evolution model to perform evolution prediction processing on the surface surveying and mapping data set of the current surveying period to generate a spatio-temporal evolution prediction result of the target geographical area, where the spatio-temporal evolution prediction result includes surface deformation prediction distribution data and surface cover change trend data; performing dynamic update processing on the target spatio-temporal evolution model according to the spatio-temporal evolution prediction result to obtain an updated spatio-temporal evolution model, and synchronizing the updated spatio-temporal evolution model to the evolution prediction processing of the next surveying period.

[0006] An embodiment of the present invention also provides a spatio-temporal evolution modeling system, including a processor, a memory and a bus connected to the processor; wherein, the processor and the memory complete communication with each other through the bus; the processor is used to call program instructions in the memory to execute the above-mentioned ground mapping-driven spatio-temporal evolution modeling method.

[0007] An embodiment of the present invention also provides a computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, the above-mentioned ground mapping-driven spatio-temporal evolution modeling method is implemented.

[0008] A ground mapping-driven spatio-temporal evolution modeling method and system provided by an embodiment of the present invention realize refined modeling and adaptive prediction capabilities of the geographical evolution process through spatio-temporal coupling analysis and dynamic model optimization mechanism of multi-period ground mapping data.

[0009] First, by integrating multi-dimensional correlation features of elevation, coverage type and deformation data within at least three consecutive periods, the limitations of single time point or single parameter analysis are broken through, and the non-linear spatio-temporal coupling law in ground evolution can be accurately captured. Secondly, based on the spatio-temporal model construction method of dynamic parameter optimization, the geographical space correlation features and coverage evolution trends are effectively integrated, enabling the model to have the adaptive representation ability for complex geographical processes. In addition, by establishing a closed-loop feedback mechanism between the prediction results and the model parameters, the real-time iterative optimization of the model parameters is realized, significantly improving the long-term prediction stability under different geographical scenarios.

[0010] Designed in this way, the problem that traditional static models are difficult to adapt to dynamic geographical environment changes is solved, and through the deep coupling of spatio-temporal features of multi-source data, the associated evolution patterns of potential ground deformation and coverage changes are effectively identified, providing high-confidence decision-making support for applications such as geological disaster warning and ecological evolution analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0012] Figure 1 It is a flowchart of a ground mapping-driven spatio-temporal evolution modeling method provided by an embodiment of the present invention.

[0013] Figure 2 It is a block diagram of a spatio-temporal evolution modeling system provided by an embodiment of the present invention.

[0014] Icon: 100 - Spatiotemporal Evolution Modeling System; 101 - Processor; 102 - Memory; 103 - Bus. Detailed Implementation Manner

[0015] Hereinafter, exemplary embodiments disclosed in the present invention will be described in more detail with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present invention can be more thoroughly understood and the scope of the present invention can be completely conveyed to those skilled in the art.

[0016] To better understand the above technical solutions, the technical solutions of the present invention will be described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solutions of the present invention, rather than limitations on the technical solutions of the present invention. Without conflict, the technical features in the embodiments of the present invention and the embodiments can be combined with each other.

[0017] Figure 1 FIG. is a flowchart of a ground mapping-driven spatiotemporal evolution modeling method according to an embodiment of the present invention, which is applied to a spatiotemporal evolution modeling system and includes steps 101 - 105.

[0018] Step 101: Obtain a multi-period ground mapping data set of a target geographic area, where the multi-period ground mapping data set includes at least three consecutive mapping period surface mapping data sets, and each surface mapping data set is composed of surface elevation data, surface coverage type data, and surface deformation monitoring data under the same geographic coordinate system.

[0019] In this embodiment, the target geographic area is a specific geological area in a mountain landslide monitoring scenario, such as a certain mountain slope area. The multi-period ground mapping data set refers to a plurality of time-series geospatial data sets obtained through periodic monitoring. Each mapping period can be a natural time interval or an event-driven interval. The surface elevation data refers to the digital terrain elevation information obtained through lidar or synthetic aperture radar interferometry, which is used to characterize the three-dimensional morphological changes of the surface; the surface coverage type data refers to the land cover category information obtained through multi-spectral remote sensing image interpretation or on-site investigation and classification, such as vegetation-covered areas, bare rock areas, soil erosion areas, etc.; the surface deformation monitoring data refers to the displacement rate and cumulative displacement data collected by surface displacement monitoring devices, which is used to quantify the surface stability.

[0020] Exemplarily, in the mountain landslide monitoring scenario, the target geographical area is a certain high-risk landslide body, and the surface mapping data sets for multiple consecutive mapping periods need to meet the following conditions: All data uniformly adopt the standard geographical coordinate system, and the elevation data is obtained through a high-resolution digital elevation model; the coverage type data is based on the classification results of remote sensing images and is divided into categories such as forests, shrubs, and bare land; the deformation monitoring data is collected in real time by multiple monitoring points arranged on the landslide body, and the periodic three-dimensional displacement amounts of each point are recorded. By integrating multi-source data and unifying the spatial reference, the constructed multi-period ground mapping data set can provide the basic input for subsequent spatio-temporal feature analysis.

[0021] Step 102: Perform spatio-temporal feature coupling processing on the multi-period ground mapping data set to generate a spatio-temporal evolution feature set, where the spatio-temporal evolution feature set includes surface deformation features, geospatial association features, and surface coverage evolution features corresponding to each mapping period.

[0022] In this embodiment, the spatio-temporal feature coupling processing refers to jointly analyzing multi-period data in the time dimension and the space dimension to extract cross-period and cross-region dynamic change laws. Specifically, the surface deformation features are obtained by calculating the differences or statistics of deformation monitoring data for adjacent periods and are used to characterize the spatio-temporal heterogeneity of surface movement; the geospatial association features refer to the correlation indexes between deformation and terrain parameters generated through spatial autocorrelation analysis or interpolation methods; the surface coverage evolution features are extracted through the change detection algorithm of time-series remote sensing images to obtain the coverage type conversion matrix.

[0023] In the mountain landslide monitoring scenario, the generation process of the spatio-temporal evolution feature set is exemplified as follows: For multiple consecutive mapping periods, first perform spatio-temporal interpolation on the displacement data to generate a landslide body global displacement rate distribution map, and calculate the displacement acceleration rate as the deformation feature; secondly, extract slope partitions based on the elevation data, and statistically analyze the spatial aggregation of displacement rates within each slope interval through spatial overlay analysis to generate geospatial association features; at the same time, use time-series remote sensing images to identify vegetation degradation areas and bare land expansion ranges to construct surface coverage evolution features. By fusing the coupling features of deformation, terrain, and coverage changes, the spatio-temporal evolution feature set can reveal the correlation pattern between the precursors of landslide body instability and surface environmental changes.

[0024] Step 103: Based on a preset basic spatio-temporal evolution model, perform dynamic parameter optimization processing on the spatio-temporal evolution feature set to generate a target spatio-temporal evolution model, where the target spatio-temporal evolution model is used to characterize the mapping relationship between surface mapping data and geospatial evolution patterns.

[0025] In this embodiment, the basic spatio-temporal evolution model can be a landslide dynamics model based on physical mechanisms or a data-driven machine learning model. The dynamic parameter optimization process refers to adjusting the model parameters according to the current spatio-temporal evolution feature set to make it adapt to the specific geological conditions of the target geographical area. For example, in a physical model, the optimized parameters may include geotechnical mechanical parameters; in a machine learning model, the optimized parameters may involve network weights or hyperparameters.

[0026] Exemplarily, in a mountain landslide monitoring scenario, the preset basic model is a prediction model that fuses spatio-temporal features, and its initial parameters are obtained through training based on historical data. The dynamic parameter optimization process includes: inputting the spatio-temporal evolution feature set into the model, calculating the deformation prediction error through an adaptive optimization algorithm, and updating the model weights; at the same time, constraining the parameter search space according to the geological exploration data of the target geographical area to ensure that the optimized target spatio-temporal evolution model can accurately map the co-evolution law of landslide body deformation-topography-coverage. The optimized model can output the landslide stability probability distribution map, providing a basis for risk grading.

[0027] Step 104: Invoke the target spatio-temporal evolution model to perform evolution prediction processing on the surface survey data set of the current survey period, and generate the spatio-temporal evolution prediction result of the target geographical area. The spatio-temporal evolution prediction result includes surface deformation prediction distribution data and surface coverage change trend data.

[0028] In this embodiment, the evolution prediction processing refers to using the optimized target spatio-temporal evolution model to deduce the surface state within a certain future time window. The surface deformation prediction distribution data can be spatial raster data, and each pixel value represents the predicted displacement amount or instability probability; the surface coverage change trend data is generated through a probability model to form a coverage type transition probability matrix.

[0029] In a mountain landslide monitoring scenario, when invoking the target spatio-temporal evolution model for prediction, the input data includes elevation data, displacement data, and coverage classification results of the current period. The model output results are: 1) The surface deformation prediction distribution data within a certain future time period shows that the displacement rate in a specific area increases significantly, corresponding to a higher instability probability; 2) The surface coverage change trend data indicates that due to environmental factors, the bare land area is expected to expand, and the risk of degradation in the vegetation-covered area increases. This prediction result provides a quantitative basis for disaster warning.

[0030] Step 105: Perform dynamic update processing on the target spatio-temporal evolution model according to the spatio-temporal evolution prediction result to obtain an updated spatio-temporal evolution model, and synchronize the updated spatio-temporal evolution model to the evolution prediction processing of the next survey period.

[0031] In this embodiment, the dynamic update process refers to iteratively correcting the model parameters or structure based on the difference between the latest prediction results and the actual observed data. The specific methods include incremental learning or transfer learning. For example, when the deviation between the measured data in the new cycle and the predicted value exceeds the preset threshold, the model retraining process is triggered to adjust the model parameters to reduce the prediction error.

[0032] For example, in the scenario of mountain landslide monitoring, the measured data in the subsequent survey cycle shows that the displacement rate in a certain area exceeds the predicted value, and the model prediction error triggers the dynamic update mechanism. The update process includes: adding the new data to the training set and using the sliding window method to retain the recent data to optimize the model generalization ability; adaptively adjusting the model parameters by comparing the spatial correlation between the prediction and the measured results to enhance the response ability to emergencies. The updated spatio-temporal evolution model will be used as the basic model for subsequent prediction tasks to continuously optimize the model prediction ability.

[0033] With such a design, based on steps 101 - 105, through multi-cycle data fusion, spatio-temporal feature coupling, model dynamic optimization, and prediction-update closed-loop, the timeliness and accuracy of landslide evolution prediction are significantly improved. At the same time, it is ensured that the model can adapt to the changes in complex geological environments, providing reliable technical support for geological disaster prevention and control.

[0034] It is worth mentioning that in the embodiment of the present invention, the target spatio-temporal evolution model is a dynamic prediction framework constructed based on multi-source geospatial data, aiming to reveal the complex correlation laws between surface deformation and geographical environment elements. The target spatio-temporal evolution model takes the integration of physical mechanisms and data-driven as the design principle. By integrating multi-dimensional spatio-temporal features such as surface elevation, coverage type, and deformation monitoring, a quantitative mapping relationship between the surface evolution process and the driving factors is established. The core architecture of the target spatio-temporal evolution model consists of a basic spatio-temporal evolution model and a dynamic optimization mechanism: the basic model can select the landslide dynamics equation based on geomechanics principles, or use machine learning models such as Long Short-Term Memory (LSTM), Spatio-Temporal Graph Convolutional Network (ST-GCN), etc. The initial parameters are obtained through pre-training with historical multi-cycle data and have the basic modeling ability for basic processes such as terrain evolution and mass movement.

[0035] In addition, the target spatio-temporal evolution model can enhance regional adaptability through dynamic parameter optimization, and use the real-time updated spatio-temporal evolution feature set for iterative parameter tuning. In the mountain landslide monitoring scenario, the optimization process adopts the adaptive particle swarm algorithm or Bayesian inference method, and applies physical constraints to the parameter search space by combining the geological exploration data of the target area (such as rock layer dip angle, soil moisture content), ensuring that the model can not only capture the non-linear relationship between displacement acceleration rate and slope change, but also reflect the coupling effect of vegetation degradation and surface seepage. For example, the optimization results for a certain landslide show that the model significantly reduces the displacement prediction error and accurately identifies the deformation correlation characteristics of potential slip surfaces and elevation mutation areas at the same time.

[0036] In addition, the target spatio-temporal evolution model also has multi-scale prediction capabilities, and can output surface deformation prediction distribution data with a spatial resolution of up to the meter level and a coverage type transfer probability matrix. The deformation prediction uses spatio-temporal sequence extrapolation technology, inputs features such as the elevation change gradient and displacement vector field of the current period into the trained network, and generates the expected displacement rate and instability probability heat map of each grid cell in the next few days. The prediction of the coverage change trend is achieved by constructing a hidden Markov model, and combining the historical coverage conversion rules and environmental factors (such as rainfall, human activity intensity) to calculate the spatial probability distribution of bare land expansion and vegetation degradation.

[0037] Finally, the target spatio-temporal evolution model can adopt a closed-loop update mechanism to ensure the timeliness of prediction. By comparing the spatial correlation index between the predicted deformation field and GNSS monitoring data, the parameter update process is dynamically triggered. When the dynamic time warping (DTW) distance between the new cycle data and the prediction result exceeds the threshold, the system automatically starts the incremental learning module, and uses the elastic weight consolidation algorithm to incorporate new features while retaining historical knowledge, avoiding model degradation caused by sudden environmental changes. In a landslide warning case, through continuous updates in multiple cycles, the model incorporates the pore water pressure changes caused by sudden rainfall into the feature system, greatly improving the determination coefficient between the prediction result and the real displacement curve, and significantly enhancing the prediction robustness under complex meteorological conditions. This "prediction-validation-update" closed-loop architecture enables the model to continuously adapt to the dynamic evolution of the geological system and provides an intelligent analysis tool with spatio-temporal generalization capabilities for geological disaster risk warning.

[0038] In one implementation, the spatio-temporal feature coupling process for the multi-cycle ground survey data set in step 102 to generate a spatio-temporal evolution feature set includes: Step 1021: Perform surface deformation feature extraction processing on the surface survey data set of each survey cycle to obtain the surface deformation rate data and surface deformation direction data corresponding to each survey cycle.

[0039] In specific implementation, the extraction and processing of surface deformation features are achieved by fusing lidar point cloud data and synthetic aperture radar interferometry data. First, differential interferometry processing is performed on the surface elevation data at different time nodes within the same mapping period to generate a differential interferogram; subsequently, a phase unwrapping algorithm is used to eliminate the phase ambiguity caused by elevation changes and extract the elevation change amount of each pixel within the mapping period; based on the spatial correlation between the elevation change amount and the time-series monitoring data, the three-dimensional displacement vector of each monitoring point is calculated, where the horizontal displacement component is obtained by calculating the coordinate change amount of the GNSS monitoring station, and the vertical displacement component is directly obtained from the differential elevation data; finally, the three-dimensional displacement vector is decomposed into horizontal and vertical deformation rates to form a surface deformation rate data raster covering the target geographical area, and at the same time, a surface deformation direction data matrix is generated according to the azimuth angle distribution of the displacement vector. For example, in the feature extraction of the third mapping period, by processing the radar image data from May to August of a certain year, it is found that the horizontal deformation rate of a certain monitoring point reaches 12.3 mm / month, the direction is 15° south-southeast, and the vertical deformation rate is -5.6 mm / month, indicating that there is a significant oblique sliding trend in this area.

[0040] Step 1022: Perform a coverage change comparison process on the surface coverage type data of adjacent mapping periods to generate surface coverage evolution path data, where the surface coverage evolution path data includes the conversion sequence of surface coverage types and the conversion time interval.

[0041] Among them, the coverage change comparison process adopts a technical solution combining time-series remote sensing image supervised classification and change detection algorithms, specifically including: performing radiometric correction and geometric registration on the multi-spectral images of the same area in adjacent periods to ensure spatial resolution and pixel alignment; using a random forest classifier to perform land cover classification on the two-phase images respectively to generate a coverage type distribution map; detecting the areas where the coverage type has changed based on the post-classification comparison method, recording the change type, area proportion, and occurrence time period; constructing a coverage type conversion chain for each changed pixel, such as the degradation path from "forest → shrub → bare land", and counting the time interval between adjacent conversion steps. In a typical example, by comparing the data of the second and third mapping periods of a certain year, it is found that the coverage type in the middle of a certain slope changes from "dense forest" to "sparse shrub" within two months, and then further degrades to "bare rock" within one month. This evolution path is encoded as the sequence code F → S → B, and the time interval parameter is recorded as [60, 30] days, and finally integrated into the surface coverage evolution path database.

[0042] Step 1023: Based on the geospatial topological relationship, perform a spatial correlation calculation process on the surface deformation rate data, surface deformation direction data, and surface coverage evolution path data to generate geospatial correlation features, where the geospatial correlation features include a deformation-coverage coupling coefficient and a spatial evolution synchronization index.

[0043] In this step, the spatial correlation degree calculation process is realized by constructing a spatial weight matrix and covariance analysis. First, a spatial topological connection relationship between monitoring points is established using a Delaunay triangulation network; then, the spatial autocorrelation Moran's I index is used to quantify the spatial aggregation degree of the deformation rate and coverage evolution, and the deformation-coverage coupling coefficient is calculated; at the same time, the time-phase relationship between the change in the deformation direction and the conversion of the coverage type is analyzed through cross-wavelet transform to generate a spatial evolution synchronization index. For example, in the posterior edge area of a landslide body, in the area where the deformation rate exceeds 20 mm / month, 85% of the pixels undergo coverage type degradation during the same period, the deformation-coverage coupling coefficient reaches 0.78, and the spatial evolution synchronization index shows that there is a significant positive correlation between the acceleration of deformation and the degradation of coverage and the time lag is less than 7 days, indicating that the loss of surface vegetation may be an important inducement for the intensification of deformation.

[0044] Step 1024: Perform multi-dimensional fusion processing on the surface deformation rate data, the surface deformation direction data, the surface coverage evolution path data, and the geospatial correlation characteristics to generate the spatio-temporal evolution feature set.

[0045] In this embodiment, the multi-dimensional fusion processing adopts a feature-level fusion strategy, which specifically includes: performing Z-score standardization on the deformation rate data to eliminate the dimension difference; converting the deformation direction data into azimuth sine and cosine components to avoid the angle cycle problem; performing one-hot encoding on the coverage evolution path data to generate a multi-dimensional category feature vector; and finally integrating the standardized deformation rate, direction components, encoded coverage path, and correlation coefficient into a unified spatio-temporal feature tensor through a feature splicing layer. For example, in a landslide monitoring case, the dimension of the fused feature tensor is [256×256×8], where the spatial resolution is 10 m, and the 8 channels respectively correspond to the standardized deformation rate, sine and cosine direction components, 3D coverage path encoding, and 2 correlation feature indicators. This feature set completely characterizes the spatio-temporal dynamic characteristics of the landslide body in the third quarter of a certain year.

[0046] In another implementation manner, the dynamically optimizing the parameters of the spatio-temporal evolution feature set based on a preset basic spatio-temporal evolution model in step 103 to generate a target spatio-temporal evolution model includes: Step 1031: Divide the spatio-temporal evolution feature set into a training feature subset and a verification feature subset. The training feature subset contains the spatio-temporal evolution features of the first N - 1 surveying and mapping periods, and the verification feature subset contains the spatio-temporal evolution features of the Nth surveying and mapping period.

[0047] In specific implementation, the partitioning operation follows the principle of temporal cross-validation. For example, when the total number of mapping cycles N = 5, the feature tensors of the first 4 cycles (2022Q4 - 2023Q3) are used as the training feature subset, and the features of the 5th cycle (2023Q4) are used as the validation feature subset. During the partitioning process, it is necessary to ensure that each subset maintains spatial range consistency to avoid introducing biases due to spatial sampling differences. At the same time, the time span of the validation feature subset should be equal to the model prediction step. For example, when the model predicts the evolution in the next three months, the validation subset should contain exactly three months of monitoring data.

[0048] Step 1032: Input the training feature subset into the basic spatio-temporal evolution model for iterative training processing to generate an initial spatio-temporal evolution model. The iterative training processing includes an operation for adjusting the weights of surface deformation features and an operation for matching the coverage evolution path.

[0049] In this embodiment, the iterative training processing adopts a spatio-temporal convolutional neural network architecture. The specific process is as follows: First, spatio-temporal features are extracted through a three-dimensional convolutional layer, and the size of its convolutional kernel is set to 5×5×3 (spatial 5×5, time 3 cycles); in the feature weight adjustment stage, the attention mechanism is used to calculate the importance scores of the deformation rate features and the coverage path features, and dynamically adjust the channel weights of the feature maps; the coverage evolution path matching operation calculates the path similarity loss function by comparing the edit distance between the predicted path and the real path. For example, during the training process, the model automatically enhances the feature weights in the high deformation rate area to 1.5 times the original value, and at the same time sets the weight coefficient of the coverage path matching loss function to 0.3 to ensure that the model synchronously optimizes the deformation prediction and coverage change prediction performance.

[0050] Step 1033: Invoke the validation feature subset to perform a prediction accuracy verification process on the initial spatio-temporal evolution model to obtain model error distribution data. The model error distribution data includes the deformation prediction error rate and the coverage trend deviation degree.

[0051] In practical applications, the validation process adopts the leave-one-out cross-validation strategy, which specifically includes: Input the validation feature subset into the initial spatio-temporal evolution model to generate a predicted deformation rate field and a coverage path transfer matrix; calculate the deformation prediction error rate of each pixel as (predicted value - measured value) / measured value × 100%; the coverage trend deviation degree is measured by the Jaccard similarity coefficient to calculate the intersection-to-union ratio of the predicted path and the real path. For example, the deformation prediction error rate in the middle area of the landslide body ≤ 8%, but in the front edge area, due to the influence of sudden rainfall, the error rate reaches 23%; the coverage path deviation degree reaches 0.85 in the vegetation stable area and drops to 0.62 in the human activity interference area. This error distribution data reveals the prediction limitations of the model in complex interference areas.

[0052] Step 1034: Perform reverse correction processing on the feature mapping parameters of the initial spatio-temporal evolution model according to the model error distribution data to generate the target spatio-temporal evolution model.

[0053] In this step, the reverse correction processing adopts the adaptive moment estimation algorithm, which specifically includes: adjusting the convolution kernel weights according to the deformation error rate distribution, and applying a larger learning rate to the feature channels corresponding to the high-error regions; setting a threshold trigger mechanism for the coverage deviation degree, and starting the dual-path training mode when the deviation degree exceeds 0.4, and optimizing the parameters of the main network and the auxiliary correction network at the same time. For example, during the parameter correction process, the update amount of the convolution kernel weights in the leading edge region is 2.3 times that in the middle region. At the same time, prior knowledge constraints of the coverage type are introduced, and the upper limit of the transfer probability from the bare rock area to the soil erosion area is set to 0.15 exemplarily. The average error rate of the finally generated target spatio-temporal evolution model on the validation set is significantly reduced.

[0054] In another implementation, the step of calling the target spatio-temporal evolution model to perform evolution prediction processing on the surface mapping data set of the current mapping period to generate the spatio-temporal evolution prediction result of the target geographical region in step 104 includes: Step 1041: Perform real-time feature extraction processing on the surface mapping data set of the current mapping period to obtain the current surface deformation features and the current coverage evolution features.

[0055] It can be understood that the real-time feature extraction processing adopts a streaming computing framework, and the specific process is as follows: access the radar interferometry data and remote sensing images of the latest period (such as 2023Q4), and perform online phase unwrapping and land classification with a 5-minute time window; eliminate short-term noise interference through the moving average method to generate the current deformation rate isoline map; correct the deformation direction vector field by combining real-time GNSS displacement monitoring data; at the same time, update the coverage type distribution by using an incremental classifier to detect newly emerged coverage change regions within the last week. For example, the features extracted on December 15 of a certain year showed that the deformation rate of a certain monitoring point suddenly increased to 35 mm / day, the direction was adjusted to 8° east of due south, and at the same time, there were shrub degradation patches of hundreds of square meters on the northeast slope.

[0056] Step 1042: Input the current surface deformation features and the current coverage evolution features into the target spatio-temporal evolution model, and generate a preliminary prediction result through multi-level feature mapping processing.

[0057] In actual implementation, the multi-level feature mapping process follows the inherent architecture of the model, specifically including: compressing the input features into a high-dimensional latent space representation in the spatial encoding layer; capturing temporal dependencies through a temporal gated recurrent unit; and performing a deconvolution operation in the decoding layer to reconstruct the deformation rate distribution and coverage path transition probability for the next month. For example, the preliminary prediction results show that the current high-deformation area will expand to the adjacent valley area within the next half month, the predicted peak deformation rate is 48 mm / day, the area of the bare rock area is expected to increase by 15%, and the spatial overlap rate between the newly degraded area and the predicted high-deformation area reaches 82%.

[0058] Step 1043: Perform geospatial constraint processing on the preliminary prediction results. The geospatial constraint processing includes a terrain continuity verification operation and a coverage type logic verification operation to obtain the verified prediction results.

[0059] Exemplarily, the terrain continuity verification calculates slope threshold constraints through a digital elevation model to force the deformation direction in areas with a slope > 35° to be consistent with the terrain trend; the coverage type logic verification applies a Markov chain state transition matrix to prohibit conversion types that violate the ecological succession law, such as "bare rock → forest". For example, after verification, the deformation direction in a steep slope area is corrected from the originally predicted vertical downward to the inclined direction along the slope surface, and the abnormal conversion of "bare rock → forest" in the coverage path is replaced with a reasonable path of "bare rock → soil erosion".

[0060] Step 1044: Generate the surface deformation prediction distribution data and the surface coverage change trend data based on the verified prediction results, and perform trend coherence analysis processing on the surface deformation prediction distribution data and the historical deformation data to generate a trend correction coefficient.

[0061] In the embodiment of the present invention, the trend coherence analysis uses the dynamic time warping algorithm, specifically including: performing optimal path matching between the current prediction curve and the actual deformation curves of the past 6 cycles to calculate the slope change consistency coefficient; determining the trend correction coefficient based on the relationship between the consistency coefficient and a preset threshold, and starting a trend compensation mechanism when the coefficient < 0.7. For example, it is found through analysis that the deviation between the currently predicted deformation acceleration trend and the average trend of historical data reaches 28%, and accordingly a trend correction coefficient of 0.85 is generated for subsequent calibration processing.

[0062] Step 1045: Perform dynamic calibration processing on the surface deformation prediction distribution data based on the trend correction coefficient to obtain the spatio-temporal evolution prediction results of the target geographical area.

[0063] Exemplarily, the dynamic calibration process can adopt the weighted average method. For example, the original predicted deformation rate of a certain pixel is 50 mm / day, and the extrapolated value of the historical trend is 42 mm / day. After calibration with a coefficient of 0.85, the final predicted value is 48.8 mm / day. The calibrated prediction result not only retains the short-term mutation signal driven by the current features but also takes into account the long-term trend reflected by the historical rules. For example, calibration can improve the correlation coefficient between the deformation prediction result and the measured value to a certain extent.

[0064] In a preferred embodiment, the dynamically updating the target spatio-temporal evolution model according to the spatio-temporal evolution prediction result in step 105 to obtain an updated spatio-temporal evolution model includes: Step 1051: Obtain the updated actual surface mapping data set after the end of the current mapping period, perform feature extraction processing on the actual surface mapping data set, and generate actual spatio-temporal evolution features.

[0065] In a specific implementation, the actual surface mapping data set includes newly acquired radar interferometry data, multi-spectral images, and on-site monitoring records. The feature extraction processing adopts the same processing flow as steps 1021 - 1024 but is executed based on the latest data: First, perform unwrapping processing on the differential interferometric phase diagram to generate the actual surface deformation rate data after eliminating the atmospheric delay error; synchronously perform supervised classification on the remote sensing image to obtain the current coverage type distribution map, and generate the actual coverage evolution path data by comparing with the data of the previous period; then, through spatial autocorrelation analysis and time series alignment, construct the actual deformation spatial correlation feature and the actual coverage time correlation feature; finally, couple the two types of features to form an actual spatio-temporal evolution feature tensor with a dimension of [256×256×6].

[0066] For example, after the end of the mapping period in the fourth quarter of a certain year, the actual feature extraction results show that the deformation rate in the trailing edge area of a certain landslide reaches 41 mm / month, the coverage path shows an abnormal degradation mode of "dense forest → bare rock", and its spatial correlation feature shows that the spatial overlap rate between the deformation center and the coverage degradation area reaches 92%.

[0067] Step 1052: Compare the spatio-temporal evolution prediction result with the actual spatio-temporal evolution features to generate a model error feature set, and the model error feature set includes a deformation prediction error vector and a coverage trend offset.

[0068] In this embodiment, the differential comparison process adopts a multi-scale verification method: in the spatial dimension, the predicted deformation rate field and the actual deformation rate field are compared pixel by pixel after being gridded, and the root mean square error is calculated as the deformation prediction error vector; in the time dimension, the time nodes of the predicted coverage path and the actual evolution path are aligned through the dynamic time warping algorithm, and the difference in the type conversion time interval is calculated as the coverage trend offset. For example, the predicted deformation rate in a certain landslide front area is 38 mm / month, the actual monitored value is 45 mm / month, and the error vector value is 7 mm / month; in the coverage path prediction, the conversion of "shrub → bare land" is expected to occur on the 25th day, and the actual observation is on the 18th day, resulting in a positive time offset of 7 days. This difference data is encoded into the error feature set.

[0069] Step 1053: Perform incremental adjustment processing on the mapping parameters of the target spatio-temporal evolution model according to the model error feature set to obtain the adjusted mapping parameters. The incremental adjustment processing adopts a sliding window weight update mechanism.

[0070] In this embodiment, the specific implementation of the incremental adjustment processing includes: setting the time window size to 3 mapping periods, and only retaining the error feature data of the most recent period and the previous two periods; calculating the parameter update amount through the elastic weight consolidation algorithm to ensure that the mapping relationship of existing important features is not damaged when new knowledge is incorporated; implementing differential adjustment on the kernel weights of the third and fourth convolutional layers in the convolutional neural network, and the adjustment amplitude is proportional to the spatial distribution gradient of the error vector. For example, for the average error of 6.5 mm / month in the middle area of the landslide, the update coefficient of the corresponding convolution kernel weight is set to 0.32, while the update coefficient of the error area of 12 mm / month in the front area is increased to 0.56. At the same time, the parameters of the fully connected layer related to the coverage path prediction are frozen to avoid over-adjustment.

[0071] Step 1054: Synchronize the adjusted mapping parameters into the target spatio-temporal evolution model to generate the updated spatio-temporal evolution model, and delete the original mapping parameters in the basic spatio-temporal evolution model that exceed the preset time threshold.

[0072] It can be understood that the parameter synchronization process adopts a version control strategy: establish a model parameter repository, mark the new mapping parameters as version V2.1, and retain the parameter history of the most recent 5 versions; physically delete the original parameters (such as version V1.3 and earlier) that exceed the 24-month validity period; at the same time, update the model metadata record, and establish an index association between the new parameters and the current geographical environment feature library. For example, in the model update in the first quarter of 202X, 12 parameter files of version V1.2 trained in 202(X - 3) were deleted, 8 parameter files of version V2.1 were added, and the deformation prediction error of the updated model showed a significant decrease on the test set.

[0073] In a preferred embodiment, the feature extraction process on the actual surface mapping data set in step 1051 to generate actual spatio-temporal evolution features includes: Step 10511: Perform a deformation gradient calculation process on the surface elevation data in the actual surface mapping data set to generate actual surface deformation rate data.

[0074] For example, the deformation gradient calculation process uses synthetic aperture radar interferometry. The specific process is as follows: Perform small baseline set processing on 30 radar images obtained in the current cycle to construct 63 interferometric pairs; Use StaMPS software for permanent scatterer identification and phase unwrapping to extract the pixel-level deformation time series; Calculate the monthly average deformation rate of each pixel through linear regression analysis to generate a deformation rate grid map with a spatial resolution of 10 meters. For example, the deformation time series of a monitoring point (coordinates E112.35°, N28.76°) shows that the cumulative settlement from October to December of a certain year reached 58 millimeters, and the actual deformation rate was calculated to be 19.3 millimeters per month.

[0075] Step 10512: Perform a coverage change path analysis process on the surface cover type data in the actual surface mapping data set to generate actual coverage evolution path data.

[0076] Among them, the coverage change path analysis process uses an improved hidden Markov model, which specifically includes: Continuously classify the multi-spectral images with a week as the time unit to generate a coverage type time series; Decode the most likely type conversion path through the Viterbi algorithm; Record the type conversion order of each pixel and the time interval between adjacent conversions. For example, the type change of a certain pixel from week 45 to week 48 of a certain year is "shrub → bare land → soil erosion", and the time intervals are 14 days and 21 days respectively, and the path code is S → B → E_14_21.

[0077] Step 10513: Perform a spatial overlay process on the actual surface deformation rate data and the historical surface deformation rate data to generate actual deformation spatial correlation features.

[0078] In this step, the spatial overlay process uses a spatio-temporal cube analysis method: Construct a three-dimensional data cube (X, Y, T), where X / Y are spatial coordinates and T is the time dimension; Calculate the autocorrelation of each spatial unit in the time series through the Moran index to generate a spatial correlation heat map; Use the Getis-Ord Gi* statistic to identify the hot spots of the deformation rate. For example, the overlay analysis shows that the west side area of a landslide body shows a high-high spatial clustering pattern (Moran's I = 0.67, p < 0.01) in three consecutive cycles, indicating that there is a persistent deformation spatial correlation characteristic in this area.

[0079] Step 10514: Perform time series alignment processing on the actual coverage evolution path data and the historical coverage evolution path data to generate actual coverage time correlation features.

[0080] Among them, the time series alignment processing uses the dynamic time warping algorithm: perform time warping path matching on the coverage paths of each spatial unit, and calculate the cumulative distance of the optimal alignment path; extract the periodic features of the path sequence through Fourier transform to generate time correlation indicators including frequency amplitude and phase angle. For example, the matching result of the coverage path in a certain slope area shows that its evolution period has a significant correlation with the rainfall period (phase difference < 5 days), and this feature is encoded as the time correlation parameter θ = 0.78.

[0081] Step 10515: Perform coupling processing on the actual deformation space correlation features and the actual coverage time correlation features to generate the actual spatio-temporal evolution features.

[0082] In this embodiment, the coupling processing uses tensor fusion technology: resample the space correlation heat map to a resolution of 256×256 and convert it into a 32-bit floating-point matrix; perform normalization processing on the time correlation parameter matrix; fuse the two types of feature matrices into a 6-channel spatio-temporal feature tensor through Kronecker product operation. For example, the fused feature tensor shows strong coupling characteristics of a space correlation value of 0.82 and a time correlation value of 0.75 in the posterior edge area of the landslide, and this feature is used by the subsequent model update process to correct geotechnical mechanics parameters.

[0083] In a preferred embodiment, the step of performing difference comparison processing on the spatio-temporal evolution prediction result and the actual spatio-temporal evolution features in Step 1052 to generate a model error feature set includes: Step 10521: Extract the predicted surface deformation rate data in the spatio-temporal evolution prediction result, and perform spatial grid comparison processing on it and the actual surface deformation rate data to generate the deformation rate difference value corresponding to each spatial grid unit. The grid comparison processing uses bilinear interpolation to unify the predicted grid and the actual grid to the same spatial reference system, and calculate the absolute error and relative error pixel by pixel: absolute error ΔV = |V_pred - V_actual|, relative error δV = ΔV / V_actual × 100%. For example, the predicted value of a 10×10 meter grid unit (row 153, column 227) is 32.5 mm / month, and the actual value is 28.7 mm / month, and it is calculated that ΔV = 3.8 mm / month, δV = 13.2%.

[0084] Step 10522: Extract the predicted coverage change trend data in the spatio-temporal evolution prediction result, and perform time node comparison processing on it and the actual coverage evolution path data to generate the coverage type matching degree corresponding to each time node.

[0085] For example, the time node comparison process adopts a sliding window matching method: taking 7 days as the time window unit, the type consistency ratio of the predicted path and the actual path within the corresponding window is statistically calculated; the matching degree is calculated through the Jaccard similarity coefficient. For example, within the time window from the 15th to the 21st day, the predicted path is "shrub→bare land", and the actual path is "shrub→soil erosion". The intersection is {shrub}, and the union is {shrub, bare land, soil erosion}, and the matching degree J = 1 / 3 ≈ 0.33.

[0086] Step 10523: Perform spatial clustering analysis on the deformation rate difference value to generate a spatial distribution pattern of deformation errors.

[0087] In this embodiment, the spatial clustering analysis adopts the DBSCAN algorithm: set the neighborhood radius ε = 150 meters and the minimum number of samples MinPts = 5; identify high-error clustering regions (ΔV ≥ 8 mm / month) and low-error dispersion regions (ΔV ≤ 3 mm / month); calculate the centroid coordinates and error means of each cluster. For example, in the error analysis of 2023Q4, 3 high-error clusters are identified. The largest cluster covers an area of 1.2 hectares, with an average ΔV = 11.4 mm / month, and its spatial distribution highly coincides with the known geological fault zone trend.

[0088] Step 10524: Perform time series analysis on the coverage type matching degree to generate a time evolution pattern of coverage errors.

[0089] Exemplarily, the time series analysis adopts the ARIMA model: construct a matching degree time series in weeks, and determine the model order through the autocorrelation function (p = 2, d = 1, q = 1); fit to obtain the trend term and the periodic term, and calculate the distribution characteristics of the prediction residuals. For example, the analysis results show that the matching degree sequence has a significant period of 14 days (p < 0.05), and the residual standard deviation is 0.21. This pattern indicates that there is a systematic time lag problem in the coverage prediction of the model.

[0090] Step 10525: Perform joint encoding on the spatial distribution pattern of deformation errors and the time evolution pattern of coverage errors to generate the model error feature set.

[0091] In this step, the joint encoding process adopts a multimodal fusion method: convert the spatial clustering result into a GeoJSON format spatial index file; encode the time series analysis result into a high-dimensional feature vector; generate a joint feature matrix through a spatial-temporal cross-attention mechanism. For example, the encoded feature set contains 12 spatial cluster attribute fields, 26 time pattern parameters, and 18 cross-correlation indicators, providing multi-dimensional error representations for model parameter optimization.

[0092] In a preferred embodiment, the incremental adjustment process of the mapping parameters of the target spatio-temporal evolution model according to the model error feature set in step 1053 to obtain the adjusted mapping parameters includes: Step 10531: Convert the deformation error spatial distribution pattern into a deformation error weight matrix, where each element in the deformation error weight matrix corresponds to an error influence coefficient of a spatial grid cell.

[0093] In a specific implementation, the conversion process uses spatial interpolation and normalization methods: First, determine the spatial boundary of the high-error region based on the DBSCAN clustering result, and use the inverse distance weighting method to calculate the error influence coefficient of each 10×10 meter grid cell. The formula is W_ij = exp(-ΔV_ij / σ), where ΔV_ij is the deformation rate difference value of the grid, and σ is the regional error standard deviation. Subsequently, normalize the calculation result to the [0, 1] interval to generate a 256×256 floating-point weight matrix. For example, for a grid (row 89, column 156) at the center of a high-error cluster, ΔV = 12.3 mm / month, and the regional σ = 4.7 mm / month. Calculate W_ij = 0.08, indicating that the parameter adjustment priority in this region is the highest.

[0094] Step 10532: Convert the coverage error time evolution pattern into a coverage error time vector, where each element in the coverage error time vector corresponds to the error accumulation amount of a time node.

[0095] In this embodiment, the conversion process uses an exponential decay model: Taking a week as the time unit, accumulate the Jaccard matching degree error of each time node. For example, the matching degree J_15 = 0.33 in the 15th week, calculate its error contribution to this week and subsequent times, and the accumulated amount E_20 = 0.41 calculated at the 20th week, forming the error distribution characteristics in the time dimension.

[0096] Step 10533: Perform gradient descent adjustment on the deformation mapping layer parameters in the target spatio-temporal evolution model based on the deformation error weight matrix.

[0097] In this embodiment, a differential learning rate strategy is implemented in the adjustment process: The deformation mapping layer contains 3 three-dimensional convolution kernel groups, and the weight update amount ΔW = η×W_ij×∇L, where η = 0.001 is the base learning rate, W_ij is the corresponding grid error weight, and ∇L is the loss function gradient. For example, the learning rate in the high-weight region (W_ij > 0.7) is increased to 0.0015, while the learning rate in the low-weight region (W_ij < 0.3) is decreased to 0.0005 to ensure that the model focuses on correcting the parameter deviation in the region with significant spatial correlation.

[0098] Step 10534: Perform time attenuation adjustment processing on coverage mapping layer parameters in the target spatiotemporal evolution model based on the coverage error time vector.

[0099] Among them, the adjustment process adopts a sliding time window mechanism: the weight update of the long short-term memory network unit covering the mapping layer follows Δt=β×E_t, where β=0.002 is the time decay coefficient and E_t is the accumulated error of the time node. For example, the weight adjustment amount of the time unit corresponding to the 20th week is Δt=0.002×0.41=0.00082, and the parameters of ±2 weeks before and after the time window are adjusted in a linked manner to enhance the continuity of time dimension prediction.

[0100] Step 10535: jointly optimize the adjusted deformation mapping layer parameters and the coverage mapping layer parameters to generate adjusted mapping parameters.

[0101] In this embodiment, the joint optimization adopts the alternating direction multiplier method: set the coupling constraints of the deformation parameters and the coverage parameters, and minimize the joint loss function through iterative solution. Specifically, it includes 10 alternating iterations. In each iteration, the coverage parameters are first fixed to optimize the deformation parameters, and then the deformation parameters are fixed to optimize the coverage parameters. For example, after 3 rounds of iterations, the deformation prediction error decreased by 12%, and the coverage matching degree increased by 9%. The final parameter file size is 48MB, including 12288 deformation convolution kernel weights and 9216 coverage LSTM unit parameters.

[0102] In a preferred embodiment, the step 1054 of synchronizing the adjusted mapping parameters to the target spatiotemporal evolution model to generate the updated spatiotemporal evolution model includes: Step 10541: Create a mapping parameter version identifier corresponding to the adjusted mapping parameter, wherein the mapping parameter version identifier includes a parameter generation timestamp and a spatial grid code.

[0103] In this embodiment, the version identification construction rule is: the timestamp adopts the ISO8601 standard format, and the spatial grid code generates 12 characters based on the Geohash algorithm. For example, the 2024-03-15T14:23:05Z timestamp and the "ws101mvt6s9c" grid code are combined to form the version ID "20240315142305_ws101mvt6s9c", ensuring that the parameters can be traced back to a specific time and space range.

[0104] Step 10542: Bind the adjusted mapping parameters with the mapping parameter version identifier to generate a mapping parameter set with a version identifier.

[0105] In this step, the binding process adopts binary encapsulation technology: insert a 128-byte identification segment at the head of the parameter file, which includes the ASCII code of the timestamp (24 bytes), grid encoding (12 bytes), and checksum (4 bytes). For example, after binding, the size of a certain parameter file increases to 48.2MB, and the file header contains the identification information "20240315142305_ws101mvt6s9c", and the version attributes can be directly read through a hexadecimal editor.

[0106] Step 10543: Perform version timeliness detection processing on the historical mapping parameter set stored in the target spatio-temporal evolution model. The version timeliness detection processing is based on whether the time difference between the timestamp generated based on the parameter and the current mapping period exceeds a preset threshold.

[0107] Among them, 365 days is set as the expiration threshold for timeliness detection, and the difference is calculated using the UNIX timestamp. For example, it is detected that the timestamp corresponding to the version ID "20230316102231_ws101mvt5r2d" is March 16, 2023, which is 364 days different from the current date of March 15, 2024, and only 1 day remains until the expiration threshold, and it is marked as the to-be-deleted state.

[0108] Step 10544: Delete the expired mapping parameters in the historical mapping parameter set whose time difference exceeds the preset threshold, and retain the historical mapping parameters that do not exceed the threshold as the reference parameters.

[0109] In this embodiment, the deletion operation implements a dual mechanism of physical erasure and logical marking: first remove the expired records in the parameter storage index, and then perform three overwrites and erasures on the disk storage block. For example, deleting 5 expired versions releases 1.2GB of storage space, and retain the latest 3 versions (June 10, 2023, December 5, 2023, February 28, 2024) as the reference parameters, with a total size of 720MB.

[0110] Step 10545: Write the mapping parameter set with version identification into the parameter storage space of the target spatio-temporal evolution model, and perform spatial grid encoding alignment processing with the reference parameters.

[0111] For example, the storage space management adopts a sharding strategy: divide the 256×256 grid into 64 32×32 sub-regions, and the parameters of each sub-region are stored independently. The alignment processing ensures that the Geohash prefix of the new parameter is the same as the first 8 bits of the reference parameter. For example, the new parameter version "ws101mvt6s9c" is the same as the reference parameter "ws101mvt5r2d" in the first 8 bits "ws101mvt", realizing spatially consistent storage.

[0112] Step 10546: Perform compatibility verification processing on all mapping parameters in the parameter storage space. The compatibility verification processing includes spatial grid coding consistency verification and parameter generation timestamp continuity verification.

[0113] For example, the verification algorithm performs three-step verification: 1) Check the Geohash prefix consistency of parameters in the same sub-region; 2) Verify whether the timestamp sequence is continuous without breaks; 3) Detect the matching of the parameter file header checksum and the content hash value. For example, it is found that there are mixed-coded parameters of "ws101" and "ws102" in a certain sub-region, triggering an abnormal alarm and isolating the conflicting files.

[0114] Step 10547: When multiple valid mapping parameter versions are detected under the same spatial grid coding, extract the parameter generation timestamps of each version and perform a difference sorting process with the timestamp of the current surveying and mapping cycle.

[0115] In the embodiment of the present invention, the sorting algorithm is arranged in ascending order based on the absolute value of the time difference. For example, there are three versions under a certain grid coding: 2024-03-10 (time difference 5 days), 2024-02-28 (time difference 16 days), 2024-01-15 (time difference 59 days), and the sorting result is [2024-03-10, 2024-02-28, 2024-01-15].

[0116] Step 10548: Select the mapping parameter version with the smallest time difference from the current surveying and mapping cycle according to the difference sorting result, and mark the remaining version parameters as the to-be-archived state.

[0117] Exemplarily, the selection strategy sets 7 days as the valid window: if the minimum difference ≤ 7 days, directly select it; otherwise, trigger the parameter reconstruction process. For example, select the 2024-03-10 version (time difference 5 days) as the main parameter, move the other two versions to the archive area, and set the retention period to one month.

[0118] Step 10549: Perform spatial overlay and fusion processing on the selected mapping parameter version and the reference parameter to generate a fused updated mapping parameter set; overwrite the original parameter storage area of the target spatio-temporal evolution model with the updated mapping parameter set to complete the model update operation and generate the updated spatio-temporal evolution model.

[0119] In an embodiment of the present invention, the fusion process adopts the weighted average method: the weight of the new parameter is 0.7, and the weight of the reference parameter is 0.3. The formula is W_fused = 0.7W_new + 0.3W_base. For example, the original value of a convolution kernel parameter is [0.12, -0.05, 0.33], the reference value is [0.15, -0.08, 0.29], and the fused value is [0.129, -0.059, 0.318]. Among them, the updated model shows a significant improvement in prediction accuracy on the test set, and the storage occupancy is significantly optimized.

[0120] In another preferred embodiment, in the evolution prediction process of synchronizing the updated spatio-temporal evolution model to the next mapping cycle in step 105, it includes: Step 10551: At the starting time node of the next mapping cycle, extract the updated mapping parameter set from the original parameter storage area of the target spatio-temporal evolution model; identify the mapping parameter version identifier of each parameter in the extracted updated mapping parameter set, and parse the parameter generation timestamp and spatial grid code included in the mapping parameter version identifier.

[0121] In a specific implementation, the parameter extraction operation is executed through a version control interface: at 08:00 on April 1, 2024 (the starting moment of the next cycle), load all parameter files marked as "active" from the distributed storage cluster; use a regular expression matching algorithm to parse the version identifier in the file name. For example, extract the generation timestamp "2024-03-29T15:02:17Z" and the 12-bit Geohash code "ws101mvt6s9c" from "20240329150217_ws101mvt6s9c.bin". During the parsing process, data integrity verification is implemented, and the CRC32 checksum is used to verify the consistency of the file header and content, and the files that fail the verification are excluded.

[0122] Step 10552: Perform a matching degree calculation process on the parameter generation timestamp and the prediction start time of the next mapping cycle to generate a time matching coefficient matrix; perform an effectiveness screening process on the extracted updated mapping parameter set according to the time matching coefficient matrix, and retain the valid parameters with a time matching coefficient greater than the set threshold.

[0123] For example, the time matching coefficient calculation formula is: C_t = 1 - |T_gen - T_start| / (T_end - T_start).

[0124] Among them, T_gen is the parameter generation timestamp, T_start is the start time of the period, and T_end is the expiration time of the parameter validity period. Set the threshold C_threshold = 0.85, and filter out the parameter files with C_t ≥ 0.85. For example, a certain parameter is generated at 2024-03-28T14:00:00Z, the start time of the period is 2024-04-01T00:00:00Z, and the expiration date is 2024-06-30T00:00:00Z. It is calculated that C_t = 0.91, which meets the validity condition.

[0125] Step 10553: Perform a grid-by-grid matching process on the spatial grid encoding of the valid parameters and the geospatial grid division data of the next mapping period.

[0126] In the embodiment of the present invention, the matching process uses a spatial index acceleration technique: load the MBR (Minimum Bounding Rectangle) spatial index of the 256×256 grid of the target area, decode the 12-bit Geohash encoding of each parameter into a longitude and latitude range, and judge the inclusion relationship with the grid spatial range. For example, the parameter encoding "ws101mvt6s9c" corresponds to a grid cell with a longitude range of 112.34° - 112.35° and a latitude range of 28.75° - 28.76° after decoding, which completely coincides with the grid in row 153 and column 227 of the target area, and is marked as a successful match.

[0127] Step 10554: When it is detected that the target spatial grid encoding does not correspond to a valid parameter, perform spatial interpolation and completion processing using the valid parameters of adjacent spatial grid encodings; load the completed valid parameters into the real-time prediction operation module of the basic spatio-temporal evolution model in the order of spatial grid encoding.

[0128] In the embodiment of the present invention, inverse distance weighting method is used for spatial interpolation: taking the target grid as the center, search for all valid parameter grids within a radius of 300 meters. An exemplary weight calculation formula is W_d = 1 / (d^2 + ε), where d is the distance from the grid center, and ε = 1e-5 to prevent division by zero error. For example, a missing grid (row 89, column 201) is interpolated by the parameters of 4 adjacent valid grids, and the weights are 0.42, 0.31, 0.19, and 0.08 respectively. Finally, the generated parameter value is the weighted average of the adjacent parameters.

[0129] Step 10555: Generate a mapping relationship index table between the mapping parameter version identifier of the valid parameters and the time node of the next mapping period. The mapping relationship index table includes spatial grid encoding, parameter version identifier, and loading timestamp.

[0130] Among them, the index table is constructed using a columnar storage structure: each row record contains three fields: Geohash code (12 bytes), version identifier (36 bytes), and loading timestamp (8 bytes). For example, the record "ws101mvt6s9c |20240329150217_ws101mvt6s9c | 20240401080000" indicates that the grid parameter version was loaded at 08:00:00 on April 1, 2024, and the total index table size is approximately 256×256×56≈3.5MB.

[0131] Step 10556: During the evolution prediction process of the next surveying and mapping cycle, continuously monitor the time difference between the timestamps of each parameter version identifier in the mapping relationship index table and the current prediction time node.

[0132] Among them, the monitoring mechanism implements a sliding window check: every time a prediction step (such as 15 minutes) is completed, calculate the difference ΔT = |T_current - T_load| between the current time T_current and the parameter loading time T_load. For example, when making a prediction at 2024-04-01T10:30:00, the loading time of a certain parameter is 2024-04-01T08:00:00, and ΔT = 2.5 hours, which does not exceed the synchronization threshold (default 24 hours).

[0133] Step 10557: When the time difference exceeds the preset synchronization threshold, trigger a parameter version update request and extract the latest mapping parameter version from the parameter storage area of the updated spatio-temporal evolution model.

[0134] In this embodiment, the threshold trigger logic is set hierarchically: the threshold for the core area (such as the middle of the landslide) is set to 6 hours, and the threshold for the edge area is set to 24 hours. For example, when the ΔT of a certain core grid parameter reaches 6 hours and 01 minute, an update request is automatically sent to the parameter server to retrieve the parameter version with the latest generation time (such as 20240401075959_ws101mvt7a2b).

[0135] Step 10558: Replace the extracted latest mapping parameter version in the real-time prediction operation module according to the spatial grid code, and synchronously update the corresponding entry in the mapping relationship index table.

[0136] In this step, the replacement operation implements atomic update: first, lock the prediction calculation thread of the target grid, write the new parameter into the temporary memory area, verify the checksum, then switch the pointer to point to the new parameter, and finally update the timestamp field of the index table. For example, update the parameter version of grid ws101mvt7a2b from 20240329150217 to 20240401075959, and the index table loading timestamp is synchronously changed to 2024-04-01T10:31:22.

[0137] Step 10559: After completing all prediction processes in the next surveying and mapping cycle, associate and store the version identifier of the mapping parameters actually used and the corresponding prediction results in the model log database.

[0138] Exemplarily, the storage structure is designed as a relational database table: the main table records the prediction task ID, time range, and spatial range; the detail table stores the parameter version identifier, predicted value, and actual value (subsequently updated) for each grid. For example, the record of task ID#20240401_landslide contains 256×256 = 65536 detail data, and each record field includes [grid code, parameter version, predicted deformation rate, predicted coverage path]. The total data volume is approximately 65MB, and after column compression, it is stored as 8.7MB.

[0139] In an extensible embodiment, the method further includes: Extract the model error feature set in the updated spatio-temporal evolution model, perform error source tracing analysis on the model error feature set, and generate error source distribution data; Perform adaptive adjustment processing on the learning rate parameter of the updated spatio-temporal evolution model according to the error source distribution data to generate a dynamic learning rate configuration parameter; Perform incremental optimization processing on the mapping parameters of the updated spatio-temporal evolution model based on the dynamic learning rate configuration parameter to generate optimized mapping parameters; Perform fusion verification processing on the optimized mapping parameters and the historical parameter versions to generate a parameter compatibility evaluation result; Perform selective overwrite processing on the optimized mapping parameters according to the parameter compatibility evaluation result to generate the final updated spatio-temporal evolution model and synchronize it to the subsequent surveying and mapping cycle.

[0140] In the iterative optimization process of the landslide monitoring system, extract the model error feature set in the updated spatio-temporal evolution model, perform error source tracing analysis on the model error feature set, and generate error source distribution data. The error source tracing analysis process uses the random forest feature importance evaluation method. By constructing an error source classification model, calculate the contribution degree index of each input feature to the prediction error, and generate an error source distribution heat map including spatial position, time stage, and feature type.

[0141] For example, in the error analysis of the second quarter of 2024, it was found that 65% of the deformation prediction errors in the front area of the landslide were due to the fact that the mutation characteristics of the surface cover type were not effectively learned. The importance score of the cover evolution path characteristics in this area reached 0.87, which was significantly higher than that of the terrain continuity characteristics (0.23). According to the distribution data of the error sources, an adaptive adjustment process was performed on the learning rate parameters of the updated spatio-temporal evolution model to generate dynamic learning rate configuration parameters. Specifically, for the feature channels corresponding to the error sources with high contribution degrees, the initial learning rate was increased from 0.001 to 0.0025, and at the same time, the learning rate of the feature channels with low contribution degrees was attenuated to 0.0003.

[0142] Based on the dynamic learning rate configuration parameters, an incremental optimization process was performed on the mapping parameters of the updated spatio-temporal evolution model to generate optimized mapping parameters. The stochastic gradient descent algorithm with momentum was used in this process. In 8 consecutive training cycles, the update amplitude of the weights of the fully connected layer corresponding to the cover path characteristics in the front area reached 1.8 times the original value. The optimized mapping parameters were fused and verified with the historical parameter versions to generate a parameter compatibility evaluation result. The verification method included: calculating the cosine similarity index between the old and new parameter vectors, and setting a threshold > 0.75 as the compatibility standard; the verification of the parameters in the middle area of the landslide showed that the similarity reached 0.82, while in the front area, due to the drastic adjustment of the feature weights, the similarity dropped to 0.68, triggering the parameter rollback protection mechanism. According to the parameter compatibility evaluation result, a selective coverage process was performed on the optimized mapping parameters to generate the final updated spatio-temporal evolution model and synchronize it to the subsequent surveying and mapping cycles. Specifically, the new parameters were directly covered in the compatible areas, and a mixed configuration of 70% old parameters and 30% new parameters was retained in the incompatible areas, ensuring that in the prediction task of the third quarter of 2024, the prediction error in the front area was reduced to 9.3% and there was no parameter conflict warning.

[0143] In an extensible embodiment, the method further includes: Obtaining the geospatial grid division data of the next surveying and mapping cycle, performing a multi-scale recombination process on it to generate a nested spatial grid structure; Based on the time node division data, the prediction time step was segmented to generate a multi-granularity time prediction sequence; Aligning the nested spatial grid structure with the multi-granularity time prediction sequence to generate a spatio-temporal matching index table; Invoking the updated spatio-temporal evolution model to perform an independent prediction process on the mapping relationships in the index table to generate a multi-scale spatio-temporal prediction result set; Performing a cross-scale fusion process on the multi-scale spatio-temporal prediction result set to generate target spatio-temporal prediction data with a unified resolution; Perform real-time comparison processing on the target spatio-temporal prediction data and the actual data to generate a cross-scale error distribution map and calibrate the hierarchical weights.

[0144] In the extended scheme for improving the spatio-temporal prediction accuracy of landslides, obtain the geospatial grid division data for the next survey period, perform multi-scale recombination processing on it to generate a nested spatial grid structure. The multi-scale recombination uses a quadtree space segmentation algorithm to divide the basic 10-meter grid into three-level sub-grid systems of 5 meters, 20 meters, and 40 meters, forming a multi-layer nested structure including 1024×1024 (5 meters), 512×512 (10 meters), 256×256 (20 meters), and 128×128 (40 meters).

[0145] Perform segmented processing on the predicted time step based on the time node division data to generate a multi-granularity time prediction sequence. The specific implementation is as follows: decompose the three-month prediction period into three-level time series of daily granularity (90 steps), weekly granularity (12 steps), and monthly granularity (3 steps), and each level of the sequence maintains temporal continuity through linear interpolation. Align the nested spatial grid structure with the multi-granularity time prediction sequence to generate a spatio-temporal matching index table. This index table contains spatial level encoding, time granularity markers, and data storage pointers. For example, the index entry "L3_T2_S45" represents the storage address of level 3 (20-meter grid), time granularity 2 (weekly step), and spatial block 45. Call the updated spatio-temporal evolution model to perform independent prediction processing on the mapping relationships in the index table to generate a multi-scale spatio-temporal prediction result set. The prediction process implements parallel computing, and each scale prediction task is assigned to different computing units of the GPU cluster. The global trend prediction is preferentially completed at the 40-meter coarse-grained level, and its results provide boundary constraints for the fine-grained prediction.

[0146] Perform cross-scale fusion processing on the multi-scale spatio-temporal prediction result set to generate target spatio-temporal prediction data with a unified resolution. The fusion algorithm uses the multi-resolution analysis method of wavelet transform to perform weighted synthesis on the frequency domain features of the prediction results at each scale. The weight coefficient of the 20-meter level is set to 0.6, the 10-meter level is 0.3, and the 5-meter level is 0.1. Perform real-time comparison processing on the target spatio-temporal prediction data and the actual data to generate a cross-scale error distribution map and calibrate the hierarchical weights. The calibration process dynamically adjusts the fusion weights by calculating the determination coefficient between the prediction results at each scale and the actual values. For example, when the error at the 5-meter level exceeds 15%, its weight is automatically reduced to 0.05 to ensure the stable maintenance of the global deformation prediction accuracy in the flood season prediction in 2024.

[0147] In an extensible embodiment, the method further includes: Extract the historical prediction result set and the actual survey data set, and perform spatio-temporal dimension expansion processing on them respectively to generate a prediction result spatio-temporal cube and an actual data spatio-temporal cube; Perform per - unit difference calculation on two spatio - temporal cubes to generate an error spatio - temporal cube; Perform spatio - temporal pattern mining on the error spatio - temporal cube to generate an error spatio - temporal evolution pattern; Perform spatio - temporal adaptive correction on the model mapping parameters according to the error spatio - temporal evolution pattern to generate a set of spatio - temporal correction parameters; Perform superposition and fusion on the set of spatio - temporal correction parameters and the original parameters to generate a spatio - temporal enhanced update model and synchronize it to subsequent cycles.

[0148] In the improvement scheme for enhancing the spatio - temporal generalization ability of the model, it involves extracting a set of historical prediction results and a set of actual surveying and mapping data, and performing spatio - temporal dimension expansion processing on them respectively to generate a prediction result spatio - temporal cube and an actual data spatio - temporal cube.

[0149] The spatio - temporal dimension expansion processing uses tensor reshaping technology to reconstruct the data of 12 consecutive surveying cycles into a four - dimensional tensor [N, H, W, T], where N = 256×256 spatial grids, T = 12 time nodes, and each unit stores a composite value of deformation rate and coverage path encoding. Perform per - unit difference calculation on the two spatio - temporal cubes to generate an error spatio - temporal cube. The difference calculation implements a double - threshold mechanism: units with a deformation rate difference > 8 mm / month or a coverage path matching degree < 0.6 are marked as significant error units. Perform spatio - temporal pattern mining on the error spatio - temporal cube to generate an error spatio - temporal evolution pattern. The mining algorithm uses three - dimensional density clustering (3D - DBSCAN), sets the neighborhood radius ε = 3 grids×3 grids×2 cycles, and identifies 6 persistent error clusters. The largest error cluster continuously appears in the landslide trailing edge area during the period from 2023Q4 to 2024Q2, with a spatial range of 1.2 hectares and a time span of 9 months.

[0150] Perform spatio - temporal adaptive correction on the model mapping parameters according to the error spatio - temporal evolution pattern to generate a set of spatio - temporal correction parameters. The correction processing uses a spatio - temporal attention mechanism, adding a spatio - temporal position encoding layer in the Transformer architecture to increase the attention of the model to high - error spatio - temporal units to 2.3 times the baseline value. Perform superposition and fusion on the set of spatio - temporal correction parameters and the original parameters to generate a spatio - temporal enhanced update model and synchronize it to subsequent cycles. The fusion method uses a residual connection structure, and the formula is W_new = W_orig+λΔW, where λ = 0.35 is the spatio - temporal correction coefficient and ΔW is the correction parameter increment. After analysis in the third quarter of 2024, this scheme significantly improves the prediction accuracy in the area of persistent error clusters. The correlation coefficient between the spatio - temporal attention weights and the measured error distribution remains at a high level, significantly enhancing the model's ability to capture complex spatio - temporal evolution laws.

[0151] In the embodiments of the present invention, through the spatio-temporal coupling analysis and dynamic model optimization mechanism of multi-period ground surveying data, the refined modeling and adaptive prediction capabilities of the geographical evolution process are realized.

[0152] First, by integrating the multi-dimensional correlation features of elevation, coverage type, and deformation data within at least three consecutive periods, the limitations of single-time-point or single-parameter analysis are broken through, and the non-linear spatio-temporal coupling law in surface evolution can be accurately captured.

[0153] Secondly, based on the spatio-temporal model construction method with dynamic parameter optimization, the geographical spatial correlation features and coverage evolution trends are effectively integrated, enabling the model to have the adaptive representation ability for complex geographical processes.

[0154] In addition, by establishing a closed-loop feedback mechanism between the prediction results and the model parameters, the real-time iterative optimization of the model parameters is realized, significantly improving the long-term prediction stability under different geographical scenarios.

[0155] Designed in this way, the problem that traditional static models are difficult to adapt to the changes in the dynamic geographical environment is solved. Moreover, through the deep coupling of spatio-temporal features of multi-source data, the associated evolution patterns of potential surface deformation and coverage change are effectively identified, providing high-confidence decision-making support for applications such as geological disaster early warning and ecological evolution analysis.

[0156] The embodiments of the present invention provide a computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, the spatio-temporal evolution modeling method driven by ground surveying is implemented.

[0157] The embodiments of the present invention provide a processor, and the processor is used to run a program, wherein when the program runs, the spatio-temporal evolution modeling method driven by ground surveying is executed.

[0158] In the embodiments of the present invention, as Figure 2 shown, the spatio-temporal evolution modeling system 100 includes at least one processor 101, at least one memory 102 connected to the processor 101, and a bus 103; wherein, the processor 101 and the memory 102 complete communication with each other through the bus 103; the processor 101 is used to call the program instructions in the memory 102 to execute the above-mentioned spatio-temporal evolution modeling method driven by ground surveying.

[0159] The present invention is described with reference to the flowcharts and / or block diagrams of methods, spatio-temporal evolution modeling systems (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices generate means for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or multiple blocks.

[0160] In a typical configuration, a spatio-temporal evolution modeling system includes one or more processors (CPUs), a memory, and a bus. The spatio-temporal evolution modeling system may also include an input / output interface, a network interface, etc.

[0161] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM), and the memory includes at least one storage chip. The memory is an example of computer-readable media.

[0162] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can store information by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage, computer-readable storage media, or any other non-transmission media that can be used to store information accessible by a spatio-temporal evolution modeling system. As defined herein, computer-readable media does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0163] It should also be noted that the term "includes", "comprising" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or computer-readable storage medium comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or computer-readable storage medium. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, commodity or computer-readable storage medium comprising the element.

[0164] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0165] The above are only embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for spatio-temporal evolution modeling driven by ground surveying and mapping, characterized in that The method includes: Obtaining a multi-period ground surveying data set of a target geographical area, where the multi-period ground surveying data set includes surface surveying data sets of at least three consecutive surveying periods, and each surface surveying data set is composed of surface elevation data, surface coverage type data, and surface deformation monitoring data under the same geographical coordinate system; Performing spatio-temporal feature coupling processing on the multi-period ground surveying data set to generate a spatio-temporal evolution feature set, where the spatio-temporal evolution feature set includes surface deformation features, geospatial association features, and surface coverage evolution features corresponding to each surveying period; Based on a preset basic spatio-temporal evolution model, performing dynamic parameter optimization processing on the spatio-temporal evolution feature set to generate a target spatio-temporal evolution model, where the target spatio-temporal evolution model is used to represent the mapping relationship between surface surveying data and geospatial evolution patterns; Invoking the target spatio-temporal evolution model to perform evolution prediction processing on the surface surveying data set of the current surveying period to generate a spatio-temporal evolution prediction result of the target geographical area, where the spatio-temporal evolution prediction result includes surface deformation prediction distribution data and surface coverage change trend data; Performing dynamic update processing on the target spatio-temporal evolution model according to the spatio-temporal evolution prediction result to obtain an updated spatio-temporal evolution model, and synchronizing the updated spatio-temporal evolution model to the evolution prediction processing of the next surveying period.

2. The method according to claim 1, wherein The performing spatio-temporal feature coupling processing on the multi-period ground surveying data set to generate a spatio-temporal evolution feature set includes: Performing surface deformation feature extraction processing on the surface surveying data set of each surveying period to obtain surface deformation rate data and surface deformation direction data corresponding to each surveying period; Performing coverage change comparison processing on the surface coverage type data of adjacent surveying periods to generate surface coverage evolution path data, where the surface coverage evolution path data includes the conversion sequence of surface coverage types and the conversion time interval; Based on the geospatial topological relationship, performing spatial correlation degree calculation processing on the surface deformation rate data, surface deformation direction data, and surface coverage evolution path data to generate geospatial association features, where the geospatial association features include deformation-coverage coupling coefficients and spatial evolution synchronization indexes; Performing multi-dimensional fusion processing on the surface deformation rate data, the surface deformation direction data, the surface coverage evolution path data, and the geospatial association features to generate the spatio-temporal evolution feature set.

3. The method according to claim 2, wherein The based on a preset basic spatio-temporal evolution model, performing dynamic parameter optimization processing on the spatio-temporal evolution feature set to generate a target spatio-temporal evolution model includes: Dividing the spatio-temporal evolution feature set into a training feature subset and a verification feature subset, where the training feature subset includes spatio-temporal evolution features of the first N-1 surveying periods, and the verification feature subset includes spatio-temporal evolution features of the Nth surveying period; Inputting the training feature subset into the basic spatio-temporal evolution model for iterative training processing to generate an initial spatio-temporal evolution model, where the iterative training processing includes surface deformation feature weight adjustment operations and coverage evolution path matching operations; Call the validation feature subset to perform prediction accuracy verification processing on the initial spatio-temporal evolution model, and obtain model error distribution data, where the model error distribution data includes deformation prediction error rate and coverage trend deviation degree; Perform reverse correction processing on the feature mapping parameters of the initial spatio-temporal evolution model according to the model error distribution data to generate the target spatio-temporal evolution model.

4. The method according to claim 3, wherein The calling the target spatio-temporal evolution model to perform evolution prediction processing on the surface mapping data set of the current mapping period to generate the spatio-temporal evolution prediction result of the target geographical area includes: Perform real-time feature extraction processing on the surface mapping data set of the current mapping period to obtain the current surface deformation feature and the current coverage evolution feature; Input the current surface deformation feature and the current coverage evolution feature into the target spatio-temporal evolution model, and generate a preliminary prediction result through multi-level feature mapping processing; Perform geospatial constraint processing on the preliminary prediction result, where the geospatial constraint processing includes terrain continuity verification operation and coverage type logic verification operation to obtain the verified prediction result; Generate the surface deformation prediction distribution data and the surface coverage change trend data according to the verified prediction result, and perform trend coherence analysis processing on the surface deformation prediction distribution data and the historical deformation data to generate a trend correction coefficient; Perform dynamic calibration processing on the surface deformation prediction distribution data based on the trend correction coefficient to obtain the spatio-temporal evolution prediction result of the target geographical area.

5. The method according to claim 1, characterized in that The performing dynamic update processing on the target spatio-temporal evolution model according to the spatio-temporal evolution prediction result to obtain an updated spatio-temporal evolution model includes: Obtain the updated actual surface mapping data set after the end of the current mapping period, and perform feature extraction processing on the actual surface mapping data set to generate actual spatio-temporal evolution features; Perform difference comparison processing on the spatio-temporal evolution prediction result and the actual spatio-temporal evolution features to generate a model error feature set, where the model error feature set includes a deformation prediction error vector and a coverage trend offset; Perform incremental adjustment processing on the mapping parameters of the target spatio-temporal evolution model according to the model error feature set to obtain the adjusted mapping parameters, and the incremental adjustment processing adopts a sliding window weight update mechanism; Synchronize the adjusted mapping parameters to the target spatio-temporal evolution model to generate the updated spatio-temporal evolution model, and delete the original mapping parameters in the basic spatio-temporal evolution model that exceed the preset time threshold.

6. The method according to claim 5, wherein The performing feature extraction processing on the actual surface mapping data set to generate actual spatio-temporal evolution features includes: Perform deformation gradient calculation processing on the surface elevation data in the actual surface mapping data set to generate actual surface deformation rate data; Perform coverage change path analysis processing on the surface coverage type data in the actual surface mapping data set to generate actual coverage evolution path data; Perform spatial superposition processing on the actual surface deformation rate data and the historical surface deformation rate data to generate actual deformation spatial correlation features; Perform time series alignment processing on the actual coverage evolution path data and the historical coverage evolution path data to generate actual coverage time correlation features; Perform coupling processing on the actual deformation space correlation features and the actual coverage time correlation features to generate the actual spatio-temporal evolution features.

7. The method according to claim 6, wherein The difference comparison processing of the spatio-temporal evolution prediction result and the actual spatio-temporal evolution features generates a model error feature set, including: Extract the predicted ground deformation rate data in the spatio-temporal evolution prediction result, and perform spatial grid comparison processing with the actual ground deformation rate data to generate the deformation rate difference value corresponding to each spatial grid unit; Extract the predicted coverage change trend data in the spatio-temporal evolution prediction result, and perform time node comparison processing with the actual coverage evolution path data to generate the coverage type matching degree corresponding to each time node; Perform spatial clustering analysis processing on the deformation rate difference values to generate a deformation error spatial distribution pattern; Perform time series analysis processing on the coverage type matching degrees to generate a coverage error time evolution pattern; Perform joint coding processing on the deformation error spatial distribution pattern and the coverage error time evolution pattern to generate the model error feature set; The incremental adjustment processing of the mapping parameters of the target spatio-temporal evolution model according to the model error feature set obtains the adjusted mapping parameters, including: Convert the deformation error spatial distribution pattern into a deformation error weight matrix, and each element in the deformation error weight matrix corresponds to an error influence coefficient of a spatial grid unit; Convert the coverage error time evolution pattern into a coverage error time vector, and each element in the coverage error time vector corresponds to an error accumulation amount of a time node; Perform gradient descent adjustment processing on the deformation mapping layer parameters in the target spatio-temporal evolution model based on the deformation error weight matrix; Perform time decay adjustment processing on the coverage mapping layer parameters in the target spatio-temporal evolution model based on the coverage error time vector; Perform joint optimization processing on the adjusted deformation mapping layer parameters and coverage mapping layer parameters to generate the adjusted mapping parameters.

8. The method according to claim 5, wherein Synchronize the adjusted mapping parameters to the target spatio-temporal evolution model to generate the updated spatio-temporal evolution model, including: Create a mapping parameter version identifier corresponding to the adjusted mapping parameters, and the mapping parameter version identifier includes a parameter generation timestamp and a spatial grid code; Bind the adjusted mapping parameters and the mapping parameter version identifier to generate a set of mapping parameters with version identifiers; Perform version timeliness detection processing on the historical mapping parameter set stored in the target spatio-temporal evolution model, and the version timeliness detection processing is based on whether the time difference between the parameter generation timestamp and the current surveying and mapping period exceeds a preset threshold; Delete the expired mapping parameters in the historical mapping parameter set whose time difference exceeds the preset threshold, and retain the historical mapping parameters that do not exceed the threshold as reference parameters; Write the set of mapping parameters with version identifiers into the parameter storage space of the target spatio-temporal evolution model, and perform spatial grid coding alignment processing with the reference parameters; Perform compatibility verification processing on all mapping parameters in the parameter storage space. The compatibility verification processing includes spatial grid coding consistency verification and parameter generation timestamp continuity verification; When it is detected that there are multiple valid mapping parameter versions under the same spatial grid coding, extract the parameter generation timestamps of each version and perform difference sorting processing with the timestamp of the current surveying and mapping cycle; Select the mapping parameter version with the smallest difference from the timestamp of the current surveying and mapping cycle according to the difference sorting result, and mark the remaining version parameters as pending archival status; Perform spatial overlay fusion processing on the selected mapping parameter version and the reference parameters to generate a fused updated set of mapping parameters; Overwrite the original parameter storage area of the target spatio-temporal evolution model with the updated set of mapping parameters to complete the model update operation and generate the updated spatio-temporal evolution model.

9. The method according to claim 1, characterized in that The synchronization of the updated spatio-temporal evolution model to the evolution prediction processing of the next surveying and mapping cycle includes: At the starting time node of the next surveying and mapping cycle, extract the updated set of mapping parameters from the original parameter storage area of the target spatio-temporal evolution model; Identify the mapping parameter version identifiers of each parameter in the extracted updated set of mapping parameters, and parse the parameter generation timestamp and spatial grid coding included in the mapping parameter version identifier; Perform matching degree calculation processing on the parameter generation timestamp and the prediction starting time of the next surveying and mapping cycle to generate a time matching coefficient matrix; Perform validity screening processing on the extracted updated set of mapping parameters according to the time matching coefficient matrix, and retain the valid parameters with a time matching coefficient greater than the set threshold; Perform grid-by-grid matching processing on the spatial grid coding of the valid parameters and the geographical spatial grid division data of the next surveying and mapping cycle; When it is detected that there are no valid parameters corresponding to the target spatial grid coding, perform spatial interpolation and completion processing using the valid parameters of adjacent spatial grid codings; Load the completed valid parameters into the real-time prediction operation module of the basic spatio-temporal evolution model in the order of spatial grid coding; Generate a mapping relationship index table between the mapping parameter version identifiers of the valid parameters and the time nodes of the next surveying and mapping cycle. The mapping relationship index table includes spatial grid coding, parameter version identifier, and loading timestamp; During the evolution prediction processing of the next surveying and mapping cycle, continuously monitor the time difference between the timestamps of each parameter version identifier in the mapping relationship index table and the current prediction time node; When the time difference exceeds the preset synchronization threshold, trigger a parameter version update request and extract the latest mapping parameter version from the parameter storage area of the updated spatio-temporal evolution model; Replace the latest mapping parameter version extracted with the spatial grid coding in the real-time prediction operation module, and synchronously update the corresponding entries in the mapping relationship index table; After completing all prediction processing of the next surveying and mapping cycle, associatively store the mapping parameter version identifiers actually used and the corresponding prediction results in the model log database.

10. A spatio-temporal evolution modeling system, characterized in that, It includes a processor, a memory, and a bus connected to the processor; wherein, the processor and the memory communicate with each other through the bus; the processor is used to call program instructions in the memory to execute the spatio-temporal evolution modeling method driven by ground mapping according to any one of claims 1-9.

Images