CN-122022198-A - Landslide space prediction method and system considering space constraint and time sequence deformation

CN122022198ACN 122022198 ACN122022198 ACN 122022198ACN-122022198-A

Abstract

The application relates to the technical field of geological disaster monitoring and prediction, and discloses a landslide space prediction method and system considering space constraint and time sequence deformation. And acquiring long-term environmental factors, time sequence deformation monitoring data and historical landslide marking data of the target area, and performing multi-source data space alignment to form unit-level input data. And extracting time evolution features based on deformation time sequence data to obtain the time consistency characterization quantity and the time interval state. And determining a four-quadrant combination state according to the space and time interval state, selecting a corresponding prediction processing mode, and executing prediction under the mode to obtain landslide prediction output. And outputting landslide prediction results and corresponding mode identifiers. The method is suitable for a long-term operation scene with continuously changing geographic environment and monitoring conditions, and can simultaneously ensure the spatial consistency, the time consistency and the output stability of the prediction result.

Inventors

ZHAO ZHENG
OUYANG CHAOJUN
AN HUICONG
ZHOU SHU
YANG TAO
LIU JIE

Assignees

中国科学院、水利部成都山地灾害与环境研究所

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (10)

1. A landslide spatial prediction method considering spatial constraint and time sequence deformation, comprising: Acquiring long-term environment factor data, deformation monitoring time series data and historical landslide marking data of a target area, and executing space alignment processing on the long-term environment factor data and the deformation monitoring time series data based on a drawing unit system to form unit-level input data corresponding to each drawing unit one by one; Determining a spatial relationship between drawing units based on a drawing unit system, generating spatial association features for long-term environmental factor data according to the spatial relationship to obtain a spatial consistency characterization quantity, and determining a spatial interval state based on the spatial consistency characterization quantity, wherein the spatial interval state at least comprises a spatial high-consistency interval state and a spatial low-consistency interval state; Generating a time evolution characteristic according to the change information of the deformation monitoring time sequence data in a preset time range to obtain a time consistency characterization quantity, and determining a time interval state based on the time consistency characterization quantity, wherein the time interval state at least comprises a time high-consistency interval state and a time low-consistency interval state; Determining a four-quadrant combination state corresponding to each drawing unit based on the space interval state and the time interval state, and selecting a prediction processing mode corresponding to the four-quadrant combination state according to the four-quadrant combination state, wherein: Selecting a first prediction processing mode under the condition that the space interval state is a space high consistency interval state and the time interval state is a time high consistency interval state; Selecting a second prediction processing mode under the condition that the space interval state is a space high-consistency interval state and the time interval state is a time low-consistency interval state; selecting a third prediction processing mode under the condition that the space interval state is a space low consistency interval state and the time interval state is a time high consistency interval state; selecting a fourth prediction processing mode under the condition that the space interval state is a space low consistency interval state and the time interval state is a time low consistency interval state; Under the selected prediction processing mode, performing prediction processing based on the spatial consistency characterization quantity and the temporal consistency characterization quantity to obtain landslide prediction output of each drawing unit; and outputting landslide prediction output and a prediction processing mode identifier corresponding to the landslide prediction output as a landslide space prediction result.
2. A landslide spatial prediction method according to claim 1 comprising, in terms of the determining of spatial relationships between cartographic units: Determining a first type of spatial relationship based on the spatial neighboring relationship of the drawing unit; Determining a second type of spatial relationship based on the similarity of the long-term environmental factor data corresponding to the drawing unit; And constructing a set of directionally weighted spatial relationships for generating spatial correlation features based on the first class of spatial relationships and the second class of spatial relationships.
3. The landslide spatial prediction method of claim 2, further comprising, in terms of said constructing a set of directionally weighted spatial relationships for generating spatially-correlated features: Determining a second order adjacency relationship between drawing units based on the spatial adjacency relationship to define a candidate associated drawing unit set; and screening the associated drawing units in the candidate associated drawing unit set according to the similarity of the long-term environment factor data so as to limit the connection range and the connection mode of the spatial relationship.
4. The landslide spatial prediction method of claim 1, characterized in that in terms of the generating time evolution features, it comprises: Extracting a first time-varying feature for representing short-term variation trend from deformation monitoring time series data within the same preset time range, and extracting a second time-dependent feature for representing long-term variation association; And determining a time coherence token based on the first time-varying feature and the second time-dependent feature together.
5. The landslide spatial prediction method of claim 1, characterized by comprising, in the execution of the prediction processing: splicing the spatial correlation features and the time evolution features to form a unit cell cascade feature; And determining a landslide prediction output based on the cell-cascade characteristics.
6. The landslide spatial prediction method of claim 1 wherein the determined spatial interval state and the determined temporal interval state are both determined based on a double boundary decision rule wherein: Setting a space low boundary and a space high boundary for the space consistency characterization quantity, and setting a time low boundary and a time high boundary for the space consistency characterization quantity, wherein the space high boundary is larger than the space low boundary and the time high boundary is larger than the time low boundary; Determining a space low-consistency interval state when the space consistency characterization quantity is smaller than or equal to a space low boundary, determining a space high-consistency interval state when the space consistency characterization quantity is larger than or equal to a space high boundary, and keeping the space interval state determined in the previous updating period unchanged when the space consistency characterization quantity is larger than the space low boundary and smaller than the space high boundary; The method comprises the steps of determining a time low-coherence interval state when a time coherence characterization quantity is smaller than or equal to a time low boundary, determining a time high-coherence interval state when the time coherence characterization quantity is larger than or equal to a time high boundary, keeping the time interval state determined in a previous updating period unchanged when the time coherence characterization quantity is larger than the time low boundary and smaller than the time high boundary, and determining a current time interval state according to a preset adjacent boundary classification rule when the time coherence characterization quantity is larger than the time low boundary and smaller than the time high boundary in an initial updating period.
7. The landslide spatial prediction method of claim 1, wherein in selecting a prediction processing mode, a prediction processing parameter set corresponding to the selected prediction processing mode is determined, and the prediction processing parameter set includes at least a temporal processing parameter and a spatial processing parameter, wherein: In a first prediction processing mode, performing prediction processing using a first set of temporal processing parameters and a first set of spatial processing parameters; in the second prediction processing mode, performing prediction processing by adopting a second group of time processing parameters and a first group of space processing parameters, wherein the length of a time range corresponding to the second group of time processing parameters is longer than that of a time range corresponding to the first group of time processing parameters; in the third prediction processing mode, performing prediction processing by adopting a first group of time processing parameters and a second group of space processing parameters, wherein the space relation range corresponding to the second group of space processing parameters is larger than the space relation range corresponding to the first group of space processing parameters; in the fourth prediction processing mode, the prediction processing is performed using the second set of temporal processing parameters and the second set of spatial processing parameters.
8. The landslide spatial prediction method of claim 1 further comprising a mode-switching robust constraint in terms of selecting a prediction processing mode, wherein: respectively determining four-quadrant combination states in at least two continuous updating periods; only when the four-quadrant combination states in at least two continuous updating periods are consistent, determining the corresponding prediction processing mode as a current execution mode; And when the four-quadrant combination states in at least two continuous updating periods are inconsistent, keeping the prediction processing mode of the last updating period unchanged.
9. The landslide spatial prediction method of claim 6 wherein the updating of the spatial low boundary, spatial high boundary, temporal low boundary, and temporal high boundary satisfies: triggering updating at preset time intervals or triggering updating when the number of newly-increased deformation monitoring calendar elements reaches a preset number; Reconstructing a reference statistical distribution for determining a boundary based on a latest preset time window after triggering the update; and updating the space low boundary, the space high boundary, the time low boundary and the time high boundary based on the reconstructed reference statistical distribution, wherein each updated boundary is limited in a preset effective value range.
10. A landslide spatial prediction system taking into account spatial constraints and temporal deformations, comprising: The system comprises a data acquisition module, a data alignment module, a space characterization module, a time characterization module, a mode selection module, a prediction processing module and a result output module; The modules are in data interaction connection according to the prediction processing flow, the output of the preamble module is used for the call of the following related module, and the prediction processing mode identifier output by the mode selection module is used for the call of the result output module; The data acquisition module is used for acquiring long-term environment factor data, deformation monitoring time sequence data and historical landslide marking data of the target area; the data alignment module is used for performing space alignment processing on the long-term environment factor data and the deformation monitoring time sequence data based on a drawing unit system so as to form unit-level input data corresponding to each drawing unit one by one; The space characterization module is used for determining the space relation among the drawing units based on the drawing unit system, generating space correlation characteristics for the long-term environment factor data according to the space relation to obtain a space consistency characterization quantity, and determining a space interval state based on the space consistency characterization quantity; the space interval state at least comprises a space high-consistency interval state and a space low-consistency interval state; the time characterization module is used for generating a time evolution characteristic according to the change information of the deformation monitoring time sequence data in a preset time range so as to obtain a time consistency characterization quantity, and determining a time interval state based on the time consistency characterization quantity, wherein the time interval state at least comprises a time high-consistency interval state and a time low-consistency interval state; The mode selection module is configured to determine a four-quadrant combination state corresponding to each drawing unit based on the space interval state and the time interval state, and select a prediction processing mode corresponding to the four-quadrant combination state, where: Selecting a first prediction processing mode under the condition that the space interval state is a space high consistency interval state and the time interval state is a time high consistency interval state; Selecting a second prediction processing mode under the condition that the space interval state is a space high-consistency interval state and the time interval state is a time low-consistency interval state; selecting a third prediction processing mode under the condition that the space interval state is a space low consistency interval state and the time interval state is a time high consistency interval state; Selecting a fourth prediction processing mode under the condition that the space interval state is a space low consistency interval state and the time interval state is a time low consistency interval state; the prediction processing module is used for executing prediction processing based on the spatial consistency characterization quantity and the time consistency characterization quantity under the selected prediction processing mode so as to obtain landslide prediction output of each drawing unit; The result output module is used for outputting the landslide prediction output and a prediction processing mode identifier corresponding to the landslide prediction output as a landslide space prediction result.

Description

Landslide space prediction method and system considering space constraint and time sequence deformation Technical Field The embodiment of the invention relates to the technical field of geological disaster monitoring and prediction, in particular to a landslide space prediction method and system considering space constraint and time sequence deformation. Background Landslide is one of typical geological disaster types, and has the characteristics of strong burst, wide influence range, obvious threat to engineering facilities and personnel safety and the like. In order to meet engineering requirements of geological disaster prevention and risk management, a landslide space prediction system is generally used for continuously evaluating landslide risks of different space positions in a target area and outputting prediction results facing a drawing unit. The system is required to continuously run for a long time in actual deployment, continuously receives long-term environment factor data and deformation monitoring time sequence data from a target area, and combines historical landslide marking data to form prediction output so as to support risk identification and dynamic management of area scale. In the long-term operation process, the geographic environment and the monitoring condition of the target area have complexity and continuous variability, and different drawing units have association relations in space, and meanwhile, the monitoring data have a requirement of continuity in time. Because the running period is long, the data sources are more, the environmental condition changes frequently, the landslide space prediction system is easy to generate the condition that the prediction output is difficult to be consistent between the space dimension and the time dimension in long-term use, and the prediction results of adjacent drawing units in the same area lack of coordination, or the prediction results of the same drawing unit in the continuous monitoring period lack of consistency. The unstable output can cause that the prediction result is difficult to form stable and usable space distribution situation in engineering application, and the use requirements of early warning and risk identification under continuous operation conditions are difficult to be met, so that the long-term deployment and engineering popularization of the landslide space prediction system are limited. Disclosure of Invention The invention provides a landslide space prediction method and a landslide space prediction system considering space constraint and time sequence deformation, which aim to solve the technical problem that a landslide space prediction system is difficult to stably output in long-term operation and simultaneously meets the prediction results of space consistency and time consistency under the condition of complex and continuously changing geographic environment and monitoring. To achieve the above object, according to a first aspect of the present invention, there is provided a landslide spatial prediction method considering spatial constraint and temporal deformation, comprising: Acquiring long-term environment factor data, deformation monitoring time series data and historical landslide marking data of a target area, and executing space alignment processing on the long-term environment factor data and the deformation monitoring time series data based on a drawing unit system to form unit-level input data corresponding to each drawing unit one by one; Determining a spatial relationship between drawing units based on a drawing unit system, generating spatial association features for long-term environmental factor data according to the spatial relationship to obtain a spatial consistency characterization quantity, and determining a spatial interval state based on the spatial consistency characterization quantity, wherein the spatial interval state at least comprises a spatial high-consistency interval state and a spatial low-consistency interval state; Generating a time evolution characteristic according to the change information of the deformation monitoring time sequence data in a preset time range to obtain a time consistency characterization quantity, and determining a time interval state based on the time consistency characterization quantity, wherein the time interval state at least comprises a time high-consistency interval state and a time low-consistency interval state; Determining a four-quadrant combination state corresponding to each drawing unit based on the space interval state and the time interval state, and selecting a prediction processing mode corresponding to the four-quadrant combination state according to the four-quadrant combination state, wherein: Selecting a first prediction processing mode under the condition that the space interval state is a space high consistency interval state and the time interval state is a time high consistency interval state; Selecting a second prediction proces