CN-122019994-A - Landslide monitoring data processing method and system based on time sequence matrix completion

CN122019994ACN 122019994 ACN122019994 ACN 122019994ACN-122019994-A

Abstract

The invention discloses a landslide monitoring data processing method and system based on time sequence matrix completion, in particular to the technical field of data completion and machine learning, which are used for solving the problem that the uncertainty evaluation distortion of a downstream early warning model is caused by the fact that a deterministic value is output in the existing completion method; the method comprises the steps of obtaining a time sequence observation matrix containing missing values, dividing the missing position subsets according to a missing mode, determining a differentiated completion uncertainty evaluation criterion by combining a time evolution stage of landslide deformation, generating completion uncertainty measurement of each subset based on space-time neighborhood data consistency, and finally carrying out iterative weighted completion according to the measurement to realize outputting a complete data matrix with quantized uncertainty information for each completion value.

Inventors

GE TAO
HU CHEN
LIU ZHIWEI
SUN MIN
SHI LIANGYU
WU WEIJUN

Assignees

安徽省地质实验研究所(国土资源部合肥矿产资源监督检测中心)

Dates

Publication Date: 20260512
Application Date: 20260115

Claims (10)

1. The landslide monitoring data processing method based on time sequence matrix completion is characterized by comprising the following steps of: S1, acquiring a time sequence observation matrix which is formed by observing a plurality of landslide monitoring points on a continuous time sequence and contains missing values; S2, extracting positions of all missing values from the time sequence observation matrix to form a missing position set, and dividing the missing position set according to a missing mode of the time sequence observation matrix to obtain a plurality of missing position subsets; S3, identifying the current time evolution stage of landslide deformation based on the existing observation value sequence in the time sequence observation matrix; s4, determining a corresponding complement uncertainty evaluation criterion according to the current time evolution stage; S5, generating a complement uncertainty measure of the missing position subset based on the consistency degree of the existing observation values of the missing position subset in the corresponding space-time neighborhood in the time sequence observation matrix and according to a complement uncertainty evaluation criterion; S6, iteration complement is carried out on the time sequence observation matrix according to the complement uncertainty measurement, the complement attention weights of the subsets of different missing positions are dynamically adjusted in the iteration process, and finally a complete landslide monitoring data matrix with uncertainty quantization information is generated and output.
2. The landslide monitoring data processing method based on time series matrix completion of claim 1, wherein S1 comprises: acquiring original observation data acquired by a plurality of monitoring sensors distributed on the surface of a landslide body at continuous time points; each row in the time sequence observation matrix corresponds to a landslide monitoring point, each column corresponds to a continuous time point, and element values in the time sequence observation matrix comprise original observation data and missing values formed by data missing.
3. The landslide monitoring data processing method based on time series matrix completion of claim 1, wherein S2 comprises: Scanning a time sequence observation matrix to identify landslide monitoring point identifiers and observation time point identifiers corresponding to each missing value; Combining all the identified landslide monitoring point identifiers with the observing time point identifiers to construct a missing position set; dividing the missing position set into a plurality of missing position subsets according to the spatial adjacency and the time continuity characteristics of each missing value in the missing position set in the time sequence observation matrix.
4. The landslide monitoring data processing method based on time series matrix completion of claim 1, wherein S3 comprises: Extracting effective observation data of each landslide monitoring point along the time dimension from the time sequence observation matrix to form an existing observation value sequence of each landslide monitoring point; analyzing the existing observation value sequence of each landslide monitoring point, and calculating the deformation change rate and change trend of each landslide monitoring point in a preset time window; And identifying the current time evolution stage of landslide deformation as one of an initial deformation stage, a constant-speed deformation stage or an acceleration deformation stage based on comprehensive judgment of deformation change rates and change trends of all landslide monitoring points.
5. The landslide monitoring data processing method based on time series matrix completion of claim 1, wherein S4 comprises: Establishing a mapping relation between a time evolution stage and a complement uncertainty evaluation criterion, wherein the time evolution stage comprises an initial deformation stage, a constant-speed deformation stage and an acceleration deformation stage; and selecting a corresponding complement uncertainty evaluation criterion from the mapping relation according to the identified current time evolution stage.
6. The method of claim 5, wherein the supplemental uncertainty evaluation criteria selected for the acceleration deformation phase is lower than the supplemental uncertainty evaluation criteria selected for the isokinetic deformation phase or the initial deformation phase.
7. The landslide monitoring data processing method based on time series matrix completion of claim 1, wherein S5 comprises: Determining the corresponding space neighborhood radius and the time window length of each missing position subset in the time sequence observation matrix; Extracting all the existing observation values within the range defined by the space neighborhood radius and the time window length; calculating the statistical dispersion of the extracted existing observation values to be used as the consistency degree of the existing observation values in the space-time adjacent domains corresponding to the missing position subsets; And mapping the statistical dispersion into a corresponding complement uncertainty measure according to an inconsistency tolerance threshold set in the complement uncertainty evaluation criterion.
8. The landslide monitoring data processing method based on time series matrix completion of claim 1, wherein S6 comprises: Calculating a weight adjustment coefficient corresponding to each missing position subset in the iterative complement process according to the complement uncertainty measurement of the missing position subset; In each iterative completion process, carrying out weighted update estimation on the missing values in each missing position subset according to the weight adjustment coefficient obtained by current calculation; And when the iterative completion process meets a preset convergence condition, generating a completion value for each missing position based on the final estimated value, and taking the completion uncertainty measurement of the missing position subset to which the missing position belongs as uncertainty quantization information of the completion value, and outputting the uncertainty quantization information together to form a complete landslide monitoring data matrix.
9. A landslide monitoring data processing method based on time series matrix completion of claim 8 wherein a subset of missing locations with higher completion uncertainty measure is given a greater weight adjustment factor.
10. Landslide monitoring data processing system based on time series matrix completion for implementing a landslide monitoring data processing method based on time series matrix completion according to any one of claims 1-9, characterized by comprising the following modules: the matrix acquisition module is used for acquiring a time sequence observation matrix which is formed by observing a plurality of landslide monitoring points on a continuous time sequence and contains missing values; the subset obtaining module is used for extracting the positions of all the missing values from the time sequence observation matrix to form a missing position set, and dividing the missing position set according to the missing mode of the time sequence observation matrix to obtain a plurality of missing position subsets; The phase identification module is used for identifying the current time evolution phase of landslide deformation based on the existing observation value sequence in the time sequence observation matrix; The criterion determining module is used for determining a corresponding complement uncertainty evaluation criterion according to the current time evolution stage; the measurement generation module is used for generating the complement uncertainty measurement of the missing position subset based on the consistency degree of the existing observation values of the missing position subset in the corresponding space-time neighborhood in the time sequence observation matrix and according to the complement uncertainty evaluation criterion; The matrix output module is used for carrying out iterative complement on the time sequence observation matrix according to the complement uncertainty measurement, dynamically adjusting the complement attention weights of the subsets of different missing positions in the iterative process, and finally generating and outputting a complete landslide monitoring data matrix with uncertainty quantization information.

Description

Landslide monitoring data processing method and system based on time sequence matrix completion Technical Field The invention relates to the technical field of data complement and machine learning, in particular to a landslide monitoring data processing method and system based on time sequence matrix complement. Background In the landslide monitoring field, by continuously observing a plurality of sensor nodes arranged on the surface of a side slope, time sequence data reflecting the deformation state of the side slope can be obtained, and the time sequence data is always missing to different degrees under the influence of natural geographic environment and equipment factors, so that a time sequence data matrix with missing values is formed, and the missing values can be estimated by adopting a time sequence matrix complement technology to restore the integrity of the data. The existing completion method aims at learning space-time correlation from known observation data and calculating a determined estimated value for each missing position so as to generate a complete data matrix which can be used for subsequent landslide stability evaluation and early warning models. However, when the decision task of landslide early warning is oriented, the essence of landslide early warning is to perform risk assessment and decision based on incomplete and noisy observation information, the effectiveness of a decision model (especially a probability-based machine learning model) depends on accurate quantification of uncertainty of data and model inference, a completion process is regarded as an independent data repair link in the prior art, an output certainty completion value covers the estimated uncertainty introduced by data deletion, and when the data are regarded as accurate completion data to be input into a downstream early warning model, uncertainty propagation and assessment mechanisms in the model are distorted, reliability assessment distortion of early warning signals is caused, mismatch between a target of a data processing stage and a requirement of a final application stage is caused, and the decision reliability of the whole monitoring early warning system is weakened. Disclosure of Invention In order to overcome the above-mentioned drawbacks of the prior art, the present invention provides a landslide monitoring data processing method and system based on time series matrix completion to solve the problems set forth in the above-mentioned background art. In order to achieve the above purpose, the present invention provides the following technical solutions: a landslide monitoring data processing method based on time sequence matrix completion comprises the following steps: S1, acquiring a time sequence observation matrix which is formed by observing a plurality of landslide monitoring points on a continuous time sequence and contains missing values; S2, extracting positions of all missing values from the time sequence observation matrix to form a missing position set, and dividing the missing position set according to a missing mode of the time sequence observation matrix to obtain a plurality of missing position subsets; S3, identifying the current time evolution stage of landslide deformation based on the existing observation value sequence in the time sequence observation matrix; s4, determining a corresponding complement uncertainty evaluation criterion according to the current time evolution stage; S5, generating a complement uncertainty measure of the missing position subset based on the consistency degree of the existing observation values of the missing position subset in the corresponding space-time neighborhood in the time sequence observation matrix and according to a complement uncertainty evaluation criterion; S6, iteration complement is carried out on the time sequence observation matrix according to the complement uncertainty measurement, the complement attention weights of the subsets of different missing positions are dynamically adjusted in the iteration process, and finally a complete landslide monitoring data matrix with uncertainty quantization information is generated and output. Further, S1 includes: acquiring original observation data acquired by a plurality of monitoring sensors distributed on the surface of a landslide body at continuous time points; each row in the time sequence observation matrix corresponds to a landslide monitoring point, each column corresponds to a continuous time point, and element values in the time sequence observation matrix comprise original observation data and missing values formed by data missing. Further, S2 includes: Scanning a time sequence observation matrix to identify landslide monitoring point identifiers and observation time point identifiers corresponding to each missing value; Combining all the identified landslide monitoring point identifiers with the observing time point identifiers to construct a missing position set; dividing the miss