CN-122020404-A - Rainfall landslide susceptibility evaluation method, system, equipment and medium

CN122020404ACN 122020404 ACN122020404 ACN 122020404ACN-122020404-A

Abstract

The invention provides a rainfall landslide susceptibility evaluation method, a rainfall landslide susceptibility evaluation system, rainfall landslide susceptibility evaluation equipment and rainfall landslide susceptibility evaluation medium, and belongs to the technical field of geological disaster risk evaluation and spatial information modeling. The method comprises the steps of collecting and preprocessing multisource environment and rainfall data of a target area, constructing a multiscale rainfall factor system, reducing variable collinearity through intra-group screening and a multi-stage characteristic engineering method, modeling a static environment factor by utilizing a random forest model, obtaining landslide background probability and converting the landslide background probability into a logarithmic probability bias term, constructing a Bayesian space statistical model under the constraint of the bias term, introducing a nonlinear effect term and a space random effect term of the rainfall factor, carrying out joint modeling and Bayesian inference on the landslide probability, and realizing dynamic prediction and space drawing of landslide probability. According to the invention, by fusing the nonlinear prediction capability of machine learning and the physical constraint and space modeling capability of Bayesian statistics, the stability, prediction accuracy and mechanism interpretation of an evaluation result are improved.

Inventors

MIAO ZELANG
XIONG YAOPENG
LI LANGPING
WANG XUYUAN
YANG ZEFA

Assignees

中南大学

Dates

Publication Date: 20260512
Application Date: 20260126

Claims (10)

1. The rainfall landslide susceptibility evaluation method is characterized by comprising the following steps: s1, obtaining topographic data, geological and engineering geological data, hydrological data, land utilization and coverage data, historical landslide disaster cataloging data and rainfall data of a target area; S2, constructing a rainfall factor system with multiple time scales based on rainfall data, and performing factor screening by an intra-group screening and multi-stage characteristic engineering method; s3, constructing a mixed landslide vulnerability evaluation model: step S31, modeling the screened topography, geological conditions and land utilization factors by utilizing a random forest model to obtain landslide background vulnerability probability of each space unit; s32, converting the landslide background probability into a logarithmic probability form as a bias item; S33, under the constraint of a bias term, a Bayesian space statistical model is constructed, a nonlinear effect term and a space random effect term of rainfall factors are introduced, and joint modeling is carried out on landslide occurrence probability; step S34, obtaining the posterior probability of landslide occurrence of each space unit through Bayesian inference, and carrying out explanatory analysis by combining with a SHAP method; and S4, predicting landslide occurrence probability of the target area under different time scales or rainfall scenes by using the trained mixed model, and generating a landslide susceptibility spatial distribution map.
2. The method for evaluating the susceptibility to rainfall landslide according to claim 1, wherein in the step S2, the rainfall factor system comprises an early-stage accumulated rainfall ARA and a rainfall intensity index, and the calculation formula of the early-stage accumulated rainfall ARA is as follows: ; Wherein, the Indicating the day before landslide occurrence The daily rainfall on the day, Representing the cumulative time window length; The in-group screening comprises the steps of dividing rainfall factors into early-stage accumulated rainfall ARA and rainfall intensity groups according to physical meanings, and evaluating the interpretation ability of variables to landslide occurrence in the groups respectively.
3. The method for evaluating the susceptibility to rainfall landslide of claim 1, wherein in step S2, the multi-stage feature engineering method comprises: S221, performing linear correlation analysis between variables by adopting a Pearson correlation coefficient, and eliminating redundant variables of which the absolute values of the correlation coefficients exceed a threshold value; Step S222, performing multiple collinearity analysis by using a variance expansion factor VIF, and removing variables with VIF values higher than a preset threshold; and S223, performing recursive feature elimination RFE based on the random forest model, and gradually removing variables with low contribution degree to obtain an optimal feature subset.
4. The method for evaluating the susceptibility to rainfall landslide of claim 1, wherein in step S32, the formula of calculating the bias term is: ; Wherein, the The term of the bias is indicated, And the background susceptibility probability output by the random forest model is represented.
5. The method for evaluating the susceptibility to rainfall landslide of claim 1, wherein in step S33, the bayesian space statistical model is in the form of: ; Wherein, the Representing a logarithmic probability function, The intercept term of the model is represented, A background landslide susceptibility log probability term generated by a random forest model is represented, Representing a non-linear effect function of the rainfall factor, Representing the effect of spatial randomness.
6. The rainfall landslide susceptibility evaluation method according to claim 1, wherein in step S34, bayesian inference is performed by adopting an integrated nested laplace approximation INLA method to perform posterior distribution calculation, so as to obtain posterior mean, variance and trusted interval of landslide occurrence probability.
7. The method according to claim 1, wherein in step S4, the spatial distribution map of landslide susceptibility is classified according to the prediction probability, including very low, medium, high and very high susceptibility levels, and is compared with actual landslide hazard points.
8. A rainfall landslide vulnerability assessment system, comprising: The data collection and preprocessing module is used for acquiring the topography data, the geology and engineering geology data, the hydrologic data, the land utilization and coverage data, the historical landslide disaster cataloging data and the rainfall data of the target area and preprocessing the data; The rainfall factor constructing and screening module is used for constructing a rainfall factor system with multiple time scales based on rainfall data, and performing factor screening by an in-group screening and multi-stage characteristic engineering method so as to reduce variable collinearity; The hybrid model construction module is used for modeling the screened topography, geological conditions and land utilization factors by utilizing a random forest model to obtain landslide background probability of each space unit, converting the landslide background probability into a logarithmic probability form to serve as a bias term, constructing a Bayesian space statistical model under the constraint of the bias term, introducing a nonlinear effect term and a space random effect term of rainfall factors, obtaining the posterior probability of landslide occurrence of each space unit through Bayesian inference, and carrying out interpretative analysis by combining with a SHAP method; and the landslide susceptibility simulation and drawing module is used for predicting landslide occurrence probability of the target area under different time scales or rainfall scenes by utilizing the trained mixed model to generate a landslide susceptibility spatial distribution map.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, carries out the steps of a method for evaluating the susceptibility to rainfall landslide according to any one of claims 1-7.
10. A computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of a rainfall landslide susceptibility evaluation method according to any one of claims 1-7.

Description

Rainfall landslide susceptibility evaluation method, system, equipment and medium Technical Field The invention relates to the technical field of geological disaster risk assessment and spatial information modeling, in particular to a rainfall landslide susceptibility evaluation method, a rainfall landslide susceptibility evaluation system, rainfall landslide susceptibility evaluation equipment and rainfall landslide susceptibility evaluation medium. Background Landslide is a typical geological disaster commonly affected by various factors such as topography, geological conditions, rainfall and the like, and the occurrence process of landslide has obvious nonlinear characteristics and spatial heterogeneity. The reliable landslide susceptibility evaluation is developed, and the method has important significance for regional disaster prevention and reduction, infrastructure site selection and homeland space planning. The existing landslide susceptibility evaluation method mainly comprises an empirical statistics-based model and a machine learning-based prediction model. While the traditional statistical method generally relies on linear or weak nonlinear assumption, it is difficult to fully describe landslide generation mechanism under the action of multi-factor coupling, the random forest and support vector machine learning method has certain advantages in prediction accuracy, but the internal decision process is complex, the model result is difficult to explain, and the application of the method in engineering practice and risk management is limited. In the prior art, most methods focus on improving prediction precision, and tend to unify static factors such as topography, geology and the like and dynamic trigger factors such as rainfall into a single model for training, so that different action mechanisms of landslide background susceptibility and rainfall triggering process cannot be effectively distinguished, and the model has no physical consistency in response to time-varying rainfall conditions. Meanwhile, the rainfall indexes with multiple time scales have stronger correlation in practical application, and if a screening and constraint mechanism of a system is lacked, the problem of multiple collinearity is easily introduced, so that the stability and generalization capability of a model are reduced. In addition, the existing landslide vulnerability evaluation method still has obvious defects in terms of result interpretation. On one hand, the contribution of static factors such as terrain, geology and the like lacks quantitative interpretation means, the relative effect of different factors in landslide background formation is difficult to be clarified, on the other hand, most methods only give a single weight or experience threshold value for the influence of dynamic factors such as rainfall, the nonlinear response characteristics of the rainfall triggering effect changing along with the intensity or accumulation process are difficult to be revealed, and the spatial difference of the nonlinear response characteristics cannot be further described. This problem of insufficient interpretation capability makes it difficult for the model results to support refined disaster prevention decisions and dynamic risk assessment. Therefore, there is a need for a landslide vulnerability assessment method that can distinguish landslide background vulnerability from rainfall dynamic triggering mechanism under physical and statistical constraint conditions, give consideration to prediction accuracy and interpretability, and quantitatively reveal static factor contribution, rainfall nonlinear threshold effect and spatial heterogeneity, so as to make up for the shortages of the prior art in stability, interpretability and practical application. Disclosure of Invention The invention aims to provide a rainfall landslide susceptibility evaluation method, a rainfall landslide susceptibility evaluation system, rainfall factor susceptibility evaluation equipment and a rainfall factor susceptibility evaluation medium, so as to solve the problems that in the prior art, a model is unstable due to high collinearity of rainfall factors, nonlinear threshold characteristics and spatial heterogeneity which are difficult to explicitly describe a rainfall triggering effect are difficult to describe, and the model prediction result is insufficient in physical interpretability. In order to achieve the above purpose, the invention provides a rainfall landslide susceptibility evaluation method, which comprises the following steps: s1, obtaining topographic data, geological and engineering geological data, hydrological data, land utilization and coverage data, historical landslide disaster cataloging data and rainfall data of a target area; S2, constructing a rainfall factor system with multiple time scales based on rainfall data, and performing factor screening by an intra-group screening and multi-stage characteristic engineering metho