Landslide early warning method based on physical acquisition of multivariate data for time series analysis

Abstract

The application discloses a landslide early warning method based on physical collection of multivariate data for timing analysis, comprising the following steps: S1, basic data collection: synchronously collecting multivariate environmental parameters and uniformly recording a time axis; S2, comprehensive judgment of rainfall state: sequentially completing rainfall existence judgment, rainfall amount estimation and rainfall form judgment; S3, characteristic processing and relationship analysis: standardizing the collected data, and identifying dominant variables and response time delay through causal test and correlation analysis; the application relates to the technical field of geological disaster monitoring and early warning. The landslide early warning method based on physical collection of multivariate data for timing analysis dynamically adjusts characteristic weights and response time delay through Granger causal test, cross-correlation analysis and a vector autoregressive model, accurately adapts to differences in different climates and geological environments, effectively solves the problems of weak generalization ability and poor adaptability to complex environments of traditional models, and ensures the stability and sensitivity of early warning indexes in various scenes.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring and early warning technology, specifically a landslide early warning method based on time-series analysis of multivariate data collected physically. Background Technology

[0002] Landslides are a common and sudden geological disaster in mountainous areas. Their occurrence is related to a variety of factors such as meteorological conditions, geological structure, and soil characteristics, and they seriously threaten the safety of residents' lives and property as well as the stability of infrastructure such as transportation and water conservancy.

[0003] The reference patent title is: A Regional Rainfall-Induced Landslide Monitoring and Early Warning Method Based on Multi-Source Data Fusion (Authorization Announcement No.: CN116167498A, Authorization Announcement Date: 2023.05.26), which includes: constructing a landslide spatiotemporal database; acquiring multi-source landslide data and adding the multi-source data to the landslide spatiotemporal database; performing correlation analysis on the landslide spatiotemporal database with added multi-source data to obtain all highly correlated factors of landslide disasters; constructing a regional rainfall-induced landslide spatiotemporal prediction model, inputting the highly correlated factors into the regional rainfall-induced landslide spatiotemporal prediction model, and obtaining the occurrence time and spatial region of landslide events. This method can contribute to the sustainable economic development of the study area, especially in the effective management of geological disaster risks.

[0004] Based on the aforementioned document, current mainstream landslide early warning technologies are mainly divided into two categories: physical model methods and data-driven methods, but both have significant drawbacks. Physical modeling methods obtain physical parameters of mountains through large-scale laboratory experiments and in-situ monitoring, and then simulate landslide displacement behavior. However, such models require a lot of manpower and resources, are costly and difficult to reproduce, and have poor adaptability to complex geological environments, with prediction results deviating significantly from actual landslide situations. In data-driven approaches, traditional machine learning models (such as support vector regression and random forest) are static models that do not consider the dynamic characteristics of landslide evolution, resulting in limited prediction accuracy. While time series algorithms based on deep learning (such as LSTM and GRU) can capture some dynamic features, they have high computational complexity and stringent hardware requirements, making it difficult to meet the real-time requirements of landslide early warning. Furthermore, they do not fully address the prediction error problem caused by external triggering factors (such as precipitation and human activities). Therefore, this invention provides a landslide early warning method based on time series analysis of physically collected multivariate data. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a landslide early warning method based on time-series analysis of multivariate data collected physically. This method solves the problems of high cost, large prediction bias and difficulty in reproducing existing physical models, static nature of traditional machine learning models, computational complexity and poor real-time performance of deep learning models, and difficulty in adapting existing methods to different environments and prediction errors caused by external triggering factors.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a landslide early warning method based on time-series analysis of physically acquired multivariate data, comprising the following steps: S1. Basic Data Acquisition: Synchronously collect multiple environmental parameters and record them on a unified timeline; S2. Comprehensive judgment of precipitation status: Sequentially complete the judgment of precipitation existence, precipitation amount estimation and precipitation pattern judgment; S3. Feature Processing and Relationship Analysis: Standardize the collected data and identify the dominant variables and response delays through causal testing and correlation analysis. S4. Dynamic early warning model construction: Establish a quantitative relationship between variables and landslide risk, characterize the dynamic interaction mechanism of multiple variables and achieve adaptive learning; S5. Tiered early warning output: Calculates real-time risk scores and outputs corresponding level early warning signals based on preset thresholds.

[0007] Preferably, the basic data collection in S1 is synchronously collected at a frequency of 5 minutes / time. The collected parameters include air temperature, relative humidity, absolute humidity, soil temperature, relative humidity, absolute humidity and atmospheric pressure. All data are synchronously stored according to a unified timeline.

[0008] Preferably, the logic for determining the existence of precipitation in S2 is as follows: When the air humidity is ≥90%, the average absolute humidity is above the 85th percentile, and the duration exceeds 30 minutes, it is considered that there is precipitation. When air humidity is high but soil-related parameters do not change significantly, it is judged as fog, dew, or frost that has not reached the ground. When all collected parameters remain stable, it is determined that there is no precipitation.

[0009] Preferably, the precipitation estimation in S2 is based on the correspondence between precipitation duration and soil moisture increase: a short-term soil moisture increase of 5% to 10% corresponds to light rain or light snow, a medium-term soil moisture increase of 10% to 20% corresponds to moderate rain or moderate snow, and a long-term soil moisture increase of ≥20% corresponds to heavy rain or heavy snow. When the soil is frozen, the snowfall is estimated by combining the duration of air humidity saturation and temperature parameters.

[0010] Preferably, the precipitation pattern determination in S2 is based on the trends of temperature and soil moisture changes: Rainfall is defined as an air temperature > 0℃, a soil temperature > 0℃, and a rapid increase in soil moisture. Snowfall is defined as an air temperature below 0°C, a soil temperature below 0°C, and no significant increase in soil moisture. When the air temperature is around 0℃, the soil temperature is around 0℃, and the soil moisture is slowly increasing, it is determined to be sleet. Freezing rain is defined as an air temperature < 0℃, a soil temperature < 0℃, an air humidity saturation, and a soil humidity that remains unchanged.

[0011] Preferably, the feature processing and relationship analysis in S3 specifically include: S3-1. Extract features from the collected data and remove outliers, then perform standardization to construct a multidimensional time series feature set; S3-2. Granger causality test was used to analyze the causal relationship between various environmental factors and landslide risk, and to identify the dominant variables and their lag periods. S3-3. By using cross-correlation analysis to capture the lag effects of each dominant variable, the optimal response delay of the model at different early warning stages is determined based on the correlation coefficients under different time lags.

[0012] Preferably, the construction of the dynamic early warning model in S4 specifically includes: S4-1. Use ordinary least squares regression to establish a quantitative mapping relationship between the dominant variables and the landslide risk index, and calculate the risk contribution weight of each variable. S4-2. Introduce a vector autoregression model to characterize the dynamic interaction and coupling mechanism among multiple variables, integrate Granger causal analysis and cross-correlation analysis results, and dynamically adjust feature weights according to real-time changes in environmental parameters to achieve adaptive learning of the model for different climates and geological environments.

[0013] Preferably, in the graded early warning output of S5, three levels of early warning thresholds are preset: green, orange, and red. Green early warning corresponds to a low-risk state, orange early warning corresponds to a medium-risk state, and red early warning corresponds to a high-risk state. The system continuously calculates the real-time landslide risk score. When the score reaches the corresponding early warning threshold, the corresponding level of early warning signal is output immediately to realize continuous monitoring and real-time early warning of landslide risk.

[0014] Beneficial effects This invention provides a landslide early warning method based on time-series analysis of physically acquired multivariate data. Compared with existing technologies, it has the following advantages: 1. This landslide early warning method based on time-series analysis of physically collected multivariate data dynamically adjusts feature weights and response delays through Granger causality test, cross-correlation analysis and vector autoregression model, accurately adapting to differences in climate and geological environment, effectively solving the problems of weak generalization ability and poor adaptability to complex environments of traditional models, and ensuring the stability and sensitivity of early warning indicators in diverse scenarios.

[0015] 2. This landslide early warning method based on time-series analysis of multi-dimensional data collected physically only requires the collection of seven types of conventional environmental parameters, such as air temperature and soil moisture. The data requirements are small and the types are few. There is no need to monitor complex geological and physical parameters. With the help of sensors, it can be used immediately, which greatly reduces the cost of data collection, deployment and maintenance. At the same time, it avoids the high computational overhead of deep learning models. Through simple algorithms, it can quickly calculate and meet the time sensitivity requirements of landslide early warning, and achieve real-time response.

[0016] 3. This landslide early warning method, based on time-series analysis of physically collected multivariate data, constructs the causal relationship between multivariate data and early warning targets. Combining a multivariate dynamic interaction and coupling mechanism, it comprehensively considers the complexity and dynamism of landslide triggering factors through steps such as three-level judgment of precipitation status and identification of dominant variables, significantly reducing prediction errors. The three-level graded early warning system can output accurate early warning signals in advance, providing reliable support for geological disaster prevention and control, infrastructure safety operation and maintenance, and emergency decision-making. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of the landslide early warning method of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 The present invention provides a technical solution: A landslide early warning method based on time-series analysis of physically acquired multivariate data includes the following steps: S1. Basic Data Acquisition: Synchronously collect multiple environmental parameters and record them on a unified timeline; S2. Comprehensive judgment of precipitation status: Sequentially complete the judgment of precipitation existence, precipitation amount estimation and precipitation pattern judgment; S3. Feature Processing and Relationship Analysis: Standardize the collected data and identify the dominant variables and response delays through causal testing and correlation analysis. S4. Dynamic early warning model construction: Establish a quantitative relationship between variables and landslide risk, characterize the dynamic interaction mechanism of multiple variables and achieve adaptive learning; S5. Tiered early warning output: Calculates real-time risk scores and outputs corresponding level early warning signals based on preset thresholds.

[0020] By using Granger causality tests, cross-correlation analysis, and vector autoregression models, feature weights and response delays are dynamically adjusted to accurately adapt to differences in climate and geological environments. This effectively solves the problems of weak generalization ability and poor adaptability to complex environments in traditional models, ensuring the stability and sensitivity of early warning indicators in diverse scenarios.

[0021] In this embodiment, basic data collection in S1 is synchronously collected at a frequency of 5 minutes / time. The collected parameters include air temperature, relative humidity, absolute humidity, soil temperature, relative humidity, absolute humidity, and atmospheric pressure. All data are synchronously stored according to a unified timeline.

[0022] The collected data is uploaded to the early warning platform in real time via a 5G wireless transmission module and stored according to a unified timeline. During the data collection process, ensure the consistency and integrity of the data sequence to provide a reliable data foundation for subsequent analysis; In this embodiment, the logic for determining the existence of precipitation in S2 is as follows: When the air humidity is ≥90%, the average absolute humidity is above the 85th percentile, and the duration exceeds 30 minutes, it is considered that there is precipitation. When air humidity is high but soil-related parameters do not change significantly, it is judged as fog, dew, or frost that has not reached the ground. When all collected parameters remain stable, it is determined that there is no precipitation.

[0023] The specific soil-related parameters are soil moisture and soil temperature; In this embodiment, the precipitation estimation in S2 is based on the correspondence between precipitation duration and soil moisture increase: a short-term soil moisture increase of 5% to 10% corresponds to light rain or light snow, a medium-term soil moisture increase of 10% to 20% corresponds to moderate rain or moderate snow, and a long-term soil moisture increase of ≥20% corresponds to heavy rain or heavy snow. When the soil is frozen, the snowfall is estimated by combining the duration of air humidity saturation and temperature parameters.

[0024] The estimation process needs to eliminate interference from humidity changes caused by soil freezing to ensure the accuracy of snowfall estimation; Short duration (usually <1 hour), medium duration (usually 1-6 hours), long duration (usually >6 hours); Light rain or light snow (0.1-9.9mm), moderate rain or moderate snow (10-24.9mm); In this embodiment, the precipitation pattern determination in S2 is based on the trends of temperature and soil moisture changes: Rainfall is defined as an air temperature > 0℃, a soil temperature > 0℃, and a rapid increase in soil moisture. Snowfall is defined as an air temperature below 0°C, a soil temperature below 0°C, and no significant increase in soil moisture. When the air temperature is around 0℃, the soil temperature is around 0℃, and the soil moisture is slowly increasing, it is determined to be sleet. Freezing rain is defined as an air temperature < 0℃, a soil temperature < 0℃, an air humidity saturation, and a soil humidity that remains unchanged.

[0025] In this embodiment, feature processing and relationship analysis in S3 specifically include: S3-1. Extract features from the collected data and remove outliers, then perform standardization to construct a multidimensional time series feature set; S3-2. Granger causality test was used to analyze the causal relationship between various environmental factors and landslide risk, and to identify the dominant variables and their lag periods. S3-3. By using cross-correlation analysis to capture the lag effects of each dominant variable, the optimal response delay of the model at different early warning stages is determined based on the correlation coefficients under different time lags.

[0026] The dominant variables are soil moisture, air humidity, and atmospheric pressure. The optimal response time is the time delay corresponding to the peak value of the correlation coefficient; In this embodiment, the construction of the dynamic early warning model in S4 specifically includes: S4-1. Use ordinary least squares regression to establish a quantitative mapping relationship between the dominant variables and the landslide risk index, and calculate the risk contribution weight of each variable. S4-2. Introduce a vector autoregression model to characterize the dynamic interaction and coupling mechanism among multiple variables, integrate Granger causal analysis and cross-correlation analysis results, and dynamically adjust feature weights according to real-time changes in environmental parameters to achieve adaptive learning of the model for different climates and geological environments.

[0027] Contribution weights are determined by quantification through regression coefficients; The basis for dynamically adjusting the feature weights is the real-time change magnitude of each variable and its correlation strength with landslide risk; In this embodiment, in the graded early warning output in S5, three levels of early warning thresholds are preset: green, orange, and red. Green early warning corresponds to a low-risk state, orange early warning corresponds to a medium-risk state, and red early warning corresponds to a high-risk state. The system continuously calculates the real-time landslide risk score. When the score reaches the corresponding early warning threshold, the system immediately outputs an early warning signal of the corresponding level, thereby realizing continuous monitoring and real-time early warning of landslide risk.

[0028] Warning signals include audible and visual alarms, SMS notifications, and platform push notifications; It only needs to collect seven types of conventional environmental parameters such as air temperature and soil moisture. The data requirements are small and the types are few. There is no need to monitor complex geological and physical parameters. With the help of sensors, it can be used immediately. This greatly reduces the cost of data collection, deployment and maintenance. At the same time, it avoids the high computational overhead of deep learning models. It can quickly calculate through simple algorithms to meet the time sensitivity requirements of landslide early warning and achieve real-time response.

[0029] By constructing a causal relationship between multivariate data and early warning targets, and combining a multivariate dynamic interaction and coupling mechanism, the system comprehensively considers the complexity and dynamism of landslide triggering factors through steps such as three-level judgment of precipitation status and identification of dominant variables. This significantly reduces prediction errors, and the three-level early warning system can output accurate early warning signals in advance, providing reliable support for geological disaster prevention and control, infrastructure safety operation and maintenance, and emergency decision-making.

[0030] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0031] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0032] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A landslide early warning method based on time-series analysis of physically acquired multivariate data, characterized in that, Includes the following steps: S1. Basic Data Acquisition: Synchronously collect multiple environmental parameters and record them on a unified timeline; S2. Comprehensive judgment of precipitation status: Sequentially complete the judgment of precipitation existence, precipitation amount estimation and precipitation pattern judgment; S3. Feature Processing and Relationship Analysis: Standardize the collected data and identify the dominant variables and response delays through causal testing and correlation analysis. S4. Dynamic early warning model construction: Establish a quantitative relationship between variables and landslide risk, characterize the dynamic interaction mechanism of multiple variables and achieve adaptive learning; S5. Tiered early warning output: Calculates real-time risk scores and outputs corresponding level early warning signals based on preset thresholds.

2. The landslide early warning method based on time-series analysis of physically acquired multivariate data according to claim 1, characterized in that: The basic data in S1 is collected synchronously at a frequency of 5 minutes / time. The collected parameters include air temperature, relative humidity, absolute humidity, soil temperature, relative humidity, absolute humidity and atmospheric pressure. All data are stored synchronously according to a unified timeline.

3. The landslide early warning method based on time-series analysis of physically acquired multivariate data according to claim 1, characterized in that: The logic for determining the existence of precipitation in S2 is as follows: When the air humidity is ≥90%, the average absolute humidity is above the 85th percentile, and the duration exceeds 30 minutes, it is considered that there is precipitation. When air humidity is high but soil-related parameters do not change significantly, it is judged as fog, dew, or frost that has not reached the ground. When all collected parameters remain stable, it is determined that there is no precipitation.

4. The landslide early warning method based on time-series analysis of physically acquired multivariate data according to claim 1, characterized in that: The precipitation estimation in S2 is based on the correspondence between precipitation duration and soil moisture increase: a short-term soil moisture increase of 5% to 10% corresponds to light rain or light snow, a medium-term soil moisture increase of 10% to 20% corresponds to moderate rain or moderate snow, and a long-term soil moisture increase of ≥20% corresponds to heavy rain or heavy snow. When the soil is frozen, the snowfall is estimated by combining the duration of air humidity saturation and temperature parameters.

5. The landslide early warning method based on time-series analysis of physically acquired multivariate data according to claim 1, characterized in that: The precipitation pattern determination in S2 is based on the trends of temperature and soil moisture changes: Rainfall is defined as an air temperature > 0℃, a soil temperature > 0℃, and a rapid increase in soil moisture. Snowfall is defined as an air temperature below 0°C, a soil temperature below 0°C, and no significant increase in soil moisture. When the air temperature is around 0℃, the soil temperature is around 0℃, and the soil moisture is slowly increasing, it is determined to be sleet. Freezing rain is defined as an air temperature < 0℃, a soil temperature < 0℃, an air humidity saturation, and a soil humidity that remains unchanged.

6. The landslide early warning method based on time-series analysis of physically acquired multivariate data according to claim 1, characterized in that: The feature processing and relation analysis in S3 specifically include: S3-1. Extract features from the collected data and remove outliers, then perform standardization to construct a multidimensional time series feature set; S3-2. Granger causality test was used to analyze the causal relationship between various environmental factors and landslide risk, and to identify the dominant variables and their lag periods. S3-3. By using cross-correlation analysis to capture the lag effects of each dominant variable, the optimal response delay of the model at different early warning stages is determined based on the correlation coefficients under different time lags.

7. The landslide early warning method based on time-series analysis of physically acquired multivariate data according to claim 1, characterized in that: The construction of the dynamic early warning model in S4 specifically includes: S4-1. Use ordinary least squares regression to establish a quantitative mapping relationship between the dominant variables and the landslide risk index, and calculate the risk contribution weight of each variable. S4-2. Introduce a vector autoregression model to characterize the dynamic interaction and coupling mechanism among multiple variables, integrate Granger causal analysis and cross-correlation analysis results, and dynamically adjust feature weights according to real-time changes in environmental parameters to achieve adaptive learning of the model for different climates and geological environments.

8. The landslide early warning method based on time-series analysis of physically acquired multivariate data according to claim 1, characterized in that: In the S5 tiered early warning output, three levels of early warning thresholds are preset: green, orange, and red. Green early warning corresponds to a low-risk state, orange early warning corresponds to a medium-risk state, and red early warning corresponds to a high-risk state. The system continuously calculates the real-time landslide risk score. When the score reaches the corresponding early warning threshold, the system immediately outputs an early warning signal of the corresponding level, thereby realizing continuous monitoring and real-time early warning of landslide risk.

Images