Seasonal decomposition residual error-based landslide mass deformation anomaly identification and early warning method

Abstract

The invention discloses a landslide mass deformation abnormity identification and early warning method based on seasonal decomposition residual errors, and belongs to the technical field of slope deformation monitoring. The method comprises the steps of data acquisition and preprocessing, seasonal decomposition, residual error anomaly identification and comprehensive early warning and attribution study and judgment. According to the method, a set of complete landslide deformation intelligent analysis and early warning technical system is established through technical means such as time series data seasonal decomposition, residual component extraction, machine learning anomaly recognition and multi-source information fusion study and judgment, and anomaly accurate recognition and graded early warning on the premise of effectively stripping seasonal interference are achieved. And a reliable technical guarantee is provided for early prevention and control of landslide disasters.

Description

Technical Field

[0001] This invention relates to the field of slope deformation monitoring technology, and in particular to a method for identifying and warning of landslide deformation anomalies based on seasonal decomposition residuals. Background Technology

[0002] Landslide deformation monitoring is a core task in slope engineering and geological disaster prevention, and its accuracy and timeliness directly affect the safety of people's lives and property in landslide-affected areas. Currently, modern monitoring technologies can achieve high-frequency and accurate acquisition of landslide displacement and deformation rates, accumulating massive amounts of data for stability assessment. However, in the crucial link from data to decision-making—deformation anomaly identification and early warning—the existing technological system still has serious shortcomings.

[0003] Current landslide deformation early warning mainly relies on static threshold methods or simple statistical models based on the overall deformation rate, which suffers from three major technical bottlenecks: First, it cannot effectively separate trend and seasonal components from deformation data, misinterpreting normal fluctuations such as seasonal expansion and contraction as instability signals, leading to frequent false alarms. Second, it is insensitive to slowly accumulating creep trends and sudden accelerated deformation, resulting in delayed anomaly identification and often missing the optimal early warning window. Third, the early warning triggering mechanism is singular, lacking the ability to quickly assist in judging the causes of abnormal events, affecting the accuracy of emergency decision-making. Although some existing technologies have attempted to introduce complex time series models, they are generally computationally complex, have poor practicality, and have failed to establish a complete technical system from "data seasonal decomposition" to "residual anomaly identification" to "intelligent early warning and attribution," thus failing to meet the urgent need for early and accurate early warning of landslide disasters. (Invention Content)

[0004] This invention proposes a method for identifying and warning of landslide deformation anomalies based on seasonal decomposition residuals. It achieves accurate identification and graded warning of anomalies under the premise of effectively removing seasonal interference, providing reliable technical support for the early prevention and control of landslide disasters.

[0005] A method for identifying and warning of landslide deformation anomalies based on seasonal decomposition residuals includes the following steps:

[0006] S1 Data Acquisition and Preprocessing: Collect time-series deformation data from landslide monitoring points, and perform data cleaning and initial standardization to form time-regular deformation sequence data;

[0007] S2 seasonal decomposition: The seasonal decomposition algorithm is used to decompose the standardized deformed sequence data to obtain the trend term reflecting long-term changes, the seasonal term reflecting periodic fluctuations, and the residual term remaining after removing the trend term and the seasonal term.

[0008] S3 Residual Anomaly Identification: Analyze the characteristics of the residual sequence, identify anomalies in the residual sequence formed by the residual terms, and identify abnormal residual points that exceed the normal fluctuation range as criteria for landslide deformation anomalies.

[0009] S4 Integrated Early Warning and Attribution Analysis: When an anomaly is detected, the risk level of the landslide is assessed and an early warning is issued by combining the long-term trend of the integrated trend item and the periodic pattern of the seasonal item. At the same time, geological and meteorological data from the same period are used to assist in the judgment of the cause of the anomaly.

[0010] The method of this invention establishes a complete intelligent analysis and early warning technology system for landslide deformation through techniques such as seasonal decomposition of time-series data, extraction of residual components, machine learning anomaly identification, and multi-source information fusion and analysis. It achieves accurate anomaly identification and graded early warning under the premise of effectively removing seasonal interference, providing reliable technical support for the early prevention and control of landslide disasters. Attached Figure Description

[0011] Figure 1 This is a flowchart of the landslide anomaly identification and early warning system of the present invention.

[0012] Figure 2 This is a schematic diagram of the layout of surface displacement observation points for the landslide body in Example 1.

[0013] Figure 3 This is a schematic diagram of the displacement change at a certain measuring point in the landslide body in Example 1;

[0014] In the figure, J03X represents the horizontal displacement in the X direction, J03Y represents the horizontal displacement in the Y direction, and J03H represents the vertical displacement.

[0015] Figure 4 This is a schematic diagram of the seasonal decomposition of the X-direction landslide body in Example 1.

[0016] Figure 5 This is a schematic diagram of the seasonal decomposition of the Y-direction landslide body in Example 1.

[0017] Figure 6 This is a schematic diagram of the seasonal decomposition of the H-direction landslide body in Example 1.

[0018] Figure 7 This is a schematic diagram of automatic X-direction residual identification in Example 1.

[0019] Figure 8 This is a schematic diagram of automatic Y-axis residual identification in Example 1.

[0020] Figure 9 This is a schematic diagram of automatic H-direction residual identification in Example 1.

[0021] Figure 10 This is a schematic diagram of the comprehensive early warning and attribution analysis process of the present invention. Detailed Implementation

[0022] Example 1: A method for identifying and warning of landslide deformation anomalies based on seasonal decomposition residuals, comprising the following steps:

[0023] S1 Data Acquisition and Preprocessing: Collect time-series deformation data from landslide monitoring points, and perform data cleaning and initial standardization to form time-regular deformation sequence data;

[0024] S2 seasonal decomposition: The seasonal decomposition algorithm is used to decompose the standardized deformed sequence data to obtain the trend term reflecting long-term changes, the seasonal term reflecting periodic fluctuations, and the residual term remaining after removing the trend term and the seasonal term.

[0025] S3 Residual Anomaly Identification: Analyze the characteristics of the residual sequence, identify anomalies in the residual sequence formed by the residual terms, and identify abnormal residual points that exceed the normal fluctuation range as criteria for landslide deformation anomalies.

[0026] S4 Integrated Early Warning and Attribution Analysis: When an anomaly is detected, the risk level of the landslide is assessed and an early warning is issued by combining the long-term trend of the integrated trend item and the periodic pattern of the seasonal item. At the same time, geological and meteorological data from the same period are used to assist in the judgment of the cause of the anomaly.

[0027] The monitoring points for the landslide body can determine the monitoring content according to the monitoring design principles. When surface deformation monitoring is the main focus, surface deformation monitoring captures the range of surface displacement and deformation rate.

[0028] In S1, data cleaning and initial standardization are performed using the Z-Score standardization method to linearly interpolate and impute missing values. The calculation formula is as follows:

[0029] ;

[0030] in, This is the original data. The mean of the sequence. The standard deviation of the sequence. This is the standardized data.

[0031] The time-series deformation data in S1 includes the displacement and monthly average deformation rate data of the landslide monitoring points. The monitoring frequency is no less than once a month, and the data collection period is no less than 12 months to cover the complete seasonal cycle. Taking the J03 monitoring point as an example, the horizontal displacement data in the X direction (downstream), the horizontal displacement data in the Y direction, and the vertical displacement data of this point are monitored and listed in Tables 1-1, 1-2, and 1-3, respectively, forming a time-regular deformation sequence data, specifically a monthly average deformation rate sequence. When the monitoring data is cumulative displacement data, the monthly average deformation rate is calculated by performing time-series differencing on the displacement data.

[0032] Table 1-1 Horizontal Displacement Data of Measurement Point J03 in the X Direction

[0033] Observation date Monthly average deformation rate Observation date Monthly average deformation rate July 2023 0.05 September 2024 0.10 August 2023 -0.06 October 2024 -0.22 September 2023 0.04 November 2024 0.41 October 2023 -0.01 December 2024 -0.26 November 2023 0.16 January 2025 -0.13 December 2023 -0.02 February 2025 0.06 January 2024 0.03 March 2025 0.25 February 2024 0.04 April 2025 -0.02 March 2024 0.26 May 2025 0.05 April 2024 -0.04 June 2025 0.21 May 2024 0.19 July 2025 0.29 June 2024 -0.16 August 2025 -0.04 July 2024 0.03 September 2025 -0.05 August 2024 -0.01 October 2025 0.26

[0034] Table 1-2 Horizontal Displacement Data of Measuring Point J03 in the Y Direction

[0035] Observation date Monthly average deformation rate Observation date Monthly average deformation rate July 2023 0.32 September 2024 0.16 August 2023 0.34 October 2024 -0.01 September 2023 0.25 November 2024 0.08 October 2023 0.32 December 2024 0.34 November 2023 0.45 January 2025 0.39 December 2023 0.12 February 2025 0.57 January 2024 0.28 March 2025 0.31 February 2024 0.45 April 2025 0.30 March 2024 0.56 May 2025 0.20 April 2024 -0.12 June 2025 0.22 May 2024 0.31 July 2025 0.64 June 2024 0.08 August 2025 0.00 July 2024 -0.02 September 2025 0.47 August 2024 0.46 October 2025 -0.41

[0036] Table 1-3 Vertical Displacement Data of J03 Measuring Point

[0037] Observation date Monthly average deformation rate Observation date Monthly average deformation rate July 2023 0.30 September 2024 0.13 August 2023 0.08 October 2024 0.16 September 2023 0.27 November 2024 0.18 October 2023 0.30 December 2024 0.11 November 2023 0.18 January 2025 0.48 December 2023 0.37 February 2025 0.27 January 2024 0.15 March 2025 0.22 February 2024 0.56 April 2025 0.24 March 2024 0.21 May 2025 0.20 April 2024 0.32 June 2025 0.28 May 2024 0.04 July 2025 1.11 June 2024 0.11 August 2025 0.05 July 2024 0.20 September 2025 0.32 August 2024 0.34 October 2025 0.63

[0038] The seasonal decomposition algorithm in S2 decomposes the standardized deformed sequence into trend, seasonality, and residual terms, and the calculation formula is as follows:

[0039] ;

[0040] in, The data is the standardized deformed sequence data at time t. The value of the trend term at time t. The seasonality value at time t. Let t be the value of the residual term at time t.

[0041] The trend term represents the non-periodic long-term evolution direction and rate dominated by the internal geological conditions and long-term load of the landslide body. Its magnitude directly reflects the stability state and potential risk level of the landslide body.

[0042] The seasonality term characterizes the deformation fluctuations with a fixed frequency that are periodically driven by seasonal environmental factors such as atmospheric rainfall and temperature changes. The magnitude of the fluctuations reflects the sensitivity of the landslide body to external disturbances.

[0043] The residual term, which is the sequence component remaining after stripping the trend and seasonal terms from the original data, contains measurement noise, model errors, and abnormal deformation signals that cannot be explained by trends and periodic patterns. It is a key indicator for identifying sudden or non-periodic instability of landslide bodies.

[0044] Taking the J03 measuring point as an example, the seasonal decomposition algorithm of S2 is used to standardize the displacements in the X, Y, and vertical directions of this point, and the standardized deformation sequence is decomposed into trend, seasonal, and residual terms, such as... Figure 4 , Figure 5 , Figure 6 As shown.

[0045] The residual anomaly identification in S3 is performed automatically using an isolated forest approach, and the steps include:

[0046] S3-1 uses the residual sequence of each monitoring point as the input feature;

[0047] S3-2 recursively and randomly splits the samples by constructing 100-200 isolated trees;

[0048] S3-3 calculates the average path length of each residual data point in all isolated trees by recursively and randomly partitioning the sample space.

[0049] S3-4 calculates the outlier score for each residual data point, and determines and outputs residual outlier points based on the preset outlier score threshold.

[0050] The formula for calculating the anomaly score is:

[0051] ;

[0052] in, The aforementioned abnormal scores, For the sample The number of samples in the dataset. The path length of a single residual sample in the isolated tree. According to The calculated average path length, This represents the average path length of each sample point in the sample dataset.

[0053] Based on the above calculation method, the abnormal scores in the X direction, Y direction and vertical direction (H) are calculated respectively. The residual abnormal points are determined and output according to the preset abnormal score threshold and listed in Table 2-1, Table 2-2 and Table 2-3.

[0054] Table 2-1 Data Table of Automatic Identification Results of X-Direction Residual Anomalies

[0055] Number (serial number) Residual Scores Anomaly 15 -0.3 -0.124 -1 16 0.19 -0.041 -1 17 -0.3 -0.124 -1 22 -0.18 -0.085 -1 23 0.22 -0.056 -1 26 -0.25 -0.073 -1

[0056] Table 2-2 Data Table of Automatic Identification Results of Residual Anomalies in the Y Direction

[0057] Number (serial number) Residual Scores Anomaly 12 -0.55 -0.152 -1 15 -0.57 -0.156 -1 16 -0.58 -0.157 -1 25 -0.71 -0.176 -1 27 -1.13 -0.214 -1

[0058] Table 2-3 Automatic Identification Results of H-Direction Residual Anomalies

[0059] Number (serial number) Residual Scores Anomaly 15 -0.26 -0.058 -1 17 -0.37 -0.097 -1 24 0.63 -0.177 -1

[0060] Based on the data in Tables 2-1, 2-2, and 2-3, a schematic diagram of the automatic residual identification results is generated, as shown below. Figure 7 , Figure 8 , Figure 9 .

[0061] The integrated early warning and attribution analysis in S4 includes the following specific steps:

[0062] S4-1 Risk Level Multidimensional Assessment: Based on the magnitude of the abnormal scores and spatial distribution of the residual abnormal points, as well as whether the trend of the point accelerates during the same period, the landslide risk level is divided into three levels: attention, warning, and alert.

[0063] S4-2 Abnormal Situation Attribution: Abnormal situations are divided into two types:

[0064] Externally triggered deformation judgment: When residual abnormal points are identified, the synchronous rainfall data is automatically retrieved to determine the amount of rainfall and whether there is continuous rainfall, monitor earthquake activity with a magnitude greater than 3 on the Richter scale, and comprehensively judge the construction situation near the landslide body. Then, the abnormality is attributed to deformation triggered by external environmental disturbance.

[0065] Spontaneous instability determination: When there is no obvious external disturbance as described above, if the rate of change of the trend term exceeds twice the standard deviation of its historical average, the anomaly is attributed to spontaneous instability caused by the deterioration of the internal structure of the landslide body.

[0066] S4-3 Tiered Early Warning Issuance and Feedback: Based on the determined risk level and attribution results, corresponding early warning information is issued; for anomalies judged as "spontaneous instability", their risk level is automatically raised by one level, and a comprehensive assessment report containing anomaly location, risk level, cause inference and handling suggestions is generated.

[0067] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The embodiments and features described in these embodiments can be arbitrarily combined without conflict. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

[0068] Example 2: The landslide deformation anomaly identification and early warning method based on seasonal decomposition residuals in Example 1 above can be used to form a computer system for early warning. The system includes the following modules:

[0069] Data acquisition and preprocessing module: used to acquire, clean, and standardize time-series deformed data;

[0070] Seasonal decomposition module: Used by the seasonal decomposition algorithm to decompose standardized time series data and output trend, seasonality and residual terms;

[0071] Residual Anomaly Identification Module: Used to automatically identify outliers in residual sequences using the isolated forest algorithm, and to identify and mark residual anomalies;

[0072] Integrated Early Warning and Analysis Module: When an anomaly is detected, it combines residual characteristics, trend term status, and rainfall, earthquake, and construction data to assess the risk level, determine the cause of the anomaly, and issue early warning information.

Claims

1. A method for identifying and warning of landslide deformation anomalies based on seasonal decomposition residuals, characterized in that... The method includes the following steps: S1 Data Acquisition and Preprocessing: Collect time-series deformation data from landslide monitoring points, and perform data cleaning and initial standardization to form time-regular deformation sequence data; S2 seasonal decomposition: The seasonal decomposition algorithm is used to decompose the standardized deformed sequence data to obtain the trend term reflecting long-term changes, the seasonal term reflecting periodic fluctuations, and the residual term remaining after removing the trend term and the seasonal term. S3 Residual Anomaly Identification: Analyze the characteristics of the residual sequence, identify anomalies in the residual sequence formed by the residual terms, and identify abnormal residual points that exceed the normal fluctuation range as criteria for landslide deformation anomalies. S4 Integrated Early Warning and Attribution Analysis: When an anomaly is detected, the risk level of the landslide is assessed and an early warning is issued by combining the long-term trend of the integrated trend item and the periodic pattern of the seasonal item. At the same time, geological and meteorological data from the same period are used to assist in the judgment of the cause of the anomaly.

2. The method for identifying and warning of landslide deformation anomalies based on seasonal decomposition residuals as described in claim 1, characterized in that... The data cleaning and initial standardization process uses the Z-Score standardization method to linearly interpolate and impute missing values. The calculation formula is as follows: ； in, This is the original data. The mean of the sequence. The standard deviation of the sequence. This is the standardized data.

3. The method for identifying and warning of landslide deformation anomalies based on seasonal decomposition residuals as described in claim 1, characterized in that... The seasonal decomposition algorithm decomposes the standardized deformed sequence into trend, seasonality, and residual terms, and the calculation formula is as follows: ； in, The data is the standardized deformed sequence data at time t. The value of the trend term at time t. The seasonality value at time t. Let t be the value of the residual term at time t.

4. The method for identifying and warning of landslide deformation anomalies based on seasonal decomposition residuals as described in claim 1, characterized in that... The identification of the defect anomalies is performed automatically using an isolated forest approach, and the steps include: S3-1 uses the residual sequence of each monitoring point as the input feature; S3-2 recursively and randomly splits the samples by constructing 100-200 isolated trees; S3-3 calculates the average path length of each residual data point in all isolated trees by recursively and randomly partitioning the sample space. S3-4 calculates the outlier score for each residual data point, and determines and outputs residual outlier points based on the preset outlier score threshold. The formula for calculating the anomaly score is: ； in, The aforementioned abnormal scores, For the sample The number of samples in the dataset. The path length of a single residual sample in the isolated tree. According to The calculated average path length, This represents the average path length of each sample point in the sample dataset.

5. The method for identifying and warning of landslide deformation anomalies based on seasonal decomposition residuals as described in claim 1, characterized in that... The comprehensive early warning and attribution analysis includes the following specific steps: S4-1 Risk Level Multidimensional Assessment: Based on the magnitude of the abnormal scores and spatial distribution of the residual abnormal points, as well as whether the trend of the point accelerates during the same period, the landslide risk level is divided into three levels: attention, warning, and alert. S4-2 Abnormal Situation Attribution: Abnormal situations are divided into two types: Externally triggered deformation judgment: When residual abnormal points are identified, the synchronous rainfall data is automatically retrieved to determine the amount of rainfall and whether there is continuous rainfall, monitor earthquake activity with a magnitude greater than 3 on the Richter scale, and comprehensively judge the construction situation near the landslide body. Then, the abnormality is attributed to deformation triggered by external environmental disturbance. Spontaneous instability determination: When there is no obvious external disturbance as described above, if the rate of change of the trend term exceeds twice the standard deviation of its historical average, the anomaly is attributed to spontaneous instability caused by the deterioration of the internal structure of the landslide body. S4-3 Tiered Early Warning Issuance and Feedback: Based on the determined risk level and attribution results, corresponding early warning information is issued; for anomalies judged as "spontaneous instability", their risk level is automatically raised by one level, and a comprehensive assessment report containing the anomaly location, risk level, cause inference and handling suggestions is generated.

6. The system formed by the landslide deformation anomaly identification and early warning method based on seasonal decomposition residuals as described in claim 1, characterized in that... The system includes the following modules: Data acquisition and preprocessing module: used to acquire, clean, and standardize time-series deformed data; Seasonal decomposition module: Used by the seasonal decomposition algorithm to decompose standardized time series data and output trend, seasonality and residual terms; Residual Anomaly Identification Module: Used to automatically identify outliers in residual sequences using the isolated forest algorithm, and to identify and mark residual anomalies; Integrated Early Warning and Analysis Module: When an anomaly is detected, it combines residual characteristics, trend term status, and rainfall, earthquake, and construction data to assess the risk level, determine the cause of the anomaly, and issue early warning information.

Images