Landslide instability time prediction method and system based on historical landslide standard curve
By constructing a dimensionless standard curve library and combining the Voight model and dynamic time warping algorithm, the problems of accuracy and universality in landslide instability prediction were solved, realizing real-time and accurate prediction of landslide instability time, which is applicable to landslide monitoring under different geological conditions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHONGQING INST OF GEOLOGY & MINERAL RESOURCES
- Filing Date
- 2026-06-15
- Publication Date
- 2026-07-14
Smart Images
Figure CN122389671A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of geological early warning technology, specifically to a method and system for predicting landslide instability time based on historical landslide standard curves. Background Technology
[0002] Currently, global creep-type landslide instability prediction technologies are all based on the Voight creep model (a classic Voight model in the field of geology / landslide prediction). However, existing technologies have the following shortcomings: First, they are purely theoretical analytical models without constraints from real landslide instability monitoring data throughout the entire cycle, resulting in significant deviations from the actual nonlinear deformation process of soil and rock masses and insufficient prediction accuracy in field applications. Second, they cannot eliminate the geological heterogeneity of landslides caused by lithology, scale, and geological conditions, and the models can only be customized for a single landslide, lacking cross-regional and cross-type universality. Third, dimensionless processing relies on future information such as total acceleration time and total displacement increment, which can only be used for post-event retrospective analysis and cannot achieve real-time early warning at engineering sites. Fourth, the operation process is complex, requiring high professional modeling capabilities, making it difficult for grassroots geological disaster monitoring units to implement. Summary of the Invention
[0003] This invention aims to provide a method and system for predicting landslide instability time based on historical landslide standard curves. By reconstructing dimensionless parameter formulas using real unstable landslides, a dimensionless standard curve library is constructed to eliminate landslide geological heterogeneity and achieve universal, accurate, and real-time prediction of creep-type landslide instability time. This addresses the technical problems of poor universality, insufficient accuracy, and difficulty in implementation of existing landslide instability prediction technologies.
[0004] The basic solution provided by this invention is: a method for predicting landslide instability time based on historical landslide standard curves, comprising: S1. Based on the Voight model, the dimensionless parameter formula is derived to characterize the mapping relationship between the dimensionless instability evolution parameters and the creep velocity, creep acceleration and landslide sample-specific nonlinear parameters, and to generate a single-sample dimensionless evolution curve that characterizes the mapping relationship between the dimensionless instability evolution parameters and the remaining time to instability. S2 uses the curve shape similarity calculated by the dynamic time warping algorithm as the core indicator, and uses the single sample-specific nonlinear parameter value to help determine the number of clusters. It clusters the dimensionless evolution curves of all historical unstable landslide samples and performs interval comparison calibration for each class. S3. The dimensionless evolution curves of similar historical unstable landslide samples are fused point by point within a unified time interval before instability to generate standard curves of this type. Power-law function fitting is performed on these standard curves to obtain normalized constants that are related to the nonlinear parameters specific to the landslide samples and are independent of the landslide geological conditions. This is used to construct a dimensionless standard curve library. S4. Real-time acquisition of the dimensionless evolution curve of the target landslide; and the use of dynamic time warping algorithm to perform real-time similarity matching between the dimensionless evolution curve of the target landslide and the standard curve, and to dynamically determine the optimal matching standard curve in real time. S5, based on the optimal matching standard curve determined in real time, calculates the remaining time to instability and the predicted instability time for each current monitoring moment of the target landslide according to a preset strategy, for dynamic early warning.
[0005] This invention also provides a landslide instability time prediction system based on historical landslide standard curves, which executes any of the landslide instability time prediction methods based on historical landslide standard curves described above; the system includes: The historical sample library module is used for screening and preprocessing historical unstable landslide samples to build a historical sample library; The standard curve library construction module is used to generate single-sample dimensionless evolution curves based on the dimensionless parameter formulas and dynamic time warping algorithm derived from the Voight model, and cluster them to form several standard curves, thereby constructing a dimensionless standard curve library. The real-time data processing module is used to acquire target landslide data in real time and call the historical sample library module for preprocessing, and call the standard curve library construction module to obtain the dimensionless evolution curve of the target landslide. The curve matching and prediction module is used to perform real-time similarity matching between the dimensionless evolution curve and the standard curve of the target landslide using a dynamic time warping algorithm, and to dynamically determine the optimal matching standard curve in real time. The early warning release module is used to calculate the remaining time to instability of the target landslide and the predicted instability time for each current monitoring moment based on the optimal matching standard curve determined in real time and according to a preset strategy, so as to provide dynamic early warning. The self-learning optimization module is used to incorporate the complete data of the target landslide into the historical sample library after the landslide becomes unstable, and to perform self-learning closed-loop optimization of the dimensionless standard curve library.
[0006] The working principle and advantages of this invention are as follows: This invention uses real full-cycle monitoring data of unstable landslides as its core foundation. Through a dimensionless transformation formula rigorously derived based on the Voight model and relying on the Dynamic Time Warping (DTW) algorithm, it constructs a standardized dimensionless instability evolution curve library with real geological constraints and covering different instability modes. By dynamically matching the similarity between the target landslide and the standard curve, it can achieve rapid and accurate prediction and early warning of the instability time of the target landslide without relying on prior geological attributes.
[0007] This invention eliminates the challenge of geological heterogeneity caused by different landslide lithology and scales, and changes the traditional prediction logic centered on idealized theoretical formulas. It relies on the Voight creep model to ensure theoretical rigor, and uses real geological data constraints to solve the problem of the disconnect between pure theoretical models and actual engineering. This enables cross-type and cross-regional accurate prediction of the instability time of creep-type landslides, providing real-time rolling early warnings without relying on future information, and significantly improving prediction accuracy compared to traditional methods. Furthermore, by eliminating geological heterogeneity through dimensionless mapping and extracting general evolution patterns of similar landslides based on curve morphology clustering, it eliminates the need for the ultra-large-scale datasets required by traditional machine learning, thus ensuring prediction accuracy with only a small sample dataset of existing landslides.
[0008] This invention employs Dynamic Time Warping (DTW) in both the historical sample aggregation to form a standard curve and the target landslide matching stages. Compared to existing technologies, DTW directly performs shape matching on the original sequence. DTW allows for non-linear scaling of the time axis, enabling accurate matching even when two landslides have similar overall morphology but different evolution rates (e.g., one accelerates faster than the other). This resistance to time axis scaling interference is particularly crucial for creeping landslides. By using shape matching to substitute function fitting, it avoids the defects of information loss and fitting errors caused by forced fitting, fully preserving key dynamic information such as acceleration patterns, fluctuation cycles, and local mutations, thus avoiding information loss and significantly improving prediction accuracy. Both stages use the same DTW distance metric for sample clustering and real-time matching, ensuring mathematical consistency in calculations across different processes and avoiding scale bias and inconsistent criteria caused by different processes and standards in traditional methods.
[0009] This invention improves the adaptability of new landslide scenarios in several ways: First, when performing clustering, it considers both the curve shape similarity calculated by DTW and the unique nonlinear parameters of a single sample. The number of clusters can be flexibly set according to the range of nonlinear parameters and the fineness of application, allowing for both coarse and fine division to adapt to different engineering scenarios. Second, the early warning mechanism abandons the traditional fixed threshold and pre-segmentation method, only extracting effective data from the accelerated creep stage. By calculating the dimensionless evolution curve and matching it with the standard curve through DTW, it directly calculates the time to instability for early warning, completely avoiding the problems of poor system adaptability and inaccurate warning caused by historical prior thresholds and hard boundary warnings of manual segmentation. It can be quickly deployed in new landslide scenarios with no or scarce samples, and is accurately applicable to a new landslide early warning with vastly different geological conditions.
[0010] This invention combines rigorous theoretical foundation with strong engineering practicality, effectively solving the core pain points of existing technologies. It significantly improves prediction accuracy compared to traditional methods, has a standardized and low-threshold operation process, and can be quickly implemented by grassroots monitoring units. At the same time, it establishes a complete closed loop for accuracy verification and self-learning optimization, which can be widely adapted to various creep-type landslide monitoring and early warning scenarios, providing stable and reliable technical support for geological disaster prevention and control. Attached Figure Description
[0011] Figure 1 This is a flowchart illustrating the landslide instability time prediction method based on historical landslide standard curves provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the landslide instability time prediction system based on historical landslide standard curves provided in an embodiment of the present invention. Figure 3 This is an example diagram of the standard curve library provided in an embodiment of the present invention. Detailed Implementation
[0012] The following detailed explanation illustrates the specific implementation methods: The basic implementation examples are as follows: Figure 1 As shown: A method for predicting landslide instability time based on historical landslide standard curves, including: S1. Based on the Voight model, the dimensionless parameter formula is derived to characterize the mapping relationship between the dimensionless instability evolution parameters and the creep velocity, creep acceleration and landslide sample-specific nonlinear parameters, and to generate a single-sample dimensionless evolution curve that characterizes the mapping relationship between the dimensionless instability evolution parameters and the remaining time to instability. S2 uses the curve shape similarity calculated by the dynamic time warping algorithm as the core indicator, and uses the single sample-specific nonlinear parameter value to help determine the number of clusters. It clusters the dimensionless evolution curves of all historical unstable landslide samples and performs interval comparison calibration for each class. S3. The dimensionless evolution curves of similar historical unstable landslide samples are fused point by point within a unified time interval before instability to generate standard curves of this type. Power-law function fitting is performed on these standard curves to obtain normalized constants that are related to the nonlinear parameters specific to the landslide samples and are independent of the landslide geological conditions. This is used to construct a dimensionless standard curve library. S4. Real-time acquisition of the dimensionless evolution curve of the target landslide; and the use of dynamic time warping algorithm to perform real-time similarity matching between the dimensionless evolution curve of the target landslide and the standard curve, and to dynamically determine the optimal matching standard curve in real time. S5, based on the optimal matching standard curve determined in real time, calculates the remaining time to instability and the predicted instability time for each current monitoring moment of the target landslide according to a preset strategy, for dynamic early warning.
[0013] Specifically, this method includes two stages: constructing a dimensionless standard curve library constrained by historical instability landslides and predicting and warning of the real-time instability time of target landslides.
[0014] Phase 1: Constructing a dimensionless standard curve library with universal predictive capabilities based on historical unstable landslide samples, unaffected by geological conditions, and performing operations based on rigorous theoretical derivations of the Voight creep model. This includes: S0, Screening, Collection, and Preprocessing of Historical Instability and Landslide Samples: S01. Screen and collect historical unstable landslide samples (historical samples) that meet the admission criteria, and extract the complete monitoring dataset from the acceleration creep start point to the instability time of each sample.
[0015] In this embodiment, the admission criteria include: (1) a progressive creep landslide that has fully experienced the entire accelerated creep process to the final instability and failure, in order to exclude sudden and strongly artificially disturbed landslides; (2) having continuous temporal displacement monitoring data covering the entire accelerated creep stage, with a monitoring frequency of not less than once per day and a data loss rate of ≤5% during the accelerated stage; (3) having clear records of geological attributes such as lithology, landslide scale, and triggering conditions, as well as accurate records of the actual instability time and the total duration of the accelerated creep stage; (4) having ≥20 typical creep landslide samples that meet (1)-(3), covering different lithologies, triggering conditions and scales, covering at least three major lithologies: soft rock, hard rock, and sedimentary bodies, and covering landslide scales of small, medium, large and extra-large.
[0016] For each sample, a complete dataset from the start of accelerated creep to the moment of instability is extracted using standardization, including monitoring time t, corresponding displacement u, and geological attribute information, forming a historical sample database in a unified format.
[0017] S02, perform standardized preprocessing on each historical sample to screen effective data for the accelerated creep stage; the preprocessing includes removing outliers, smoothing and denoising, and calculating creep rate and acceleration.
[0018] In this embodiment, the 3σ criterion is used to remove outliers, the mean and standard deviation of the sample displacement sequence are calculated, outlier values exceeding "mean ± 3 times standard deviation" are removed, and linear interpolation of adjacent data is used to fill in missing values.
[0019] The displacement sequence is smoothed using a 5-point moving average method to eliminate the interference of monitoring noise on subsequent rate and acceleration calculations.
[0020] The creep rate was calculated time-by-time using the first-order finite difference method. With creep acceleration .
[0021] By using the criterion of "continuously increasing rate and continuously positive acceleration", the start and end points of the accelerated creep stage are reconfirmed, invalid data from the initial creep and constant-rate creep stages are removed, and only valid data from the accelerated creep stage are retained.
[0022] S1, Perform dimensionless transformation of samples: dimensionless parameter formula derived from the Voight model. Fitting single-sample specific nonlinear parameters Calculate dimensionless instability evolution parameters Convert the standardized time axis to the time remaining until instability. Generate a single sample - Dimensionless evolution curves. Includes: S11, according to the basic expression of the Voight creep instability criterion, we have: (1) In the formula: The landslide creep rate; This refers to landslide creep acceleration; It is a proportionality constant and is related to geological conditions such as landslide scale, lithology, and structure; It is a nonlinear creep parameter of soil and rock mass, which is only related to the creep characteristics of the soil and rock mass itself. Based on the current global historical samples of unstable creep-type landslides, The value remained stable within the range of 1.5 to 2.2.
[0023] By performing an identity transformation on the formula (1), we have: (2) For a given landslide, during the accelerated creep stage, the proportionality constant related to geological conditions... It is a constant value.
[0024] Define dimensionless instability evolution parameters ,make: (3) The differences between different landslides are extremely large The values are converted into standardized dimensionless parameters to eliminate the heterogeneity of geological conditions for different landslides.
[0025] According to the theory of creep instability in soil and rock masses, after a landslide enters the accelerated creep stage, the rate evolution strictly follows a power law relationship: (4) (5) In the formula, The time since instability This marks the final moment of instability in the landslide. For the current monitoring time, This is a constant related to landslide characteristics.
[0026] Taking the time derivative of formula (4), we have: (6) Substituting formulas (4) and (6) into formula (3), the power-law function is: (7) In the formula, , for only with The relevant normalization constants are independent of the landslide geological conditions.
[0027] Therefore, the defined dimensionless instability evolution parameter Π eliminates the proportionality constant A, which is strongly correlated with geological conditions in the Voight model, thus eliminating the heterogeneity of geological conditions among different landslides. Therefore, similar creep characteristics ( For landslides with similar values, their dimensionless parameters Remaining time until instability Their evolutionary patterns are completely consistent.
[0028] S12, based on the above theoretical derivation, performs a dimensionless transformation on each historical landslide sample: S121, Sample-specific Value fitting: For effective data in the accelerated creep stage of the sample, the Voight model is transformed into a log-linear model. ,by For independent variable, Perform linear least squares fitting on the dependent variable, and then use the fitted slope... Inverse calculation of sample-specific nonlinear parameters .
[0029] S122, dimensionless parameter Calculation: Calculate the dimensionless instability evolution parameters of the sample at each time step. .
[0030] S123, Standardized Time Axis Transformation: Converts the absolute monitoring time of the sample into the remaining time until instability. ,in This represents the actual moment of instability of the sample (historical samples are known).
[0031] S124, Generation of dimensionless curves for single samples: For the horizontal axis, Using the vertical axis as the ordinate, generate the dimensionless evolution curve for the complete acceleration phase of this sample. .
[0032] S2 uses the curve shape similarity calculated by the Dynamic Time Warping (DTW) algorithm as the core indicator, and uses the single-sample exclusive nonlinear parameter value to help determine the number of clusters. It clusters the dimensionless evolution curves of all historical unstable landslide samples and performs interval comparison calibration for each class. In this embodiment, the single-sample specific nonlinear parameter value The values range from 1.5 to 2.2, with a step size of 0.1. The dimensionless curves of all historical samples are clustered into 7 classes according to the following method.
[0033] set up The range of values is [ and step size Calculate the number of clusters And determine indivual Value interval; where: (8) For all dimensionless evolution curves, calculate the DTW distance between each pair of dimensionless instability evolution parameters and the time series remaining after instability, and construct a distance matrix.
[0034] K-means clustering is performed using DTW distance as the sole similarity measure: in the assignment phase, each sample is assigned to the cluster center with the nearest DTW distance; in the update phase, the cluster centers are updated using DTW centroid average (DBA). The entire clustering process is driven entirely by the curve shape.
[0035] Therefore, we obtain A set of dimensionless curves, and the dimensionless curves of each set... Value range and each category The value intervals are compared and verified, but the clustering is not adjusted in reverse.
[0036] Each category is labeled with corresponding geological attributes and instability patterns. That is, the common geological characteristics of all samples in each cluster are statistically analyzed, and each category is given a clear geological meaning, including typical lithology, average acceleration time, etc. Lithological information is only recorded and is not used in the prediction project.
[0037] S3, using dimensionless curves of similar historical samples in a unified manner. Point-by-point fusion calculations within the interval generate a standard curve entirely determined by real monitoring data. A power-law function is then applied to the standard curve to obtain a normalization constant for simplified calculations. This integrates and forms a dimensionless standard curve library. Specifically, it includes: S31, take the dimensionless evolution curves of all historical samples of this type, and at a uniform time remaining until instability. Within the interval 0d to 30d, with a step size of 0.01d, for the same... All historical samples of the points The values are calculated as a point-by-point arithmetic average to directly generate a representative standard curve for this type of landslide. The shape of this curve is entirely determined by the actual monitoring data of this type of landslide, fully preserving the true evolution characteristics of similar landslides before instability. There are no pre-existing constraints or artificial corrections from theoretical formulas, thus solving the problem of the disconnect between pure theoretical models and actual engineering.
[0038] S32, For the seven true standard curves generated above, apply the power law formula derived above. Perform least-squares fitting to obtain the fitting normalization constants corresponding to this type of curve. In this embodiment, the seven standard curves corresponding to 26 typical historical unstable landslide samples are fitted together. The values are, in order: 0.1868, 0.2064, 0.2243, 0.2406, 0.2555, 0.2692, and 0.2818. Those skilled in the art should understand that the above... The value is the optimal value obtained by fitting typical samples, and can be updated adaptively according to the expansion of the historical sample library (such as adding the target landslide to the historical sample library after instability) and the adjustment of sample type.
[0039] S33 integrates all standard curves to form the final dimensionless standard curve library.
[0040] The shape of the resulting standard curve is entirely determined by real monitoring data from similar historical samples; the power-law function fitting is only used to obtain the normalization constant for simplified calculations. It does not change the true shape of the standard curve.
[0041] The second stage involves real-time prediction and early warning of the target landslide's instability time, including: S4, Determine the optimal matching standard curve based on DTW: S41: Collect real-time displacement monitoring data of the target landslide, perform standardized preprocessing according to the S0 process, identify the accelerated creep stage and initiate prediction; S42, perform dimensionless transformation on the target landslide according to S12, based on the formula... Real-time computation Current value, construct the real-time dimensionless evolution curve of the target landslide; S43 uses the DTW algorithm to perform similarity matching between the curve segments of the target landslide in the most recent five consecutive monitoring periods and the standard curve library constructed in S3, and determines the optimal matching standard curve.
[0042] S44, As new monitoring data is added, the sampling algorithm is dynamically optimized using a preset sampling algorithm (in this embodiment, Bayesian MCMC sampling). The value is used to recalculate the real-time dimensionless evolution curve of the target landslide and rematch it with the standard curve. S441, for each new set of real-time monitoring data, the new monitoring data is included in the effective dataset of the target landslide accelerated creep stage, and S121 is repeated to obtain the updated initial data. value.
[0043] S442, obtaining the initial from S441 The values are assumed to be a prior distribution. A likelihood function is constructed using newly added monitoring data. The Metropolis-Hastings algorithm is used for MCMC sampling, iterating 5000 times until convergence, to obtain the optimal value that best suits the current deterioration characteristics of the landslide soil and rock mass. value.
[0044] S443, the optimal result obtained based on S442 The value is used to recalculate the real-time dimensionless evolution curve of the target landslide, and then perform a similarity match with the standard curve library to determine the optimal matching standard curve, which is used to update the calculation of the remaining time to instability and the predicted instability time in real time.
[0045] S5, Real-time Instability Time Prediction and Early Warning of Target Landslides: S51, calculate the remaining time to instability and the predicted instability time for each current monitoring moment of the target landslide according to the preset strategy: The preset strategy includes calculating the remaining time to instability using at least one of the following methods: Option 1: Precise Prediction Scheme - Extract dimensionless instability evolution parameters of the target landslide at the latest monitoring time. To match the standard curve Corresponding time remaining until instability The value is taken as the current moment of the target landslide. Time remaining until instability; Option 2: Rapid estimation method: The normalization constant corresponding to the matched standard curve is used. Value, substitute into the formula Using T as the estimated target landslide current time Time remaining until instability; The preset strategy also includes the final prediction of the instability time. .
[0046] S52 dynamically updates forecast results based on newly added monitoring data and determines the tiered early warning levels based on the remaining time. The tiered early warning levels are divided into four levels: blue, yellow, orange, and red, each corresponding to different emergency response actions.
[0047] Based on the remaining time to instability updated in real time as monitoring data is updated, four warning levels are defined and corresponding warning signals are issued: ① Blue Warning: ① Continue routine monitoring; ② Yellow alert: ③ Encrypted monitoring frequency; Orange alert: ④ Red Alert: Activate emergency preparedness; Personnel evacuation and emergency response were initiated.
[0048] It also includes S6, which, after the target landslide becomes unstable, uses its complete dataset as a sample of historical unstable landslides to perform self-learning closed-loop optimization of the standard curve library.
[0049] In S6, if the target landslide eventually becomes unstable, based on the admission conditions in S0, the instability of the target landslide is determined to be a historical unstable landslide sample. A complete dataset of the target landslide from the start of accelerated creep to the moment of instability is extracted and standardized, forming a unified format as a new historical sample, which is then added to the historical sample database. Then, standardized preprocessing is performed on the new historical sample to filter out valid data from the accelerated creep stage. Based on S1, a single sample of the new historical sample is generated. - Dimensionless evolution curve. Based on S2, single samples of new historical samples... - The dimensionless evolution curves are re-clustered with the dimensionless evolution curves of all previous historical unstable landslide samples. Based on S3, a new standard curve is generated. Subsequent S4 and S5 use the new standard curve to predict the new target landslide. After S6 confirms the instability of the new target landslide, this process is repeated to achieve self-learning optimization of the standard curve library.
[0050] like Figure 2 This embodiment also provides a landslide instability time prediction system based on historical landslide standard curves to execute any of the landslide instability time prediction methods based on historical landslide standard curves described above; the system includes: The historical sample library module is used for screening and preprocessing historical unstable landslide samples to build a historical sample library; The standard curve library construction module is used to derive dimensionless parameter formulas and dynamic time warping algorithms based on the Voight model, generate single-sample dimensionless evolution curves, cluster them to form several standard curves, and thus construct a dimensionless standard curve library. The real-time data processing module is used to acquire target landslide data in real time and call the historical sample library module for preprocessing, and call the standard curve library construction module to obtain the dimensionless evolution curve of the target landslide. The curve matching and prediction module is used to perform real-time similarity matching between the dimensionless evolution curve and the standard curve of the target landslide using a dynamic time warping algorithm, and to dynamically determine the optimal matching standard curve in real time. The early warning release module is used to calculate the remaining time to instability of the target landslide and the predicted instability time for each current monitoring moment based on the optimal matching standard curve determined in real time and according to a preset strategy, so as to provide dynamic early warning. The self-learning optimization module is used to incorporate the complete data of the target landslide into the historical sample library after the landslide becomes unstable, and to perform self-learning closed-loop optimization of the dimensionless standard curve library.
[0051] It is understandable that the above system can fully execute the above method, with the same specific process and effect, to achieve real-time prediction and graded early warning of creep-type landslide instability time, which will not be elaborated here.
[0052] The following is a schematic diagram of the construction process of a dimensionless standard curve library constrained by historical instability and landslides: 1. Historical Sample Collection and Screening: 26 typical creep landslide samples from around the world were collected. All samples met the inclusion criteria, covering three major types of lithology: soft rock, hard rock, and sedimentary bodies, and two triggering conditions: reservoir water and rainfall, with a scale ranging from 100,000 cubic meters to 12 million cubic meters.
[0053] 2. Standardized preprocessing: All 26 samples were preprocessed uniformly: outliers were removed using the 3σ criterion, displacement sequences were smoothed using the 5-point moving average, rate and acceleration were calculated using the first-order difference method, and valid data in the accelerated creep stage were screened. The data processing rules for all samples were completely consistent.
[0054] 3. Dimensionless transformation: Fitting a specific model for each sample. Value, calculate dimensionless parameter The time axis is converted to the time remaining until instability. 26 results generated - The single-sample dimensionless evolution curve is used to unify all samples into the same dimensionless space.
[0055] 4. Clustering and Geological Calibration: Based on The values ranged from 1.5 to 2.2, with a step size of 0.1, and were clustered into 7 classes (as shown in Table 1). Clustering was completed using DTW curve similarity as the sole morphological similarity metric. The clustering results were compared with... Interval comparison verification.
[0056] 5. Standard curve generation and parameter fitting: For the 7 clustering results, in Averaging over the interval point by point generates 7 true standard curves (e.g. Figure 3 As shown, from bottom to top, these are standard curves 1-7, corresponding to each category. (Value interval); the optimal value for each curve is obtained by fitting using a power-law formula. The values are shown in Table 1: Table 1: Optimal Selection of Dimensionless Standard Curve Library for Historical Instability and Landslide Constraints Value table
[0057] The following is a schematic diagram of the instability prediction process for a landslide in a mountainous area: The landslide in this mountainous area is a typical layered soft rock landslide, with interbedded Jurassic mudstone and sandstone. The landslide volume is approximately 3 million cubic meters. It entered the accelerated creep stage in February 2024 and experienced overall instability on June 12, 2024. This method was used for real-time prediction, and the specific steps are as follows: 1. Data Acquisition and Preprocessing: Daily-scale displacement monitoring data from the landslide GNSS monitoring station were collected, and the preprocessing process was performed in complete agreement with the database construction phase. On February 15, 2024, it was identified that the landslide rate was continuously increasing and the acceleration was continuously positive, confirming that it had entered the accelerated creep phase, and the prediction process was initiated.
[0058] 2. Dimensionless transformation: Using monitoring data from February 15, 2024 to June 5, 2024, an initial value was obtained through fitting. =1.68, calculated time-by-time The value was used to construct the real-time dimensionless evolution curve of the landslide.
[0059] 3. Curve Matching and Instability Time Prediction: Curve segments from the most recent 5 days (June 5, 2024) were selected. The DTW algorithm was used to calculate similarity with 7 standard curves, and the optimal match was found to be standard curve number 2. , Current moment The value is 0.0382, and the remaining time can be directly obtained by matching the standard curve. The predicted instability period is June 10th to June 11th. As monitoring data is continuously updated, a rematch will be performed on June 10th, 2024, with the optimal match becoming standard curve #3. , ),current The value is 0.201, which indicates the remaining time. The predicted period of instability is June 11-12.
[0060] 4. Warning Levels and Emergency Response: On June 5, 2024, with a predicted remaining time of ≤7 days, an orange warning was issued and emergency preparedness was initiated; on June 10, 2024, with a predicted remaining time of ≤2 days, a red warning was issued and personnel evacuation was initiated; the landslide became unstable on June 12, 2024, without causing any casualties.
[0061] 5. Self-learning optimization: After a landslide becomes unstable, its complete monitoring data, geological attributes, and actual instability time are included in the historical sample library, and the standard curve library is re-optimized to complete the self-learning closed loop.
[0062] The landslide instability time prediction method and system based on historical landslide standard curves provided in this embodiment constructs a dimensionless standard curve library using real unstable landslides, eliminating the geological heterogeneity of landslides and realizing universal, accurate, and real-time prediction of creep-type landslide instability time.
[0063] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics of the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent.
Claims
1. A method for predicting landslide instability time based on historical landslide standard curves, characterized in that, include: S1. Based on the Voight model, the dimensionless parameter formula is derived to characterize the mapping relationship between the dimensionless instability evolution parameters and the creep velocity, creep acceleration and landslide sample-specific nonlinear parameters, and to generate a single-sample dimensionless evolution curve that characterizes the mapping relationship between the dimensionless instability evolution parameters and the remaining time to instability. S2 uses the curve shape similarity calculated by the dynamic time warping algorithm as the core indicator, and uses the single sample-specific nonlinear parameter value to help determine the number of clusters. It clusters the dimensionless evolution curves of all historical unstable landslide samples and performs interval comparison calibration for each class. S3. The dimensionless evolution curves of similar historical unstable landslide samples are fused point by point within a unified time interval before instability to generate standard curves of this type. Power-law function fitting is performed on these standard curves to obtain normalized constants that are related to the nonlinear parameters specific to the landslide samples and are independent of the landslide geological conditions. This is used to construct a dimensionless standard curve library. S4. Real-time acquisition of the dimensionless evolution curve of the target landslide; and the use of dynamic time warping algorithm to perform real-time similarity matching between the dimensionless evolution curve of the target landslide and the standard curve, and to dynamically determine the optimal matching standard curve in real time. S5, based on the optimal matching standard curve determined in real time, calculates the remaining time to instability and the predicted instability time for each current monitoring moment of the target landslide according to a preset strategy, for dynamic early warning.
2. The method for predicting landslide instability time based on historical landslide standard curves according to claim 1, characterized in that, This also includes S0, the screening, collection, and preprocessing of historical unstable landslide samples: S01, Screen and collect historical unstable landslide samples that meet the admission criteria, and extract the complete monitoring dataset from the acceleration creep initiation point to the instability time of each sample; S02. Perform standardized preprocessing on each historical sample. The preprocessing includes removing outliers, smoothing and reducing noise, and calculating creep rate and acceleration. Valid data in the accelerated creep stage are selected based on the judgment rule that the creep rate continues to increase and the acceleration continues to be positive.
3. The method for predicting landslide instability time based on historical landslide standard curves according to claim 1, characterized in that, The formula for the dimensionless parameter in S1 is: ,in These are dimensionless instability evolution parameters. This represents the landslide creep rate. For landslide creep acceleration, Nonlinear parameters specific to landslide samples.
4. The method for predicting landslide instability time based on historical landslide standard curves according to claim 3, characterized in that, Based on effective data from the accelerated creep stage of landslide samples, single-sample specific nonlinear parameters are fitted. ,use Calculate dimensionless instability evolution parameters The absolute monitoring time t of the sample is converted into the remaining time until instability. ,in Generate a single sample at the actual instability moment of the sample. - Dimensionless evolution curve .
5. The method for predicting landslide instability time based on historical landslide standard curves according to claim 1, characterized in that, In S2, let the landslide sample-specific nonlinear parameters be defined. The range of values is [ and step size Calculate the number of clusters And determine There are α-value sub-intervals; where: For all dimensionless evolution curves, calculate the DTW distance between each pair of dimensionless instability evolution parameters and the time series remaining until instability, and construct a distance matrix. Using DTW distance as the sole similarity metric, K-means clustering was performed to obtain... A set of dimensionless curves; The range of α values for each dimensionless curve set is compared and verified with each sub-interval of α values in the classification.
6. The method for predicting landslide instability time based on historical landslide standard curves according to claim 1, characterized in that, In S3, the power-law functions are: in, To characterize dimensionless instability evolution parameters for single samples and the remaining time from instability The dimensionless evolution curve of the mapping relationship, The normalization constant is Here, C is a nonlinear parameter specific to the landslide sample, and C is a constant related to the landslide characteristics.
7. The method for predicting landslide instability time based on historical landslide standard curves according to claim 1, characterized in that, In S4, as new monitoring data is added for the target landslide, the value of the nonlinear parameter α specific to the landslide sample is dynamically optimized using a preset sampling algorithm, the real-time dimensionless evolution curve of the target landslide is recalculated, and the standard curve is rematched.
8. The method for predicting landslide instability time based on historical landslide standard curves according to claim 1, characterized in that, In S5, the preset strategy includes calculating the remaining time before instability using at least one of the following methods: Option 1: Extract the dimensionless instability evolution parameters of the target landslide at the latest monitoring time. To match the standard curve Corresponding time remaining until instability The value is taken as the current moment of the target landslide. Time remaining until instability; Option 2: Use the normalization constant corresponding to the matched standard curve. Value, substitute into the formula Using T as the estimated target landslide current time Time remaining until instability; The preset strategy also includes the final prediction of the instability time. .
9. The method for predicting landslide instability time based on historical landslide standard curves according to claim 1, characterized in that, It also includes S6, which uses the complete data of the target landslide as a sample of historical unstable landslides to perform self-learning closed-loop optimization of the standard curve library after the target landslide becomes unstable.
10. A landslide instability time prediction system based on historical landslide standard curves, characterized in that, The system comprises the method for predicting landslide instability time based on historical landslide standard curves as described in any one of claims 1-9; the system includes: The historical sample library module is used for screening and preprocessing historical unstable landslide samples to build a historical sample library; The standard curve library construction module is used to generate single-sample dimensionless evolution curves based on the dimensionless parameter formulas and dynamic time warping algorithm derived from the Voight model, and cluster them to form several standard curves, thereby constructing a dimensionless standard curve library. The real-time data processing module is used to acquire target landslide data in real time and call the historical sample library module for preprocessing, and call the standard curve library construction module to obtain the dimensionless evolution curve of the target landslide. The curve matching and prediction module is used to perform real-time similarity matching between the dimensionless evolution curve and the standard curve of the target landslide using a dynamic time warping algorithm, and to dynamically determine the optimal matching standard curve in real time. The early warning release module is used to calculate the remaining time to instability of the target landslide and the predicted instability time for each current monitoring moment based on the optimal matching standard curve determined in real time and according to a preset strategy, so as to provide dynamic early warning. The self-learning optimization module is used to incorporate the complete data of the target landslide into the historical sample library after the landslide becomes unstable, and to perform self-learning closed-loop optimization of the dimensionless standard curve library.