Method and device for constructing event-level rainfall sample of rainfall-type landslide

Abstract

The application discloses a kind of event-level rainfall sample construction methods and devices of rainfall type landslide, it is related to geological disaster risk analysis field, wherein the method includes: obtaining the daily rainfall data in the range of study area and corresponding landslide historical record data;With rainfall site as foundation, thieven polygon is constructed, and the spatial matching of landslide historical record data is carried out;Based on matching result and rainfall event division comprehensive evaluation function, optimal rainfall event division parameter is searched using particle swarm optimization algorithm, and the daily rainfall data is divided into rainfall event, and continuous rainfall process is identified as rainfall event, and a plurality of sub-events with different cumulative characteristics are obtained by decomposing rainfall event;Disaster-causing label is carried out to independent rainfall event and its decomposed sub-event, and event-level rainfall sample set containing disaster-causing and non-disaster-causing event is constructed.The application can improve the reliability of rainfall type landslide risk assessment and early warning by reasonably constructing event-level rainfall event sample.

Description

Technical Field

[0001] This invention relates to the field of geological disaster risk analysis technology, and in particular to a method and apparatus for constructing event-level rainfall samples for rainfall-induced landslides. Background Technology

[0002] This section is intended to provide background or context for the invention. The description herein is not intended to imply that it is prior art simply because it is included in this section.

[0003] The occurrence of rainfall-induced landslides is typically influenced by the combined effects of rainfall intensity, duration, and prior accumulated rainfall conditions. Accurately characterizing rainfall processes and quantifying the risk of rainfall-triggered landslides are key technical challenges in landslide monitoring, early warning, and risk assessment.

[0004] Existing studies on rainfall-induced landslides typically construct rainfall threshold models or empirical discrimination models based on daily rainfall data to determine the likelihood of a landslide. These methods use daily or cumulative rainfall as the basic analytical unit and can reflect the impact of rainfall on landslides to some extent. However, daily rainfall sequences themselves are difficult to directly characterize the overall features of continuous rainfall processes, especially under conditions of multi-day continuous rainfall or intermittent rainfall, where daily analysis methods struggle to accurately depict the complete structure of the rainfall process.

[0005] In recent years, some studies have identified continuous rainfall events as physically meaningful samples by dividing daily rainfall sequences into rainfall events to support landslide triggering analysis and risk assessment. Currently, commonly used rainfall event segmentation methods are mostly based on fixed time intervals or fixed rainfall thresholds, such as using the number of consecutive rainless days or a fixed rainfall intensity threshold as the start and end conditions of events. While these methods are computationally simple and easy to implement, their parameter settings often rely on experience and are difficult to adapt to differences in rainfall rhythms and climatic conditions in different regions. This can easily lead to an overestimation or underestimation of rainfall events, thus affecting the stability and universality of landslide triggering thresholds. Therefore, existing methods for constructing event-level rainfall samples for rainfall-induced landslides are unreasonable, resulting in low reliability of rainfall-induced landslide risk assessment and early warning. Summary of the Invention

[0006] This invention provides a method for constructing event-level rainfall samples for rainfall-induced landslides. Based on daily rainfall data, this method improves the reliability of risk assessment and early warning for rainfall-induced landslides by rationally constructing event-level rainfall event samples. The method includes:

[0007] Obtain daily rainfall data and corresponding historical landslide data within the study area;

[0008] Based on rainfall stations, Thiessen polygons are constructed, and based on the Thiessen polygons, spatial matching is performed on landslide historical data to determine the rainfall station corresponding to each landslide historical record, thus forming a correspondence between historical landslide events and rainfall stations.

[0009] Based on the correspondence between historical landslide events and rainfall stations, and a pre-constructed comprehensive evaluation function for rainfall event classification, the particle swarm optimization algorithm is used to search for the optimal rainfall event classification parameters within a preset parameter range.

[0010] Based on the optimal rainfall event segmentation parameters, the daily rainfall data is segmented into rainfall events, continuous rainfall processes are identified as independent rainfall events, and the independent rainfall events are decomposed to obtain several decomposed sub-events with different cumulative characteristics.

[0011] Based on historical landslide data, disaster-causing markers are assigned to independent rainfall events and their decomposed sub-events, and an event-level rainfall sample set containing both disaster-causing and non-disaster-causing events is constructed.

[0012] This invention also provides a device for constructing event-level rainfall samples for rainfall-induced landslides, which improves the reliability of risk assessment and early warning for rainfall-induced landslides by rationally constructing event-level rainfall event samples based on daily rainfall data. The device includes:

[0013] The acquisition unit is used to acquire daily rainfall data and corresponding historical landslide data within the study area;

[0014] The matching unit is used to construct Thiessen polygons based on rainfall stations, and to perform spatial matching on landslide historical data based on the Thiessen polygons to determine the rainfall station corresponding to each landslide historical record, thus forming a correspondence between historical landslide events and rainfall stations.

[0015] The optimal rainfall event classification parameter determination unit is used to search for the optimal rainfall event classification parameters within a preset parameter range based on the correspondence between historical landslide events and rainfall stations, as well as a pre-constructed comprehensive evaluation function for rainfall event classification, using the particle swarm optimization algorithm.

[0016] The division and decomposition unit is used to divide the daily rainfall data into rainfall events based on the optimal rainfall event division parameters, identify continuous rainfall processes as independent rainfall events, and decompose the independent rainfall events to obtain several decomposed sub-events with different cumulative characteristics.

[0017] The construction unit is used to label independent rainfall events and their decomposed sub-events based on historical landslide data, and to construct an event-level rainfall sample set containing both disaster-causing and non-disaster-causing events.

[0018] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for constructing event-level rainfall samples for rainfall-induced landslides.

[0019] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for constructing event-level rainfall samples for rainfall-induced landslides.

[0020] The present invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described method for constructing event-level rainfall samples for rainfall-induced landslides.

[0021] In this invention, the event-level rainfall sample construction scheme for rainfall-induced landslides involves: acquiring daily rainfall data and corresponding historical landslide data within the study area; constructing Thiessen polygons based on rainfall stations, and spatially matching the historical landslide data using these polygons to determine the rainfall station corresponding to each historical landslide event, thus establishing a correspondence between historical landslide events and rainfall stations; based on this correspondence and a pre-constructed comprehensive evaluation function for rainfall event classification, using a particle swarm optimization algorithm to search for optimal rainfall event classification parameters within a preset parameter range; classifying daily rainfall data into rainfall events based on these optimal parameters, identifying continuous rainfall processes as independent rainfall events, and decomposing these independent events into several sub-events with different cumulative characteristics; and, based on the historical landslide data, labeling the independent rainfall events and their decomposed sub-events as disaster-causing events, thus constructing an event-level rainfall sample set containing both disaster-causing and non-disaster-causing events. This demonstrates that by rationally constructing event-level rainfall event samples, the reliability of rainfall-induced landslide risk assessment and early warning can be improved. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0023] Figure 1 This is a flowchart illustrating the method for constructing event-level rainfall samples for rainfall-induced landslides in an embodiment of the present invention.

[0024] Figure 2This is a schematic diagram of the spatial matching of Thiessen polygons and landslide data constructed based on rainfall stations in an embodiment of the present invention;

[0025] Figure 3 This is a scoring chart of rainfall event classification parameters in an embodiment of the present invention;

[0026] Figure 4 This is a flowchart illustrating a method for constructing event-level rainfall samples for rainfall-induced landslides according to another embodiment of the present invention.

[0027] Figure 5 This is a sensitivity diagram of different combinations of rainfall variables and their sensitivity to attenuation coefficients in an embodiment of the present invention;

[0028] Figure 6 This is a baseline threshold curve obtained by fitting all rainfall samples in the embodiments of the present invention;

[0029] Figure 7 This is a unified feature binning statistical diagram in an embodiment of the present invention;

[0030] Figure 8 This is a curve showing the probability of disaster in an embodiment of the present invention.

[0031] Figure 9 This is a disaster probability diagram of the combination of rainfall variables in an embodiment of the present invention;

[0032] Figure 10 A comparison chart of ROC curves for landslide triggering discrimination between the method and the conventional method is provided for embodiments of the present invention;

[0033] Figure 11 This is a schematic diagram of the structure of the event-level rainfall sample construction device for rainfall-induced landslides in an embodiment of the present invention;

[0034] Figure 12 This is a schematic diagram of the structure of an event-level rainfall sample construction device for rainfall-induced landslides in another embodiment of the present invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0036] The acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations.

[0037] Existing rainfall-based landslide analysis and early warning technologies still have room for improvement in terms of rainfall event construction and rainfall characteristic parameter configuration. On the one hand, rainfall event classification methods often rely on empirical parameter settings, making it difficult to consider both rainfall rhythms and landslide response characteristics in different regions. On the other hand, after the construction of event-level rainfall samples, the configuration of rainfall characteristic indicators and their parameters lacks systematic optimization for the overall structure of the event samples, which restricts the stability and engineering applicability of landslide triggering discrimination or probability calculation results. In practical applications, how to synergistically optimize rainfall characteristics and achieve quantitative characterization of landslide triggering risk based on the reasonable construction of rainfall event samples remains an urgent technical problem to be solved.

[0038] Against this backdrop, it is necessary to propose a scheme for constructing event-level rainfall samples for rainfall-induced landslides. This scheme is a method for constructing rainfall events and calculating trigger probabilities for rainfall-induced landslides. Based on daily rainfall data, this method identifies rainfall event samples that conform to regional characteristics through a reasonable rainfall event classification mechanism. Furthermore, it systematically optimizes the rainfall variable indicators and their parameter configurations based on the structural characteristics of the event-level rainfall samples. On this basis, by constructing a unified expression form for rainfall event characteristics, it achieves quantitative calculation and discrimination evaluation of the landslide risk triggered by different rainfall events, thereby improving the stability and applicability of the rainfall-induced landslide triggering analysis results. The following section provides a detailed introduction to this event-level rainfall sample construction scheme for rainfall-induced landslides.

[0039] Figure 1 This is a flowchart illustrating the method for constructing event-level rainfall samples for rainfall-induced landslides in an embodiment of the present invention. Figure 1 As shown, the method includes the following steps:

[0040] Step 101: Obtain daily rainfall data and corresponding historical landslide data within the study area;

[0041] Step 102: Construct Thiessen polygons based on rainfall stations, and perform spatial matching on landslide historical data based on the Thiessen polygons to determine the rainfall station corresponding to each landslide historical record, thus forming a correspondence between historical landslide events and rainfall stations.

[0042] Step 103: Based on the correspondence between historical landslide events and rainfall stations, and the pre-constructed comprehensive evaluation function for rainfall event classification, the particle swarm optimization algorithm is used to search for the optimal rainfall event classification parameters within the preset parameter range;

[0043] Step 104: Based on the optimal rainfall event segmentation parameters, divide the daily rainfall data into rainfall events, identify continuous rainfall processes as independent rainfall events, and decompose the independent rainfall events to obtain several decomposed sub-events with different cumulative characteristics.

[0044] Step 105: Based on historical landslide data, label independent rainfall events and their decomposed sub-events as disaster-causing events, and construct an event-level rainfall sample set that includes both disaster-causing and non-disaster-causing events.

[0045] The method for constructing event-level rainfall samples for rainfall-induced landslides provided in this invention involves: acquiring daily rainfall data and corresponding historical landslide data within the study area; constructing Thiessen polygons based on rainfall stations, and spatially matching the historical landslide data using these polygons to determine the rainfall station corresponding to each historical landslide event, thus establishing a correspondence between historical landslide events and rainfall stations; based on this correspondence and a pre-constructed comprehensive evaluation function for rainfall event classification, searching for optimal rainfall event classification parameters within a preset parameter range using a particle swarm optimization algorithm; classifying daily rainfall data into rainfall events based on these optimal parameters, identifying continuous rainfall processes as independent rainfall events, and decomposing these independent events into several sub-events with different cumulative characteristics; and, according to the historical landslide data, labeling the independent rainfall events and their decomposed sub-events as disaster-causing events, thus constructing an event-level rainfall sample set containing both disaster-causing and non-disaster-causing events. This method improves the reliability of risk assessment and early warning for rainfall-induced landslides by rationally constructing event-level rainfall event samples. The following provides a detailed description of this method for constructing event-level rainfall samples for rainfall-induced landslides.

[0046] This invention provides a method for constructing event-level rainfall samples for rainfall-induced landslides. This method is a method for constructing rainfall events and calculating triggering probabilities for rainfall-induced landslides, which will be described in detail below.

[0047] In step 101 above, daily rainfall data and corresponding historical landslide data within the study area are obtained. To ensure the effectiveness of subsequent rainfall event construction and landslide trigger probability calculation, this step includes the following sub-steps:

[0048] 1.1: Obtain daily rainfall data from rainfall stations within the study area.

[0049] The daily rainfall data are diurnal rainfall observation data, including station number, location, date information, and corresponding daily rainfall values. Missing and abnormal records are screened and processed to form a basic dataset that can be used for subsequent analysis.

[0050] 1.2: Processing of historical landslide data.

[0051] The data mainly includes information such as the date of landslide occurrence, latitude and longitude coordinates, landslide type, scale level, and triggering factors. To ensure the accuracy of the analysis, the raw data was screened, records with missing time and spatial information were removed, and the coordinate and date formats were standardized.

[0052] In step 102 above, Thiessen polygon spatial units are constructed using rainfall stations as spatial control points, and based on the Thiessen polygon spatial units, landslide historical data are spatially matched to assign a corresponding rainfall station to each landslide record.

[0053] To establish a spatial correspondence between landslide events and rainfall observation data, this step includes the following sub-steps:

[0054] 2.1: Using the locations of each rainfall station within the study area as generation points, construct a Thiessen polygon covering the study area, dividing the study area into several spatial sub-regions.

[0055] 2.2: Spatially overlay the historical landslide data with the Thiessen polygon, determine the Thiessen polygon region where the landslide event is located based on the spatial location information of the landslide event, and thus identify the rainfall station corresponding to the landslide event. Figure 2 This is a schematic diagram illustrating the spatial matching of Thiessen polygons and landslide data constructed based on rainfall stations, as described in an embodiment of the present invention. Figure 2 In the diagram, small flags represent rainfall stations, and dots represent landslide events.

[0056] 2.3: Establish a one-to-one correspondence between landslide events and rainfall stations. Associate each historical landslide record with its corresponding rainfall station to form a spatial matching result between landslides and rainfall stations.

[0057] As described above, in one embodiment, Thiessen polygons are constructed based on rainfall stations, and spatial matching is performed on landslide historical data based on the Thiessen polygons to determine the rainfall station corresponding to each landslide historical record, thus forming a correspondence between historical landslide events and rainfall stations. This may include:

[0058] Using the locations of each rainfall station within the study area as generation points, a Thiessen polygon covering the study area is constructed, dividing the study area into several spatial sub-regions;

[0059] The historical landslide data is spatially overlaid with the Thiessen polygon, and the spatial sub-region of the Thiessen polygon where the landslide event is located is determined based on the spatial location information of the landslide event, and the rainfall station corresponding to the landslide event is identified.

[0060] Based on the rainfall stations corresponding to the identified landslide events, a correspondence between historical landslide events and rainfall stations is established.

[0061] In step 103 above, a comprehensive evaluation function is constructed, and the particle swarm optimization algorithm is used to search for the optimal rainfall event division parameters in the parameter space to determine a reasonable division scheme for rainfall events.

[0062] To achieve objective determination of parameters for classifying rainfall events, this step includes the following sub-steps:

[0063] 3.1: Using two key parameters, the time window length DT and the rainfall threshold ET, a sliding analysis is performed on the daily rainfall sequence to determine the start and end of rainfall events. When the cumulative rainfall within the time window of length DT is lower than the threshold ET, the rainfall process is considered to have ended and a new event has been formed.

[0064] 3.2: Construct a comprehensive evaluation function for the classification of rainfall events.

[0065] 3.2.1: Using historical rainfall data from all meteorological stations in the study area from xxxx to yyyy, the average annual number of rainy days (T) and the average longest consecutive rainy days (CWD) in the study area were calculated to be 137 days, which were used as the climate constraint benchmark.

[0066] 3.2.2: R is the landslide event coverage rate, which represents the proportion of landslide occurrence dates that fall within the defined rainfall event time range, ensuring that the defined rainfall events include more landslides.

[0067] 3.2.3: Constructing a comprehensive evaluation function based on landslide response, event rhythm, and climate constraints. F(DT,ET) Its expression is, that is, in one embodiment, the comprehensive evaluation function for classifying rainfall events is:

[0068] ;

[0069] in, , , These represent the landslide response factors, rainfall days, and event duration factors, respectively.

[0070] ;

[0071] ;

[0072] ;

[0073] In the formula: F(DT,ET) Comprehensive evaluation factors are used to classify rainfall events; R is the landslide event coverage rate; J is the average number of events per year after classification; Q is the average event duration after classification; T is the multi-year average number of rainy days per year; CWD is the multi-year average number of consecutive rainy days.

[0074] 3.3: The optimal event partitioning parameters are searched within the parameter space using the Particle Swarm Optimization (PSO) algorithm. Each parameter combination (DT, ET) is considered as a particle position, and its fitness is determined by the comprehensive evaluation function constructed in step 3.2. F(DT,ET) Calculated; through iterative updates of particle velocity and position, a search is conducted within a preset parameter range to achieve... F(DT,ET) The parameter combination with the maximum value is taken as the optimal event partitioning parameter. Figure 3 This is a parameter scoring chart for classifying rainfall events in an embodiment of the present invention.

[0075] As described above, in one embodiment, based on the correspondence between historical landslide events and rainfall stations, and a pre-constructed comprehensive evaluation function for rainfall event classification, the particle swarm optimization algorithm is used to search for optimal rainfall event classification parameters within a preset parameter range. This may include:

[0076] By using the time window length and rainfall threshold as a parameter combination, a sliding analysis is performed on the daily rainfall data series to determine the time range of rainfall events;

[0077] Using historical rainfall data from all meteorological stations in the study area within a preset time period, the multi-year average annual number of rainy days and the multi-year average longest consecutive rainfall days in the study area were determined.

[0078] Based on historical landslide data and the time range of the rainfall events, the proportion of landslide occurrence dates falling within the time range of the rainfall events is determined as the landslide event coverage rate;

[0079] Based on the landslide event coverage rate, determine the landslide response factor value; based on the multi-year average annual rainfall days, determine the rainfall days factor value; based on the multi-year average longest consecutive rainfall days, determine the event duration factor value.

[0080] Based on the landslide response factor value, the rainfall day factor value, the event duration factor value, and the comprehensive evaluation function for rainfall event classification pre-constructed based on landslide response, event rhythm, and climate constraints, determine the comprehensive evaluation factor value for rainfall event classification corresponding to each set of parameters.

[0081] Each set of parameters is regarded as a particle position. Through iterative updates of particle velocity and position, the parameter combination that maximizes the value of the comprehensive evaluation factor for rainfall event classification is searched within the preset parameter range and is taken as the optimal event classification parameter.

[0082] In step 104 above, based on the optimal rainfall event segmentation parameters determined in step 103, the daily rainfall data is segmented into rainfall events, continuous rainfall processes are identified as independent rainfall events, and these rainfall events are then decomposed. This step includes the following sub-steps:

[0083] 4.1: Based on the optimal time window length DT and rainfall threshold ET output in step 103, a sliding analysis is performed on the daily rainfall sequence of each rainfall station in the study area, and the daily rainfall data is divided into rainfall events according to the preset event judgment rules.

[0084] 4.2: In the results of rainfall event segmentation, considering that landslides may be triggered on any day within a rainfall event, using only the complete rainfall event as the analysis unit may easily overlook the triggering differences between different stages within the event. Therefore, after completing the identification of rainfall events, it is necessary to further decompose them. Specifically, for each identified rainfall event, starting from the event's start date, the rainfall process within the event is accumulated day by day in chronological order, decomposing a complete rainfall event into several sub-events with different accumulation characteristics. Each sub-event uses "event start date - current day" as a time window, including daily rainfall information and corresponding accumulated rainfall characteristics within that time window, to characterize the state of the rainfall event at different evolutionary stages.

[0085] In step 105 above, based on historical landslide data, the rainfall events and their decomposed sub-events obtained in step 104 are labeled as disaster-causing events, constructing an event-level rainfall sample set containing both disaster-causing and non-disaster-causing events. This step includes the following sub-steps:

[0086] 5.1: Based on the location of the landslide, determine the corresponding rainfall station and obtain all rainfall sub-events obtained in step 104 under that rainfall station.

[0087] 5.2: Using the landslide occurrence time as the criterion, the landslide occurrence time is matched with the time window of the rainfall sub-event. When the landslide occurrence time is the same as the end time of a certain rainfall sub-event, the rainfall sub-event is marked as a disaster-causing event.

[0088] 5.3: Rainfall sub-events that do not match the timing of any landslide occurrence are uniformly labeled as non-hazardous events.

[0089] 5.4: Summarize all disaster-causing and non-disaster-causing events from all rainfall stations, and construct an event-level rainfall sample set containing event time information, daily rainfall sequences, and disaster-causing markers for subsequent rainfall variable index calculation and trigger probability analysis.

[0090] In practice, a rainfall-induced landslide risk prediction model can be constructed using an event-level rainfall sample set. The input of this model can be daily rainfall data in the area to be predicted, and the output can be a prediction result of a disaster-causing or non-disaster-causing rainfall-induced landslide risk. By constructing a reasonable event-level rainfall event sample, the reliability of rainfall-induced landslide risk assessment and early warning can be improved.

[0091] Furthermore, after classifying rainfall events, existing technologies typically conduct landslide triggering analysis based on a limited set of rainfall characteristic indicators. These methods rely heavily on empirical settings for input feature selection and parameter configuration, lacking a systematic optimization mechanism for the overall structure of rainfall event samples. This limits the distinguishing effect of different rainfall events in the feature space, thus affecting the stability and discrimination effectiveness of the triggering probability calculation results. Considering this technical problem, the embodiments of this invention propose the following further preferred solutions.

[0092] Figure 4 This is a flowchart illustrating a method for constructing event-level rainfall samples for rainfall-induced landslides according to another embodiment of the present invention, as shown below. Figure 4 As shown, in one embodiment, the above-described method for constructing event-level rainfall samples for rainfall-induced landslides may further include the following steps:

[0093] Step 106: Based on each independent rainfall event and its decomposed sub-events, construct multiple sets of different combinations of rainfall variables and their corresponding attenuation coefficient values ​​to characterize the features of rainfall events;

[0094] Step 107: Based on the event-level rainfall sample set, use sensitivity analysis to quantitatively evaluate different combinations of rainfall variables and their corresponding attenuation coefficients, and determine the optimal combination of rainfall variables and their corresponding attenuation coefficient values ​​that can effectively distinguish between disaster-causing events and non-disaster-causing events.

[0095] Step 108: Based on the optimal combination of rainfall variables and their corresponding attenuation coefficient values, construct unified features for the event-level rainfall sample set to obtain feature values ​​that characterize the features of each rainfall event;

[0096] Step 109: Based on the feature values ​​of each rainfall event, determine the landslide trigger probability corresponding to each rainfall event under probability constraints.

[0097] In step 106 above, a combination of rainfall variables is constructed to characterize the features of rainfall events, and the rainfall is attenuated based on existing effective rainfall calculation methods. This step includes the following sub-steps:

[0098] 6.1: For each rainfall event and its sub-events, extract the total rainfall E, total effective rainfall EE, rainfall intensity I, and event duration D within the event as basic rainfall variables to describe the characteristics of the current rainfall process.

[0099] 6.2: Calculate the effective cumulative rainfall AE3 (3 days before), AE5 (5 days before), AE7 (7 days before), AE10 (10 days before), and AE15 (15 days before) at different time scales before the event to characterize the cumulative impact of the previous rainfall on the landslide triggering.

[0100] The effective rainfall amount is calculated using existing rainfall attenuation methods to characterize the actual intensity of rainfall after it decays over time. Considering the influence of factors such as slope runoff and evaporation, not all rainfall participates in the landslide triggering process, and the longer the time since the landslide occurs, the weaker its inducing effect. The calculation formula is as follows:

[0101] ;

[0102] In the formula: EE represents the effective rainfall; R i Let represent the rainfall on the i-th day before the landslide, D represent the duration of the rainfall event, and K represent the rainfall attenuation coefficient.

[0103] 6.3: The above steps yield the combination of rainfall variables for each rainfall event. For each combination of rainfall variables, a decay coefficient value is assigned to form a set of candidate input variables for subsequent sensitivity analysis and parameter optimization.

[0104] As can be seen from the above, in one embodiment, based on each independent rainfall event and its decomposed sub-events, multiple sets of different combinations of rainfall variables and their corresponding attenuation coefficient values ​​are constructed to characterize the features of the rainfall event, including:

[0105] For each independent rainfall event and its sub-events, the total rainfall, total effective rainfall, rainfall intensity, and event duration within the event are extracted as basic rainfall variables to describe the characteristics of the current rainfall process.

[0106] The effective cumulative rainfall in the preceding time scales before the event is calculated. The effective cumulative rainfall in the preceding time scale includes: effective cumulative rainfall in the preceding 3 days, effective cumulative rainfall in the preceding 5 days, effective cumulative rainfall in the preceding 7 days, effective cumulative rainfall in the preceding 10 days, and effective cumulative rainfall in the preceding 15 days. The effective cumulative rainfall in the preceding time scale is used to characterize the cumulative impact of the preceding rainfall on the landslide triggering.

[0107] Using the basic rainfall variables and the previous effective cumulative rainfall, multiple combinations of different rainfall variables used to characterize rainfall events are determined, along with the attenuation coefficient value corresponding to each combination of rainfall variables.

[0108] In step 107 above, based on the event-level rainfall sample set, a sensitivity analysis method is introduced to quantitatively evaluate the distinguishing ability of different combinations of rainfall variables and their corresponding attenuation coefficients, and to determine the optimal combination of rainfall variables and parameter configuration scheme. This step includes the following sub-steps:

[0109] 7.1: Given a combination of rainfall variables and attenuation coefficients, a sensitivity index is introduced to quantitatively assess the ability to distinguish between disaster-causing and non-disaster-causing events. Divergence is used as a distance function to measure the difference between the distributions of the two types of samples. The distance between the corresponding probability distributions of disaster-causing and non-disaster-causing rainfall events is calculated, yielding sensitivity coefficients under different attenuation coefficients. A larger sensitivity coefficient indicates a stronger ability of the rainfall variable combination to distinguish between disaster-causing and non-disaster-causing events.

[0110] The sensitivity analysis formula is as follows:

[0111]

[0112] In the formula: SA(k) The sensitivity coefficient value is calculated under the attenuation coefficient k; This indicates that the samples of rainfall that caused the disaster were concentrated, and the attenuation coefficient was high. k Under the condition of rainfall variable R 1 and R The probability distribution of 2; This indicates a concentration of non-disastrous rainfall samples, with a high attenuation coefficient. k Under the condition of rainfall variable R 1 and R The probability distribution of 2; R 1. R 2 represents two rainfall variables; L represents the disaster-causing rainfall sample set; NL represents the non-disaster-causing rainfall sample set; d is a distance function that measures the difference in distribution between the two types of samples, and this distance function is as follows: JS(p,q) The function, namely the JS divergence used in this embodiment of the invention, is defined by the formula: In one embodiment, the divergence is used as a distance function to measure the difference between the distributions of two types of samples, namely disaster-causing and non-disaster-causing events, to calculate the distance value between the corresponding probability distributions of disaster-causing rainfall events and non-disaster-causing rainfall events, including calculating the distance value according to the following formula:

[0113] ;

[0114] In the formula: p and q represent the probability distributions of two types of disaster-causing and non-disaster-causing rainfall events for which distance needs to be calculated, respectively corresponding to the above. , N is the number of histograms that divide the probability distribution; i The interval indices (i=1,2,3...N) after the probability distribution is partitioned. x i This indicates the state corresponding to the rainfall variable falling into the i-th interval; the larger the JS value, the higher the discrimination of the indicator combination (rainfall variable combination) in disaster-causing and non-disaster-causing samples, and the stronger its sensitivity.

[0115] 7.2: Introducing the particle swarm optimization algorithm, a joint search is performed on the combination of rainfall variables and attenuation coefficients within a preset parameter space. With the maximization of the sensitivity coefficient as the optimization objective, the optimal parameter combination that maximizes the distinction between disaster-causing and non-disaster-causing events is obtained. Figure 5 This is a graph showing the sensitivity of different combinations of rainfall variables and their attenuation coefficients in an embodiment of the present invention.

[0116] Based on the above sensitivity indicators, the Particle Swarm Optimization (PSO) algorithm is used to simultaneously search for combinations of rainfall variables and the attenuation coefficient k to obtain the optimal parameter combination that maximizes the distinguishability between disaster-causing and non-disaster-causing samples. The formula is as follows:

[0117] ;

[0118] In the formula: , , which represent the optimal combination of rainfall variables and their attenuation coefficients; m and k represent the combination of rainfall variables and their attenuation coefficients, respectively.

[0119] As described above, in one embodiment, based on the event-level rainfall sample set, sensitivity analysis is used to quantitatively evaluate different combinations of rainfall variables and their corresponding attenuation coefficients, determining the optimal combination of rainfall variables and their corresponding attenuation coefficient values ​​that can effectively distinguish between disaster-causing events and non-disaster-causing events, including:

[0120] Given a combination of rainfall variables and attenuation coefficient values, a sensitivity index is introduced to quantitatively evaluate the ability to distinguish between disaster-causing and non-disaster-causing events, resulting in the distinguishability between disaster-causing and non-disaster-causing events and the corresponding sensitivity coefficient values.

[0121] A particle swarm optimization algorithm is introduced to jointly search for combinations of rainfall variables and attenuation coefficients within a preset parameter space. With the maximization of the sensitivity coefficient value as the optimization objective, the optimal combination of rainfall variables and attenuation coefficients that maximizes the distinction between disaster-causing and non-disaster-causing events are obtained as the optimal combination of rainfall variables and their corresponding attenuation coefficients.

[0122] As described above, in one embodiment, given a combination of rainfall variables and attenuation coefficient values, a sensitivity index is introduced to quantitatively evaluate the ability to distinguish between disaster-causing and non-disaster-causing events, obtaining the distinguishability between disaster-causing and non-disaster-causing events and the corresponding sensitivity coefficient values, including:

[0123] Divergence is used as a distance function to measure the difference between the distributions of two types of samples, namely disaster-causing and non-disaster-causing events. The distance value between the corresponding probability distributions of disaster-causing rainfall events and non-disaster-causing rainfall events is calculated, and the distance value represents the discriminant.

[0124] Based on the distance value, the sensitivity coefficient values ​​under different attenuation coefficient conditions are obtained. The larger the sensitivity coefficient value, the stronger the ability of the combination of rainfall variables to distinguish between disaster-causing and non-disaster-causing events.

[0125] In step 108 above, based on the optimal combination of rainfall variables and their corresponding attenuation coefficients obtained in step 107, a unified feature is constructed for the event-level rainfall samples, mapping the multidimensional rainfall variables to one-dimensional feature values ​​for subsequent landslide triggering probability modeling. This step includes the following sub-steps:

[0126] 8.1: The optimal combination of rainfall variables obtained by particle swarm optimization and its corresponding attenuation coefficient are used as the preferred parameter configuration scheme to calculate the characteristics of each rainfall event. Specifically, the effective rainfall EE and the effective rainfall AE15 of the previous 15 days are calculated for each rainfall event. The EE, AE15, and attenuation coefficient k=0.728 are the preferred results obtained through parameter optimization in this embodiment.

[0127] 8.2 Using all rainfall event samples, regression fitting is performed on the preferred combination of rainfall variables in a double logarithmic coordinate system to construct a benchmark threshold function to describe the overall distribution characteristics of disaster-causing and non-disaster-causing rainfall events. The benchmark threshold function is used to characterize the empirical relationship between rainfall variables. Figure 6 This is a schematic diagram of the baseline threshold curve obtained by fitting all rainfall samples in this embodiment of the invention. The expression is:

[0128] ;

[0129] In the formula: EE represents the effective rainfall of the event; AE15 represents the effective cumulative rainfall in the 15 days prior to the event; and The parameters to be estimated are obtained by fitting using the least squares method.

[0130] 8.3 Based on the aforementioned benchmark threshold function, calculate the degree of deviation of each rainfall event sample from the rainfall variable feature space relative to the benchmark threshold function, and define this degree of deviation as a unified feature value of the rainfall event. The unified feature value is used to characterize the relative positional relationship of the rainfall event in the overall sample distribution. It can be defined as the vertical distance from the rainfall sample point to the fitted curve, expressed as follows: In one embodiment, the unified feature value is used to characterize the relative positional relationship of the rainfall event in the overall sample distribution, and is the vertical distance from the rainfall sample point to the fitted curve. The benchmark threshold function is:

[0131] ;

[0132] Here, δ represents the degree of deviation of an event from the threshold line: a larger δ indicates a higher potential for causing disaster; a negative δ indicates that the event falls below the threshold line, generally corresponding to a lower probability of causing disaster. α and β All of these are fitting parameters for a baseline threshold function between the effective rainfall of the event and the cumulative effective rainfall in the preceding period, where... α This is a scaling factor that reflects the overall horizontal position of the baseline threshold curve. β This is an exponential parameter that reflects the intensity of the response of the effective rainfall of an event to changes in the cumulative effective rainfall in the preceding period.

[0133] As described above, in one embodiment, based on the optimal combination of rainfall variables and their corresponding attenuation coefficient values, a unified feature is constructed on the event-level rainfall sample set to obtain feature values ​​characterizing the features of each rainfall event, including:

[0134] The optimal combination of rainfall variables and its corresponding attenuation coefficient obtained by particle swarm optimization are used as the preferred parameter configuration scheme to calculate the characteristics of each rainfall event.

[0135] Using all rainfall event samples, regression fitting is performed on the optimal combination of rainfall variables in a double logarithmic coordinate system to construct a benchmark threshold function to describe the overall distribution characteristics of disaster-causing and non-disaster-causing rainfall events. The benchmark threshold function is used to characterize the relationship between rainfall variables.

[0136] Based on the benchmark threshold function, the degree of deviation of each rainfall event sample from the benchmark threshold function in the rainfall variable feature space is calculated, and the degree of deviation is used as a unified feature value characterizing the features of each rainfall event.

[0137] In step 109 above, based on the unified feature values ​​of rainfall events constructed in step 108, the landslide trigger probability corresponding to each rainfall event is calculated under probabilistic constraints, and the effectiveness of the method is verified by a discriminant performance evaluation method. This step includes the following sub-steps:

[0138] 9.1: Perform binning statistics on the unified feature values ​​obtained in step 108, divide the unified feature values ​​into several continuous intervals, and statistically analyze the frequency distribution of disaster-causing rainfall events and non-disaster-causing rainfall events in each unified feature interval. Figure 7 This is a unified feature binning statistical diagram in an embodiment of the present invention.

[0139] 9.2: Based on the binning statistics results in step 9.1, under probability constraints, the landslide triggering probability is calculated for each unified feature interval, and the ratio of the frequency of disaster-causing rainfall events in each interval to the frequency of all rainfall events is used as an empirical estimate of the triggering probability. On this basis, the triggering probability corresponding to the unified feature interval is fitted by a function to construct a time probability model between the unified feature value and the landslide triggering probability. Figure 8 This is a curve showing the probability of disaster in an embodiment of the present invention.

[0140] 9.3: Based on the time probability model constructed in step 9.2, the unified feature value of each rainfall event is substituted into the time probability model to calculate the corresponding landslide triggering probability, thereby realizing a quantitative characterization of the risk of landslides triggered by rainfall events. Figure 9 This is a disaster probability diagram for the combination of rainfall variables in an embodiment of the present invention.

[0141] 9.4 The landslide triggering probability calculated by the method of the present invention is compared with the landslide triggering discrimination result constructed based on the traditional rainfall event classification rules. The performance of different discrimination results is evaluated by the area under the receiver operating characteristic curve, so as to verify the effectiveness of the method of the present invention in terms of accuracy and stability of landslide triggering discrimination. Figure 10 This is a comparison chart of the ROC curves for landslide triggering discrimination between the method of the present invention and the traditional method in this embodiment of the invention.

[0142] In practice, the relationship between the unified feature value of each rainfall event and the landslide triggering probability can be obtained by further processing the event-level rainfall sample set. This can be used to construct a rainfall-induced landslide risk prediction model (which can be called a time probability model). The input of this model can be the unified feature value of each rainfall event in the area to be predicted, and the output can be the landslide triggering probability. This model can achieve a quantitative characterization of the risk of rainfall-induced landslides and further improve the reliability of rainfall-induced landslide risk assessment and early warning.

[0143] In step 109 above, in the field of binning statistical methods, probability constraints usually refer to pre-set statistical thresholds, such as: "requiring the statistical sample size in each rainfall interval to meet a 95% confidence level" or "pre-set classification accuracy to reach more than 90%".

[0144] In summary, based on daily rainfall data, this invention enables the quantitative calculation of landslide triggering risk by rationally constructing rainfall event samples and systematically optimizing rainfall characteristic parameters, thereby improving the reliability of rainfall-induced landslide risk assessment and early warning.

[0145] This invention also provides an apparatus for constructing event-level rainfall samples for rainfall-induced landslides, as described in the following embodiments. Since the principle by which this apparatus solves the problem is similar to the method for constructing event-level rainfall samples for rainfall-induced landslides, the implementation of this apparatus can refer to the implementation of the method for constructing event-level rainfall samples for rainfall-induced landslides; repeated details will not be elaborated further.

[0146] Figure 11 This is a schematic diagram of the structure of the event-level rainfall sample construction device for rainfall-induced landslides in an embodiment of the present invention, as shown below. Figure 11 As shown, the device includes:

[0147] Acquisition Unit 01 is used to acquire daily rainfall data and corresponding historical landslide data within the study area;

[0148] Matching unit 02 is used to construct Thiessen polygons based on rainfall stations, and to perform spatial matching on landslide historical data based on the Thiessen polygons to determine the rainfall station corresponding to each landslide historical record, thus forming a correspondence between historical landslide events and rainfall stations.

[0149] The optimal rainfall event classification parameter determination unit 03 is used to search for the optimal rainfall event classification parameters within a preset parameter range based on the correspondence between historical landslide events and rainfall stations, as well as the pre-constructed comprehensive evaluation function for rainfall event classification, using the particle swarm optimization algorithm.

[0150] The division and decomposition unit 04 is used to divide the daily rainfall data into rainfall events based on the optimal rainfall event division parameters, identify continuous rainfall processes as independent rainfall events, and decompose the independent rainfall events to obtain several decomposed sub-events with different cumulative characteristics.

[0151] Building Unit 05 is used to label independent rainfall events and their decomposed sub-events with disaster-causing characteristics based on historical landslide data, and to build an event-level rainfall sample set that includes both disaster-causing and non-disaster-causing events.

[0152] In one embodiment, Figure 12 This is a schematic diagram of the structure of an event-level rainfall sample construction device for rainfall-induced landslides according to another embodiment of the present invention, as shown below. Figure 12 As shown, the above-mentioned event-level rainfall sample construction device for rainfall-induced landslides also includes:

[0153] The parameter configuration unit 06 is used to construct multiple sets of different combinations of rainfall variables and their corresponding attenuation coefficient values ​​to characterize the features of rainfall events, based on each independent rainfall event and its decomposed sub-events.

[0154] The preferred unit 07 is used to perform a quantitative evaluation of different combinations of rainfall variables and their corresponding attenuation coefficients based on the event-level rainfall sample set using sensitivity analysis, and to determine the optimal combination of rainfall variables and their corresponding attenuation coefficient values ​​that can effectively distinguish between disaster-causing events and non-disaster-causing events.

[0155] Feature value construction unit 08 is used to construct unified features for the event-level rainfall sample set based on the optimal combination of rainfall variables and their corresponding attenuation coefficient values, so as to obtain feature values ​​that characterize the features of each rainfall event.

[0156] The trigger probability determination unit 09 is used to determine the landslide trigger probability corresponding to each rainfall event based on the feature values ​​of each rainfall event characteristics, under probability constraints.

[0157] In one embodiment, the parameter configuration unit is specifically used for:

[0158] For each independent rainfall event and its sub-events, the total rainfall, total effective rainfall, rainfall intensity, and event duration within the event are extracted as basic rainfall variables to describe the characteristics of the current rainfall process.

[0159] The effective cumulative rainfall in the preceding time scales before the event is calculated. The effective cumulative rainfall in the preceding time scale includes: effective cumulative rainfall in the preceding 3 days, effective cumulative rainfall in the preceding 5 days, effective cumulative rainfall in the preceding 7 days, effective cumulative rainfall in the preceding 10 days, and effective cumulative rainfall in the preceding 15 days. The effective cumulative rainfall in the preceding time scale is used to characterize the cumulative impact of the preceding rainfall on the landslide triggering.

[0160] Using the basic rainfall variables and the previous effective cumulative rainfall, multiple combinations of different rainfall variables used to characterize rainfall events are determined, along with the attenuation coefficient value corresponding to each combination of rainfall variables.

[0161] In one embodiment, the preferred unit is specifically used for:

[0162] Given a combination of rainfall variables and attenuation coefficient values, a sensitivity index is introduced to quantitatively evaluate the ability to distinguish between disaster-causing and non-disaster-causing events, resulting in the distinguishability between disaster-causing and non-disaster-causing events and the corresponding sensitivity coefficient values.

[0163] A particle swarm optimization algorithm is introduced to jointly search for combinations of rainfall variables and attenuation coefficients within a preset parameter space. With the maximization of the sensitivity coefficient value as the optimization objective, the optimal combination of rainfall variables and attenuation coefficients that maximizes the distinction between disaster-causing and non-disaster-causing events are obtained as the optimal combination of rainfall variables and their corresponding attenuation coefficients.

[0164] In one embodiment, given a combination of rainfall variables and attenuation coefficient values, a sensitivity index is introduced to quantitatively assess the ability to distinguish between disaster-causing and non-disaster-causing events, obtaining the distinguishability between disaster-causing and non-disaster-causing events and the corresponding sensitivity coefficient values, including:

[0165] Divergence is used as a distance function to measure the difference between the distributions of two types of samples, namely disaster-causing and non-disaster-causing events. The distance value between the corresponding probability distributions of disaster-causing rainfall events and non-disaster-causing rainfall events is calculated, and the distance value represents the discriminant.

[0166] Based on the distance value, the sensitivity coefficient values ​​under different attenuation coefficient conditions are obtained. The larger the sensitivity coefficient value, the stronger the ability of the combination of rainfall variables to distinguish between disaster-causing and non-disaster-causing events.

[0167] In one embodiment, divergence is used as a distance function to measure the difference between the distributions of disaster-causing and non-disaster-causing events. The distance between the corresponding probability distributions of disaster-causing and non-disaster-causing rainfall events is calculated using the following formula:

[0168] ;

[0169] In the formula: p and q represent the probability distributions of two types of disaster-causing and non-disaster-causing rainfall events for which distance needs to be calculated, respectively corresponding to the following: , N is the number of histograms used to partition the probability distribution; i is the index of the interval after partitioning the probability distribution (i=1,2,3...N). x i This indicates the state corresponding to the rainfall variable falling into the i-th interval; the larger the JS value, the higher the discrimination of the rainfall variable combination between disaster-causing and non-disaster-causing samples, and the stronger its sensitivity.

[0170] The sensitivity analysis formula is as follows:

[0171] ;

[0172] In the formula: SA(k) The sensitivity coefficient value is calculated under the attenuation coefficient k; This indicates that the samples of rainfall that caused the disaster were concentrated, and the attenuation coefficient was high. k Under the condition of rainfall variable R 1 and R The probability distribution of 2; This indicates a concentration of non-disastrous rainfall samples, with a high attenuation coefficient. k Under the condition of rainfall variable R 1 and R The probability distribution of 2; R 1. R2 represents two rainfall variables; L represents the disaster-causing rainfall sample set; NL represents the non-disaster-causing rainfall sample set; d represents a distance function that measures the difference in distribution between the two types of samples. JS(p,q) The function.

[0173] In one embodiment, the feature value construction unit is specifically used for:

[0174] The optimal combination of rainfall variables and its corresponding attenuation coefficient obtained by particle swarm optimization are used as the preferred parameter configuration scheme to calculate the characteristics of each rainfall event.

[0175] Using all rainfall event samples, regression fitting is performed on the optimal combination of rainfall variables in a double logarithmic coordinate system to construct a benchmark threshold function to describe the overall distribution characteristics of disaster-causing and non-disaster-causing rainfall events. The benchmark threshold function is used to characterize the relationship between rainfall variables.

[0176] Based on the benchmark threshold function, the degree of deviation of each rainfall event sample from the benchmark threshold function in the rainfall variable feature space is calculated, and the degree of deviation is used as a unified feature value characterizing the features of each rainfall event.

[0177] In one embodiment, the unified feature value is used to characterize the relative position of rainfall events in the overall sample distribution, and is the vertical distance from the rainfall sample point to the fitted curve. The benchmark threshold function is:

[0178] ;

[0179] Here, δ represents the degree of deviation of the event from the threshold line: the larger the δ, the higher its potential disaster-causing capacity; if δ is negative, it indicates that the event falls below the threshold line, corresponding to a low probability of disaster; EE is the effective rainfall of the event; AE15 is the effective cumulative rainfall in the 15 days prior to the event. α This is the proportionality coefficient. β For exponential parameters.

[0180] In one embodiment, the matching unit is specifically used for:

[0181] Using the locations of each rainfall station within the study area as generation points, a Thiessen polygon covering the study area is constructed, dividing the study area into several spatial sub-regions;

[0182] The historical landslide data is spatially overlaid with the Thiessen polygon, and the spatial sub-region of the Thiessen polygon where the landslide event is located is determined based on the spatial location information of the landslide event, and the rainfall station corresponding to the landslide event is identified.

[0183] Based on the rainfall stations corresponding to the identified landslide events, a correspondence between historical landslide events and rainfall stations is established.

[0184] In one embodiment, the optimal rainfall event segmentation parameter determination unit is specifically used for:

[0185] By using the time window length and rainfall threshold as a parameter combination, a sliding analysis is performed on the daily rainfall data series to determine the time range of rainfall events;

[0186] Using historical rainfall data from all meteorological stations in the study area within a preset time period, the multi-year average annual number of rainy days and the multi-year average longest consecutive rainfall days in the study area were determined.

[0187] Based on historical landslide data and the time range of the rainfall events, the proportion of landslide occurrence dates falling within the time range of the rainfall events is determined as the landslide event coverage rate;

[0188] Based on the landslide event coverage rate, determine the landslide response factor value; based on the multi-year average annual rainfall days, determine the rainfall days factor value; based on the multi-year average longest consecutive rainfall days, determine the event duration factor value.

[0189] Based on the landslide response factor value, the rainfall day factor value, the event duration factor value, and the comprehensive evaluation function for rainfall event classification pre-constructed based on landslide response, event rhythm, and climate constraints, determine the comprehensive evaluation factor value for rainfall event classification corresponding to each set of parameters.

[0190] Each set of parameters is regarded as a particle position. Through iterative updates of particle velocity and position, the parameter combination that maximizes the value of the comprehensive evaluation factor for rainfall event classification is searched within the preset parameter range and is taken as the optimal event classification parameter.

[0191] In one embodiment, the comprehensive evaluation function for classifying rainfall events is:

[0192] ;

[0193] in, , , These represent the landslide response factors, rainfall days, and event duration factors, respectively.

[0194] ;

[0195] ;

[0196] ;

[0197] In the formula: F(DT,ET)Comprehensive evaluation factors are used to classify rainfall events; R is the landslide event coverage rate; J is the average number of events per year after classification; Q is the average event duration after classification; T is the multi-year average number of rainy days per year; CWD is the multi-year average number of consecutive rainy days.

[0198] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described method for constructing event-level rainfall samples for rainfall-induced landslides.

[0199] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for constructing event-level rainfall samples for rainfall-induced landslides.

[0200] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described method for constructing event-level rainfall samples for rainfall-induced landslides.

[0201] In this embodiment of the invention, the event-level rainfall sample construction scheme for rainfall-induced landslides involves: acquiring daily rainfall data and corresponding historical landslide data within the study area; constructing Thiessen polygons based on rainfall stations, and spatially matching the historical landslide data using these Thiessen polygons to determine the rainfall station corresponding to each historical landslide event, thus establishing a correspondence between historical landslide events and rainfall stations; based on this correspondence and a pre-constructed comprehensive evaluation function for rainfall event classification, using a particle swarm optimization algorithm to search for optimal rainfall event classification parameters within a preset parameter range; classifying daily rainfall data into rainfall events based on these optimal parameters, identifying continuous rainfall processes as independent rainfall events, and decomposing these independent rainfall events into several sub-events with different cumulative characteristics; and, based on the historical landslide data, labeling the independent rainfall events and their decomposed sub-events as disaster-causing events, thus constructing an event-level rainfall sample set containing both disaster-causing and non-disaster-causing events. This demonstrates that by rationally constructing event-level rainfall event samples, the reliability of rainfall-induced landslide risk assessment and early warning can be improved.

[0202] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0203] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0204] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0205] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0206] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing event-level rainfall samples for rainfall-induced landslides, characterized in that, include: Obtain daily rainfall data and corresponding historical landslide data within the study area; Based on rainfall stations, Thiessen polygons are constructed, and based on the Thiessen polygons, spatial matching is performed on landslide historical data to determine the rainfall station corresponding to each landslide historical record, thus forming a correspondence between historical landslide events and rainfall stations. Based on the correspondence between historical landslide events and rainfall stations, and a pre-constructed comprehensive evaluation function for rainfall event classification, the particle swarm optimization algorithm is used to search for the optimal rainfall event classification parameters within a preset parameter range. Based on the optimal rainfall event segmentation parameters, the daily rainfall data is segmented into rainfall events, continuous rainfall processes are identified as independent rainfall events, and the independent rainfall events are decomposed to obtain several decomposed sub-events with different cumulative characteristics. Based on historical landslide data, disaster-causing markers are assigned to independent rainfall events and their decomposed sub-events, and an event-level rainfall sample set containing both disaster-causing and non-disaster-causing events is constructed. Based on each independent rainfall event and its decomposed sub-events, multiple sets of different combinations of rainfall variables and their corresponding attenuation coefficients are constructed to characterize the features of rainfall events. Based on the event-level rainfall sample set, sensitivity analysis is used to quantitatively evaluate different combinations of rainfall variables and their corresponding attenuation coefficients, determining the optimal combination of rainfall variables and their corresponding attenuation coefficients that can effectively distinguish between disaster-causing events and non-disaster-causing events. This includes: under given combinations of rainfall variables and attenuation coefficient values, introducing sensitivity indicators to quantitatively evaluate the ability to distinguish between disaster-causing events and non-disaster-causing events, obtaining the distinguishability between disaster-causing and non-disaster-causing events and their corresponding sensitivity coefficient values; introducing a particle swarm optimization algorithm to jointly search for combinations of rainfall variables and attenuation coefficient values ​​within a preset parameter space, with the maximization of the sensitivity coefficient value as the optimization objective, to obtain the combination of rainfall variables and attenuation coefficient values ​​that maximizes the distinguishability between disaster-causing and non-disaster-causing events as the optimal combination of rainfall variables and their corresponding attenuation coefficient values. Based on the optimal combination of rainfall variables and their corresponding attenuation coefficients, a unified feature is constructed for the event-level rainfall sample set to obtain feature values ​​characterizing the features of each rainfall event. This includes: using the optimal combination of rainfall variables obtained through particle swarm optimization and their corresponding attenuation coefficients as the preferred parameter configuration scheme to calculate the features of each rainfall event; using all rainfall event samples, performing regression fitting on the optimal combination of rainfall variables in a double logarithmic coordinate system to construct a benchmark threshold function describing the overall distribution characteristics of disaster-causing and non-disaster-causing rainfall events, where the benchmark threshold function characterizes the relationship between rainfall variables; and based on the benchmark threshold function, calculating the degree of deviation of each rainfall event sample from the benchmark threshold function in the rainfall variable feature space, and using this degree of deviation as the unified feature value characterizing the features of each rainfall event. Based on the feature values ​​of each rainfall event, the landslide trigger probability corresponding to each rainfall event is determined under probabilistic constraints.

2. The method as described in claim 1, characterized in that, Based on each independent rainfall event and its decomposed sub-events, multiple sets of different combinations of rainfall variables and their corresponding attenuation coefficient values ​​are constructed to characterize the features of rainfall events, including: For each independent rainfall event and its sub-events, the total rainfall, total effective rainfall, rainfall intensity, and event duration within the event are extracted as basic rainfall variables to describe the characteristics of the current rainfall process. The effective cumulative rainfall in the preceding time scales before the event is calculated. The effective cumulative rainfall in the preceding time scale includes: effective cumulative rainfall in the preceding 3 days, effective cumulative rainfall in the preceding 5 days, effective cumulative rainfall in the preceding 7 days, effective cumulative rainfall in the preceding 10 days, and effective cumulative rainfall in the preceding 15 days. The effective cumulative rainfall in the preceding time scale is used to characterize the cumulative impact of the preceding rainfall on the landslide triggering. Using the basic rainfall variables and the previous effective cumulative rainfall, multiple combinations of different rainfall variables used to characterize rainfall events are determined, along with the attenuation coefficient value corresponding to each combination of rainfall variables.

3. The method as described in claim 1, characterized in that, Given a combination of rainfall variables and attenuation coefficient values, a sensitivity index is introduced to quantitatively assess the ability to distinguish between disaster-causing and non-disaster-causing events, yielding the distinguishability between disaster-causing and non-disaster-causing events and their corresponding sensitivity coefficient values, including: Divergence is used as a distance function to measure the difference between the distributions of two types of samples, namely disaster-causing and non-disaster-causing events. The distance value between the corresponding probability distributions of disaster-causing rainfall events and non-disaster-causing rainfall events is calculated, and the distance value represents the discriminant. Based on the distance value, the sensitivity coefficient values ​​under different attenuation coefficient conditions are obtained. The larger the sensitivity coefficient value, the stronger the ability of the combination of rainfall variables to distinguish between disaster-causing and non-disaster-causing events.

4. The method as described in claim 3, characterized in that, Divergence is used as a distance function to measure the difference between the distributions of disaster-causing and non-disaster-causing events. The distance between the corresponding probability distributions of disaster-causing and non-disaster-causing rainfall events is calculated, including the calculation of the distance value according to the following formula: ； In the formula: p and q represent the probability distributions of two types of disaster-causing and non-disaster-causing rainfall events for which distance needs to be calculated, respectively corresponding to the following: , N is the number of histograms that divide the probability distribution; i The interval indices (i=1,2,3...N) after the probability distribution is partitioned. x i This indicates that the rainfall variable falls into the first... i The state corresponding to each interval; the larger the JS value, the higher the discrimination of the rainfall variable combination in disaster-causing and non-disaster-causing samples, and the stronger its sensitivity. The sensitivity analysis formula is as follows: ； In the formula: SA(k) The sensitivity coefficient value is calculated under the attenuation coefficient k. R 1. R 2 represents two rainfall variables; L represents the disaster-causing rainfall sample set; NL is the non-disastrous rainfall sample set; d is a distance function that measures the difference in distribution between the two types of samples. JS(p,q) The function, This indicates that the samples of rainfall that caused the disaster were concentrated, and the attenuation coefficient was high. k Under the condition of rainfall variable R 1 and R The probability distribution of 2; This indicates a concentration of non-disastrous rainfall samples, with a high attenuation coefficient. k Under the condition of rainfall variable R 1 and R The probability distribution of 2.

5. The method as described in claim 1, characterized in that, The unified feature value is used to characterize the relative position of rainfall events in the overall sample distribution, and is the vertical distance from the rainfall sample point to the fitted curve. The benchmark threshold function is: ； Here, δ represents the degree of deviation of the event from the threshold line: the larger the δ, the higher its potential disaster-causing capacity; if δ is negative, it indicates that the event falls below the threshold line, corresponding to a low probability of disaster; EE is the effective rainfall of the event; AE15 is the effective cumulative rainfall in the 15 days prior to the event. α This is the proportionality coefficient. β For exponential parameters.

6. The method as described in claim 1, characterized in that, Based on rainfall stations, Thiessen polygons are constructed. Then, based on these Thiessen polygons, spatial matching is performed on historical landslide data to determine the rainfall station corresponding to each historical landslide event, thus establishing a correspondence between historical landslide events and rainfall stations. This includes: Using the locations of each rainfall station within the study area as generation points, a Thiessen polygon covering the study area is constructed, dividing the study area into several spatial sub-regions; The historical landslide data is spatially overlaid with the Thiessen polygon, and the spatial sub-region of the Thiessen polygon where the landslide event is located is determined based on the spatial location information of the landslide event, and the rainfall station corresponding to the landslide event is identified. Based on the rainfall stations corresponding to the identified landslide events, a correspondence between historical landslide events and rainfall stations is established.

7. The method as described in claim 1, characterized in that, Based on the correspondence between historical landslide events and rainfall stations, and a pre-constructed comprehensive evaluation function for rainfall event classification, the particle swarm optimization algorithm is used to search for optimal rainfall event classification parameters within a preset parameter range, including: By using the time window length and rainfall threshold as a parameter combination, a sliding analysis is performed on the daily rainfall data series to determine the time range of rainfall events; Using historical rainfall data from all meteorological stations in the study area within a preset time period, the multi-year average annual number of rainy days and the multi-year average longest consecutive rainfall days in the study area were determined. Based on historical landslide data and the time range of the rainfall events, the proportion of landslide occurrence dates falling within the time range of the rainfall events is determined as the landslide event coverage rate; Based on the landslide event coverage rate, determine the landslide response factor value; based on the multi-year average annual rainfall days, determine the rainfall days factor value; based on the multi-year average longest consecutive rainfall days, determine the event duration factor value. Based on the values ​​of landslide response factors, rainfall days, event duration factors, and the comprehensive evaluation function for rainfall event classification pre-constructed based on landslide response, event rhythm, and climate constraints, the comprehensive evaluation factor values ​​for rainfall event classification corresponding to each set of parameters are determined. Each set of parameters is regarded as a particle position. Through iterative updates of particle velocity and position, the parameter combination that maximizes the value of the comprehensive evaluation factor for rainfall event classification is searched within the preset parameter range and is taken as the optimal event classification parameter.

8. The method as described in claim 7, characterized in that, The comprehensive evaluation function for classifying rainfall events is: ； in, , , These represent the landslide response factors, rainfall days, and event duration factors, respectively. ； ； ； In the formula: F(DT,ET) Comprehensive evaluation factors are used to classify rainfall events; R is the landslide event coverage rate; J is the average number of events per year after classification; Q is the average event duration after classification; T is the multi-year average number of rainy days per year; CWD is the multi-year average number of consecutive rainy days.

9. A device for constructing event-level rainfall samples for rainfall-induced landslides, characterized in that, include: The acquisition unit is used to acquire daily rainfall data and corresponding historical landslide data within the study area; The matching unit is used to construct Thiessen polygons based on rainfall stations, and to perform spatial matching on landslide historical data based on the Thiessen polygons to determine the rainfall station corresponding to each landslide historical record, thus forming a correspondence between historical landslide events and rainfall stations. The optimal rainfall event classification parameter determination unit is used to search for the optimal rainfall event classification parameters within a preset parameter range based on the correspondence between historical landslide events and rainfall stations, as well as a pre-constructed comprehensive evaluation function for rainfall event classification, using the particle swarm optimization algorithm. The division and decomposition unit is used to divide the daily rainfall data into rainfall events based on the optimal rainfall event division parameters, identify continuous rainfall processes as independent rainfall events, and decompose the independent rainfall events to obtain several decomposed sub-events with different cumulative characteristics. The construction unit is used to label independent rainfall events and their decomposed sub-events based on historical landslide data, and to construct an event-level rainfall sample set containing both disaster-causing and non-disaster-causing events; The parameter configuration unit is used to construct multiple sets of different combinations of rainfall variables and their corresponding attenuation coefficient values ​​to characterize the features of each rainfall event, based on each independent rainfall event and its decomposed sub-events. The optimization unit is used to quantitatively evaluate different combinations of rainfall variables and their corresponding attenuation coefficients based on the event-level rainfall sample set using sensitivity analysis, and to determine the optimal combination of rainfall variables and its corresponding attenuation coefficient value that can effectively distinguish between disaster-causing events and non-disaster-causing events. This includes: under given combinations of rainfall variables and attenuation coefficient values, introducing a sensitivity index to quantitatively evaluate the ability to distinguish between disaster-causing events and non-disaster-causing events, obtaining the distinguishability between disaster-causing and non-disaster-causing events and the corresponding sensitivity coefficient value; introducing a particle swarm optimization algorithm to jointly search for combinations of rainfall variables and attenuation coefficient values ​​within a preset parameter space, with the maximization of the sensitivity coefficient value as the optimization objective, to obtain the combination of rainfall variables and attenuation coefficient value that maximizes the distinguishability between disaster-causing and non-disaster-causing events as the optimal combination of rainfall variables and its corresponding attenuation coefficient value; The feature value construction unit is used to construct unified features for an event-level rainfall sample set based on the optimal combination of rainfall variables and their corresponding attenuation coefficient values, obtaining feature values ​​to characterize the features of each rainfall event. This includes: using the optimal combination of rainfall variables obtained through particle swarm optimization and their corresponding attenuation coefficient values ​​as a preferred parameter configuration scheme to calculate the features of each rainfall event; using all rainfall event samples, performing regression fitting on the optimal combination of rainfall variables in a double logarithmic coordinate system to construct a benchmark threshold function to describe the overall distribution characteristics of disaster-causing and non-disaster-causing rainfall events, the benchmark threshold function being used to characterize the relationship between rainfall variables; and based on the benchmark threshold function, calculating the degree of deviation of each rainfall event sample from the benchmark threshold function in the rainfall variable feature space, and using this degree of deviation as a unified feature value characterizing the features of each rainfall event. The trigger probability determination unit is used to determine the landslide trigger probability corresponding to each rainfall event based on the feature values ​​of each rainfall event characteristics, under probability constraints.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1 to 8.

12. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method of any one of claims 1 to 8.

Images