Mud-rock flow prediction method based on substance start-propagation process

Abstract

The invention relates to the technical field of debris flow prediction, in particular to a debris flow prediction method based on a substance start-propagation process, which comprises the following steps: collecting a debris flow event and a rainfall event; acquiring geographical background conditions and rainfall excitation conditions according to the debris flow event and the rainfall event; constructing an index system for predicting debris flow according to geographical background conditions and rainfall excitation conditions; the indexes in the index system are input into a debris flow prediction model, the time-space probability value of debris flow occurrence is output, and the debris flow prediction model is constructed based on a whale optimization algorithm and a machine learning model. Triggering conditions and geographical background features are integrated through a hybrid machine learning model, and the time-space possibility of debris flow occurrence based on the hourly watershed is evaluated from the material triggering-propagation angle of debris flow. The method is implemented in a typical terrain sudden change zone, provides technical guidance for early warning and prediction, and serves prevention and disaster reduction of debris flow in a mountainous area.

Description

Technical Field

[0001] The present invention relates to the technical field of debris flow prediction, and particularly to a debris flow prediction method based on the initiation - propagation process of materials. Background Art

[0002] Many studies have been conducted on predicting the spatial likelihood of debris flow occurrence, with a focus on debris flow susceptibility. These methods include heuristic methods that highly rely on expert experience, statistical methods such as frequency ratio, principal component analysis, and logistic regression, and physical methods that require high data processing and parameter calibration, such as the triggering - propagation modeling methods supported by the TRIGRS model and the FLO - 2D model. Although these methods can identify the distribution of disaster - prone locations, they are still insufficient in disaster prevention and mitigation decision - making. The temporal probability of future debris flow occurrence still needs to be foreseen.

[0003] Many researchers and institutions have been working on using data - driven or statistical models to predict the spatio - temporal likelihood of debris flow occurrence. However, this method still has significant limitations due to its reliance on historical disaster records. That is, these methods are not suitable for areas with less historical disaster information collection because they predict the likelihood of debris flow by analyzing the relationship between historical disaster occurrences and rainfall characteristics. Generally speaking, it is challenging to promote and update the I - D threshold.

[0004] Watersheds with continuous geographical distributions tend to experience cluster debris flows, which occur simultaneously during the same rainfall event. Similar rainfall thresholds in cluster watersheds are thought to be due to similar geographical background conditions that cause disasters, such as topographic conditions, tectonic conditions, hydrological conditions, and source material conditions. Scholars have recognized from these observations that the rainfall thresholds triggering debris flows need to be updated to account for changes in the geographical background environment. Therefore, researchers have pointed out that the key factors driving debris flow formation can be divided into two categories, including the geographical background conditions that control the susceptibility of a location to debris flows, and the triggering conditions that control when debris flows occur, such as extreme rainfall and increased local soil moisture conditions. Subsequently, some studies have enhanced the rainfall threshold method by considering geographical background variables. As a result, the occurrence of debris flows can be predicted in places where disaster data records are scarce. For example, a global prediction model called LHASA version 2 is described, which uses a global landslide susceptibility map to represent the geographical background conditions and adopts a data-driven approach. Although potential hazard area grid maps within the range of 60° north and south can be generated every 3 hours by combining satellite precipitation products with the global susceptibility map, due to its coarse kilometer-scale spatial resolution and 3-hour time resolution, the ground application of this model is still limited. Bordoni, M., V. Vivaldi, L. Lucchelli, L. Ciabatta, L. Brocca, J. P. Galve, and C. Meisina. 2021. Development of a data-driven model for spatial and temporal shallow landslide probability of occurrence at catchment scale. Landslides 18(4): 1209-1229. developed a data-driven method to predict the spatio-temporal probability of rainfall-induced shallow landslides, supporting the sensitivity results calculated by geological, geomorphic, and hydrological predictors, and its time triggering conditions involve short-term cumulative rainfall, antecedent rainfall, and soil hydrological conditions. Although the above methods enable scholars to predict the spatio-temporal probability of landslides or debris flows in areas lacking disaster history records, these methods are based on grid assessment objects and are only applicable to rough regional predictions, and will be limited when used to predict debris flows developing within a watershed. This is due to the lack of consideration of the physical formation mechanism by which unstable materials supply debris flows and are transported to the gully mouth to cause damage.

[0005] Therefore, the present invention proposes a debris flow prediction method based on the material initiation - propagation process. Summary of the Invention

[0006] The object of the present invention is to provide a debris flow prediction method based on the material initiation - propagation process, which integrates triggering conditions and geographical background features through a hybrid machine learning model, and evaluates the spatio - temporal possibility of debris flow occurrence based on the hourly watershed from the perspective of debris flow material initiation - propagation. This method is implemented in a typical terrain transition zone, provides technical guidance for early warning and prediction, and serves for the prevention and mitigation of debris flows in mountainous areas.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A debris flow prediction method based on the material initiation - propagation process, comprising:

[0009] Collect debris flow events and rainfall events;

[0010] According to the debris flow events and the rainfall events, obtain geographical background conditions and rainfall triggering conditions;

[0011] Construct an index system for predicting debris flow according to the geographical background conditions and the rainfall triggering conditions;

[0012] Input the indexes in the index system into a debris flow prediction model, and output the spatio - temporal probability values of debris flow occurrence, wherein the debris flow prediction model is constructed based on the whale optimization algorithm and a machine learning model.

[0013] Optionally, obtaining the geographical background conditions includes: selecting key geographical background indexes according to the conditions for debris flow formation to obtain the geographical background conditions, wherein the key geographical background indexes include geological and tectonic indexes, topographic and geomorphic indexes, hydro - meteorological indexes, ecological and human indexes.

[0014] Optionally, obtaining the rainfall triggering conditions includes:

[0015] According to the debris flow events and the rainfall events, obtain a rainfall sequence;

[0016] Extract features from the rainfall sequence to obtain rainfall features;

[0017] Identify the unstable material distribution and the safety factor characterizing material stability through the TRIGRS model, and calculate the material initiation - propagation index;

[0018] According to the rainfall features and the material initiation - propagation index, obtain the rainfall triggering conditions.

[0019] Optionally, extracting features from the rainfall sequence to obtain rainfall features includes:

[0020] According to the traditional I-D threshold curve theory, extract traditional rainfall statistical features from the rainfall sequence, where the traditional rainfall statistical features include average rainfall intensity and rainfall event duration;

[0021] Extract features from the rainfall sequence through the Tsfresh tool to obtain rainfall driving features, where the rainfall driving features include absolute energy, sum value, approximate entropy, continuous wavelet transform coefficients, autocorrelation, upper quantile, and lower quantile;

[0022] Combine the traditional rainfall statistical features and the rainfall driving features to obtain the rainfall features.

[0023] Optionally, obtaining the average rainfall intensity includes: calculating the average rainfall intensity by dividing the cumulative rainfall by the rainfall event duration.

[0024] Optionally, calculating the material trigger-propagation index includes:

[0025] ;

[0026] ;

[0027] where MTP g represents the material trigger-propagation index at the g-th stage, A ust,g represents the unstable area within the basin at the g-th stage, V ust,g represents the volume of unstable material within the basin determined by the TRIGRS model, IC std,g represents the standardized connectivity index at the g-th stage, A wetershed represents the total area of the basin, A unit represents the area of the calculation unit, D sld,g represents the failure surface depth of the calculation unit with the minimum safety factor output by the TRIGRS model.

[0028] Optionally, the index system for predicting debris flow includes: MTP index, I-D threshold index, Tsfresh index, ENV index, where the MTP index is the material trigger-propagation index, the I-D threshold index includes average rainfall intensity and rainfall event duration, the Tsfresh index is the rainfall driving features extracted from the rainfall sequence through the Tsfresh tool, and the ENV index includes the key geographical background indicators and the effective rainfall before the debris flow outbreak.

[0029] Optionally, constructing the debris flow prediction model based on the whale optimization algorithm and the machine learning model includes: training the optimal hyperparameters of the machine learning model through the whale optimization algorithm to construct the debris flow prediction model, where the machine learning model is support vector classification, random forest, and extreme gradient boosting.

[0030] The beneficial effects of the present invention are as follows: First, from the perspective of the debris flow formation mechanism, the present invention constructs a debris flow dynamic hazard assessment index system, which consists of two parts. One part is the geographical background conditions, which are used to characterize the susceptibility of debris flow occurrence, that is, the spatial possibility of debris flow in a certain area. The other part is the triggering conditions, which are used to characterize the triggering characteristics related to rainfall. Subsequently, by integrating three advanced machine learning models and a hyperparameter optimization algorithm, a hybrid machine learning model is established, optimizing the model convergence speed and generalization ability to simulate the impact of the susceptibility differences controlled by geographical background conditions on the change of rainfall thresholds. In this way, the complex nonlinear relationship between debris flow occurrence and disaster-causing conditions can be fitted by the hybrid machine learning model, so as to predict the spatio-temporal possibility of debris flow occurrence, providing technical support for debris flow forecasting and early warning in complex mountainous areas. Brief Description of the Drawings

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0032] Figure 1 It is a schematic diagram of the infinite slope model of the embodiment of the present invention;

[0033] Figure 2 shows the comparison results of the ROC curves of different hybrid machine learning models under different parameter design schemes in the embodiment of the present invention, where (a) is the ROC verification curve of the WOA-SVC model under different parameter schemes, (b) is the ROC verification curve of the WOA-RF model under different parameter schemes, and (c) is the ROC verification curve of the WOA-XGB model under different parameter schemes;

[0034] Figure 3 shows the spatio-temporal possibility prediction of a debris flow in the target area in the embodiment of the present invention. Among them, (a) is the probability level, and (b) is the hourly rainfall conditions and material triggering-propagation conditions from the start of the rainfall event to the debris flow outbreak;

[0035] Figure 4 shows the spatio-temporal probability prediction of a debris flow in the second target area in the embodiment of the present invention. Among them, (a) is the probability level, and (b) is the hourly rainfall conditions and material triggering-propagation conditions from the start of the rainfall event to the debris flow outbreak;

[0036] Figure 5 This is a flowchart of a debris flow prediction method based on the initiation - propagation process of substances according to an embodiment of the present invention. Detailed implementation manners

[0037] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0038] To make the above - mentioned objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the drawings and specific implementation manners.

[0039] As Figure 5 shown, this embodiment provides a debris flow prediction method based on the initiation - propagation process of substances, including:

[0040] Collect debris flow events and rainfall events;

[0041] According to the debris flow events and rainfall events, obtain geographical background conditions and rainfall triggering conditions;

[0042] Construct an index system for predicting debris flow according to the geographical background conditions and rainfall triggering conditions;

[0043] Input the indexes in the index system into a debris flow prediction model, and output the spatio - temporal probability values of debris flow occurrence, where the debris flow prediction model is constructed based on the whale optimization algorithm and a machine learning model.

[0044] Specifically, the collection of debris flow events includes: The main purpose of collecting debris flow records is to train a hybrid machine learning model and evaluate its prediction accuracy and practical applicability. The collected debris flow events include comprehensive records of rainfall events, covering rainfall sequences, occurrence locations, and times. This extensive data set helps to explore the factors triggering debris flows and deeply study the physical mechanisms of their formation. In addition, supplementary information on debris flow events is retrieved from the literature database. Using 15 - year literature records collected by the Institute of Plateau Meteorology, Chengdu, China, the corresponding hourly rainfall sequences triggering debris flows are sorted out.

[0045] Rainfall event collection includes: According to the definition of rainfall events that trigger debris flows, when the hourly rainfall intensity reaches or exceeds 4 mm / h, the rainfall event is considered to start; when the rainfall intensity is continuously below 4 mm / h for 6 hours, the rainfall event is considered to end. Therefore, rainfall sequences closely related to the formation of debris flows are identified to capture the temporal variations of rainfall statistical characteristics. Subsequently, a rolling method is used to divide the rainfall sequence into subsequences at one-hour intervals. Finally, 209 rainfall subsequences are generated, and the samples are balanced by the SMOTE tool, and 360 samples are determined.

[0046] Furthermore, obtaining the geographical background conditions includes: According to the conditions for the formation of debris flows, key geographical background indicators are selected to obtain the geographical background conditions. Among them, the key geographical background indicators include geological and tectonic indicators, topographic and geomorphic indicators, hydro-meteorological indicators, ecological and anthropogenic indicators.

[0047] Specifically, "debris flow susceptibility" refers to the relative likelihood of debris flows occurring or being triggered within a specific area. This is a comprehensive concept that can be applied to various practical scenarios, such as risk assessment and disaster prevention planning. To describe and evaluate debris flow susceptibility and characterize the critical state when debris flows occur in a basin, based on the three main conditions for the formation of debris flows (steep terrain, abundant source materials, and sufficient water sources), four categories of key indicators are selected. These indicators include: (i) geological and tectonic indicators, such as lithology (LIT), soil erosion index (SE), soil depth (SD), distance from fault (DFF), distance from the epicenter of historical earthquakes (DFE) (magnitude greater than 4 since 2000), and peak ground acceleration (PSA); (ii) topographic and geomorphic indicators, such as channel gradient (CG) and relief (REL); (iii) hydro-meteorological indicators, such as aridity index (AI), annual average precipitation (AP), and annual average temperature (MAT); (iv) ecological and anthropogenic indicators, such as road density (RD), population density (PD), land use type (LD), and normalized difference vegetation index (NDVI).

[0048] Furthermore, obtaining the rainfall triggering conditions includes:

[0049] According to the debris flow events and rainfall events, obtain the rainfall sequence;

[0050] Extract the characteristics of the rainfall sequence to obtain the rainfall characteristics;

[0051] Identify the distribution of unstable materials and the safety factor characterizing material stability through the TRIGRS model, and calculate the material trigger-propagation index;

[0052] According to the rainfall characteristics and the material trigger-propagation index, obtain the rainfall triggering conditions.

[0053] Specifically, debris flow is a soil-water mixture formed by rainfall runoff. Therefore, in this embodiment, the triggering conditions of debris flow are characterized by combining the dynamic rainfall statistical characteristics and the rainfall-driven material supply characteristics. The rainfall-driven triggering characteristics are thus divided into two categories: (i) statistical characteristics extracted from the rainfall sequence, and (ii) rainfall-driven material triggering-propagation characteristics.

[0054] Furthermore, feature extraction is performed on the rainfall sequence to obtain rainfall characteristics including:

[0055] According to the traditional I-D threshold curve theory, traditional rainfall statistical characteristics are extracted from the rainfall sequence, where the traditional rainfall statistical characteristics include average rainfall intensity and rainfall event duration;

[0056] Feature extraction is performed on the rainfall sequence through the Tsfresh tool to obtain rainfall-driven characteristics, where the rainfall-driven characteristics include absolute energy, sum value, approximate entropy, continuous wavelet transform coefficient, autocorrelation, upper quantile, and lower quantile;

[0057] Combining the traditional rainfall statistical characteristics and the rainfall-driven characteristics, rainfall characteristics are obtained.

[0058] Specifically, obtaining rainfall characteristics includes:

[0059] (1) Extraction of rainfall sequence statistical characteristics:

[0060] Rainfall statistical characteristics are quantitative statistical information extracted from rainfall events. These characteristics include the spatio-temporal distribution of rainfall events, as well as rainfall intensity, rainfall amount, duration, and frequency. Analyzing these characteristics helps to determine the scale, intensity, and duration of rainfall events that trigger debris flow.

[0061] Traditional rainfall statistical characteristics:

[0062] According to the traditional I-D threshold curve theory, traditional rainfall statistical characteristics are extracted from the rainfall sequence, including average rainfall intensity (ARI) and rainfall event duration (RED). The average rainfall intensity (ARI) of each rainfall subsequence is calculated by dividing the cumulative rainfall amount by the rainfall event duration (RED).

[0063] In addition, the effective rainfall amount before the occurrence of debris flow can be quantified by a formula.

[0064] (1);

[0065] where ERA k represents the effective rainfall amount within k days before the occurrence of debris flow; K represents a reduction coefficient with a value of 0.84; k represents the number of days, and R i represents the rainfall amount on the i-th day before.

[0066] More statistical features extracted by the Tsfresh tool:

[0067] Extracting additional rainfall statistical features through the Tsfresh tool is crucial for constructing a comprehensive debris flow prediction model and can complement traditional rainfall statistical features. Relying solely on traditional metrics may not fully capture the inherent complexity and variability in rainfall events. The Tsfresh tool can automatically extract features from time series data to capture more comprehensive and rich rainfall statistical features, thereby enhancing the model's representational ability. Additionally, the process of debris flow occurrence is complex, and there is a potential non-linear relationship between rainfall and debris flow. By introducing more rainfall statistical features extracted by the Tsfresh tool, the complex correlation between rainfall and debris flow can be better captured, thus improving the prediction model's ability to identify potential non-linear relationships. Moreover, the Tsfresh tool can extract various rainfall statistical features, such as temporal features, spectral features, and information entropy features. This enables the prediction model to comprehensively analyze rainfall from multiple perspectives and dimensions, reflecting key features such as the spatio-temporal distribution and intensity changes of rainfall events, thereby enhancing the model's robustness and adaptability.

[0068] Several features that significantly contribute to debris flow formation were extracted from rainfall time series using the Python package Tsfresh. A total of 143 features were extracted from the rainfall time series, and machine learning techniques were used to analyze the relationship between these features and debris flow occurrence within the watershed. Ultimately, 7 features that significantly contribute to debris flow formation were identified. These 7 rainfall-driven features include absolute energy (AE), sum value (SV), approximate entropy (AEMR), continuous wavelet transform coefficient (CWT), autocorrelation coefficient (ATC), upper quantile (CQ1), and lower quantile (CQ2). A detailed description of these metrics. The statistical feature descriptions based on the Tsfresh tool are shown in Table 1, where the variable x represents the time series of the rainfall sequence.

[0069] Table 1

[0070]

[0071] (2) Calculation of rainfall-driven mass propagation characteristics:

[0072] The dynamic characterization of the material supply characteristics during rainfall is crucial for debris flow prediction, as it is an integral part of the debris flow composition. In the research method, by inputting sediment characteristics, spatial sediment distribution (i.e., soil depth), rainfall data, and topographic conditions, the TRIGRS model is used to obtain the distribution of unstable materials within the basin, and this is regarded as the main material source of debris flow. Subsequently, the propagation model (FLO-2D) is used to simulate the triggering-propagation process of debris flow. However, due to the limitations of the detailed numerical simulation of debris flow propagation in terms of computational resources, time cost, and application restrictions in its research method, this method has certain deficiencies in multi-basin early warning tasks.

[0073] Geomorphic connectivity refers to the probability of material transport and convergence, which can be quantified by the geomorphic connectivity index (IC). This index describes the likelihood of an object within the basin being transported from point A to point B, providing a simplified method for simulating the triggering-propagation of debris flow. In this embodiment, the TRIGRS model is used to quantify the distribution of unstable materials within the basin, and then the connectivity index is used to represent the degree of connection between the location of unstable materials and the convergence point of the debris flow basin, describing the ability of unstable materials to supply debris flow. The specific implementation process is as follows:

[0074] i. Use the TRIGRS model to identify the distribution of unstable materials:

[0075] To evaluate the distribution of unstable materials within a given basin during a specific rainfall event, the Transient Rainfall Infiltration and Grid-based Regional Slope-stability (TRIGRS) model is used through software compiled in Fortran language. By combining the output of TRIGRS with geomorphic connectivity, the transport potential of the identified unstable materials is quantified. Subsequently, this information is incorporated as a hazard factor into a hybrid machine learning model to predict the likelihood of debris flow occurrence. Figure 1 The schematic diagram of the model parameter division settings is shown.

[0076] ψ(Z,t) represents the pressure head at vertical depth Z and time t. ψ(Z,t) is calculated by formula (2).

[0077] (2);

[0078] where d is the initial water level depth, the square of the cosine of the slope β = cos²δ, δ is the topographic slope, I nZ is the surface flux of a given intensity within the nth time interval, H(·) is the Heaviside step function, t n is the time of the nth time interval in the rainfall sequence, D1 = D0 / cos²δ, D0 is the saturated hydraulic diffusivity (equal to 200 times K s ), d LZ is the thickness of the permeable soil layer. , erfc(·) is the complementary error function. K s is the saturated hydraulic conductivity, N is the total number of time intervals, m is the number of terms of the infinite series. Since the infinite series converges rapidly in this formula, only the first few terms of the series are needed in the calculation process.

[0079] This model is based on the following assumptions: The solution given by Equation (2) is only applicable to the case of initial tension saturation. Therefore, it can be inferred that the linearized form of the Richards equation holds in this case, and the hydraulic conductivity can be approximated by its saturated value. It should be particularly noted that within the framework of this embodiment, rainfall will not generate runoff exceeding the infiltration capacity. After determining the pore water pressure head through Equation (2), the safety factor (F s ) characterizing the stability of the material can be calculated based on the infinite slope model, such as Equation (3).

[0080] (3);

[0081] Among them, F s (Z,t) represents the safety factor. c′ and represent the soil cohesion and the soil internal friction angle respectively; γ w and γ s represent the unit weight of water and the unit weight of soil respectively. According to existing research, failure occurs at the base boundary Z = d LZ where the pore water pressure reaches the maximum depth. For the basic soil mechanics parameters of different soil types (such as cohesion, internal friction angle and unit weight), values are assigned according to the established mapping relationship between soil types and soil basic parameters.

[0082] ii: Calculation of the Mass Trigger-Propagation (MTP) index:

[0083] The Mass Trigger-Propagation (MTP) index is determined by calculating the average potential for unstable material to be eroded and enter the river channel through Equations (4)-(5). Among them, Equation (4) quantifies the erosion intensity of potentially unstable material, and Equation (5) evaluates its probability of migrating along the terrain towards the river channel. This index comprehensively reflects the coupling effect of topographic features, soil erosion resistance and hydrodynamical conditions on the mass transport process.

[0084] (4);

[0085] (5);

[0086] Among them, MTP g represents the Mass Trigger-Propagation index at the g-th stage, A ust,g represents the unstable area within the basin at the g-th stage, V ust,gIndicates the volume of unstable material within the watershed determined by the TRIGRS model, IC std,g Represents the normalized connectivity index at the g-th stage, A wetershed Represents the total area of the watershed, A unit Represents the area of the computational unit, D sld,g Represents the failure surface depth of the computational unit with the minimum safety factor output by the TRIGRS model.

[0087] Furthermore, the index system for debris flow prediction includes: MTP index, I-D threshold index, Tsfresh index, ENV index. Among them, the MTP index is the material trigger-propagation index, the I-D threshold index includes the average rainfall intensity and the rainfall event duration, the Tsfresh index is the rainfall-driven feature extracted from the rainfall sequence by the Tsfresh tool, and the ENV index includes the key geographical background indicators and the effective rainfall before the debris flow outbreak.

[0088] Specifically, in order to explore the optimal combination of indicators, all indicators are divided into different groups through three schemes, as shown in Table 2. The MTP index refers to the material trigger-propagation (MTP) index; the I-D threshold index includes the average rainfall intensity (ARI) and the rainfall event duration (RED); the Tsfresh index refers to the rainfall statistical features extracted by the Tsfresh tool, including the absolute energy (AE), the sum value (SV), the approximate entropy (AEMR), the continuous wavelet transform coefficient (CWT), the autocorrelation coefficient (ATC), the upper quantile (CQ1) and the lower quantile (CQ2); the ENV index refers to the geographical background environmental conditions for evaluating debris flow sensitivity, including four key indicators included in the geographical background conditions and the effective rainfall (ERA) before the debris flow outbreak. In addition, by comparing the prediction performance of the hybrid machine learning models in Plan A and Plan B, the advantage of adding the MTP index can be verified. Similarly, by comparing the prediction performance of the hybrid machine learning models in Plan B and Plan C, the advantage of using the Tsfresh index compared with the traditional I-D index can be revealed.

[0089] Table 2

[0090]

[0091] Furthermore, the debris flow prediction model constructed based on the whale optimization algorithm and the machine learning model includes: training the best hyperparameters of the machine learning model through the whale optimization algorithm to construct the debris flow prediction model, where the machine learning model is support vector classification, random forest, and extreme gradient boosting.

[0092] Specifically, machine learning (ML) is a complex data-driven method that can predict the spatio-temporal probability of debris flow occurrence by modeling the complex relationship between predictor variables and response variables. Before training a machine learning model, it is usually necessary to manually set some hyperparameters, which significantly limits the prediction accuracy and output objectivity of the model. Therefore, in this embodiment, by combining the whale optimization algorithm (WOA) with advanced machine learning algorithms, several hybrid machine learning models for predicting debris flow are established, including support vector classification (SVC), random forest (RF), and extreme gradient boosting (XGBoost). The WOA algorithm is a heuristic hyperparameter optimization algorithm that can automatically search for the optimal hyperparameters for training a machine learning model and reduce the influence of subjective factors. The above three machine learning algorithms are selected because they have performed well in previous debris flow prediction tasks. In addition, similar results obtained from multiple different machine learning models help to verify the reliability of the model results.

[0093] In summary, this embodiment proposes a debris flow prediction method based on the initiation-propagation process of materials. First, a research establishes an index system that combines rainfall triggering conditions with geographical background conditions characterizing debris flow susceptibility; based on this index system, a comprehensive parameter quantification scheme is provided for each index; then, a machine learning model with automatic hyperparameter optimization function is used to integrate these indexes to estimate the probability of debris flow occurrence.

[0094] Among the input parameters of the model, the rainfall sequence is the specific dynamic predictor of all dynamic driving features. This enables the subsequent debris flow warning task to be automatically executed because the model only needs to extract the rainfall sequence from radar rainfall products, real-time site data, and predicted precipitation to achieve dynamic debris flow warning. In addition, although different dynamic features (such as rainfall statistical features and material initiation-propagation features) have different units, they are all dimensionless processed before being input into the machine learning model to accelerate the model convergence speed and improve the robustness of the model.

[0095] The output of the model is a probability value ranging from 0 (no debris flow) to 1 (debris flow). This index is divided into five levels by setting 0.2, 0.4, 0.6, and 0.8 as breakpoints: very low, low, medium, high, and very high. These output results quantify the potential risk of debris flow occurrence through five different levels.

[0096] Before model training, there is a significant imbalance in the number of positive (debris flow occurrence) and negative (no debris flow occurrence) examples in the samples, indicating a data imbalance problem. To address this issue, oversampling operations were employed to increase the number of positive samples, generating a more balanced dataset and thereby enhancing the robustness of the model. Subsequently, the dataset was randomly divided into a training set (70%) for model training and a test set (30%) for model validation. To prevent overfitting and better control the complexity of the model, cross-validation was adopted. During the sample partitioning and cross-validation process, samples from the same watershed with similar geographical conditions were grouped, forming Q + 1 groups based on the number of debris flow events. Consequently, the region was divided into Q sub-regions, and the model was trained on Q - 1 sub-regions and validated on the Qth sub-region. This approach aims to evaluate the robustness of the model in geospatial problems.

[0097] It is worth noting that although current advanced machine learning models (such as SVC, RF, and XGBoost models) can be directly used as predictors, in this embodiment, a hybrid machine learning model was selected because they have stronger adaptability in hyperparameter optimization. Compared with manually tuning parameters during the training process, the machine learning model enhanced by the WOA hyperparameter optimization algorithm has a stronger parameter search ability.

[0098] The reliability and practicality of the debris flow occurrence prediction model were evaluated using the Receiver Operating Characteristic curve (ROC curve) and the Confusion Matrix (CM). Specifically, the performance of the model was estimated using the formula ACC = (TP + TN) / (TP + TN + FP + FN), where ACC represents the prediction accuracy rate, TP and TN represent the number of correctly predicted test points, and FP and FN represent the number of incorrectly predicted test points. The quality of the model was evaluated by calculating the Area Under the Curve (AUC) of the ROC curve. The higher the AUC value, the better the prediction accuracy of the model.

[0099] The method of this embodiment will be analyzed below according to the target area:

[0100] The target geographical area of this embodiment is located within a fault zone, with frequent tectonic activities and dense secondary faults. Lithologically, the mainly exposed strata include the Proterozoic, Sinian, Cambrian, Silurian, Devonian, Triassic, and Quaternary systems. The elevation range of the study area is extensive, ranging from 804 meters to 5933 meters, and approximately 66% of the area has a slope exceeding 30°. The study area is divided into two natural climate zones in terms of climate: the rainy zone in the south and the semi-arid river valley zone in the north. The central area of the rainy zone has an average annual rainfall of 1235 millimeters and a maximum daily rainfall of 270 millimeters.

[0101] The average annual rainfall in the semi-arid river valley region is 526 mm and 80 mm respectively. The rich research results and comprehensive disaster records in this area make it an ideal location for debris flow prediction modeling.

[0102] In this embodiment, three parameter design schemes are adopted, named Plan A, Plan B, and Plan C respectively, to train three hybrid machine learning (HML) models, namely WOA-SVC, WOA-RF, and WOA-XGBoost. The performance of the models is evaluated and quantified through the confusion matrix and ROC curve. The results show that the proposed method has high prediction accuracy, with the ACC score ranging from 0.815 to 0.917 and the AUC value ranging from 0.863 to 0.967. The comparison results of the ROC curves of different hybrid machine learning models under different parameter design schemes are shown in Figures 2(a)-2(c).

[0103] This embodiment outlines the improvement of the model from three perspectives. First, the integration of the Whale Optimization Algorithm (WOA) promotes the automation and precise identification of the hyperparameters of the machine learning model, thus significantly improving the convergence speed and prediction accuracy of the model. The model performance evaluation shows that the machine learning (ML) model optimized by WOA is superior to the independent ML model. According to the ACC score, the WOA optimization algorithm improves the performance of SVC, RF, and XGBoost by 5.64%, 3.31%, and 5.58% respectively.

[0104] Second, the hybrid machine learning (HML) model trained with the additional rainfall statistical features extracted by the Tsfresh model has better prediction performance than the model trained only based on the traditional I-D threshold features. The Tsfresh tool is used to extract a wider range of statistical features from the rainfall sequence to characterize the impact of the rainfall triggering mechanism on debris flow formation. Integrating these rainfall statistical features into the prediction model significantly improves the performance of the HML model. Specifically, under the parameter configuration of Plan C, the performance of the HML model improves compared to Plan B, with SVC increasing by 4.25%, RF increasing by 2.21%, and XGBoost increasing by 2.12% (based on the ACC score). The results show that the above-mentioned rainfall statistical features extracted by the Tsfresh tool have stronger debris flow occurrence prediction ability than the traditional I-D threshold.

[0105] The integration of indicators characterizing the material triggering - propagation (MPT) process further improves the performance of the model. After adding the MPT index to the proposed method, the performance of the hybrid machine learning (HML) model is significantly improved. Specifically, under the parameter design of Plan B, compared with Plan A, the support vector classification (SVC) of the HML model increases by 1.05%, the random forest (RF) increases by 2.26%, and the extreme gradient boosting (XGBoost) increases by 1.01% (based on the ACC score). These findings emphasize the effectiveness of the prediction method that combines rainfall statistical characteristics and the material triggering - propagation process, thus improving the debris flow prediction performance by comprehensively considering the basic components of debris flow, namely water and material. It is worth noting that the MPT index has become a key indicator in the debris flow monitoring and warning system because its derivation process takes into account the physical mechanisms of debris flow formation, including the triggering and propagation of unstable materials.

[0106] To further study the joint decision - making mechanism of rainfall statistical characteristics and material triggering - propagation characteristics on debris flow prediction results, the spatio - temporal probability prediction processes of two typical debris flow events are analyzed.

[0107] As shown in Figures 3(a) - 3(b), as the rainfall event progresses, the material triggering - propagation (MTP) index in Target Area 1 gradually increases. It initially grows rapidly, then the growth rate slows down, and stabilizes approximately 2 hours before the debris flow outbreak. The absolute energy (AE) of rainfall shows exponential growth throughout the event, with the growth rate gradually accelerating, especially showing explosive growth in the last two hours. Correspondingly, the dynamically calculated probability index initially remains at a low level (0 - 0.2), and then gradually climbs to high (0.6 - 0.8) and extremely high (0.8 - 1.0) levels approximately 2 hours before the debris flow outbreak.

[0108] As shown in Figures 4(a) - 4(b), for Target Area 2, the AE index describing material triggering and propagation grows in a step - like manner as the rainfall event progresses. It initially grows rapidly, then the growth rate slows down, and then increases rapidly again. It stabilizes approximately 2 hours before the debris flow outbreak and then increases rapidly. The MTP index shows a similar growth trend to the AE index but does not exhibit a step - like pattern. Finally, both the MTP index and the AE of rainfall increase simultaneously, causing the dynamically calculated probability index to climb from a low level (0 - 0.2) to a high level (0.6 - 0.8).

[0109] Generally speaking, the material triggering - propagation (MTP) and the absolute energy (AE) seem to predict debris flow outbreaks in a complementary way. When the material triggering - propagation conditions reach a bottleneck, a higher rainfall absolute energy may also trigger debris flow. Generally, when developing a debris flow monitoring and warning system to support mountain disaster prevention and control work, the three indicators of AE, SV, and MTP are all worthy of attention.

[0110] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A debris flow prediction method based on the initiation - propagation process of substances, characterized in that, Including: Collect debris flow events and rainfall events; According to the debris flow event and the rainfall event, obtain geographical background conditions and rainfall triggering conditions; Construct an index system for debris flow prediction based on the geographical background conditions and the rainfall triggering conditions; Input the indexes in the index system into a debris flow prediction model, and output the spatio-temporal probability values of debris flow occurrence, where the debris flow prediction model is constructed based on the whale optimization algorithm and a machine learning model.

2. The debris flow prediction method based on the substance initiation - propagation process according to claim 1, wherein Obtaining the geographical background conditions includes: according to the conditions for the formation of debris flow, selecting key geographical background indexes to obtain the geographical background conditions, where the key geographical background indexes include geological and tectonic indexes, topographical and geomorphic indexes, hydro-meteorological indexes, and ecological and anthropogenic indexes.

3. The debris flow prediction method based on the substance initiation - propagation process according to claim 2, wherein, Obtaining the rainfall triggering conditions includes: According to the debris flow event and the rainfall event, obtain a rainfall sequence; Extract features from the rainfall sequence to obtain rainfall features; Identify the distribution of unstable substances and the safety factor characterizing the stability of substances through the TRIGRS model, and calculate the substance trigger-propagation index; According to the rainfall features and the substance trigger-propagation index, obtain the rainfall triggering conditions.

4. The debris flow prediction method based on the substance initiation - propagation process according to claim 3, wherein Extracting features from the rainfall sequence to obtain rainfall features includes: According to the traditional I-D threshold curve theory, extract traditional rainfall statistical features from the rainfall sequence, where the traditional rainfall statistical features include average rainfall intensity and rainfall event duration; Extract rainfall driving features from the rainfall sequence through the Tsfresh tool, where the rainfall driving features include absolute energy, sum value, approximate entropy, continuous wavelet transform coefficient, autocorrelation, upper quantile, and lower quantile; Combine the traditional rainfall statistical features and the rainfall driving features to obtain the rainfall features.

5. The debris flow prediction method based on the substance initiation - propagation process according to claim 4, wherein Obtaining the average rainfall intensity includes: calculating the average rainfall intensity by dividing the cumulative rainfall by the rainfall event duration.

6. The debris flow prediction method based on the substance initiation - propagation process according to claim 3, wherein Calculating the substance trigger-propagation index includes: ； ； Among them, MTP g represents the material trigger - propagation index in the g - th stage, A ust,g represents the unstable area within the basin in the g - th stage, V ust,g represents the volume of unstable materials within the basin determined by the TRIGRS model, IC std,g represents the standardized connectivity index in the g - th stage, A wetershed represents the total area of the basin, A unit represents the area of the calculation unit, D sld,g represents the failure surface depth of the calculation unit with the minimum safety factor output by the TRIGRS model.

7. The debris flow prediction method based on the substance initiation - propagation process according to claim 4, characterized in that, The index system for debris flow prediction includes: MTP index, I-D threshold index, Tsfresh index, ENV index, where the MTP index is the substance trigger-propagation index, the I-D threshold index includes average rainfall intensity and rainfall event duration, the Tsfresh index is the rainfall driving features extracted from the rainfall sequence through the Tsfresh tool, and the ENV index includes the key geographical background indexes and the effective rainfall before debris flow outbreak.

8. The debris flow prediction method based on the substance initiation - propagation process according to claim 1, wherein Constructing the debris flow prediction model based on the whale optimization algorithm and a machine learning model includes: training the best hyperparameters of the machine learning model through the whale optimization algorithm to construct the debris flow prediction model, where the machine learning model is support vector classification, random forest, and extreme gradient boosting.

Images