An avalanche meteorological risk estimation method and system based on multi-source data

Abstract

The present application belongs to the technical field of avalanche meteorological risk assessment, and provides a kind of avalanche meteorological risk estimation method and system based on multi-source data, method includes: initial evaluation index data acquisition, acquisition index system setting, initial evaluation, to be collected index setting, data supplement, model prediction, secondary data supplement;The present application is through rough initial evaluation, ladder supplement index setting, model prediction and secondary supplement after prediction, on the one hand, avoid collecting a large number of indicators each time, on the one hand, the phased ladder type risk assessment can better grasp the development of risk, realizes the on-demand collection and efficient reuse of data, while balancing the evaluation efficiency and accuracy;Through dynamic index adjustment mechanism, improve the dynamic evaluation and response efficiency of avalanche meteorological risk, avoid the consequences of not being able to reduce disaster and prevent disaster due to not timely response when risk increases.

Description

Technical Field

[0001] This invention relates to the field of avalanche meteorological risk assessment technology, and in particular to a method and system for predicting avalanche meteorological risk based on multi-source data. Background Technology

[0002] Currently, avalanche risk assessment technologies are mainly divided into two categories. One type focuses on comprehensive regional avalanche susceptibility assessment, using avalanche samples and influencing factors as the core, emphasizing climate background analysis, and focusing on assessing the potential likelihood of regional avalanche occurrences. This type often combines multi-source data such as topography and meteorology, and conducts regional-scale susceptibility analysis through methods such as entropy weighting and risk zoning. The other type relies on historical data to train models to achieve real-time avalanche risk assessment. By analyzing real-time monitoring data, it quickly outputs the current avalanche risk status of the region, providing data support for short-term prevention and control. This type of method is mostly based on technologies such as deep learning, integrating meteorological time series, topographic spatial, and snow layer physical data to mine deep correlations among multi-source data to improve the accuracy of real-time assessment.

[0003] However, current technologies have many shortcomings. Publicly available technologies use fixed data collection methods, often employing a single-dimensional, full-data collection approach. This fails to select assessment indicators differentiated according to risk levels, leading to data redundancy and resource waste in low-risk areas, while key, detailed indicators are missing in high-risk areas, hindering refined prevention and control. Furthermore, existing indicator systems lack progression; higher-risk levels are not matched with more complex and comprehensive assessment indicators, making it difficult to support precise prevention and control decisions. In addition, conventional models have simplistic data processing logic, directly inputting all data into the network for training without grouping the training data according to gradient prediction. This means the model can only guarantee assessment accuracy with all indicator data input. Moreover, conventional assessment methods, using a single-assessment approach, lack a closed-loop mechanism from initial coarse judgment to gradient supplementation and then to precise detailed judgment, making it difficult to balance assessment efficiency and accuracy. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for predicting avalanche meteorological risks based on multi-source data, which solves the problems of fixed data acquisition modes, redundancy in low-risk areas, lack of detailed indicators in high-risk areas, poor adaptability, and inability to balance refined prevention and control with resource optimization in existing technologies.

[0005] To achieve the above objectives, the present invention provides the following solution: A method for predicting avalanche meteorological risk based on multi-source data includes: Collect preset indicator data for the target area to obtain initial evaluation indicator data; A preliminary assessment level and a corresponding data collection indicator for each preliminary assessment level are defined to obtain a data collection indicator system; the preliminary assessment levels include: Level 1 to Level 5; The initial assessment index data are calculated using the avalanche integrated meteorological hazard assessment formula to obtain the preliminary assessment level; When the preliminary assessment level is any one of level two to level five, the indicators corresponding to level two to the preliminary assessment level in the collection indicator system are set as the indicators to be collected; Collect data corresponding to the target region and the indicators to be collected to obtain supplementary data; The initial assessment index data and the supplementary data are input into the pre-trained avalanche meteorological risk prediction model for prediction, and the risk prediction level is obtained. When the risk prediction level is greater than the preliminary assessment level, the indicators corresponding to the preliminary assessment level to the risk prediction level in the collection indicator system are added to the indicators to be collected, and the process returns to the step "collect the data corresponding to the target area and the indicators to be collected to obtain supplementary data".

[0006] Preferably, it further includes: When the preliminary assessment level is Level 1, the initial assessment index data of the target area is collected at a fixed period, and the process returns to the step "Calculate the initial assessment index data using the avalanche integrated meteorological hazard assessment formula to obtain the preliminary assessment level"; When the risk prediction level is less than or equal to the preliminary assessment level, the indicators from level two to the risk prediction level in the data collection indicator system are set as the indicators to be collected, and the process returns to the step "collect the data corresponding to the target area and the indicators to be collected to obtain supplementary data".

[0007] Preferably, the preset index data includes: previous snow depth, altitude, slope, previous temperature, previous cumulative snowfall, daily snowfall, and daily increase in snow depth.

[0008] Preferably, the secondary indicators in the data collection indicator system include: snow density and average winter wind speed; the tertiary indicators include: longitudinal temperature gradient, lateral temperature gradient, snow stability factor, and vegetation cover; the quaternary indicators include: surface microdeformation rate, stress concentration index inside the snow layer, and frequency of historical avalanches; and the quinary indicators include: snow layer crack propagation rate, weather radar reflectivity, and intensity of potential avalanche movement paths.

[0009] Preferably, the avalanche weather risk prediction model adopts a fusion network of CNN and LSTM; the avalanche weather risk prediction model includes: an input layer, a preprocessing layer, a CNN spatial feature extraction layer, an LSTM temporal feature extraction layer, an Attention fusion layer, and a fully connected output layer; The input layer and the preprocessing layer are connected; the preprocessing layer is connected to the CNN spatial feature extraction layer and the LSTM temporal feature extraction layer, respectively; the CNN spatial feature extraction layer and the LSTM temporal feature extraction layer are connected to the Attention fusion layer, respectively; the Attention fusion layer is connected to the fully connected output layer.

[0010] Preferably, the avalanche comprehensive meteorological hazard assessment formula is as follows: ;in, ; ; The preliminary assessment level; Let j be the j-th indicator in the aforementioned indicator system; The first in the collection index system One indicator; It is one of the following: altitude and slope. It is one of the following: previous snow depth, previous temperature, previous cumulative snowfall, current snowfall, and current increase in snow depth; The environmental risk index for disaster-prone environments; This is a meteorological risk index.

[0011] Preferably, the expression for the snow stability factor is: The expression for the stress concentration index inside the snow layer is: The expression for the potential avalanche path strength is: ;in, This refers to the snow stability factor; Shear strength; Shear stress; The stress concentration index within the snow layer; This is the terrain correction factor; This represents the maximum snow layer stress in the region. The average snow layer stress in the region; The potential motion path intensity of the avalanche; The potential energy corresponding to the geopotential energy gradient; This represents the potential path area of ​​an avalanche.

[0012] Preferably, the initial assessment index data and the supplementary data are input into a pre-trained avalanche meteorological risk prediction model for prediction, to obtain a risk prediction level, including: The input layer is used to divide the initial evaluation index data and the supplementary data into spatial data and temporal data; The preprocessing layer is used to clean the spatial data and the temporal data respectively. The spatial class data output by the preprocessing layer is used to extract features using the CNN spatial feature extraction layer to obtain the first sub-feature; The LSTM temporal feature extraction layer is used to extract features from the temporal data output by the preprocessing layer to obtain the second sub-feature; The first sub-feature and the second sub-feature are simultaneously input into the Attention fusion layer for feature fusion calculation to obtain spatial temporal features; The risk prediction level is obtained by performing a level mapping calculation on the spatial temporal features using the fully connected output layer.

[0013] Preferably, the pre-training process of the avalanche meteorological risk prediction model includes: Collect data for each indicator in the aforementioned data collection system within the target research area to obtain the original dataset; The original dataset is expanded according to a preset grouping rule to obtain a full risk level dataset. The preset grouping rule is used to divide each group of full data in the original dataset into a first group to a fifth group. The data types in the first group correspond to the indicator types of Level 1 in the preliminary assessment level; the data types in the second group correspond to the indicator types of Levels 1 and 2 in the preliminary assessment level; the data types in the third group correspond to the indicator types of Levels 1 to 3 in the preliminary assessment level; the data types in the fourth group correspond to the indicator types of Levels 1 to 4 in the preliminary assessment level; and the data types in the fifth group correspond to the indicator types of Levels 1 to 5 in the preliminary assessment level. The original avalanche meteorological risk prediction model is iteratively trained using the full risk level dataset and preset label data to obtain the trained avalanche meteorological risk prediction model.

[0014] Preferably, an avalanche meteorological risk prediction system based on multi-source data includes: The data acquisition module is used to collect preset indicator data of the target area to obtain initial evaluation indicator data; The indicator system construction module is used to set the preliminary evaluation levels and the collection indicators corresponding to each preliminary evaluation level, thereby obtaining the collection indicator system; the preliminary evaluation levels include: Level 1 to Level 5; The preliminary assessment module is used to calculate the initial assessment index data using the avalanche integrated meteorological hazard assessment formula to obtain the preliminary assessment level; The supplementary indicator determination module is used to set the indicators corresponding to the preliminary assessment level from level two to level five in the collection indicator system as indicators to be collected when the preliminary assessment level is any one of level two to level five. The indicator supplementary data collection module is used to collect data corresponding to the target region and the indicator to be collected, and obtain supplementary data. The model prediction module is used to input the initial assessment index data and the supplementary data into the pre-trained avalanche meteorological risk prediction model for prediction, and obtain the risk prediction level. The secondary data collection module is used to add the indicators corresponding to the risk prediction level from the preliminary assessment level to the risk prediction level in the data collection indicator system to the indicators to be collected when the risk prediction level is greater than the preliminary assessment level, and return to the step "collect the data corresponding to the target area and the indicators to be collected to obtain supplementary data".

[0015] The present invention discloses the following technical effects: This invention provides a method and system for predicting avalanche meteorological risks based on multi-source data. Through a rough initial assessment, tiered supplementary data collection setting, model prediction, and secondary supplementary data collection after prediction, it solves the problems of fixed data collection modes, redundancy in low-risk areas, and lack of fine indicators in high-risk areas in existing technologies. It achieves on-demand data collection and efficient reuse, as well as a balance between assessment efficiency and accuracy. Through a dynamic indicator adjustment mechanism, it solves the problems of poor adaptability and inability to balance refined prevention and control with resource optimization in existing technologies, achieving the goal of refined prevention and control in high-risk areas and resource conservation in low-risk areas. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments 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.

[0017] Figure 1 This is a schematic diagram of an avalanche meteorological risk prediction process based on multi-source data, provided in an embodiment of the present invention. Figure 2 A flowchart for avalanche meteorological risk prediction provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the model architecture provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the model prediction process provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the model pre-training process provided in an embodiment of the present invention. Detailed Implementation

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

[0019] The purpose of this invention is to provide a method and system for predicting avalanche meteorological risks based on multi-source data, which solves the problems of fixed data acquisition modes, redundancy in low-risk areas, lack of detailed indicators in high-risk areas, poor adaptability, and inability to balance refined prevention and control with resource optimization in existing technologies.

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] Figure 1 This is a schematic diagram of an avalanche meteorological risk prediction process based on multi-source data, provided in an embodiment of the present invention. Figure 2 A flowchart for avalanche weather risk prediction provided in an embodiment of the present invention, such as Figure 1 and Figure 2 As shown, this invention provides a method for predicting avalanche meteorological risk based on multi-source data, including: Step 100: Collect preset indicator data for the target area to obtain initial evaluation indicator data; Step 200: Set the preliminary assessment levels and the corresponding collection indicators for each preliminary assessment level to obtain the collection indicator system; the preliminary assessment levels include: Level 1 to Level 5; Step 300: Calculate the initial assessment index data using the avalanche integrated meteorological hazard assessment formula to obtain the preliminary assessment level; Step 400: When the preliminary assessment level is any one of level two to level five, set the indicators corresponding to level two to the preliminary assessment level in the collection indicator system as the indicators to be collected; Step 500: Collect data corresponding to the target region and the indicators to be collected to obtain supplementary data; Step 600: Input the initial assessment index data and the supplementary data into the pre-trained avalanche meteorological risk prediction model for prediction, and obtain the risk prediction level; Step 700: When the risk prediction level is greater than the preliminary assessment level, add the indicators corresponding to the preliminary assessment level to the risk prediction level in the collection indicator system to the indicators to be collected, and return to the step "Collect the data corresponding to the target area and the indicators to be collected to obtain supplementary data".

[0022] Furthermore, it also includes: When the preliminary assessment level is Level 1, the initial assessment index data of the target area is collected at a fixed period, and the process returns to the step "Calculate the initial assessment index data using the avalanche integrated meteorological hazard assessment formula to obtain the preliminary assessment level"; When the risk prediction level is less than or equal to the preliminary assessment level, the indicators from level two to the risk prediction level in the data collection indicator system are set as the indicators to be collected, and the process returns to the step "collect the data corresponding to the target area and the indicators to be collected to obtain supplementary data".

[0023] Optionally, the preset index data includes: previous snow depth, altitude, slope, previous temperature, previous cumulative snowfall, daily snowfall, and daily increase in snow depth.

[0024] Preferably, the secondary indicators in the data collection indicator system include: snow density and average winter wind speed; the tertiary indicators include: longitudinal temperature gradient, lateral temperature gradient, snow stability factor, and vegetation cover; the quaternary indicators include: surface microdeformation rate, stress concentration index inside the snow layer, and frequency of historical avalanches; and the quinary indicators include: snow layer crack propagation rate, weather radar reflectivity, and intensity of potential avalanche movement paths.

[0025] refer to Figure 3 The avalanche weather risk prediction model adopts a fusion network of CNN and LSTM; the avalanche weather risk prediction model includes: input layer, preprocessing layer, CNN spatial feature extraction layer, LSTM temporal feature extraction layer, attention fusion layer, and fully connected output layer; The input layer and the preprocessing layer are connected; the preprocessing layer is connected to the CNN spatial feature extraction layer and the LSTM temporal feature extraction layer, respectively; the CNN spatial feature extraction layer and the LSTM temporal feature extraction layer are connected to the Attention fusion layer, respectively; the Attention fusion layer is connected to the fully connected output layer.

[0026] Furthermore, the formula for the comprehensive meteorological hazard assessment of avalanches is as follows: ;in, ; ; The preliminary assessment level; Let j be the j-th indicator in the aforementioned indicator system; The first in the collection index system One indicator; It is one of the following: altitude and slope. It is one of the following: previous snow depth, previous temperature, previous cumulative snowfall, current snowfall, and current increase in snow depth; The environmental risk index for disaster-prone environments; This is a meteorological risk index.

[0027] Specifically, the expression for the snow stability factor is: The expression for the stress concentration index inside the snow layer is: The expression for the potential avalanche path strength is: ;in, This refers to the snow stability factor; Shear strength; Shear stress; The stress concentration index is the internal stress concentration index of the snow layer. This is the terrain correction factor; This represents the maximum snow layer stress in the region. The average snow layer stress in the region; The potential motion path intensity of the avalanche; The potential energy corresponding to the geopotential energy gradient; This represents the potential path area of ​​an avalanche.

[0028] refer to Figure 4 The initial assessment index data and the supplementary data are input into a pre-trained avalanche meteorological risk prediction model for prediction, resulting in a risk prediction level, including: Step 601: Use the input layer to divide the initial evaluation index data and the supplementary data into spatial data and time series data; Step 602: Use the preprocessing layer to clean the spatial data and the temporal data respectively; Step 603: Use the CNN spatial feature extraction layer to extract features from the spatial class data output by the preprocessing layer to obtain the first sub-feature; Step 604: Use the LSTM temporal feature extraction layer to extract features from the temporal data output by the preprocessing layer to obtain the second sub-feature; Step 605: Simultaneously input the first sub-feature and the second sub-feature into the Attention fusion layer for feature fusion calculation to obtain spatial temporal features; Step 606: Calculate the risk prediction level by performing a level mapping on the spatial temporal features using the fully connected output layer.

[0029] refer to Figure 5 The pre-training process of the avalanche meteorological risk prediction model includes: Collect data for each indicator in the aforementioned data collection system within the target research area to obtain the original dataset; The original dataset is expanded according to a preset grouping rule to obtain a full risk level dataset. The preset grouping rule is used to divide each group of full data in the original dataset into a first group to a fifth group. The data types in the first group correspond to the indicator types of Level 1 in the preliminary assessment level; the data types in the second group correspond to the indicator types of Levels 1 and 2 in the preliminary assessment level; the data types in the third group correspond to the indicator types of Levels 1 to 3 in the preliminary assessment level; the data types in the fourth group correspond to the indicator types of Levels 1 to 4 in the preliminary assessment level; and the data types in the fifth group correspond to the indicator types of Levels 1 to 5 in the preliminary assessment level. The original avalanche meteorological risk prediction model is iteratively trained using the full risk level dataset and preset label data to obtain the trained avalanche meteorological risk prediction model.

[0030] Specifically, the risk level classification is as follows: Based on the probability and severity of avalanches, five risk levels are set: Level 1 (low risk), Level 2 (relatively low risk), Level 3 (medium risk), Level 4 (relatively high risk), and Level 5 (high risk). The higher the level, the higher the probability and severity of avalanches. The risk thresholds for each level are determined using the percentile method based on historical avalanche data. The specific threshold standards are as follows: Level 1: ≤20%; Level 2: 20%< ≤40%; Level 3: 40%< ≤60%; Level 4: 60%< ≤80%; Level 5: >80%, of which, This is the comprehensive meteorological risk index for avalanches.

[0031] Furthermore, the initial assessment criteria consisted of seven basic indicators: previous snow depth, altitude, slope, previous temperature, previous cumulative snowfall, current day's snowfall, and current day's newly accumulated snow depth. Based on these seven initial assessment indicators, the comprehensive avalanche meteorological hazard index was derived. The risk level is then determined based on a preset risk threshold; if it is determined to be Level 1, this calculation method will be continuously used for dynamic monitoring and updates. Avalanche Comprehensive Meteorological Hazard Index The calculation process is as follows: Hierarchical assignment and normalization: The seven initial indicators are assigned hierarchical values, and the normalized values ​​are recorded as follows: ( ); calculate the following two indices separately: Disaster-prone environmental risk index The corresponding terrain factors are elevation and slope, and their expressions are as follows: Meteorological Factor Risk Index The expression corresponding to the previous snow accumulation material conditions, previous temperature, previous cumulative snowfall, current day snowfall, and current day's new snow depth is: The avalanche comprehensive meteorological hazard index is obtained by multiplication. ,Right now: After normalization The value range is [0,1], and the corresponding level is determined by comparing it with a preset risk threshold. In the above expression, These are the normalized values ​​of the two terrain factors. These are the normalized values ​​of the five meteorological factors.

[0032] Specifically, this embodiment sets specific data collection indicators for each level: Level 2: Supplementary indicators include snow density. Average wind speed in winter .

[0033] Level 3: 1) Longitudinal temperature gradient : .

[0034] in, This represents the temperature difference at different depths of snow layer along the longitudinal direction. The longitudinal sensor spacing is an indicator that reflects the temperature variation in the vertical direction of the snow layer, affecting its stability.

[0035] 2) Lateral temperature gradient : .

[0036] in, This represents the temperature difference between different monitoring points in the horizontal direction. The horizontal sensor spacing is the indicator that reflects the uniformity of temperature distribution in the horizontal direction of the snow layer within the area.

[0037] 3) Snow stability factor Based on the Mohr-Coulomb criterion, the calculation formula is as follows: .

[0038] Among them, shear strength , For the cohesion of the snow layer, The internal friction angle of the snow layer and the shear stress are given. , snow density, For snow layer thickness, For slope, For gravitational acceleration, normal stress , The larger the value, the stronger the snow layer stability.

[0039] 4) Vegetation coverage : Conventional parameters, obtained through inversion of satellite remote sensing data, this indicator is used to reflect the role of regional vegetation in fixing the snow layer.

[0040] Level 4: 1) Surface micro-deformation rate : .

[0041] in, It represents the vertical or horizontal displacement of the Earth's surface per unit time. The monitoring time interval reflects the dynamic changes in the ground surface and snow layer, indirectly indicating the potential for avalanches.

[0042] 2) Stress concentration index inside the snow layer : .

[0043] in, This is a terrain correction factor, adjusted according to the terrain complexity of the target area; This represents the maximum snow layer stress in the region. The average snow layer stress in the region.

[0044] 3) Frequency of historical avalanche events .

[0045] Level 5: 1) Snow layer crack propagation rate : .

[0046] in, This represents the change in the length of snow layer cracks per unit time. The monitoring time interval is used to directly reflect the degree of damage to the snow layer structure.

[0047] 2) Weather radar reflectivity Real-time data is obtained through weather radar, reflecting the intensity of snowfall and the distribution characteristics of snow particles, and indirectly determining the rate of snow accumulation.

[0048] 3) Intensity of potential avalanche movement path : .

[0049] in, This represents the potential energy corresponding to the geopotential energy gradient. The potential path area of ​​an avalanche reflects the intensity and extent of its movement after an avalanche occurs.

[0050] Specifically, the initial risk assessment process involves: first, collecting raw data for seven initial assessment indicators. After data collection, the comprehensive avalanche meteorological hazard index is calculated. Determine the initial risk level by comparing it with the preset risk threshold. (1≤ ≤5), complete the initial rough judgment. Establish an initial judgment result repository, recording in detail the original values ​​of each indicator, the normalized values ​​after tiered assignment, and the initial risk index. and The risk level is determined by associating the data collection time with grid information of the target area, providing a clear data baseline for subsequent gradient data collection, data reuse, and level iteration, ensuring seamless integration with subsequent gradient detection processes. The initial risk assessment strictly follows conventional methods and is only used for preliminary screening of risk levels and triggering subsequent data collection processes; this method is no longer used for accurate assessment after subsequent gradient data collection. Level 1 risks will continue to be dynamically monitored and updated using this initial calculation method.

[0051] Preferably, the supplementary sampling triggering rule is: if the initial risk level... =Level 1, no supplementary sampling of indicators is triggered, the risk level is directly output, and the initial judgment process is repeated every 24 hours to dynamically update the risk level and ensure that the risk status is controllable in real time. If the initial risk level is... ≥Level 2: Immediately trigger the corresponding level's indicator supplementation process, for example, when When the level reaches level 3, indicators corresponding to levels 2 and 3 are collected, meaning all corresponding indicators from the current level to level 2 are collected. Level upgrade supplementary data collection logic: After the indicator supplementary data collection is completed, the initial indicators and the supplementary indicators are input together into the pre-trained neural network model for the first accurate determination, resulting in the accurate determined level. .

[0052] like > (Level Upgrade): Only replenish resources Level to Level-specific indicators, no need for repeated data collection and all indicators at the following levels, such as initial =Level 2, determined after supplementing Level 2 indicators. If the level is 4, then only the specific indicators corresponding to level 3 and level 4 will be collected to reduce the collection of invalid data.

[0053] like = (Level unchanged): No additional indicators need to be collected; this level will be directly output as the final risk level.

[0054] like < (Level Downgrade): Retain all collected data, and subsequently... Monitoring indicators at different levels saves monitoring resources.

[0055] Specifically, this embodiment adopts a CNN-LSTM-Attention-fully connected hybrid neural network architecture, with the specific architecture hierarchy set as follows: The input layer receives structured initial indicators and gradient sampling indicators. Based on the data characteristics, it is divided into two categories: spatial data and temporal data, which are then input into the corresponding feature extraction layers.

[0056] The preprocessing layer comprises three sub-modules: data standardization, outlier removal, and dimension alignment, which are used to standardize the input data.

[0057] The CNN spatial feature extraction layer is mainly used to extract spatial correlation features such as terrain and snow distribution, and to capture the impact of regional spatial differences on avalanche risk.

[0058] The LSTM temporal feature extraction layer is used to capture temporal dependent features such as weather changes, snow accumulation, and crack propagation, reflecting the dynamic changes of risk factors.

[0059] The Attention fusion layer dynamically assigns weights to the spatial features output by the CNN and the temporal features output by the LSTM through a cross-attention mechanism, thereby strengthening the expression of key factor features that are strongly correlated with avalanche risk and suppressing interference from irrelevant features.

[0060] The fully connected output layer uses Softmax as the activation function, and finally outputs a probability distribution of 5 risk levels. The level corresponding to the dimension with the highest probability is the final risk level.

[0061] Preferably, this embodiment designs the following model training steps for the aforementioned model functions: Given the unique and non-overlapping characteristics of indicators specific to each risk level, and to avoid insufficient prediction accuracy in low-level indicator combinations (i.e., indicators corresponding to level 4 lacking level 5 specific indicators, indicators corresponding to level 3 lacking level 4 and 5 specific indicators, etc.) due to simple training with all indicators, this embodiment adopts a level-based group training mode. It strictly adheres to the principle that each training combination includes level 1 data and only uses indicators corresponding to the target level and below, ensuring the prediction accuracy of each level combination. The specific training logic is as follows: 1) Collect all indicator data corresponding to all risk levels (level 1 to level 5) in the target area and construct a training dataset.

[0062] 2) Grouping Rules: The training dataset is divided into 4 groups based on 4 indicator combinations. Each group contains the initial 7 indicators of level 1 by default, and does not contain indicator data of a higher level than the target combination, to avoid interference from high-level indicators on the prediction of low-level combinations. The specific grouping is as follows: Group 1: Contains all indicator data of levels 1 and 2, and removes the exclusive indicator data corresponding to levels 3, 4, and 5; Group 2: Contains all indicator data of levels 1 to 3, and removes the exclusive indicator data corresponding to levels 4 and 5; Group 3: Contains all indicator data of levels 1 to 4, and removes the exclusive indicator data corresponding to level 5; Group 4: Contains all indicator data of levels 1 to 5, with no indicators removed.

[0063] 3) Initialize the parameters of each layer of the model and train group by group: Employ a combination of independent group training and parameter-sharing optimization, training the model sequentially on four groups of data. Training constraints: During each training session, strictly shield indicator data at levels higher than the target combination level to prevent high-level indicators from interfering with the prediction of low-level combinations.

[0064] 4) During inference, the model parameters of the corresponding group are automatically called according to the current combination of indicators determined by the gradient detection mechanism to ensure that the model input is completely consistent with the combination of indicators during training.

[0065] As an optional implementation, this embodiment also provides an avalanche meteorological risk prediction system based on multi-source data, including: The data acquisition module is used to collect preset indicator data of the target area to obtain initial evaluation indicator data; The indicator system construction module is used to set the preliminary evaluation levels and the collection indicators corresponding to each preliminary evaluation level, thereby obtaining the collection indicator system; the preliminary evaluation levels include: Level 1 to Level 5; The preliminary assessment module is used to calculate the initial assessment index data using the avalanche integrated meteorological hazard assessment formula to obtain the preliminary assessment level; The supplementary indicator determination module is used to set the indicators corresponding to the preliminary assessment level from level two to level five in the collection indicator system as indicators to be collected when the preliminary assessment level is any one of level two to level five. The indicator supplementary data collection module is used to collect data corresponding to the target region and the indicator to be collected, and obtain supplementary data. The model prediction module is used to input the initial assessment index data and the supplementary data into the pre-trained avalanche meteorological risk prediction model for prediction, and obtain the risk prediction level. The secondary data collection module is used to add the indicators corresponding to the risk prediction level from the preliminary assessment level to the risk prediction level in the data collection indicator system to the indicators to be collected when the risk prediction level is greater than the preliminary assessment level, and return to the step "collect the data corresponding to the target area and the indicators to be collected to obtain supplementary data".

[0066] The beneficial effects of this invention are as follows: This invention, through a rough initial assessment, tiered supplementary data collection, model prediction, and subsequent secondary data collection, avoids the need to collect a large number of indicators each time. Furthermore, the phased, tiered risk assessment better grasps the development of risks, enabling on-demand data collection and efficient reuse, while balancing assessment efficiency and accuracy. The dynamic indicator adjustment mechanism improves the dynamic assessment and response efficiency of avalanche meteorological risks, preventing untimely responses that could hinder disaster mitigation and prevention when risks increase.

[0067] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0068] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for predicting avalanche meteorological risk based on multi-source data, characterized in that, include: Collect preset indicator data for the target area to obtain initial evaluation indicator data; The preset index data include: previous snow depth, altitude, slope, previous temperature, previous cumulative snowfall, daily snowfall, and daily increase in snow depth; A preliminary assessment level and a corresponding data collection indicator for each preliminary assessment level are defined to obtain a data collection indicator system; the preliminary assessment levels include: Level 1 to Level 5; The preliminary assessment level is obtained by calculating the initial assessment index data using the comprehensive avalanche meteorological hazard assessment formula; the comprehensive avalanche meteorological hazard assessment formula is as follows: ;in, ; ; The preliminary assessment level; It is one of the altitude and slope values ​​after the hierarchical assignment and normalization process. The value is one of the following after the graded assignment and normalization process: previous snow depth, previous temperature, previous cumulative snowfall, daily snowfall, and daily new snow depth. The environmental risk index for disaster-prone environments; This is a meteorological risk index. When the preliminary assessment level is any one of level two to level five, the indicators corresponding to level two to the preliminary assessment level in the collection indicator system are set as the indicators to be collected; Collect data corresponding to the target region and the indicators to be collected to obtain supplementary data; The initial assessment index data and the supplementary data are input into the pre-trained avalanche meteorological risk prediction model for prediction, and the risk prediction level is obtained. When the risk prediction level is greater than the preliminary assessment level, the indicators corresponding to the preliminary assessment level to the risk prediction level in the collection indicator system are added to the indicators to be collected, and the process returns to the step "collect the data corresponding to the target area and the indicators to be collected to obtain supplementary data".

2. The avalanche meteorological risk prediction method based on multi-source data according to claim 1, characterized in that, Also includes: When the preliminary assessment level is Level 1, the initial assessment index data of the target area is collected at a fixed period, and the process returns to the step "Calculate the initial assessment index data using the avalanche integrated meteorological hazard assessment formula to obtain the preliminary assessment level"; When the risk prediction level is less than or equal to the preliminary assessment level, the indicators from level two to the risk prediction level in the data collection indicator system are set as the indicators to be collected, and the process returns to the step "collect the data corresponding to the target area and the indicators to be collected to obtain supplementary data".

3. The avalanche meteorological risk prediction method based on multi-source data according to claim 1, characterized in that, The secondary indicators in the data collection index system include: snow density and average winter wind speed; the tertiary indicators include: longitudinal temperature gradient, lateral temperature gradient, snow stability factor, and vegetation cover; the quaternary indicators include: surface microdeformation rate, stress concentration index inside the snow layer, and frequency of historical avalanche events; and the quinary indicators include: snow layer crack propagation rate, weather radar reflectivity, and intensity of potential avalanche movement paths.

4. The avalanche meteorological risk prediction method based on multi-source data according to claim 1, characterized in that, The avalanche meteorological risk prediction model uses a fusion network of CNN and LSTM; The avalanche meteorological risk prediction model includes: an input layer, a preprocessing layer, a CNN spatial feature extraction layer, an LSTM temporal feature extraction layer, an Attention fusion layer, and a fully connected output layer. The input layer and the preprocessing layer are connected; the preprocessing layer is connected to the CNN spatial feature extraction layer and the LSTM temporal feature extraction layer, respectively; the CNN spatial feature extraction layer and the LSTM temporal feature extraction layer are connected to the Attention fusion layer, respectively; the Attention fusion layer is connected to the fully connected output layer.

5. The avalanche meteorological risk prediction method based on multi-source data according to claim 3, characterized in that, The expression for the snow stability factor is: The expression for the stress concentration index inside the snow layer is: The expression for the potential avalanche path strength is: ;in, This refers to the snow stability factor; Shear strength; Shear stress; The stress concentration index within the snow layer; This is the terrain correction factor; This represents the maximum snow layer stress in the region. The average snow layer stress in the region; The potential motion path intensity of the avalanche; The potential energy corresponding to the geopotential energy gradient; This represents the potential path area of ​​an avalanche.

6. The avalanche meteorological risk prediction method based on multi-source data according to claim 4, characterized in that, The initial assessment index data and the supplementary data are input into a pre-trained avalanche meteorological risk prediction model for prediction, resulting in a risk prediction level, including: The input layer is used to divide the initial evaluation index data and the supplementary data into spatial data and temporal data; The preprocessing layer is used to clean the spatial data and the temporal data respectively. The spatial class data output by the preprocessing layer is used to extract features using the CNN spatial feature extraction layer to obtain the first sub-feature; The LSTM temporal feature extraction layer is used to extract features from the temporal data output by the preprocessing layer to obtain the second sub-feature; The first sub-feature and the second sub-feature are simultaneously input into the Attention fusion layer for feature fusion calculation to obtain spatial temporal features; The risk prediction level is obtained by performing a level mapping calculation on the spatial temporal features using the fully connected output layer.

7. The avalanche meteorological risk prediction method based on multi-source data according to claim 4, characterized in that, The pre-training process of the avalanche meteorological risk prediction model includes: Collect data for each indicator in the aforementioned data collection system within the target research area to obtain the original dataset; The original dataset is expanded according to a preset grouping rule to obtain a full risk level dataset. The preset grouping rule is used to divide each group of full data in the original dataset into a first group to a fifth group. The data types in the first group correspond to the indicator types of Level 1 in the preliminary assessment level; the data types in the second group correspond to the indicator types of Levels 1 and 2 in the preliminary assessment level; the data types in the third group correspond to the indicator types of Levels 1 to 3 in the preliminary assessment level; the data types in the fourth group correspond to the indicator types of Levels 1 to 4 in the preliminary assessment level; and the data types in the fifth group correspond to the indicator types of Levels 1 to 5 in the preliminary assessment level. The original avalanche meteorological risk prediction model is iteratively trained using the full risk level dataset and preset label data to obtain the trained avalanche meteorological risk prediction model.

8. An avalanche meteorological risk prediction system based on multi-source data, characterized in that, The system for implementing the avalanche meteorological risk prediction method based on multi-source data as described in claim 1 includes: The data acquisition module is used to collect preset indicator data of the target area to obtain initial evaluation indicator data; the preset indicator data includes: previous snow depth, altitude, slope, previous temperature, previous cumulative snowfall, daily snowfall, and daily increase in snow depth; The indicator system construction module is used to set the preliminary evaluation levels and the collection indicators corresponding to each preliminary evaluation level, thereby obtaining the collection indicator system; the preliminary evaluation levels include: Level 1 to Level 5; The preliminary assessment module is used to calculate the initial assessment index data using the avalanche comprehensive meteorological hazard assessment formula to obtain the preliminary assessment level; the avalanche comprehensive meteorological hazard assessment formula is: ;in, ; ; The preliminary assessment level; It is one of the altitude and slope values ​​after the hierarchical assignment and normalization process. The value is one of the following after the graded assignment and normalization process: previous snow depth, previous temperature, previous cumulative snowfall, daily snowfall, and daily new snow depth. The environmental risk index for disaster-prone environments; This is a meteorological risk index. The supplementary indicator determination module is used to set the indicators corresponding to the preliminary assessment level from level two to level five in the collection indicator system as indicators to be collected when the preliminary assessment level is any one of level two to level five. The indicator supplementary data collection module is used to collect data corresponding to the target region and the indicator to be collected, and obtain supplementary data. The model prediction module is used to input the initial assessment index data and the supplementary data into the pre-trained avalanche meteorological risk prediction model for prediction, and obtain the risk prediction level. The secondary data collection module is used to add the indicators corresponding to the preliminary assessment level to the risk prediction level in the data collection indicator system to the indicators to be collected when the risk prediction level is greater than the preliminary assessment level, and return to the step "collect the data corresponding to the target area and the indicators to be collected to obtain supplementary data".

Images