Multi-mode mountain torrent disaster in-situ early warning threshold dynamic delimiting method and system

Abstract

The invention discloses a multi-mode mountain torrent disaster in-situ early warning threshold value dynamic demarcation method and system, and the method comprises the steps: collecting the topographic data of a research region, setting the DEM precision, automatically recognizing a river in the research region, recognizing grid abnormal points, and dividing the research region into a molecular basin; the method comprises the following steps: extracting historical rainfall data of a research area, constructing a rainfall scene random construction model based on prior distribution of historical rainfall, generating a rainfall scene of the research area, inputting the rainfall scene into a pre-constructed distributed hydrological model, and simulating to obtain confluence time, a rainfall cover range and a rainfall harvesting area; constructing a mountain torrent forecasting model, inputting the rainfall scenes of the plurality of research areas into the constructed mountain torrent forecasting model, simulating and extracting disaster scenes, inputting the disaster scenes into a pre-constructed Seq2Seq model, and training; historical mountain torrent data of a research area are collected, historical mountain torrent scenes of the research area are input into the Seq2Seq models corresponding to the seven built early warning combinations, the mountain torrent disaster process is obtained through simulation, early warning thresholds of the seven early warning combinations are calculated, and the mountain torrent disaster in-situ early warning threshold is obtained through combination.

Description

Technical Field

[0001] The present invention relates to the field of flash flood disaster warning, and in particular to a method and system for dynamically defining a multi-mode flash flood disaster on-site warning threshold. Background Art

[0002] Flash flood disasters refer to major natural disasters caused by the rapid rise of mountain rivers due to meteorological factors such as heavy rainfall in a short period of time. As global climate change and hydrological cycles have changed significantly, the frequency and intensity of flash flood disasters have gradually increased. How to quickly and accurately define the warning thresholds for flash flood disasters is crucial to strengthening disaster monitoring and early warning and improving response capabilities.

[0003] At present, most flash flood disaster warning systems rely on fixed thresholds for monitoring and warning, which are often affected by multiple factors and lead to poor warning effects. In addition, most existing monitoring equipment and technologies rely on fixed-point observations and cannot achieve real-time dynamic monitoring of rainfall and flow.

[0004] The present invention proposes a method and system for dynamically defining on-site warning thresholds for multi-mode flash flood disasters to solve the above-mentioned problems that currently exist, improve warning accuracy, enhance emergency response capabilities, promote scientific management, and contribute to sustainable development. Summary of the invention

[0005] The present invention provides a method and system for dynamically defining a multi-mode flash flood disaster on-site warning threshold value to solve the above-mentioned problems existing in the prior art.

[0006] The present invention is implemented by the following technical scheme: a method for dynamically defining a multi-mode flash flood disaster on-site warning threshold, comprising the following steps: Step S1, collect data of the study area, set the DEM accuracy based on the terrain data of the study area, grid the study area and automatically identify the rivers in the study area, identify grid anomalies, divide the study area into n sub-basins, and obtain the area of ​​each sub-basin, where n is a positive integer; Step S2, extracting historical rainfall data of the study area, constructing a rainfall scenario random construction model based on the prior distribution of historical rainfall, and generating several rainfall scenarios for the study area; Step S3, inputting several rainfall scenarios of the study area into the pre-built distributed hydrological model, simulating the runoff generation process corresponding to the rainfall scenarios of the study area, calculating the corresponding confluence time, and calculating the rainwater collection area based on the rainfall coverage range and the area of ​​each sub-basin; Step S4, constructing a flash flood forecasting model, inputting several rainfall scenarios of the study area into the constructed flash flood forecasting model, simulating and extracting disaster-causing scenarios as the target sequence of the pre-constructed Seq2Seq model, and taking rainfall, water level, rising speed, rainfall and water level, water level and rising speed, rainfall and rising speed, rainfall and water level and rising speed as input sequences to train the Seq2Seq model; Step S5, collect historical flash flood data in the study area, input the historical flash flood scenarios in the study area into the Seq2Seq models corresponding to the seven warning combinations constructed, simulate the flash flood disaster process respectively, calculate the warning thresholds of the seven warning combinations, and combine them to obtain the on-site warning thresholds for flash flood disasters.

[0007] Specifically, the step S1 is further as follows: Step S11, collecting digital elevation raster data of the study area, setting the DEM accuracy based on the terrain data of the study area using an adaptive resolution digital elevation model construction method, and dividing the study area into a number of grids according to the DEM accuracy to obtain digital elevation raster data of the study area; Step S12, read the digital elevation grid data of the study area, construct a monitoring window, monitor the height relationship of adjacent windows, obtain the fastest descent vector, and determine the descent direction of each grid, that is, the water flow direction; Step S13: Each grid extends the vector along its descending direction until it intersects with the river, and the grid belongs to the first river intersecting with it; Step S14, scanning all grids, identifying grids that are different from all surrounding grids and belong to the same river as abnormal grids, calculating the descending gradients of the abnormal grid and the eight surrounding grids respectively, and the river to which the grid with the largest descending gradient belongs is the river to which the abnormal grid belongs; Step S15: Divide the study area into n sub-basins, and calculate the area of ​​each sub-basin, where n is a positive integer.

[0008] Specifically, the step S11 is further as follows: Step S11a, collecting digital elevation raster data of the study area and generating a basic DEM; Step S11b, using wavelet transform method to analyze different frequency components in DEM, identify the spatial characteristics of the terrain in the study area and extract the terrain characteristics of the study area; Step S11c, based on the spatial characteristics and terrain characteristics of the study area terrain, the adaptive resolutions of different areas are set respectively, and based on the characteristic intensity, an adaptive grid generation algorithm is used to generate a dynamic grid to obtain the digital elevation raster data of the study area.

[0009] Specifically, the step S12 is further as follows: Step S12a, using GIS software to read elevation grid data, setting the window size and sliding the window in the DEM to monitor the height relationship between adjacent grids; Step S12b, loop through each grid and collect the elevation values ​​of surrounding grids in the window; Step S12c, calculating the height difference between each grid and the surrounding eight grids based on the elevation value of the grid, and calculating the steepest descent vector based on the height difference between the grid and the surrounding eight grids; Step S12d: determine the descent direction of each grid, that is, the water flow direction, based on the steepest descent vector.

[0010] Specifically, the step S2 is further as follows: Step S21, collecting rainfall data in the study area, using the improved meta-analysis method to cyclically identify and screen rainfall indicators, and constructing a quantitative indicator set for rainfall; Step S22: based on the historical rainfall data of the study area, respectively calculating the quantitative index value of each rainfall in each historical rainfall event, and fitting the probability distribution of the quantitative index of each rainfall based on the quantitative index value of each rainfall in each historical rainfall event; Step S23: construct a multidimensional joint distribution of all rainfall quantitative indicators based on the probability distribution of each rainfall quantitative indicator, and obtain a number of rainfall scenarios by random sampling.

[0011] Specifically, the step S21 is further as follows: Step S21a, collecting rainfall data in the study area, using the improved meta-analysis method to identify rainfall indicators, and constructing an initial rainfall indicator set; specifically, the steps include: Step S21a1, collecting rainfall data in the study area, using keywords and database search to screen literature, and extracting indicators related to rainfall based on the screened literature; Step S21a2: using a Bayesian model to estimate the effect of each rainfall-related indicator and setting a confidence interval; Step S21a3: considering the flash flood risk probability of the rainfall index identified by each document in the screened documents as a prior distribution, using the Markov Chain Monte Carlo method for posterior inference, obtaining the posterior distribution of the rainfall index identified by each document, and screening out the rainfall index based on the confidence interval to construct an initial rainfall index set; Step S21b, counting the number of occurrences of each initial rainfall indicator to calculate the occurrence frequency and the average frequency, identifying the initial rainfall indicator with a higher occurrence frequency than the average frequency as a rainfall indicator, using the K-means algorithm to identify similar rainfall indicators and cluster them, and constructing a rainfall indicator set; Step S21c, extracting keywords of the rainfall index set and using the improved meta-analysis method again based on the keywords to identify new rainfall indicators, and coupling the rainfall index set to construct a quantitative rainfall index set; specifically includes the following steps: Step S21c1, calculating the TF-IDF value of each rainfall indicator in the rainfall indicator set in turn, and converting the TF-IDF value into a numerical feature, and dividing the rainfall indicator set into M categories based on the numerical feature, each category has its own classification keyword, and M is a positive integer greater than 2; Step S21c2: Based on the M-class classification keywords, the improved meta-analysis method is used again to identify new rainfall indicators, and the frequency values ​​of the new rainfall indicators are calculated; Step S21c3, compare the frequency values ​​of each newly appeared rainfall index with the rainfall index with the smallest frequency value in turn, and classify the new rainfall index that is not less than the minimum rainfall index frequency value in the rainfall index set into the rainfall index set. The obtained new rainfall index set is the quantitative indicator set of rainfall.

[0012] Specifically, the step S3 is further as follows: Step S31, constructing a distributed hydrological model; Step S32, inputting several rainfall scenarios of the study area into the distributed hydrological model, simulating and obtaining the runoff generation process and confluence time corresponding to the rainfall scenarios of the study area; Step S33: Calculate the rainfall coverage area corresponding to a number of rainfall scenarios, and respectively calculate the union of the rainfall coverage area with the area of ​​each sub-basin, and the obtained union is the rainwater collection area corresponding to the number of rainfall scenarios.

[0013] Specifically, the step S4 is further as follows: Step S41, constructing a one-dimensional and a two-dimensional hydrodynamic model, and coupling the constructed distributed hydrological model to obtain a flash flood forecasting model; Step S42, inputting several rainfall scenarios of the study area into the flash flood forecasting model, and simulating and obtaining a flash flood disaster forecast; Step S43, extracting disaster-causing scenarios from flash flood disaster forecasts as target sequences of the pre-constructed Seq2Seq model, taking rainfall, water level, rising speed, rainfall and water level, water level and rising speed, rainfall and rising speed, rainfall and water level and rising speed as input sequences, respectively, training the constructed LSTM model, and obtaining seven LSTM models; specifically comprising the following steps: Step S43a, constructing a Seq2Seq model; Step S43b, extract disaster scenarios based on flash flood disaster forecasts, collect rainfall and water level data, calculate rising speed based on water level changes, obtain data sets of rainfall, water level and rising speed, generate sequence data of rainfall, water level and rising speed, and combine rainfall, water level and rising speed and their two-by-two and three-way combinations to obtain sequence data corresponding to the combination of rainfall and water level, the combination of water level and rising speed, the combination of rainfall and rising speed, and the combination of rainfall, water level and rising speed; Step S43c, sequentially divide the sequence data of seven warning combinations, namely, rainfall, water level, rising speed, rainfall and water level, water level and rising speed, rainfall and rising speed, and rainfall, water level and rising speed, into training sets and validation sets in chronological order, wherein the first 70% is the training set and the last 30% is the validation set; Step S43d, train the Seq2Seq model respectively using the training sets of rainfall, water level, rising rate, rainfall and water level, water level and rising rate, rainfall and rising rate, and rainfall, water level and rising rate, to obtain the prediction results, de-normalize and then map the predicted values ​​back to the original data range, draw a comparison chart of the predicted values ​​and the validation set, evaluate the model performance and optimize the model parameters and hyperparameters based on the comparison results of the predicted values ​​and the validation set until the difference between the predicted values ​​and the validation set is less than the threshold, stop the iteration, and obtain the Seq2Seq models corresponding to the seven warning combinations.

[0014] Specifically, the step S5 is further as follows: Step S51, collecting historical flash flood data of the study area, including: historical flash flood scenario data, confluence time, catchment area and flash flood disaster process; Step S52, inputting the historical flash flood scenarios of the study area into the Seq2Seq models corresponding to the seven warning combinations constructed, and simulating the flash flood disaster processes respectively; Step S53, setting the model fitting confidence interval, respectively calculating the similarity between the flash flood disaster process simulated by the Seq2Seq model corresponding to the seven warning combinations and the historical flash flood disaster process of the real study area, respectively extracting the historical flash flood scenarios of the study area whose similarity is within the confidence interval for the seven Seq2Seq models, and then the warning combination corresponding to the Seq2Seq model is the best warning combination of the extracted historical flash flood scenarios of the study area; Step S54, respectively count the confluence time and catchment area corresponding to the historical flash flood scenarios in the study area extracted by each Seq2Seq model, and use the maximum value as the warning threshold of the warning combination corresponding to the Seq2Seq model, so as to obtain the on-site warning threshold of flash flood disasters, that is, the warning threshold of the seven warning combinations.

[0015] The present invention also discloses a multi-mode flash flood disaster on-site warning threshold dynamic delineation system, which includes the following units: Divide the study area into units: collect data on the study area, set the DEM accuracy based on the terrain data of the study area, grid the study area and automatically identify the rivers in the study area, identify grid anomalies, divide the study area into n sub-basins, and obtain the area of ​​each sub-basin, where n is a positive integer; Generate rainfall scenario unit: extract historical rainfall data of the study area, build a rainfall scenario random construction model based on the prior distribution of historical rainfall, and generate several rainfall scenarios in the study area; Construct a distributed hydrological model unit: input several rainfall scenarios of the study area into the pre-constructed distributed hydrological model, simulate the runoff generation process corresponding to several rainfall scenarios of the study area, calculate the corresponding confluence time, and calculate the catchment area based on the rainfall coverage range and the area of ​​each sub-basin; Construct flash flood forecast model unit: construct a flash flood forecast model, input several rainfall scenarios in the study area into the constructed flash flood forecast model, simulate and extract disaster-causing scenarios as the target sequence of the pre-constructed Seq2Seq model, and use rainfall, water level, rising speed, rainfall and water level, water level and rising speed, rainfall and rising speed, rainfall and water level and rising speed as input sequences to train the Seq2Seq model; Simulation calculation warning threshold unit: collect historical flash flood data in the study area, input the historical flash flood scenarios in the study area into the Seq2Seq models corresponding to the seven warning combinations constructed, simulate the flash flood disaster process respectively, calculate the warning thresholds of the seven warning combinations, and combine them to obtain the on-site warning thresholds for flash flood disasters.

[0016] The present invention also provides a multi-mode flash flood disaster on-site warning threshold dynamic delineation system, characterized in that the system includes the following units: Divide the study area into units: collect data on the study area, set the DEM accuracy based on the terrain data of the study area, grid the study area and automatically identify the rivers in the study area, identify grid anomalies, divide the study area into n sub-basins, and obtain the area of ​​each sub-basin, where n is a positive integer; Generate rainfall scenario unit: extract historical rainfall data of the study area, build a rainfall scenario random construction model based on the prior distribution of historical rainfall, and generate several rainfall scenarios in the study area; Construct a distributed hydrological model unit: input several rainfall scenarios of the study area into the pre-constructed distributed hydrological model, simulate the runoff generation process corresponding to several rainfall scenarios of the study area, calculate the corresponding confluence time, and calculate the catchment area based on the rainfall coverage range and the area of ​​each sub-basin; Construct flash flood forecast model unit: construct a flash flood forecast model, input several rainfall scenarios in the study area into the constructed flash flood forecast model, simulate and extract disaster-causing scenarios as the target sequence of the pre-constructed Seq2Seq model, and use rainfall, water level, rising speed, rainfall and water level, water level and rising speed, rainfall and rising speed, rainfall and water level and rising speed as input sequences to train the Seq2Seq model; Simulation calculation warning threshold unit: collect historical flash flood data in the study area, input the historical flash flood scenarios in the study area into the Seq2Seq models corresponding to the seven warning combinations constructed, simulate the flash flood disaster process respectively, calculate the warning thresholds of the seven warning combinations, and combine them to obtain the on-site warning thresholds for flash flood disasters.

[0017] The beneficial effects of the present invention are as follows: a multi-mode flash flood disaster on-site warning threshold dynamic delineation method is adopted to improve warning accuracy, enhance emergency response capabilities, promote scientific management, and assist sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of the method of the present invention; Figure 2 It is the system process module of the present invention. DETAILED DESCRIPTION

[0019] like Figure 1 According to one aspect of the present application, a method for dynamically defining a multi-mode flash flood disaster on-site warning threshold is provided, comprising the following steps: Step S1, collect data of the study area, set the DEM accuracy based on the terrain data of the study area, grid the study area and automatically identify the rivers in the study area, identify grid anomalies, divide the study area into n sub-basins, and obtain the area of ​​each sub-basin, where n is a positive integer; Step S2, extracting historical rainfall data of the study area, constructing a rainfall scenario random construction model based on the prior distribution of historical rainfall, and generating several rainfall scenarios for the study area; Step S3, inputting several rainfall scenarios of the study area into the pre-built distributed hydrological model, simulating the runoff generation process corresponding to the rainfall scenarios of the study area, calculating the corresponding confluence time, and calculating the rainwater collection area based on the rainfall coverage range and the area of ​​each sub-basin; Step S4, constructing a flash flood forecasting model, inputting several rainfall scenarios of the study area into the constructed flash flood forecasting model, simulating and extracting disaster-causing scenarios, and inputting a pre-constructed Seq2Seq model, respectively training the constructed Seq2Seq model with rainfall, water level, rising speed, rainfall and water level, water level and rising speed, rainfall and rising speed, and rainfall, water level and rising speed; Step S5, collect historical flash flood data in the study area, input the historical flash flood scenarios in the study area into the Seq2Seq models corresponding to the seven warning combinations constructed, simulate the flash flood disaster process respectively, calculate the warning thresholds of the seven warning combinations, and combine them to obtain the on-site warning thresholds for flash flood disasters.

[0020] "On-site" in "on-site warning of flash flood disasters" means at the scene or local area where the flood occurs. This warning system aims to use real-time monitoring and data to promptly warn of flash flood disasters occurring in specific areas so that appropriate measures can be taken quickly to protect the safety of residents and property. Compared with general weather warnings, on-site warnings place more emphasis on real-time risk assessment and response capabilities for specific locations.

[0021] According to one aspect of the present application, step S1 further comprises: Step S11, collecting digital elevation raster data of the study area, setting the DEM accuracy based on the terrain data of the study area using an adaptive resolution digital elevation model construction method, and dividing the study area into a number of grids according to the DEM accuracy to obtain digital elevation raster data of the study area; Digital elevation grids provide high-precision terrain data, which enables the structure and morphology of river networks to be accurately identified. Compared with traditional manual measurements or low-resolution terrain data, digital elevation grids can reflect the subtle ups and downs of the surface in more detail. By analyzing elevation data, the hydrological process of the basin can be considered, and river flow, runoff paths and hydrological characteristics can be better simulated.

[0022] Step S12, read the digital elevation grid data of the study area, construct a monitoring window, monitor the height relationship of adjacent windows, obtain the fastest descent vector, and determine the descent direction of each grid, that is, the water flow direction; The steepest descent vector, that is, the direction from the high point in space to the low point, is also the direction of water flow; Step S13: Each grid extends the vector along its descending direction until it intersects with the river, and the grid belongs to the first river intersecting with it; Step S14, scanning all grids, identifying grids that are different from all surrounding grids and belong to the same river as abnormal grids, calculating the descending gradients of the abnormal grid and the eight surrounding grids respectively, and the river to which the grid with the largest descending gradient belongs is the river to which the abnormal grid belongs; Step S15: Divide the study area into n sub-basins, and calculate the area of ​​each sub-basin, where n is a positive integer.

[0023] In the extraction of digital river network based on DEM, different DEM resolutions will lead to different extraction results of digital river network. Choosing the appropriate threshold is crucial to the accuracy of digital river network and watershed extraction. A smaller threshold can extract more small rivers, but it will increase the amount of calculation and data storage requirements; a larger threshold is more suitable for extracting large rivers and can improve calculation efficiency. At the same resolution, as the threshold increases, the river network becomes sparse. Topographic factors will have a greater impact on the selection of thresholds, and DEM resolution directly affects the extraction and analysis of terrain factors. High-resolution DEM can provide more reliable terrain information, thereby improving the precision and accuracy of the results. Low-resolution DEM may cause the digital river network extraction results to be inaccurate, but too high a resolution is also more susceptible to noise interference.

[0024] According to one aspect of the present application, step S11 further comprises: Step S11a, collecting digital elevation raster data of the study area and generating a basic DEM; Step S11b, using wavelet transform method to analyze different frequency components in DEM, identify the spatial characteristics of the terrain in the study area and extract the terrain characteristics of the study area; Step S11c, based on the spatial characteristics and terrain characteristics of the study area terrain, the adaptive resolutions of different areas are set respectively, and based on the characteristic intensity, an adaptive grid generation algorithm is used to generate a dynamic grid to obtain the digital elevation raster data of the study area.

[0025] The adaptive resolution digital elevation model construction method combines multi-scale analysis and dynamic grid generation algorithm, and adjusts the grid resolution by judging the intensity of terrain features. It can ensure that more accurate elevation data can be provided when dealing with terrains of different complexities. In this embodiment, it can not only improve the effectiveness and adaptability of DEM data, but also provide more reliable basic data for subsequent calculations.

[0026] According to one aspect of the present application, step S12 further comprises: Step S12a, using GIS software to read elevation grid data, setting the window size and sliding the window in the DEM to monitor the height relationship between adjacent grids; Step S12b, loop through each grid and collect the elevation values ​​of surrounding grids in the window; Step S12c, calculating the height difference between each grid and the surrounding eight grids based on the elevation value of the grid, and calculating the steepest descent vector based on the height difference between the grid and the surrounding eight grids; Step S12d: determine the descent direction of each grid, that is, the water flow direction, based on the steepest descent vector.

[0027] In this embodiment, the digital elevation grid data is read and a monitoring window is constructed, and the height relationship of adjacent windows is monitored to determine the water flow direction of each grid, specifically: Use GIS software to read elevation raster data. When operating in Python, use the rasterio library to read DEM. Choose a suitable window size, and the window will slide in the DEM to monitor the height relationship of adjacent pixels, i.e., grids; Loop through each grid, get the elevation values ​​of the surrounding grids in the window, and use the sliding window technology to calculate the surrounding pixels of each pixel; Analyze the elevation data of the current pixel and its adjacent pixels, and calculate the height difference between the adjacent pixels and the current pixel; Compare the elevation value of the current grid with the elevation values ​​of the adjacent 8 directions to find the fastest descending direction, which is represented by an angle or vector. The final flow direction vector is calculated based on the descending directions obtained from 8 adjacent pixels, and each direction is represented by a two-dimensional array; Store the calculated flow direction results as new raster data.

[0028] According to one aspect of the present application, step S2 further comprises: Step S21, collecting rainfall data in the study area, using the improved meta-analysis method to cyclically identify and screen rainfall indicators, and constructing a quantitative indicator set for rainfall; Step S22: based on the historical rainfall data of the study area, respectively calculating the quantitative index value of each rainfall in each historical rainfall event, and fitting the probability distribution of the quantitative index of each rainfall based on the quantitative index value of each rainfall in each historical rainfall event; Step S23: construct a multidimensional joint distribution of all rainfall quantitative indicators based on the probability distribution of each rainfall quantitative indicator, and obtain a number of rainfall scenarios by random sampling.

[0029] In this embodiment, the initial indicators are first identified by using the meta-analysis method. After insensitive indicators are screened out, the keywords of sensitive indicators are used for meta-analysis again, and the newly identified indicators are compared with the screened indicators. The same indicators are used as quantitative indicators of rainfall, and the sensitivity of the newly emerged indicators is calculated. The indicators with a sensitivity greater than a threshold are also used as quantitative indicators of rainfall. According to one aspect of the present application, the step S21 is further as follows: Step S21a, collecting rainfall data in the study area, using an improved meta-analysis method to identify rainfall indicators, and constructing an initial rainfall indicator set; Step S21b, counting the number of occurrences of each initial rainfall indicator to calculate the occurrence frequency and the average frequency, identifying the initial rainfall indicator with a higher occurrence frequency than the average frequency as a rainfall indicator, using the K-means algorithm to identify similar rainfall indicators and cluster them, and constructing a rainfall indicator set; Step S21c, extracting keywords of the rainfall indicator set and using the improved meta-analysis method again based on the keywords to identify new rainfall indicators, and coupling the rainfall indicator set to construct a quantitative indicator set for rainfall.

[0030] In this embodiment, the frequency of the initial rainfall index is obtained by calculating the number of occurrences of the initial rainfall index, the average frequency is calculated, and the initial rainfall index with an occurrence frequency far lower than the average frequency is removed to obtain a quantitative index set of rainfall.

[0031] According to one aspect of the present application, step S21a is further: Step S21a1, collecting rainfall data in the study area, using keywords and database search to screen literature, and extracting indicators related to rainfall based on the screened literature; Step S21a2: using a Bayesian model to estimate the effect of each rainfall-related indicator and setting a confidence interval; Step S21a3: The flash flood risk probability of the rainfall index identified by each document in the screened documents is regarded as a prior distribution, and the Markov Chain Monte Carlo method is used for posterior inference to obtain the posterior distribution of the rainfall index identified by each document and screen out the rainfall index based on the confidence interval to construct an initial rainfall index set.

[0032] According to one aspect of the present application, the step S21c is further: Step S21c1, calculating the TF-IDF value of each rainfall indicator in the rainfall indicator set in turn, and converting the TF-IDF value into a numerical feature, and dividing the rainfall indicator set into M categories based on the numerical feature, each category has its own classification keyword, and M is a positive integer greater than 2; TF-IDF is a weighted technology commonly used in text mining and information retrieval. It aims to evaluate the importance of a word in a document collection or corpus. It is composed of word frequency and inverse document frequency. In this embodiment, by calculating the TF-IDF value of each indicator and converting the TF-IDF value into a numerical feature, the word embedding technology is used to map similar indicators to a close vector space, and then the hierarchical clustering method is used to classify similar rainfall indicators into the same category, so as to obtain the classification of rainfall indicators with keywords.

[0033] Step S21c2: Based on the M-class classification keywords, the improved meta-analysis method is used again to identify new rainfall indicators, and the frequency values ​​of the new rainfall indicators are calculated; Step S21c3, compare the frequency values ​​of each newly appeared rainfall index with the rainfall index with the smallest frequency value in turn, and classify the new rainfall index that is not less than the minimum rainfall index frequency value in the rainfall index set into the rainfall index set. The obtained new rainfall index set is the quantitative indicator set of rainfall.

[0034] According to one aspect of the present application, step S3 further comprises: Step S31, constructing a distributed hydrological model; Step S32, inputting several rainfall scenarios of the study area into the distributed hydrological model, simulating and obtaining the runoff generation process and confluence time corresponding to the rainfall scenarios of the study area; Step S33: Calculate the rainfall coverage area corresponding to a number of rainfall scenarios, and respectively calculate the union of the rainfall coverage area with the area of ​​each sub-basin, and the obtained union is the rainwater collection area corresponding to the number of rainfall scenarios.

[0035] According to one aspect of the present application, step S4 is further: Step S41, constructing a one-dimensional and a two-dimensional hydrodynamic model, and coupling the constructed distributed hydrological model to obtain a flash flood forecasting model; Step S42, inputting several rainfall scenarios of the study area into the flash flood forecasting model, and simulating and obtaining a flash flood disaster forecast; Step S43, extract disaster scenarios based on flash flood disaster forecasts and input them into a pre-built LSTM model optimized based on the Attention mechanism, and train the constructed LSTM models with seven warning combinations of rainfall, water level, rising speed, combination of rainfall and water level, combination of water level and rising speed, combination of rainfall and rising speed, and combination of rainfall, water level and rising speed, to obtain seven LSTM models.

[0036] According to one aspect of the present application, step S43 is further: Step S43a, extracting disaster scenarios based on flash flood disaster forecasts, collecting rainfall and water level data, calculating the rising speed based on water level changes, and obtaining respective data sets of rainfall, water level, and rising speed; Step S43b, construct a Seq2Seq model to generate sequence data of rainfall, water level and rising speed respectively, combine rainfall, water level and rising speed in pairs and in pairs, and obtain sequence data corresponding to the combination of rainfall and water level, the combination of water level and rising speed, the combination of rainfall and rising speed, and the combination of rainfall, water level and rising speed; Step S43c, sequentially divide the sequence data of seven warning combinations, namely, rainfall, water level, rising speed, rainfall and water level, water level and rising speed, rainfall and rising speed, and rainfall, water level and rising speed, into training sets and validation sets in chronological order, wherein the first 70% is the training set and the last 30% is the validation set; Step S43d, train the Seq2Seq model respectively using the training sets of rainfall, water level, rising rate, rainfall and water level, water level and rising rate, rainfall and rising rate, and rainfall, water level and rising rate, to obtain the prediction results, de-normalize and then map the predicted values ​​back to the original data range, draw a comparison chart of the predicted values ​​and the validation set, evaluate the model performance and optimize the model parameters and hyperparameters based on the comparison results of the predicted values ​​and the validation set until the difference between the predicted values ​​and the validation set is less than the threshold, stop the iteration, and obtain the Seq2Seq models corresponding to the seven warning combinations.

[0037] The step S43b is further as follows: Step S43b1, using Keras to build a Seq2Seq model, including an encoder and a decoder, setting model parameters, and introducing an Attention mechanism; Step S43b2, extracting the data sets of rainfall, water level and rising speed and inputting them into the decoder, and sequentially using the sliding window to generate the sequence data of rainfall, water level and rising speed; Step S43b3, combine rainfall, water level and rising speed in pairs and in combination to obtain sequence data corresponding to the combination of rainfall and water level, the combination of water level and rising speed, the combination of rainfall and rising speed, and the combination of rainfall, water level and rising speed, for a total of seven warning combinations.

[0038] In this embodiment, the Seq2Seq model combining LSTM and Attention can effectively capture the complex nonlinear relationship between input and output, which is particularly important for complex meteorological phenomena such as flash floods. It can improve the prediction performance of the model. At the same time, the model can flexibly support variable-length input and output sequences. In the process of dealing with flash floods, since rainfall, water level and rising speed have irregular time intervals, the model can adapt to a variety of different types of time series and can better fit the scenario.

[0039] Through the Attention mechanism, the model can automatically determine which features have a greater impact on predictions at different time points, which helps reduce overfitting and improve the model's generalization ability on unseen data. At the same time, the Attention weight can provide visualization of the model's decision-making process, increasing the model's interpretability, which is particularly important in climate disaster warning.

[0040] According to one aspect of the present application, step S5 is further: Step S51, collecting historical flash flood data of the study area, including: historical flash flood scenario data, confluence time, catchment area and flash flood disaster process; Step S52, inputting the historical flash flood scenarios of the study area into the Seq2Seq models corresponding to the seven warning combinations constructed, and simulating the flash flood disaster processes respectively; Step S53, setting the model fitting confidence interval, respectively calculating the similarity between the flash flood disaster process simulated by the Seq2Seq model corresponding to the seven warning combinations and the historical flash flood disaster process of the real study area, respectively extracting the historical flash flood scenarios of the study area whose similarity is within the confidence interval for the seven Seq2Seq models, and then the warning combination corresponding to the Seq2Seq model is the best warning combination of the extracted historical flash flood scenarios of the study area; Step S54, respectively count the confluence time and catchment area corresponding to the historical flash flood scenarios in the study area extracted by each Seq2Seq model, and use the maximum value as the warning threshold of the warning combination corresponding to the Seq2Seq model, so as to obtain the on-site warning threshold of flash flood disasters, that is, the warning threshold of the seven warning combinations.

[0041] like Figure 2 As shown, the present invention also provides a multi-mode flash flood disaster on-site warning threshold dynamic delineation system, characterized in that the system includes the following units: Divide the study area into units: collect data on the study area, set the DEM accuracy based on the terrain data of the study area, grid the study area and automatically identify the rivers in the study area, identify grid anomalies, divide the study area into n sub-basins, and obtain the area of ​​each sub-basin, where n is a positive integer; Generate rainfall scenario unit: extract historical rainfall data of the study area, build a rainfall scenario random construction model based on the prior distribution of historical rainfall, and generate several rainfall scenarios in the study area; Construct a distributed hydrological model unit: input several rainfall scenarios of the study area into the pre-constructed distributed hydrological model, simulate the runoff generation process corresponding to several rainfall scenarios of the study area, calculate the corresponding confluence time, and calculate the catchment area based on the rainfall coverage range and the area of ​​each sub-basin; Construct flash flood forecast model unit: construct a flash flood forecast model, input several rainfall scenarios in the study area into the constructed flash flood forecast model, simulate and extract disaster-causing scenarios as the target sequence of the pre-constructed Seq2Seq model, and use rainfall, water level, rising speed, rainfall and water level, water level and rising speed, rainfall and rising speed, rainfall and water level and rising speed as input sequences to train the Seq2Seq model; Simulation calculation warning threshold unit: collect historical flash flood data in the study area, input the historical flash flood scenarios in the study area into the Seq2Seq models corresponding to the seven warning combinations constructed, simulate the flash flood disaster process respectively, calculate the warning thresholds of the seven warning combinations, and combine them to obtain the on-site warning thresholds for flash flood disasters.

[0042] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A method for dynamically defining a multi-mode flash flood disaster on-site warning threshold, characterized in that: The steps include: Step S1, collect data of the study area, set the DEM accuracy based on the terrain data of the study area, grid the study area and automatically identify the rivers in the study area, identify grid anomalies, divide the study area into n sub-basins, and obtain the area of ​​each sub-basin, where n is a positive integer; Step S2, extracting historical rainfall data of the study area, constructing a rainfall scenario random construction model based on the prior distribution of historical rainfall, and generating several rainfall scenarios for the study area; Step S3, inputting several rainfall scenarios of the study area into the pre-built distributed hydrological model, simulating the runoff generation process corresponding to the rainfall scenarios of the study area, calculating the corresponding confluence time, and calculating the rainwater collection area based on the rainfall coverage range and the area of ​​each sub-basin; Step S4, constructing a flash flood forecasting model, inputting several rainfall scenarios of the study area into the constructed flash flood forecasting model, simulating and extracting disaster-causing scenarios as the target sequence of the pre-constructed Seq2Seq model, and taking rainfall, water level, rising speed, rainfall and water level, water level and rising speed, rainfall and rising speed, rainfall and water level and rising speed as input sequences to train the Seq2Seq model; Step S5, collect historical flash flood data in the study area, input the historical flash flood scenarios in the study area into the Seq2Seq models corresponding to the seven warning combinations constructed, simulate the flash flood disaster process respectively, calculate the warning thresholds of the seven warning combinations, and combine them to obtain the on-site warning thresholds for flash flood disasters.

2. The method for dynamically defining the multi-mode flash flood disaster on-site warning threshold according to claim 1, characterized in that: The step S1 is further as follows: Step S11, collecting digital elevation raster data of the study area, setting the DEM accuracy based on the terrain data of the study area using an adaptive resolution digital elevation model construction method, and dividing the study area into a number of grids according to the DEM accuracy to obtain digital elevation raster data of the study area; Step S12, read the digital elevation grid data of the study area, construct a monitoring window, monitor the height relationship of adjacent windows, obtain the fastest descent vector, and determine the descent direction of each grid, that is, the water flow direction; Step S13: Each grid extends the vector along its descending direction until it intersects with the river, and the grid belongs to the first river intersecting with it; Step S14, scanning all grids, identifying grids that are different from all surrounding grids and belong to the same river as abnormal grids, calculating the descending gradients of the abnormal grid and the eight surrounding grids respectively, and the river to which the grid with the largest descending gradient belongs is the river to which the abnormal grid belongs; Step S15: Divide the study area into n sub-basins, and calculate the area of ​​each sub-basin, where n is a positive integer.

3. The method for dynamically defining the multi-mode flash flood disaster on-site warning threshold according to claim 2, characterized in that: The step S11 is further as follows: Step S11a, collecting digital elevation raster data of the study area and generating a basic DEM; Step S11b, using wavelet transform method to analyze different frequency components in DEM, identify the spatial characteristics of the terrain in the study area and extract the terrain characteristics of the study area; Step S11c, based on the spatial characteristics and terrain characteristics of the study area terrain, the adaptive resolutions of different areas are set respectively, and based on the characteristic intensity, an adaptive grid generation algorithm is used to generate a dynamic grid to obtain the digital elevation raster data of the study area.

4. The method for dynamically defining the multi-mode flash flood disaster on-site warning threshold according to claim 2, characterized in that: The step S12 is further as follows: Step S12a, using GIS software to read elevation grid data, setting the window size and sliding the window in the DEM to monitor the height relationship between adjacent grids; Step S12b, loop through each grid and collect the elevation values ​​of surrounding grids in the window; Step S12c, calculating the height difference between each grid and the surrounding eight grids based on the elevation value of the grid, and calculating the steepest descent vector based on the height difference between the grid and the surrounding eight grids; Step S12d: determine the descent direction of each grid, that is, the water flow direction, based on the steepest descent vector.

5. The method for dynamically defining the multi-mode flash flood disaster on-site warning threshold according to claim 1, characterized in that: The step S2 is further as follows: Step S21, collecting rainfall data in the study area, using the improved meta-analysis method to cyclically identify and screen rainfall indicators, and constructing a quantitative indicator set for rainfall; Step S22: based on the historical rainfall data of the study area, respectively calculating the quantitative index value of each rainfall in each historical rainfall event, and fitting the probability distribution of the quantitative index of each rainfall based on the quantitative index value of each rainfall in each historical rainfall event; Step S23: construct a multidimensional joint distribution of all rainfall quantitative indicators based on the probability distribution of each rainfall quantitative indicator, and obtain a number of rainfall scenarios by random sampling.

6. The method for dynamically defining the multi-mode flash flood disaster on-site warning threshold according to claim 5, characterized in that: The step S21 is further as follows: Step S21a, collecting rainfall data in the study area, using the improved meta-analysis method to identify rainfall indicators, and constructing an initial rainfall indicator set; specifically, the steps include: Step S21a1, collecting rainfall data in the study area, using keywords and database search to screen literature, and extracting indicators related to rainfall based on the screened literature; Step S21a2: using a Bayesian model to estimate the effect of each rainfall-related indicator and setting a confidence interval; Step S21a3: considering the flash flood risk probability of the rainfall index identified by each document in the screened documents as a prior distribution, using the Markov Chain Monte Carlo method for posterior inference, obtaining the posterior distribution of the rainfall index identified by each document, and screening out the rainfall index based on the confidence interval to construct an initial rainfall index set; Step S21b, counting the number of occurrences of each initial rainfall indicator to calculate the occurrence frequency and the average frequency, identifying the initial rainfall indicator with a higher occurrence frequency than the average frequency as a rainfall indicator, using the K-means algorithm to identify similar rainfall indicators and cluster them, and constructing a rainfall indicator set; Step S21c, extracting keywords of the rainfall index set and using the improved meta-analysis method again based on the keywords to identify new rainfall indicators, and coupling the rainfall index set to construct a quantitative rainfall index set; specifically includes the following steps: Step S21c1, calculating the TF-IDF value of each rainfall indicator in the rainfall indicator set in turn, and converting the TF-IDF value into a numerical feature, and dividing the rainfall indicator set into M categories based on the numerical feature, each category has its own classification keyword, and M is a positive integer greater than 2; Step S21c2: Based on the M-class classification keywords, the improved meta-analysis method is used again to identify new rainfall indicators, and the frequency values ​​of the new rainfall indicators are calculated; Step S21c3, compare the frequency values ​​of each newly appeared rainfall index with the rainfall index with the smallest frequency value in turn, and classify the new rainfall index that is not less than the minimum rainfall index frequency value in the rainfall index set into the rainfall index set. The obtained new rainfall index set is the quantitative indicator set of rainfall.

7. The method for dynamically defining the multi-mode flash flood disaster on-site warning threshold according to claim 1, characterized in that: The step S3 is further as follows: Step S31, constructing a distributed hydrological model; Step S32, inputting several rainfall scenarios of the study area into the distributed hydrological model, simulating and obtaining the runoff generation process and confluence time corresponding to the rainfall scenarios of the study area; Step S33: Calculate the rainfall coverage area corresponding to a number of rainfall scenarios, and respectively calculate the union of the rainfall coverage area with the area of ​​each sub-basin, and the obtained union is the rainwater collection area corresponding to the number of rainfall scenarios.

8. The method for dynamically defining the multi-mode flash flood disaster on-site warning threshold according to claim 1, characterized in that: The step S4 is further as follows: Step S41, constructing a one-dimensional and a two-dimensional hydrodynamic model, and coupling the constructed distributed hydrological model to obtain a flash flood forecast model; Step S42, inputting several rainfall scenarios of the study area into the flash flood forecasting model, and simulating and obtaining a flash flood disaster forecast; Step S43, extracting disaster-causing scenarios from flash flood disaster forecasts as target sequences of the pre-constructed Seq2Seq model, taking rainfall, water level, rising speed, rainfall and water level, water level and rising speed, rainfall and rising speed, rainfall and water level and rising speed as input sequences, respectively, training the constructed LSTM model, and obtaining seven LSTM models; specifically comprising the following steps: Step S43a, constructing a Seq2Seq model; Step S43b, extract disaster scenarios based on flash flood disaster forecasts, collect rainfall and water level data, calculate rising speed based on water level changes, obtain data sets of rainfall, water level and rising speed, generate sequence data of rainfall, water level and rising speed, and combine rainfall, water level and rising speed and their two-by-two and three-way combinations to obtain sequence data corresponding to the combination of rainfall and water level, the combination of water level and rising speed, the combination of rainfall and rising speed, and the combination of rainfall, water level and rising speed; Step S43c, sequentially divide the sequence data of seven warning combinations, namely, rainfall, water level, rising speed, rainfall and water level, water level and rising speed, rainfall and rising speed, and rainfall, water level and rising speed, into training sets and validation sets in chronological order, wherein the first 70% is the training set and the last 30% is the validation set; Step S43d, train the Seq2Seq model respectively using the training sets of rainfall, water level, rising rate, rainfall and water level, water level and rising rate, rainfall and rising rate, and rainfall, water level and rising rate, to obtain the prediction results, de-normalize and then map the predicted values ​​back to the original data range, draw a comparison chart of the predicted values ​​and the validation set, evaluate the model performance and optimize the model parameters and hyperparameters based on the comparison results of the predicted values ​​and the validation set until the difference between the predicted values ​​and the validation set is less than the threshold, stop the iteration, and obtain the Seq2Seq models corresponding to the seven warning combinations.

9. The method for dynamically defining the multi-mode flash flood disaster on-site warning threshold according to claim 1, characterized in that: The step S5 is further as follows: Step S51, collecting historical flash flood data of the study area, including: historical flash flood scenario data, confluence time, catchment area and flash flood disaster process; Step S52, inputting the historical flash flood scenarios of the study area into the Seq2Seq models corresponding to the seven warning combinations constructed, and simulating the flash flood disaster processes respectively; Step S53, setting the model fitting confidence interval, respectively calculating the similarity between the flash flood disaster process simulated by the Seq2Seq model corresponding to the seven warning combinations and the historical flash flood disaster process of the real study area, respectively extracting the historical flash flood scenarios of the study area whose similarity is within the confidence interval for the seven Seq2Seq models, and then the warning combination corresponding to the Seq2Seq model is the best warning combination of the extracted historical flash flood scenarios of the study area; Step S54, respectively count the confluence time and catchment area corresponding to the historical flash flood scenarios in the study area extracted by each Seq2Seq model, and use the maximum value as the warning threshold of the warning combination corresponding to the Seq2Seq model, so as to obtain the on-site warning threshold of flash flood disasters, that is, the warning threshold of the seven warning combinations.

10. A multi-mode flash flood disaster on-site warning threshold dynamic delineation system, characterized in that: The system includes the following units: Divide the study area into units: collect data on the study area, set the DEM accuracy based on the terrain data of the study area, grid the study area and automatically identify the rivers in the study area, identify grid anomalies, divide the study area into n sub-basins, and obtain the area of ​​each sub-basin, where n is a positive integer; Generate rainfall scenario unit: extract historical rainfall data of the study area, build a rainfall scenario random construction model based on the prior distribution of historical rainfall, and generate several rainfall scenarios in the study area; Construct a distributed hydrological model unit: input several rainfall scenarios of the study area into the pre-constructed distributed hydrological model, simulate the runoff generation process corresponding to several rainfall scenarios of the study area, calculate the corresponding confluence time, and calculate the catchment area based on the rainfall coverage range and the area of ​​each sub-basin; Construct flash flood forecast model unit: construct a flash flood forecast model, input several rainfall scenarios in the study area into the constructed flash flood forecast model, simulate and extract disaster-causing scenarios as the target sequence of the pre-constructed Seq2Seq model, and use rainfall, water level, rising speed, rainfall and water level, water level and rising speed, rainfall and rising speed, rainfall and water level and rising speed as input sequences to train the Seq2Seq model; Simulation calculation warning threshold unit: collect historical flash flood data in the study area, input the historical flash flood scenarios in the study area into the Seq2Seq models corresponding to the seven warning combinations constructed, simulate the flash flood disaster process respectively, calculate the warning thresholds of the seven warning combinations, and combine them to obtain the on-site warning thresholds for flash flood disasters.

Images