A flash flood disaster early warning method and system based on multi-mode combination optimization

Abstract

The present invention discloses a flash flood disaster early warning method and system based on multi-mode combination optimization. The method adopts a comprehensive adaptive index screening framework method to identify flash flood disaster indicators, constructs a key flash flood disaster indicator set, calculates its indicator weights, and configures a pre-constructed flash flood disaster forecasting model; extracts the rainwater collection area and confluence time corresponding to each historical flash flood disaster from the historical flash flood disaster data of the study area, fits the distribution of the rainwater collection area and the confluence time respectively, and constructs the joint distribution of the rainwater collection area and the confluence time, and randomly samples to obtain a number of flash flood scenarios; fits the flash flood disaster process and inputs it into a pre-constructed early warning combination optimization model, solves the model to obtain the optimized early warning combination corresponding to each flash flood scenario and fits the flash flood disaster process, calculates the key flash flood disaster indicator values ​​and inputs them into the constructed flash flood disaster forecasting model, and solves the model to obtain a flash flood disaster forecast; the present invention can provide accurate flash flood disaster early warning and forecast that changes with actual conditions.

Description

Technical Field

[0001] The present invention relates to the field of flash flood disaster warning, and in particular to a flash flood disaster warning method and system based on multi-mode combination optimization. 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. Summary of the invention

[0004] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a flash flood disaster warning method and system based on multi-mode combination optimization, so as to solve the technical problem that most of the monitoring equipment and technologies in the prior art rely on fixed-point observation and cannot realize real-time dynamic monitoring of rainfall and flow.

[0005] The present invention is implemented by the following technical scheme: a flash flood disaster early warning method based on multi-mode combination optimization, comprising the following steps:

[0006] Step S1, collect flash flood disaster data in the study area, use the comprehensive adaptive index screening framework method to identify flash flood disaster indicators, construct a key flash flood disaster indicator set, calculate the key flash flood disaster indicator weights and configure a pre-built flash flood disaster forecast model;

[0007] Step S2, extracting the catchment area and confluence time corresponding to each historical flash flood disaster from the historical flash flood disaster data of the study area, fitting the distribution of the catchment area and confluence time respectively and constructing the joint distribution of the catchment area and confluence time, and randomly sampling to obtain several flash flood scenarios;

[0008] Step S3, constructing several warning combinations based on rainfall, water level and rising speed, using several warning combinations in turn for several flash flood scenarios to fit the flash flood disaster process respectively and inputting the pre-constructed warning combination optimization model, solving the model to obtain the optimal warning combination corresponding to each flash flood scenario;

[0009] Step S4: For each flash flood scenario, the flash flood disaster process is obtained by using the optimal warning combination fitting, and the key flash flood disaster index values ​​are calculated and input into the constructed flash flood disaster forecasting model, and the model is solved to obtain the flash flood disaster forecast.

[0010] Specifically, the step S1 is further as follows:

[0011] Step S11, collecting flash flood disaster data in the study area, including: historical flash flood disaster data and flash flood disaster literature in the study area;

[0012] Step S12: using the comprehensive adaptability index screening framework method to identify flash flood disaster indicators, construct a flash flood disaster indicator set, and calculate the weight of each flash flood disaster indicator;

[0013] Step S13: construct a flash flood disaster forecasting model, and configure the flash flood disaster forecasting model using flash flood disaster index weights.

[0014] Specifically, the step S12 is further as follows:

[0015] Step S12a, extracting flash flood disaster literature, identifying various indicators that affect flash flood disasters, constructing an initial flash flood disaster indicator set, and using a circular feedback mechanism in combination with expert opinions to screen the initial flash flood disaster indicators to construct a flash flood disaster indicator set;

[0016] Step S12b, using a network analysis method to construct a relationship diagram between flash flood disaster indicators, and identifying key flash flood disaster indicators and interdependencies;

[0017] Step S12c, using an adaptive algorithm to assign a weight to each flash flood disaster indicator based on dynamic feedback and the interdependence between the flash flood disaster indicators, and comprehensively ranking each flash flood disaster indicator;

[0018] Step S12d, calculating the weighted average of the flash flood disaster indicators, screening out the flash flood disaster indicators whose weights are far lower than the weighted average, constructing the remaining flash flood disaster indicators into a key flash flood disaster indicator set, and obtaining the weights of the key flash flood disaster indicators.

[0019] Specifically, the step S2 is further as follows:

[0020] Step S21, extracting the catchment area and confluence time corresponding to each historical flash flood disaster from the historical flash flood disaster data of the study area, and using the kernel density estimation method to fit the distribution of the catchment area and confluence time respectively;

[0021] Step S22: constructing a joint distribution of the catchment area and the runoff time using an improved Bayesian network model based on the distribution of the catchment area and the runoff time;

[0022] Step S23: Use a generative adversarial network to perform random sampling to obtain several flash flood scenarios.

[0023] Specifically, the step S21 is further as follows:

[0024] Step S21a, extracting the catchment area and confluence time corresponding to each historical flash flood disaster from the historical flash flood disaster data of the study area;

[0025] Step S21b, using a histogram to view the preliminary distribution of the catchment area and the confluence time, and setting kernel density estimation parameters based on the preliminary distribution;

[0026] Step S21c, using the KDE function to perform kernel density estimation on the rainwater collection area and the runoff time, and draw a fitting curve;

[0027] Step S21d, analyzing the KDE results, adjusting the bandwidth parameters based on the distribution characteristics of the area and the confluence time, optimizing the fitting curve, and obtaining the distribution of the rainwater collection area and the confluence time.

[0028] Specifically, the step S22 is further as follows:

[0029] Step S22a, constructing a Bayesian network model, using a structural learning algorithm to automatically learn the network structure and determine the dependencies between nodes;

[0030] Step S22b, using Bayesian estimation to define a conditional probability table for each node, and using cross-validation to optimize parameter estimation;

[0031] Step S22c: using the sample method to generate samples of the joint distribution of the catchment area and the runoff time from the Bayesian network.

[0032] Specifically, the step S23 is further as follows:

[0033] Step S23a, constructing a generator and a discriminator of a generative adversarial network model;

[0034] Step S23b, training the discriminator and the generator in sequence to obtain a trained generative adversarial network model;

[0035] Step S23c: Based on the joint distribution of the catchment area and the runoff time, a trained generative adversarial network model is used to generate new catchment area and runoff time samples to obtain several flash flood scenarios.

[0036] Specifically, the step S3 is further as follows:

[0037] Step S31, construct seven warning combinations based on rainfall, water level and rising speed, including three single warnings and four combined warnings, the single warnings are respectively: rainfall warning, water level warning and rising speed warning, the combined warnings are respectively: rainfall-water level combined warning, rainfall rising speed combined warning, water level rising speed combined warning and rainfall-water level rising speed combined warning;

[0038] Step S32, constructing an early warning combination optimization model, wherein the objective function is to maximize the early warning combination forecast accuracy;

[0039] Step S33, using seven warning combinations in sequence to fit the flash flood disaster-causing processes for a number of flash flood scenarios, and obtaining seven disaster-causing processes corresponding to each flash flood scenario;

[0040] Step S34, input the seven disaster-causing processes corresponding to each flash flood scenario into the early warning combination optimization model, and use the Enhanced MOEA / D algorithm to solve the model to obtain the optimized early warning combination corresponding to each flash flood scenario; the step S34 is further as follows:

[0041] Step S34a, randomly generate a population, automatically generate a weight vector based on the target number, and adjust the field structure based on the current population characteristics;

[0042] Step S34b, dynamically adjusting the learning rate based on individual fitness for updating;

[0043] Step S34c, generating new individuals based on the domain individuals, and updating the population with the new individuals;

[0044] Step S34d, iterate repeatedly until complete convergence, and the obtained warning combination is the optimized warning combination corresponding to the current flash flood scenario.

[0045] Specifically, the step S4 is further as follows:

[0046] Step S41, sequentially fitting each flash flood scenario with the optimal early warning combination to obtain the flash flood disaster process, and obtaining the key flash flood disaster index value based on the flash flood disaster process;

[0047] Step S42: input the key flash flood disaster index values ​​corresponding to each flash flood scenario into the constructed flash flood disaster forecasting model in turn, and solve the model to obtain the flash flood disaster forecast corresponding to each flash flood scenario.

[0048] The present invention also discloses a flash flood disaster early warning system based on multi-mode combination optimization, comprising the following modules:

[0049] Pre-built flash flood disaster forecast model module: collect flash flood disaster data in the study area, use the comprehensive adaptive index screening framework method to identify flash flood disaster indicators, build a key flash flood disaster indicator set, calculate the weights of key flash flood disaster indicators and configure the pre-built flash flood disaster forecast model;

[0050] Construct a joint distribution module: extract the catchment area and confluence time corresponding to each historical flash flood disaster from the historical flash flood disaster data of the study area, fit the distribution of the catchment area and confluence time respectively, and construct a joint distribution of the catchment area and confluence time, and randomly sample to obtain several flash flood scenarios;

[0051] Solving the optimization model module: construct several warning combinations based on rainfall, water level and rising speed, use several warning combinations to fit the flash flood disaster process for several flash flood scenarios, input the pre-built warning combination optimization model, and solve the optimization model to obtain the optimal warning combination corresponding to each flash flood scenario;

[0052] Module for solving flash flood disaster forecasting model: For each flash flood scenario, the flash flood disaster process is obtained by fitting the optimal warning combination, and the key flash flood disaster indicator values ​​are calculated and input into the constructed flash flood disaster forecasting model, and the flash flood disaster forecasting model is solved to obtain the flash flood disaster forecast.

[0053] The beneficial effects of the present invention are as follows: a flash flood disaster early warning method based on multi-mode combination optimization is adopted to improve early warning accuracy, enhance emergency response capabilities, promote scientific management, and assist sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a flow chart of the present invention;

[0055] Figure 2 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0056] like Figure 1 According to one aspect of the present application, a flash flood disaster early warning method based on multi-mode combination optimization is provided, comprising the following steps:

[0057] Step S1, collect flash flood disaster data in the study area, use the comprehensive adaptive index screening framework method to identify flash flood disaster indicators, construct a key flash flood disaster indicator set, calculate the key flash flood disaster indicator weights and configure a pre-built flash flood disaster forecast model;

[0058] Step S2, extracting the catchment area and confluence time corresponding to each historical flash flood disaster from the historical flash flood disaster data of the study area, fitting the distribution of the catchment area and confluence time respectively and constructing the joint distribution of the catchment area and confluence time, and randomly sampling to obtain several flash flood scenarios;

[0059] Step S3, constructing several warning combinations based on rainfall, water level and rising speed, using several warning combinations in turn for several flash flood scenarios to fit the flash flood disaster process respectively and inputting the pre-constructed warning combination optimization model, solving the model to obtain the optimal warning combination corresponding to each flash flood scenario;

[0060] Step S4: For each flash flood scenario, the flash flood disaster process is obtained by using the optimal warning combination fitting, and the key flash flood disaster index values ​​are calculated and input into the constructed flash flood disaster forecasting model, and the model is solved to obtain the flash flood disaster forecast.

[0061] "On-site" in "On-site Flash Flood Disaster Warning" 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.

[0062] The present invention uses different warning combinations to forecast flash flood disasters for rainfall scenarios with different catchment areas and confluence times, compares the flash flood disaster forecasts obtained with different warning combinations with the actual flash flood disaster process to find the optimal warning combination for different rainfall scenarios, and sets respective thresholds. When subsequently conducting flash flood disaster warnings, the warning combination can be directly selected through the calculated catchment area and confluence time data to forecast flash flood disasters, thereby improving the rate of on-site warnings and strengthening real-time risk assessment and response capabilities.

[0063] According to one aspect of the present application, step S1 further comprises:

[0064] Step S11, collecting flash flood disaster data in the study area, including: historical flash flood disaster data and flash flood disaster literature in the study area;

[0065] Step S12: using the comprehensive adaptability index screening framework method to identify flash flood disaster indicators, construct a flash flood disaster indicator set, and calculate the weight of each flash flood disaster indicator;

[0066] Step S13: construct a flash flood disaster forecasting model, and configure the flash flood disaster forecasting model using flash flood disaster index weights.

[0067] According to one aspect of the present application, step S12 further comprises:

[0068] Step S12a, extracting flash flood disaster literature, identifying various indicators that affect flash flood disasters, constructing an initial flash flood disaster indicator set, and using a circular feedback mechanism in combination with expert opinions to screen the initial flash flood disaster indicators to construct a flash flood disaster indicator set;

[0069] Step S12b, using a network analysis method to construct a relationship diagram between flash flood disaster indicators, and identifying key flash flood disaster indicators and interdependencies;

[0070] Step S12c, using an adaptive algorithm to assign a weight to each flash flood disaster indicator based on dynamic feedback and the interdependence between the flash flood disaster indicators, and comprehensively ranking each flash flood disaster indicator;

[0071] Step S12d, calculating the weighted average of the flash flood disaster indicators, screening out the flash flood disaster indicators whose weights are far lower than the weighted average, constructing the remaining flash flood disaster indicators into a key flash flood disaster indicator set, and obtaining the weights of the key flash flood disaster indicators.

[0072] The comprehensive adaptive index screening framework method collects and evaluates indicators, combines dynamic expert feedback, mutual influence analysis and weight optimization, and selects the most effective flash flood disaster indicators through important component identification and final evaluation. In this embodiment, the use of this method not only emphasizes the scientific nature of data-driven, but also incorporates the experience of experts and the understanding of the relationship between indicators, ensuring that the final selected indicators are both scientific and meet the needs of practical applications.

[0073] The comprehensive adaptability index screening framework is a systematic method that comprehensively evaluates and screens influencing factors from a multi-dimensional perspective. Specifically:

[0074] Preliminary identification of potential indicators affecting flash flood hazards based on flash flood literature;

[0075] Divide the selected indicators into different levels and construct an indicator system, including: basic data layer, characteristic indicator layer and final evaluation layer;

[0076] By evaluating the adaptability of indicators, the contribution of each indicator to flash flood disasters is analyzed;

[0077] Assign weights based on the importance of the indicators;

[0078] By combining the weights of each indicator and the adaptability analysis results, a comprehensive evaluation score of the indicator is formed;

[0079] Based on the comprehensive scores of the indicators, indicators are screened and the indicator system is adjusted and optimized in combination with newly collected data and event feedback.

[0080] In this embodiment, the comprehensive adaptive index screening framework method is used to identify flash flood disaster indicators, which can comprehensively consider the relationship between multiple factors and avoid the one-sidedness of focusing on a single indicator. At the same time, indicators can be evaluated and screened from multiple dimensions, making the results more credible and applicable. This method allows dynamic adjustment of indicators based on new data, which can ensure that the indicators always reflect the current characteristics of flash flood disasters and are more suitable for rapidly changing environmental backgrounds.

[0081] According to one aspect of the present application, step S2 further comprises:

[0082] Step S21, extracting the catchment area and confluence time corresponding to each historical flash flood disaster from the historical flash flood disaster data of the study area, and using the kernel density estimation method to fit the distribution of the catchment area and confluence time respectively;

[0083] Step S22: constructing a joint distribution of the catchment area and the runoff time using an improved Bayesian network model based on the distribution of the catchment area and the runoff time;

[0084] Step S23: Use a generative adversarial network to perform random sampling to obtain several flash flood scenarios.

[0085] According to one aspect of the present application, step S21 is further:

[0086] Step S21a, extracting the catchment area and confluence time corresponding to each historical flash flood disaster from the historical flash flood disaster data of the study area;

[0087] Step S21b, using a histogram to view the preliminary distribution of the catchment area and the confluence time, and setting kernel density estimation parameters based on the preliminary distribution;

[0088] Step S21c, using the KDE function to perform kernel density estimation on the rainwater collection area and the runoff time, and draw a fitting curve;

[0089] Step S21d, analyzing the KDE results, adjusting the bandwidth parameters based on the distribution characteristics of the area and the confluence time, optimizing the fitting curve, and obtaining the distribution of the rainwater collection area and the confluence time.

[0090] Kernel density estimation is a nonparametric statistical method used to estimate the probability density function of a random variable. It generates a smooth density estimate by placing a kernel function on the data points. In hydrology and water resources management, catchment area and runoff time are two important parameters. Kernel density estimation can effectively analyze and model these two variables.

[0091] The catchment area refers to the area that can collect and affect water flow in a specific rainfall event. By performing kernel density estimation on the catchment area data in historical rainfall events, the probability density distribution of the catchment area can be obtained, which helps to identify the common range of the catchment area and evaluate the changing trend of the catchment area under different rainfall events.

[0092] Confluence time refers to the time from the beginning of rainfall to the time when the water reaches a certain point. The probability distribution of confluence time can be obtained by kernel density estimation of historical confluence time data;

[0093] Kernel density estimation does not require any assumptions about the data distribution, so it is very effective when dealing with data of unknown distribution. By selecting the appropriate kernel function bandwidth, kernel density estimation can generate a smooth density curve for easy visualization and analysis. When the data volume is small or the distribution is uneven, kernel density estimation can still provide effective density estimation and the generated probability density map is intuitive and easy to understand. It can clearly show the distribution characteristics of the data and help decision makers understand the changes in the catchment area and runoff time.

[0094] According to one aspect of the present application, step S22 is further:

[0095] Step S22a, constructing a Bayesian network model, using a structural learning algorithm to automatically learn the network structure and determine the dependencies between nodes;

[0096] Step S22b, using Bayesian estimation to define a conditional probability table for each node, and using cross-validation to optimize parameter estimation;

[0097] Step S22c: using the sample method to generate samples of the joint distribution of the catchment area and the runoff time from the Bayesian network.

[0098] According to one aspect of the present application, step S23 is further:

[0099] Step S23a, constructing a generator and a discriminator of a generative adversarial network model;

[0100] Step S23b, training the discriminator and the generator in sequence to obtain a trained generative adversarial network model;

[0101] Step S23c: Based on the joint distribution of the catchment area and the runoff time, a trained generative adversarial network model is used to generate new catchment area and runoff time samples to obtain several flash flood scenarios.

[0102] Generative adversarial network is a deep learning model, mainly composed of generator and discriminator. The generator is responsible for generating fake data samples. The generator accepts random noise as input and generates samples similar to real data through a series of neural network layers. The discriminator makes complex judgments on whether the input data is real or generated. The discriminator learns the distribution of real data through training and tries to distinguish between real samples and generated samples.

[0103] The training process of the generative adversarial network model is a game process. The generator and the discriminator compete with each other. The generator tries to generate more and more real data to deceive the discriminator. The discriminator improves the classification accuracy by comparing the generated samples with the real samples. The specific process of random sampling through the generative adversarial network model is as follows:

[0104] Randomly initialize the parameters of the generator and discriminator;

[0105] Perform iterative training, train the discriminator with real samples, generate random noise, and use the generator to generate corresponding fake samples to update the discriminator;

[0106] Update the generator and improve the generation network by generating fake samples to provide feedback to the discriminator;

[0107] After training, the generator can accept new random noise as input and generate multiple possible scenarios.

[0108] The use of generative adversarial networks for random sampling can obtain a large amount of simulated scenario data in an efficient, flexible and economical way.

[0109] According to one aspect of the present application, step S3 further comprises:

[0110] Step S31, construct seven warning combinations based on rainfall, water level and rising speed, including three single warnings and four combined warnings, the single warnings are respectively: rainfall warning, water level warning and rising speed warning, the combined warnings are respectively: rainfall-water level combined warning, rainfall rising speed combined warning, water level rising speed combined warning and rainfall-water level rising speed combined warning;

[0111] In this embodiment, seven warning combinations are constructed based on rainfall, water level, rising speed and their combinations. By calculating the flash flood disaster process simulated by the seven warning combinations for a certain scenario and calculating the accuracy respectively, the optimal warning combination corresponding to each flash flood scenario is found to improve the accuracy of subsequent forecasts.

[0112] Step S32, constructing an early warning combination optimization model, wherein the objective function is to maximize the early warning combination forecast accuracy;

[0113] Step S33, using seven warning combinations in sequence to fit the flash flood disaster-causing processes for a number of flash flood scenarios, and obtaining seven disaster-causing processes corresponding to each flash flood scenario;

[0114] Step S34: input the seven disaster-causing processes corresponding to each flash flood scenario into the early warning combination optimization model, and use the Enhanced MOEA / D algorithm to solve the model to obtain the optimized early warning combination corresponding to each flash flood scenario.

[0115] According to one aspect of the present application, step S34 is further:

[0116] Step S34a, randomly generate a population, automatically generate a weight vector based on the target number, and adjust the field structure based on the current population characteristics;

[0117] Step S34b, dynamically adjusting the learning rate based on individual fitness for updating;

[0118] Step S34c, generating new individuals based on the domain individuals, and updating the population with the new individuals;

[0119] Step S34d, iterate repeatedly until complete convergence, and the obtained warning combination is the optimized warning combination corresponding to the current flash flood scenario.

[0120] Enhanced MOEA / D algorithm is a method that transforms multi-objective optimization problems into multiple single-objective optimization problems. By decomposing the objective space, each individual is responsible for optimizing one or more objectives and cooperating with its neighboring individuals. Unlike the commonly used MOEA / D algorithm, Enhanced MOEA / D algorithm does not have a fixed neighborhood structure, but dynamically adjusts the distribution of individuals in the field according to the distribution and diversity of the population, increasing the ability to fully explore the global optimal solution.

[0121] At the same time, the Enhanced MOEA / D algorithm also dynamically adjusts the weight vector according to the importance of each goal, making the balance between multiple goals more flexible. The Enhanced MOEA / D algorithm will dynamically adjust the parameters in the mutation and crossover process according to the running stage of the algorithm and the performance of the current population to avoid falling into the local optimum at the beginning. Specifically:

[0122] Randomly generate the initial individuals of the population, each of which represents a possible solution;

[0123] Generate a weight vector for each individual, ensuring that it fits the target number;

[0124] Selecting neighboring individuals based on their distance or similarity;

[0125] Evaluate the fitness of each individual through the objective function calculation;

[0126] Select individuals with better performance and those in their neighborhood, using a selection mechanism based on non-dominated sorting and crowding distance;

[0127] Generate new individuals using crossover and mutation operations;

[0128] Merge the newly generated individuals with the current population and use the non-dominated sorting method to select the next generation;

[0129] Repeat the iteration until complete convergence.

[0130] In a certain embodiment, specifically:

[0131] Collect rainfall data, topographic data, and historical disaster data;

[0132] Constructing adaptive weights based on the number of targets, this embodiment is three,initializing the population, and randomly initializing the parameters of the early warning system;

[0133] The fitness of each individual is evaluated, the values ​​of the three objectives are calculated, and the performance of each warning combination is determined by simulating flash flood disaster responses based on historical data;

[0134] The dynamic neighborhood structure of Enhanced MOEA / D is used to introduce adaptive mutation rate to generate offspring;

[0135] Update the population with the newly generated individuals and retain the top N high-quality solutions.

[0136] The Enhanced MOEA / D algorithm uses adaptive weights and dynamic domain structures, allowing the algorithm to be flexibly adjusted during exploration to find a better solution. The algorithm can find a balance between multiple goals to ensure that the accuracy of early warnings is improved without significantly increasing early warning costs. As the performance of the algorithm improves, the response time can be accelerated. Therefore, in actual operations, faster decision-making can be achieved to effectively respond to emergencies.

[0137] According to one aspect of the present application, step S4 is further:

[0138] Step S41, sequentially fitting each flash flood scenario with the optimal early warning combination to obtain the flash flood disaster process, and obtaining the key flash flood disaster index value based on the flash flood disaster process;

[0139] Step S42: input the key flash flood disaster index values ​​corresponding to each flash flood scenario into the constructed flash flood disaster forecasting model in turn, and solve the model to obtain the flash flood disaster forecast corresponding to each flash flood scenario.

[0140] The on-site warning calculates the upstream surface rainfall, analyzes and measures the upstream and downstream disaster-causing water levels, and analyzes the relationship between surface rainfall and river water level changes based on the rainfall, upstream and downstream water levels, and rising speed detection values ​​in different warning periods. In this embodiment, the optimal warning combination is calculated through the current catchment area and confluence time of the study area, and then the optimized warning combination is used to simulate the evolution of flash floods, and the corresponding values ​​of key flash flood disaster indicators are obtained and input into the constructed flash flood disaster forecasting model, and finally a flash flood disaster forecast that changes with the warning period is obtained.

[0141] like Figure 2 As shown, according to another aspect of the present application, a flash flood disaster warning system based on multi-mode combination optimization is also provided, comprising the following modules:

[0142] Pre-built flash flood disaster forecast model module: collect flash flood disaster data in the study area, use the comprehensive adaptive index screening framework method to identify flash flood disaster indicators, build a key flash flood disaster indicator set, calculate the weights of key flash flood disaster indicators and configure the pre-built flash flood disaster forecast model;

[0143] Construct a joint distribution module: extract the catchment area and confluence time corresponding to each historical flash flood disaster from the historical flash flood disaster data of the study area, fit the distribution of the catchment area and confluence time respectively, and construct a joint distribution of the catchment area and confluence time, and randomly sample to obtain several flash flood scenarios;

[0144] Solving the optimization model module: construct several warning combinations based on rainfall, water level and rising speed, use several warning combinations to fit the flash flood disaster process for several flash flood scenarios, input the pre-built warning combination optimization model, and solve the optimization model to obtain the optimal warning combination corresponding to each flash flood scenario;

[0145] Module for solving flash flood disaster forecasting model: For each flash flood scenario, the flash flood disaster process is obtained by fitting the optimal warning combination, and the key flash flood disaster indicator values ​​are calculated and input into the constructed flash flood disaster forecasting model, and the flash flood disaster forecasting model is solved to obtain the flash flood disaster forecast.

[0146] The present invention also provides an electronic device, comprising

[0147] At least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement any of the above-mentioned flash flood disaster warning methods based on multi-mode combination optimization.

[0148] 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 flash flood disaster early warning method based on multi-mode combination optimization, characterized in that: The steps include: Step S1, collect flash flood disaster data in the study area, use the comprehensive adaptive index screening framework method to identify flash flood disaster indicators, construct a key flash flood disaster indicator set, calculate the key flash flood disaster indicator weights and configure a pre-built flash flood disaster forecast model; Step S2, extracting the catchment area and confluence time corresponding to each historical flash flood disaster from the historical flash flood disaster data of the study area, fitting the distribution of the catchment area and confluence time respectively and constructing the joint distribution of the catchment area and confluence time, and randomly sampling to obtain several flash flood scenarios; the step S2 is further as follows: Step S21, extracting the catchment area and confluence time corresponding to each historical flash flood disaster from the historical flash flood disaster data of the study area, and using the kernel density estimation method to fit the distribution of the catchment area and confluence time respectively; Step S22: constructing a joint distribution of the catchment area and the runoff time using an improved Bayesian network model based on the distribution of the catchment area and the runoff time; Step S23, using a generative adversarial network for random sampling to obtain several flash flood scenarios; Step S3, constructing several warning combinations based on rainfall, water level and rising speed, using several warning combinations in turn to fit the flash flood disaster process for several flash flood scenarios and inputting the pre-constructed warning combination optimization model, solving the model to obtain the optimal warning combination corresponding to each flash flood scenario; Step S3 is further: Step S31, construct seven warning combinations based on rainfall, water level and rising speed, including three single warnings and four combined warnings, the single warnings are respectively: rainfall warning, water level warning and rising speed warning, the combined warnings are respectively: rainfall-water level combined warning, rainfall rising speed combined warning, water level rising speed combined warning and rainfall-water level rising speed combined warning; Step S32, constructing an early warning combination optimization model, wherein the objective function is to maximize the early warning combination forecast accuracy; Step S33, using seven warning combinations in sequence to fit the flash flood disaster-causing processes for a number of flash flood scenarios, and obtaining seven disaster-causing processes corresponding to each flash flood scenario; Step S34, input the seven disaster-causing processes corresponding to each flash flood scenario into the early warning combination optimization model, and use the Enhanced MOEA / D algorithm to solve the model to obtain the optimized early warning combination corresponding to each flash flood scenario; the step S34 is further as follows: Step S34a, randomly generate a population, automatically generate a weight vector based on the target number, and adjust the field structure based on the current population characteristics; Step S34b, dynamically adjusting the learning rate based on individual fitness for updating; Step S34c, generating new individuals based on the domain individuals, and updating the population with the new individuals; Step S34d, iterate repeatedly until complete convergence, and the obtained warning combination is the optimized warning combination corresponding to the current flash flood scenario; Step S4: For each flash flood scenario, the flash flood disaster process is obtained by using the optimal warning combination fitting, and the key flash flood disaster index values ​​are calculated and input into the constructed flash flood disaster forecasting model, and the model is solved to obtain the flash flood disaster forecast.

2. The flash flood disaster early warning method based on multi-mode combination optimization as claimed in claim 1, characterized in that: The step S1 is further as follows: Step S11, collecting flash flood disaster data in the study area, including: historical flash flood disaster data and flash flood disaster literature in the study area; Step S12: using the comprehensive adaptability index screening framework method to identify flash flood disaster indicators, construct a flash flood disaster indicator set, and calculate the weight of each flash flood disaster indicator; Step S13: construct a flash flood disaster forecasting model, and configure the flash flood disaster forecasting model using flash flood disaster index weights.

3. The flash flood disaster early warning method based on multi-mode combination optimization as claimed in claim 2, characterized in that: The step S12 is further as follows: Step S12a, extracting flash flood disaster literature, identifying various indicators that affect flash flood disasters, constructing an initial flash flood disaster indicator set, and using a circular feedback mechanism in combination with expert opinions to screen the initial flash flood disaster indicators to construct a flash flood disaster indicator set; Step S12b, using a network analysis method to construct a relationship diagram between flash flood disaster indicators, and identifying key flash flood disaster indicators and interdependencies; Step S12c, using an adaptive algorithm to assign a weight to each flash flood disaster indicator based on dynamic feedback and the interdependence between the flash flood disaster indicators, and comprehensively ranking each flash flood disaster indicator; Step S12d, calculating the weighted average of the flash flood disaster indicators, screening out the flash flood disaster indicators whose weights are far lower than the weighted average, constructing the remaining flash flood disaster indicators into a key flash flood disaster indicator set, and obtaining the weights of the key flash flood disaster indicators.

4. The flash flood disaster early warning method based on multi-mode combination optimization as claimed in claim 1, characterized in that: The step S21 is further as follows: Step S21a, extracting the catchment area and confluence time corresponding to each historical flash flood disaster from the historical flash flood disaster data of the study area; Step S21b, using a histogram to view the preliminary distribution of the catchment area and the confluence time, and setting kernel density estimation parameters based on the preliminary distribution; Step S21c, using the KDE function to perform kernel density estimation on the rainwater collection area and the runoff time, and draw a fitting curve; Step S21d, analyzing the KDE results, adjusting the bandwidth parameters based on the distribution characteristics of the area and the confluence time, optimizing the fitting curve, and obtaining the distribution of the catchment area and the confluence time.

5. The flash flood disaster early warning method based on multi-mode combination optimization as claimed in claim 1, characterized in that: The step S22 is further as follows: Step S22a, constructing a Bayesian network model, using a structural learning algorithm to automatically learn the network structure and determine the dependencies between nodes; Step S22b, using Bayesian estimation to define a conditional probability table for each node, and using cross-validation to optimize parameter estimation; Step S22c: using the sample method to generate samples of the joint distribution of the catchment area and the runoff time from the Bayesian network.

6. The flash flood disaster early warning method based on multi-mode combination optimization according to claim 1, characterized in that: The step S23 is further as follows: Step S23a, constructing a generator and a discriminator of a generative adversarial network model; Step S23b, training the discriminator and the generator in sequence to obtain a trained generative adversarial network model; Step S23c: Based on the joint distribution of the catchment area and the runoff time, a trained generative adversarial network model is used to generate new catchment area and runoff time samples to obtain several flash flood scenarios.

7. The flash flood disaster early warning method based on multi-mode combination optimization according to claim 1, characterized in that: The step S4 is further as follows: Step S41, sequentially fitting each flash flood scenario with the optimal early warning combination to obtain the flash flood disaster process, and obtaining the key flash flood disaster index value based on the flash flood disaster process; Step S42: input the key flash flood disaster index values ​​corresponding to each flash flood scenario into the constructed flash flood disaster forecasting model in turn, and solve the model to obtain the flash flood disaster forecast corresponding to each flash flood scenario.

8. A flash flood disaster warning system based on multi-mode combination optimization, characterized in that: Includes the following modules: Pre-built flash flood disaster forecast model module: collect flash flood disaster data in the study area, use the comprehensive adaptive index screening framework method to identify flash flood disaster indicators, build a key flash flood disaster indicator set, calculate the weights of key flash flood disaster indicators and configure the pre-built flash flood disaster forecast model; Constructing a joint distribution module: extracting the catchment area and confluence time corresponding to each historical flash flood disaster from the historical flash flood disaster data of the study area, fitting the distribution of the catchment area and confluence time respectively, and constructing the joint distribution of the catchment area and confluence time, and randomly sampling to obtain several flash flood scenarios; specifically, extracting the catchment area and confluence time corresponding to each historical flash flood disaster from the historical flash flood disaster data of the study area, and using the kernel density estimation method to fit the distribution of the catchment area and confluence time respectively; based on the distribution of the catchment area and confluence time, using the improved Bayesian network model to construct the joint distribution of the two; using the generative adversarial network for random sampling, several flash flood scenarios were obtained; Solving the optimization model module: construct several warning combinations based on rainfall, water level and rising speed, and use several warning combinations in turn for several flash flood scenarios to fit the flash flood disaster process respectively and input the pre-constructed warning combination optimization model, and solve the optimization model to obtain the optimal warning combination corresponding to each flash flood scenario; specifically, construct seven warning combinations based on rainfall, water level and rising speed, including three single warnings and four combined warnings. The single warnings are: rainfall warning, water level warning and rising speed warning, and the combined warnings are: rainfall-water level combination warning, rainfall rising speed combination warning, water level rising speed combination warning and rainfall-water level rising speed combination warning; construct a warning combination optimization model, and the objective function is that the warning combination has the highest forecast accuracy; for several flash flood scenarios, use seven warning combinations in turn to fit the flash flood disaster process respectively, and obtain the seven disaster-causing processes corresponding to each flash flood scenario; input the seven disaster-causing processes corresponding to each flash flood scenario into the warning combination optimization model, and use Enhanced The MOEA / D algorithm solves the model to obtain the optimal warning combination corresponding to each flash flood scenario, specifically: randomly generate populations, automatically generate weight vectors based on the target number, and adjust the field structure based on the current population characteristics; Dynamically adjust the learning rate based on individual fitness for updating; generate new individuals based on domain individuals, and use the new individuals to update the population; repeatedly iterate until complete convergence, and the obtained warning combination is the optimal warning combination corresponding to the current flash flood scenario; Module for solving flash flood disaster forecasting model: For each flash flood scenario, the flash flood disaster process is obtained by fitting the optimal warning combination, and the key flash flood disaster indicator values ​​are calculated and input into the constructed flash flood disaster forecasting model, and the flash flood disaster forecasting model is solved to obtain the flash flood disaster forecast.

Images