Comprehensive Risk Assessment Method and System for Multi-Level Progressive Rainstorm-Mountain Torrent Chain Disasters

Abstract

The present invention discloses a comprehensive risk assessment method and system for multiple progressive rainstorm-mountain flood chain disasters. The method constructs a multi-band weather radar networking multi-objective optimization model, and couples a radar rainfall measurement simulation model to obtain an optimized networking model. Taking the collected solution set as input, the optimal networking scheme and rainfall forecast are obtained by solving. A flood forecast model is constructed, and the rainfall forecast is extracted as the model input to obtain an initial flood forecast. The initial flood forecast is corrected and optimized to obtain an optimized flood forecast model, and the flood forecast is simulated. Initial mountain flood risk factors are screened to construct a mountain flood risk factor set, and the mountain flood risk factors are dimensionally reduced to obtain a rainstorm-mountain flood chain comprehensive risk index. Based on the flood forecast, the comprehensive risk index value is calculated and input into a mountain flood risk assessment model constructed by using the XGBoost model, and the comprehensive risk forecast is obtained by solving the model. The present invention enhances the scientificity and timeliness of disaster emergency decision-making, and improves the early warning ability and risk management level of rainstorm-mountain flood disasters.

Description

Technical Field

[0001] The present invention relates to the field of flash flood disaster warning, and particularly to a comprehensive risk assessment method and system for multi-element progressive rainstorm-flash flood chain disasters. Background Art

[0002] At present, there are still many technical problems in the risk forecasting and comprehensive prevention and control of rainstorm-flash flood chain disasters in small mountainous watersheds. On the one hand, under the significant spatio-temporal variability of short-term heavy rainfall, the accuracy and response speed of short-term near-term rainfall forecasting are insufficient, and it is difficult to provide accurate rainfall spatial distribution and temporal evolution. On the other hand, the spatial and temporal resolutions of the flash flood disaster forecasting models in the current comprehensive prevention and control system do not meet the requirements of "rapid response and timely response" in disaster prevention. Affected by the multi-disaster chain transfer effect, there is a lack of an effective comprehensive risk assessment method for rainstorm and flash flood chain disasters. Existing models are difficult to accurately capture the interactions between different disasters. Under the requirements of multi-sector collaborative prevention and control, the existing information sharing mechanism and collaborative prevention and control mechanism are not yet perfect, which affects the coordination and efficiency of disaster prevention and control. Summary of the Invention

[0003] Aiming at the deficiencies of the prior art, the purpose of the present invention is to provide a comprehensive risk assessment method and system for multi-element progressive rainstorm-flash flood chain disasters to solve the above problems existing in the prior art.

[0004] The technical solution of the present invention is realized through the following technical solutions:

[0005] A comprehensive risk assessment method for multi-element progressive rainstorm-flash flood chain disasters includes the following steps:

[0006] Step S1: Collect multi-band weather radar data in the study area, extract historical measurement data of the multi-band weather radar in the study area and input it into a pre-constructed radar rainfall simulation model to obtain historical rainfall forecasts, compare them with the real historical rainfall to obtain the forecast errors of the multi-band weather radar in the study area, collect all multi-band weather radar networking schemes, and based on different lead times, respectively screen the multi-band weather radar networking schemes to construct an optimized multi-band weather radar networking scheme set corresponding to different lead times;

[0007] Step S2: Construct a multi-objective optimization model for multi-band weather radar networking, couple it with the radar rainfall simulation model to obtain an optimized multi-band rainfall radar networking model, input the optimized multi-band weather radar networking scheme set corresponding to different lead times into the optimized multi-band rainfall radar networking model, solve the model to obtain the optimal solution, that is, the optimal networking scheme of the multi-band rainfall radar, and obtain rainfall forecasts based on the radar data of the optimal networking scheme of the multi-band rainfall radar;

[0008] Step S3: Construct a flood forecasting model based on the spatial combination mechanism of the AI coupled runoff generation model, extract rainfall forecasts as the model input to obtain initial flood forecasts, correct the initial flood forecasts using no less than two correction models, optimize the parameters of the flood forecasting model by coupling multiple correction results, and input the rainfall forecasts into the optimized flood forecasting model to simulate and obtain flood forecasts;

[0009] Step S4: Collect mountain flood literature, identify initial mountain flood risk factors, calculate the sensitivity of the initial mountain flood risk factors, screen the initial mountain flood risk factors with sensitivity greater than the threshold to construct a mountain flood risk factor set, reduce the dimension of the mountain flood risk factors using the projection pursuit method optimized by t-SNE to obtain the rainstorm-mountain flood chain comprehensive risk index, and construct a mountain flood risk assessment model using the XGBoost model. Calculate the rainstorm-mountain flood chain comprehensive risk index value based on the flood forecast and input it into the mountain flood risk assessment model to solve the model and obtain the comprehensive risk forecast of the rainstorm-mountain flood disaster.

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

[0011] Step S11: Collect multi-band weather radar data in the study area, including: weather radar types, weather radar quantities, weather radar distributions, historical measurement data of multi-band weather radars in the study area, and real historical rainfall data;

[0012] Step S12: Construct a radar rainfall measurement simulation model, extract the historical measurement data of multi-band weather radars in the study area and input it into the radar rainfall measurement simulation model, calculate the historical rainfall forecast, and compare it with the real historical rainfall data to obtain the rainfall forecast error of multi-band weather radars in the study area;

[0013] Step S13: Collect multi-band weather radar networking data, extract all multi-band weather radar networking schemes, and screen the multi-band weather radar networking schemes based on different lead times respectively to construct a set of optimized multi-band weather radar networking schemes corresponding to different lead times. The different lead times include: short-term forecast, medium-short-term forecast, and long-term forecast.

[0014] Specifically, step S2 is further as follows:

[0015] Step S21: Construct a multi-objective optimization model for multi-band weather radar networking, and the objective function is: the highest predictable accuracy and the lowest cost for different lead times;

[0016] Step S22: Couple the radar rainfall measurement simulation model and the multi-objective optimization model for multi-band weather radar networking to obtain an optimized multi-band rainfall measurement radar networking model;

[0017] Step S23: Input the multi-band weather radar optimized networking solution sets corresponding to different prediction periods into the multi-band rainfall radar optimized networking model, and use the comprehensive multi-objective decision-making and priority evaluation method to solve the model to obtain the optimal networking solution for the multi-band rainfall radar; The step S23 is further as follows:

[0018] Step S23a: Determine the weight of each objective in the multi-objective optimization model for the band weather radar networking based on expert opinions, and allocate the weights to the corresponding objectives;

[0019] Step S23b: Determine the ideal solution and the negative ideal solution, calculate the distances between each multi-band weather radar optimized networking solution and the ideal solution and the negative ideal solution in turn, and use the TOPSIS method to select the solution that is closest to the ideal solution and farthest from the negative ideal solution in the multi-band weather radar optimized networking solution set as the preliminary preferred solution;

[0020] Step S23c: Set the minimum acceptable standard of each objective based on the preliminary preferred solution as a constraint condition, and impose it on each objective, re-evaluate each solution in the multi-band weather radar optimized networking solution set, and select the solutions that meet the acceptable ranges of all objectives as candidate solutions;

[0021] Step S23d: For each candidate solution, calculate the score of each candidate solution using the weighted method in turn, and sort them according to the score size, and select the candidate solution with the highest score as the optimal solution, that is, the optimal multi-band rainfall radar optimized networking solution;

[0022] Step S24: Obtain the rainfall forecast based on the radar data of the optimal networking solution of the multi-band rainfall radar.

[0023] Specifically, the step S3 is further as follows:

[0024] Step S31: Construct a flood forecast model based on the space combination mechanism of the AI-coupled runoff generation model; The step S31 is further as follows:

[0025] Step S31a: Construct a hydrological model based on HEC-HMS and a flood prediction model based on a convolutional neural network;

[0026] Step S31b: Couple the hydrological model and the flood prediction model to obtain a flood forecast model, including: a precipitation-runoff module, that is, the HEC-HMS hydrological model, and a flood prediction module, that is, the flood prediction model;

[0027] Step S31c: Determine the over-storage combination mode based on the space combination mechanism of the runoff generation mode, and optimize the parameters of the flood forecast model;

[0028] Step S32: Extract the rainfall forecast and input it into the flood forecast model to obtain the initial flood forecast;

[0029] Step S33: Calibrate the initial flood forecast using a linear regression model, a random forest regression model, and a physical calibration model in sequence to obtain the calibration results of the three calibration models. Couple the calibration results of the three calibration models and optimize the parameters of the flood forecast model. Input the rainfall forecast into the optimized flood forecast model to simulate the flood forecast.

[0030] Specifically, step S33 is further as follows:

[0031] Step S33a: Extract the flow values predicted by the flood forecast model as independent variables, collect the corresponding actual observed flow values as dependent variables, construct a linear regression model using historical data, and input the initial flood forecast into the linear regression model to obtain the statistically calibrated flow prediction values.

[0032] Step S33b: Extract the flow values predicted by the flood forecast model, real-time meteorological data, the characteristics of the study area's watershed, and the actual observed flow to train a random forest model, and use cross-validation to evaluate the model performance. Input the initial flood forecast into the random forest regression model to obtain the machine learning calibrated flow prediction values.

[0033] Step S33c: Use the SCS-CN method to construct a hydrological model, extract the flow prediction values as model inputs, and compare them with the actual observed values to dynamically adjust the parameters of the hydrological model to generate the physically model calibrated flow prediction values.

[0034] Step S33d: Assign weights to the statistically calibrated flow prediction values, the machine learning calibrated flow prediction values, and the physically model calibrated flow prediction values respectively, and adjust the weights based on the actual flow to obtain the comprehensively calibrated flow prediction values and optimize the parameters of the flood forecast model. Input the rainfall forecast into the optimized flood forecast model to simulate the flood forecast.

[0035] Specifically, step S4 is further as follows:

[0036] Step S41: Collect mountain flood literature and use the meta-analysis method to circularly identify mountain flood risk factors, including: hazard factors, disaster-forming factors, and disaster-bearing factors, to construct an initial mountain flood risk factor set.

[0037] Step S42: Calculate the sensitivity of each initial mountain flood risk factor in the initial mountain flood risk factor set in sequence, and extract the initial mountain flood risk factors with sensitivity greater than the threshold as mountain flood risk factors. Use the projection pursuit method optimized by t-SNE to reduce the dimension of the mountain flood risk factors to obtain the rainstorm-mountain flood chain comprehensive risk index.

[0038] Step S43: Construct a mountain flood risk assessment model using the XGBoost model based on the rainstorm-mountain flood chain comprehensive risk index.

[0039] Step S44: Calculate the comprehensive risk index value of the rainstorm-mountain flood chain based on flood forecasting and input it into the mountain flood risk assessment model, and solve the model to obtain the comprehensive risk forecast of the rainstorm-mountain flood chain disaster.

[0040] Specifically, the step S41 is further as follows:

[0041] Step S41a: Collect mountain flood literature, conduct a meta-analysis on the mountain flood literature using the rainstorm-mountain flood chain disaster as the keyword to obtain the mountain flood risk factors corresponding to the rainstorm-mountain flood chain disaster.

[0042] Step S41b: Conduct a meta-analysis on the mountain flood literature using rainstorm, rainstorm-mountain flood runoff, and mountain flood disaster as keywords respectively to obtain the corresponding mountain flood risk factors for each.

[0043] Step S41c: Classify the mountain flood risk factors corresponding to the rainstorm-mountain flood chain disaster according to rainstorm, rainstorm-mountain flood runoff, mountain flood disaster, and others to obtain four categories of mountain flood risk factors. Compare the three categories of mountain flood factors classified based on rainstorm, rainstorm-mountain flood runoff, and mountain flood disaster with the mountain flood risk factors corresponding to rainstorm, rainstorm-mountain flood runoff, and mountain flood disaster respectively, and screen out the same mountain flood factors.

[0044] Step S41d: Extract the mountain flood risk factors obtained from the classification of others, extract keywords using the semantic recognition method, and identify mountain flood literature again using the meta-analysis method based on the extracted keywords. Merge the obtained mountain flood risk factors and the screened-out same mountain flood factors to obtain the initial mountain flood risk factor set.

[0045] Specifically, the step S42 is further as follows:

[0046] Step S42a: Calculate the sensitivity of each initial mountain flood risk factor in the initial mountain flood risk factor set in turn, and extract the initial mountain flood risk factors with sensitivity greater than the threshold as mountain flood risk factors.

[0047] Step S42b: Set the initial objective function, initialize the projection vector and perform data projection, calculate the projection result, and iterate in the loop of the multi-dimensional projection direction to obtain multiple optimal projection directions.

[0048] Step S42c: Extract the low-dimensional data obtained by the projection pursuit method as the input of t-SNE, and set the parameters of t-SNE. The parameters include: balancing local and global data, learning rate, and number of iterations.

[0049] Step S42d: Run t-SNE to obtain the mountain flood risk factors after dimensionality reduction, that is, the comprehensive risk index of the rainstorm-mountain flood chain.

[0050] Specifically, the step S43 is further as follows:

[0051] Step S43a: Construct a mountain flood risk assessment model;

[0052] Step S43b: Use the grid search method to determine the hyperparameters of XGBoost, collect historical mountain flood data in the study area, calculate the corresponding rainstorm-mountain flood chain comprehensive risk index values, and input them into the mountain flood risk assessment model for training;

[0053] Step S43c: Based on the rainstorm-mountain flood chain comprehensive risk index, use the forward cloud generator program to draw the evaluation index cloud diagram and construct the comprehensive cloud model of the evaluation object;

[0054] Step S43d: Calculate the similarity between the parameters of the comprehensive cloud model of the evaluation object and the evaluation index cloud diagram, and extract the parameter configuration of the mountain flood risk assessment model corresponding to the evaluation level with the highest similarity.

[0055] The second aspect of the present invention: A multi - progressive rainstorm - mountain flood chain disaster comprehensive risk assessment system, including the following modules:

[0056] Collection and construction module: Collect multi - band weather radar data in the study area, extract the historical measurement data of the multi - band weather radar in the study area and input it into the pre - constructed radar rainfall simulation model to obtain the historical rainfall forecast, compare it with the real historical rainfall to obtain the forecast error of the multi - band weather radar in the study area, collect all multi - band weather radar networking schemes, and based on different lead times, respectively screen the multi - band weather radar networking schemes to construct the multi - band weather radar optimized networking scheme sets corresponding to different lead times;

[0057] Solution model module: Construct a multi - objective optimization model for multi - band weather radar networking, couple the radar rainfall simulation model to obtain a multi - band rainfall radar optimized networking model, input the multi - band weather radar optimized networking scheme sets corresponding to different lead times into the multi - band rainfall radar optimized networking model, solve the model to obtain the optimal solution, that is, the optimal networking scheme of the multi - band rainfall radar, and obtain the rainfall forecast based on the radar data of the optimal networking scheme of the multi - band rainfall radar;

[0058] Optimization model module: Based on the space combination mechanism of AI - coupled runoff generation models, construct a flood forecast model, extract the rainfall forecast as the model input to obtain the initial flood forecast, use no less than two calibration models to calibrate the initial flood forecast, couple multiple calibration results to optimize the parameters of the flood forecast model, and input the rainfall forecast into the optimized flood forecast model to simulate and obtain the flood forecast;

[0059] Solving risk forecasting module: Collect mountain flood literature, identify initial mountain flood risk factors, calculate the sensitivity of initial mountain flood risk factors, screen out initial mountain flood risk factors with sensitivity greater than the threshold, construct a mountain flood risk factor set, reduce the dimension of mountain flood risk factors using the projection pursuit method optimized based on t-SNE to obtain the rainstorm-mountain flood chain comprehensive risk index, and use the XGBoost model to construct a mountain flood risk assessment model. Based on flood forecasting, calculate the rainstorm-mountain flood chain comprehensive risk index value and input it into the mountain flood risk assessment model, and solve the model to obtain the comprehensive risk forecast of rainstorm-mountain flood disasters.

[0060] The beneficial effects of the present invention are as follows:

[0061] Adopt a multivariate progressive rainstorm-mountain flood chain disaster comprehensive risk assessment method to enhance the scientificity and timeliness of disaster emergency decision-making, and improve the early warning ability and risk management level of rainstorm-mountain flood disasters. Description of the Drawings

[0062] Figure 1 It is the flowchart of the present invention. Detailed Embodiments

[0063] As Figure 1 shown, the following technical solutions are proposed. According to one aspect of the present application, a multivariate progressive rainstorm-mountain flood chain disaster comprehensive risk assessment method is provided, including the following steps:

[0064] Step S1: Collect multi-band weather radar data in the study area, extract the historical measurement data of the multi-band weather radar in the study area and input it into the pre-constructed radar rainfall simulation model to obtain the historical rainfall forecast and compare it with the real historical rainfall to obtain the forecast error of the multi-band weather radar in the study area. Collect all multi-band weather radar networking schemes, and based on different lead times, respectively screen the multi-band weather radar networking schemes to construct a set of optimized multi-band weather radar networking schemes corresponding to different lead times;

[0065] Step S2: Construct a multi-objective optimization model for multi-band weather radar networking, and couple the radar rainfall simulation model to obtain an optimized multi-band rainfall radar networking model. Input the set of optimized multi-band weather radar networking schemes corresponding to different lead times into the optimized multi-band rainfall radar networking model, solve the model to obtain the optimal solution, that is, the optimal networking scheme of the multi-band rainfall radar, and obtain the rainfall forecast based on the radar data of the optimal networking scheme of the multi-band rainfall radar;

[0066] Step S3: Construct a flood forecasting model based on the spatial combination mechanism of AI-coupled runoff generation models, extract the rainfall forecast as the model input to obtain the initial flood forecast, use no less than two correction models to correct the initial flood forecast, couple multiple correction results to optimize the parameters of the flood forecasting model, and input the rainfall forecast into the optimized flood forecasting model to simulate and obtain the flood forecast;

[0067] Step S4: Collect mountain flood literature, identify initial mountain flood risk factors, calculate the sensitivity of the initial mountain flood risk factors, screen out the initial mountain flood risk factors with sensitivity greater than the threshold, construct a mountain flood risk factor set, reduce the dimension of the mountain flood risk factors by using the projection pursuit method optimized by t-SNE to obtain a rainstorm-mountain flood chain comprehensive risk index, and use the XGBoost model to construct a mountain flood risk assessment model. Calculate the value of the rainstorm-mountain flood chain comprehensive risk index based on flood forecasting and input it into the mountain flood risk assessment model, and solve the model to obtain the comprehensive risk forecast of the rainstorm-mountain flood disaster.

[0068] High spatio-temporal resolution precipitation information is the basis for flash flood early warning and refined forecasting in small watersheds. Weather radars can achieve area monitoring, make up for the deficiencies of ground stations, and have unique advantages in precipitation monitoring and short-term and nowcasting in mountainous small watersheds. Radars have been widely used in disaster weather monitoring and other aspects. Many countries choose S-band and C-band radars as important observation instruments for weather monitoring and weather forecasting to form a national radar monitoring network. However, due to the influence of the earth's curvature and terrain occlusion, during the detection process of S-band and C-band radars near an altitude of 1 kilometer, only 30% of the weather targets in the area can be captured, seriously affecting refined detection and weather feature identification. X-band radars can overcome the limitations of the resolution and detection range of traditional weather radar networks and have great advantages in detecting low-altitude mesoscale systems.

[0069] Therefore, the present invention introduces distributed cooperative adaptive observation, forms an X-band radar observation network to achieve high spatio-temporal resolution observation, which will be more advantageous in capturing rapidly developing and moving severe convective systems. Focusing on the requirements of refined monitoring and forecasting of mountain flood rainfall, it is carried out from multiple aspects such as S- and X-band radar precipitation inversion, short-term and nowcasting technology, and identification of characteristics of short-term heavy rainfall, so as to improve the forecasting and early warning ability of heavy precipitation in mountainous small watersheds and achieve precipitation monitoring and forecasting with high spatio-temporal resolution.

[0070] According to one aspect of the present application, the step S1 is further as follows:

[0071] Step S11: Collect multi-band weather radar data in the study area, including: types of weather radars, numbers of weather radars, distributions of weather radars, historical measurement data of multi-band weather radars in the study area, and real historical rainfall data;

[0072] Step S12: Construct a radar rainfall simulation model, extract the historical measurement data of multi-band weather radars in the study area and input it into the radar rainfall simulation model, calculate the historical rainfall forecast, and compare it with the real historical rainfall data to obtain the forecast error of multi-band weather radars in the study area;

[0073] Step S13: Collect the data of multi-band weather radar networking, extract all the multi-band weather radar networking schemes, and screen the multi-band weather radar networking schemes based on different forecast periods respectively to construct the optimized multi-band weather radar networking scheme sets corresponding to different forecast periods. The different forecast periods include: short-term forecast, medium-short-term forecast, and long-term forecast.

[0074] According to one aspect of the present application, the step S2 is further as follows:

[0075] Step S21: Construct a multi-objective optimization model for multi-band weather radar networking, and the objective function is: the highest predictable accuracy and the lowest cost for different forecast periods;

[0076] Step S22: Couple the radar rainfall simulation model and the multi-objective optimization model for multi-band weather radar networking to obtain an optimized multi-band rainfall radar networking model;

[0077] Step S23: Input the optimized multi-band weather radar networking scheme sets corresponding to different forecast periods into the optimized multi-band rainfall radar networking model, and use the comprehensive multi-objective decision-making and priority evaluation method to solve the model to obtain the optimal networking scheme of the multi-band rainfall radar;

[0078] Step S24: Obtain the rainfall forecast based on the radar data of the optimal networking scheme of the multi-band rainfall radar.

[0079] According to one aspect of the present application, the step S23 is further as follows:

[0080] Step S23a: Determine the weight of each objective in the multi-objective optimization model for multi-band weather radar networking based on expert opinions, and assign the weights to the corresponding objectives;

[0081] Step S23b: Determine the ideal solution and the negative ideal solution, calculate the distances between each optimized multi-band weather radar networking scheme and the ideal solution and the negative ideal solution in turn, and use the TOPSIS method to select the solution that is closest to the ideal solution and farthest from the negative ideal solution in the optimized multi-band weather radar networking scheme set as the preliminary preferred solution;

[0082] Step S23c: Set the minimum acceptable standard of each objective based on the preliminary preferred solution as a constraint condition, and impose it on each objective, re-evaluate each solution in the optimized multi-band weather radar networking scheme set, and select the solutions that meet the acceptable ranges of all objectives as candidate solutions;

[0083] Step S23d: For each candidate solution, calculate the scores of each candidate solution in turn using the weighted method, sort them according to the scores, and select the candidate solution with the highest score as the optimal solution, that is, the optimal networking scheme of the multi-band rainfall radar.

[0084] The comprehensive multi-objective decision-making and priority evaluation method is a comprehensive method that combines multiple decision-making techniques, integrating the priority ranking method, the weighting method, TOPSIS, and the ε-constraint method. By defining decision-making objectives and criteria, constructing a non-dominated solution set, priority ranking and weight assignment, distance evaluation and TOPSIS implementation, and applying the ε-constraint method, the optimal solution is finally obtained. This method can combine different decision-making methods and evaluation techniques to achieve a comprehensive evaluation and effective selection of multiple objectives. The COMPA method can synthesize multiple different indicators into a unified score, enabling decision-makers to clearly see the advantages and disadvantages of different solutions at a glance, and is particularly suitable for the analysis of complex problems. This method allows different indicators and weights to be set according to specific problems, can adapt to different decision-making backgrounds and requirements, and enhances applicability.

[0085] According to one aspect of the present application, the step S3 is further as follows:

[0086] Step S31: Construct a flood forecasting model based on the spatial combination mechanism of the AI-coupled runoff generation model;

[0087] Step S32: Extract rainfall forecasts and input them into the flood forecasting model to obtain initial flood forecasts;

[0088] Step S33: Successively correct the initial flood forecasts using a linear regression model, a random forest regression model, and a physical correction model to obtain the correction results of the three correction models. Couple the correction results of the three correction models and optimize the parameters of the flood forecasting model. Input the rainfall forecasts into the optimized flood forecasting model to simulate and obtain flood forecasts.

[0089] According to one aspect of the present application, the step S31 is further as follows:

[0090] Step S31a: Construct a hydrological model based on HEC-HMS and a flood prediction model based on a convolutional neural network;

[0091] Step S31b: Couple the hydrological model and the flood prediction model to obtain a flood forecasting model, including: a precipitation-runoff module, i.e., the HEC-HMS hydrological model, and a flood prediction module, i.e., the flood prediction model;

[0092] Step S31c: Determine the over-storage combination mode based on the spatial combination mechanism of the runoff generation model and optimize the parameters of the flood forecasting model.

[0093] According to one aspect of the present application, the step S33 is further as follows:

[0094] Step S33a: Extract the flow values predicted by the flood forecasting model as independent variables, collect the corresponding actual observed flow values as dependent variables, construct a linear regression model using historical data, and input the initial flood forecast into the linear regression model to obtain the statistically corrected flow prediction values;

[0095] Step S33b: Extract the flow values predicted by the flood forecasting model, real-time meteorological data, the characteristics of the study area's watershed, and the actual observed flow to train a random forest model, and use cross-validation to evaluate the model performance. Input the initial flood forecast into the random forest regression model to obtain the machine learning-corrected flow prediction values;

[0096] Step S33c: Construct a hydrological model using the SCS-CN method, extract the flow prediction values as model inputs, and compare them with the actual observed values to dynamically adjust the hydrological model parameters to generate the physically model-corrected flow prediction values;

[0097] Step S33d: Assign weights to the statistically corrected flow prediction values, the machine learning-corrected flow prediction values, and the physically model-corrected flow prediction values respectively, and adjust the weights based on the actual flow to obtain the comprehensively corrected flow prediction values and optimize the flood forecasting model parameters. Input the rainfall forecast into the optimized flood forecasting model to simulate the flood forecast.

[0098] According to one aspect of the present application, the step S4 is further as follows:

[0099] Step S41: Collect mountain flood literature, and use the meta-analysis method to cyclically identify mountain flood risk factors, including: hazard factors, disaster-forming factors, and disaster-bearing factors, to construct an initial mountain flood risk factor set;

[0100] Step S42: Calculate the sensitivity of each initial mountain flood risk factor in the initial mountain flood risk factor set in turn, and extract the initial mountain flood risk factors with sensitivity greater than the threshold as mountain flood risk factors. Use the projection pursuit method optimized by t-SNE to reduce the dimension of the mountain flood risk factors to obtain the rainstorm-mountain flood chain comprehensive risk index;

[0101] Step S43: Construct a mountain flood risk assessment model using the XGBoost model based on the rainstorm-mountain flood chain comprehensive risk index;

[0102] Step S44: Calculate the rainstorm-mountain flood chain comprehensive risk index value based on the flood forecast and input it into the mountain flood risk assessment model, and solve the model to obtain the rainstorm-mountain flood chain disaster comprehensive risk forecast.

[0103] According to one aspect of the present application, the step S41 is further as follows:

[0104] Step S41a: Collect mountain flood literature, conduct a meta-analysis of the mountain flood literature using rainstorm-mountain flood chain disasters as keywords, and obtain the mountain flood risk factors corresponding to rainstorm-mountain flood chain disasters;

[0105] Step S41b: Conduct a meta-analysis of the mountain flood literature using rainstorm, rainstorm-mountain flood runoff, and mountain flood disasters as keywords in sequence, and obtain the corresponding mountain flood risk factors for each;

[0106] Step S41c: Classify the mountain flood risk factors corresponding to rainstorm-mountain flood chain disasters according to rainstorm, rainstorm-mountain flood runoff, mountain flood disasters, and others to obtain four categories of mountain flood risk factors. Compare the three categories of mountain flood factors classified based on rainstorm, rainstorm-mountain flood runoff, and mountain flood disasters with the mountain flood risk factors corresponding to rainstorm, rainstorm-mountain flood runoff, and mountain flood disasters respectively, and screen out the same mountain flood factors;

[0107] Step S41d: Extract the mountain flood risk factors obtained by classification based on others, extract keywords using the semantic recognition method, identify mountain flood literature again using the meta-analysis method based on the extracted keywords, and merge the obtained mountain flood risk factors and the screened-out same mountain flood factors to obtain the initial mountain flood risk factor set.

[0108] For mountain flood disasters, according to the availability and relevance of data, select elevation, slope, aspect, plane curvature, stream power index, terrain wetness index, sediment transport index, land use type, soil type, normalized difference vegetation index, lithology, distance from the river, average annual maximum 24-hour precipitation, and population density as mountain flood disaster impact factors. Considering the differences in the selection of mountain flood disaster impact factors due to different research areas, determine the final mountain flood disaster impact factors through multiple collinearity analysis and feature selection;

[0109] The index factors affecting landslide development mainly include topography, geological conditions, and human engineering activities. In this embodiment, based on the distinct geographical conditions, select land use, slope structure, rock formation, slope, distance from the river channel, thickness of the accumulation layer, vegetation normalized index, slope form, and catchment area to construct an evaluation index system for the susceptibility of the study area. To study the correlation between multiple variables, the present invention uses multiple collinearity, which means there is an exact correlation or a high degree of correlation between the variables in the model. In the modeling of the susceptibility of mountain flood and landslide disasters, if there is a correlation between the disaster impact factors, it is intuitively not easy to distinguish the individual impact of the impact factors on the susceptibility of mountain flood disasters, and in terms of calculation, it will make the variance of the model parameter estimates larger, resulting in poor model stability. To eliminate the factors that may have multiple collinearity, perform multiple collinearity analysis using the variance inflation factor, and the calculation formula is: ; where: is the variance inflation factor; is the correlation coefficient of the regression equation fitted by taking the i-th mountain flood disaster impact factor as the dependent variable and the remaining mountain flood disaster impact factors as independent variables; The closer the value is to 1, the weaker the multicollinearity; conversely, the stronger it is. Taking 10 as the judgment threshold, when F < 10, it can be considered that there is no multicollinearity; otherwise, there is multicollinearity.

[0110] According to one aspect of the present application, the step S42 is further as follows:

[0111] Step S42a: Calculate the sensitivity of each initial mountain flood risk factor in the initial mountain flood risk factor set in turn, and extract the initial mountain flood risk factors with sensitivity greater than the threshold as mountain flood risk factors;

[0112] Step S42b: Set the initial objective function, initialize the projection vector and perform data projection, calculate the projection result, and iterate in the loop of the multi-dimensional projection direction to obtain multiple optimal projection directions;

[0113] Step S42c: Extract the low-dimensional data obtained by the projection pursuit method as the input of t-SNE, and set the parameters of t-SNE. The parameters include: balancing local and global data, learning rate, and number of iterations;

[0114] Step S42d: Run t-SNE to obtain the mountain flood risk factors after dimensionality reduction, that is, the rainstorm-mountain flood chain comprehensive risk index.

[0115] According to one aspect of the present application, the step S43 is further as follows:

[0116] Step S43a: Construct a mountain flood risk assessment model;

[0117] Step S43b: Use the grid search method to determine the hyperparameters of XGBoost, collect historical mountain flood data in the study area, calculate the corresponding rainstorm-mountain flood chain comprehensive risk index values, and input them into the mountain flood risk assessment model for training;

[0118] Step S43c: Draw an evaluation index cloud diagram based on the rainstorm-mountain flood chain comprehensive risk index by using a forward cloud generator program, and construct a comprehensive cloud model of the evaluation object;

[0119] Step S43d: Calculate the similarity between the parameters of the comprehensive cloud model of the evaluation object and the evaluation index cloud diagram, and extract the parameter configuration of the evaluation level with the highest similarity to configure the mountain flood risk assessment model. The XGBoost algorithm can be attributed to "solving an optimization problem". In order to obtain the optimal solution, a criterion needs to be constructed. This algorithm is jointly decided by multiple related CART trees. Based on this integration, the prediction result of a certain sample can be expressed as: ; where: For the sample The prediction result; is the total number of decision trees; is the prediction result of the k-th tree; Based on this, the objective function can be expressed as:

[0120] ; In the formula: For the sample The prediction error; is the regularization term of the k-th tree; There is a correlation between decision trees. From the perspective of model optimization, the prediction error of the previous tree will become the reference value for the next tree. According to the above formula, when k = t, the prediction result of the sample is: ; Then the objective function can be expressed as: ; The regularization term is considered in units of a certain tree. After specifying the regularization term and substituting it into the objective function, we get: .

[0121] In this embodiment, the weights of geological disaster factors are determined by the XGBoost algorithm. On this basis, the cloud model is used to convert the fuzziness problem of factor grading into a quantitative problem. The cloud model uses three numerical characteristics, namely expectation, entropy, and hyperentropy, to quantitatively describe an uncertain concept. The cloud generator is the specific implementation method of the cloud. Among them, the algorithm for generating cloud droplets from the numerical characteristics of the cloud is called the forward generator, which is used to realize the uncertain conversion between quantitative values and their qualitative language values. Conversely, the algorithm for effectively converting a series of precise data into a qualitative concept represented by numerical characteristics is called the reverse cloud generation.

[0122] In a certain embodiment, specifically:

[0123] Suppose there are n clouds in the set , and the weight vector is . If the indicators are independent of each other and the correlation is small, then the comprehensive cloud algorithm in the virtual cloud theory is used. The formula for the comprehensive cloud is as follows:

[0124] ; In the formula, is the weight of indicator i; , , are the numerical characteristic parameters of the corresponding indicator cloud model; The cloud model parameters of the evaluation object are compared with the evaluation indicator cloud, and the cloud vectors and are defined respectively. The cosine angle between them is called the similarity , and the formula is as follows:

[0125] ; Use the above formula to calculate the similarity between the evaluation index cloud and the evaluation index cloud, and according to the principle of maximum similarity, the evaluation level corresponding to the highest similarity is the final evaluation level.

[0126] According to another aspect of the present application, a multi - level progressive rainstorm - mountain flood chain disaster comprehensive risk assessment system, characterized by including the following modules:

[0127] Collection and construction module: Collect multi - band weather radar data in the study area, extract historical measurement data of multi - band weather radar in the study area and input it into a pre - constructed radar rainfall simulation model to obtain historical rainfall forecasts, compare with the real historical rainfall to obtain the forecast error of multi - band weather radar in the study area, collect all multi - band weather radar networking schemes, and based on different lead times, screen the multi - band weather radar networking schemes respectively to construct an optimized multi - band weather radar networking scheme set corresponding to different lead times;

[0128] Solution model module: Construct a multi - objective optimization model for multi - band weather radar networking, couple it with the radar rainfall simulation model to obtain an optimized multi - band rainfall radar networking model, input the optimized multi - band weather radar networking scheme set corresponding to different lead times into the optimized multi - band rainfall radar networking model, solve the model to obtain the optimal solution, that is, the optimal networking scheme of multi - band rainfall radar, and obtain rainfall forecasts based on the radar data of the optimal networking scheme of multi - band rainfall radar;

[0129] Optimization model module: Construct a flood forecast model based on the spatial combination mechanism of AI - coupled runoff generation patterns, extract rainfall forecasts as model inputs to obtain initial flood forecasts, use no less than two correction models to correct the initial flood forecasts, couple multiple correction results to optimize the parameters of the flood forecast model, input the rainfall forecasts into the optimized flood forecast model, and simulate to obtain flood forecasts;

[0130] Solution risk forecast module: Collect mountain flood literature, identify initial mountain flood risk factors, calculate the sensitivity of initial mountain flood risk factors, screen the initial mountain flood risk factors with sensitivity greater than the threshold, construct a mountain flood risk factor set, use the projection pursuit method optimized based on t - SNE to reduce the dimension of mountain flood risk factors to obtain rainstorm - mountain flood chain comprehensive risk indicators, and use the XGBoost model to construct a mountain flood risk assessment model. Based on the flood forecasts, calculate the rainstorm - mountain flood chain comprehensive risk indicator values and input them into the mountain flood risk assessment model, and solve the model to obtain the comprehensive risk forecast of rainstorm - mountain flood chain disasters.

[0131] The present invention also provides an electronic device,

[0132] At least one processor; and

[0133] A memory communicatively connected to at least one of the processors; wherein,

[0134] The memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the multi-level progressive rainstorm-mountain flood chain disaster comprehensive risk assessment method described in any one of the above.

[0135] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of 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 fall within the protection scope of the present invention.

Claims

1. A multivariate progressive rainstorm-flash flood chain disaster comprehensive risk assessment method, characterized in that: The steps include: Step S1, collecting multi-band weather radar data in the study area, extracting historical measurement data of multi-band weather radar in the study area and inputting it into a pre-built radar rainfall measurement simulation model, obtaining historical rainfall forecasts and comparing them with actual historical rainfall to obtain the forecast errors of multi-band weather radar in the study area, collecting all multi-band weather radar networking schemes, and screening multi-band weather radar networking schemes based on different forecast periods, and constructing a set of multi-band weather radar optimization networking schemes corresponding to different forecast periods; Step S2, constructing a multi-band weather radar networking multi-objective optimization model, and coupling the radar rain measurement simulation model to obtain a multi-band rain measurement radar optimization networking model, inputting a multi-band weather radar optimization networking solution set corresponding to different forecast periods into the multi-band rain measurement radar optimization networking model, solving the model to obtain an optimal solution, that is, a multi-band rain measurement radar optimal networking solution, and obtaining a rainfall forecast based on radar data of the multi-band rain measurement radar optimal networking solution; the step S2 is further as follows: Step S21, constructing a multi-objective optimization model for a multi-band weather radar network, wherein the objective function is: the highest predictable accuracy and the lowest cost for different forecast periods; Step S22, coupling the radar rain measurement simulation model and the multi-band weather radar networking multi-objective optimization model to obtain a multi-band rain measurement radar optimization networking model; Step S23, inputting the multi-band weather radar optimization networking solution set corresponding to different forecast periods into the multi-band rain measuring radar optimization networking model, and solving the model by using the comprehensive multi-objective decision-making and priority evaluation method to obtain the multi-band rain measuring radar optimal networking solution; the step S23 is further as follows: Step S23a, determining the weight of each target in the multi-target optimization model of the multi-band weather radar network based on expert opinions, and assigning the weight to the corresponding target; Step S23b, determining an ideal solution and a negative ideal solution, calculating the distance between each multi-band weather radar optimization networking solution and the ideal solution and the negative ideal solution in turn, and using the TOPSIS method to select a solution that is closest to the ideal solution and farthest from the negative ideal solution from the multi-band weather radar optimization networking solution set as a preliminary optimal solution; Step S23c, based on the preliminary optimal solution, the minimum acceptable standard of each target is set as a constraint condition, and is applied to each target, and each solution in the multi-band weather radar optimization networking solution set is re-evaluated, and the solution that meets the acceptable range of all targets is selected as a candidate solution; Step S23d, for each candidate solution, a weighted method is used to calculate the score of each candidate solution in turn, and the candidate solutions with the highest scores are selected as the optimal solution, that is, the optimal networking solution of the multi-band rainfall radar; Step S24, obtaining rainfall forecast based on radar data of the optimal networking solution of multi-band rainfall radar; Step S3, constructing a flood forecasting model based on the spatial combination mechanism of AI-coupled runoff patterns, extracting rainfall forecast as model input, obtaining an initial flood forecast, using no less than two correction models to correct the initial flood forecast, coupling multiple correction results to optimize flood forecasting model parameters, inputting rainfall forecast into the optimized flood forecasting model, and simulating to obtain a flood forecast; the step S3 is further: Step S31, constructing a flood forecasting model based on the spatial combination mechanism of AI coupled runoff patterns; the step S31 is further as follows: Step S31a, constructing a hydrological model based on HEC-HMS, and constructing a flood prediction model based on a convolutional neural network; Step S31b, coupling the hydrological model and the flood prediction model to obtain a flood forecast model, including: a precipitation-runoff module, i.e., a HEC-HMS hydrological model, and a flood prediction module, i.e., a flood prediction model; Step S31c, determining the super-storage combination mode based on the spatial combination mechanism of runoff generation modes, and optimizing the flood forecasting model parameters; Step S32, extracting rainfall forecast and inputting it into the flood forecast model to obtain an initial flood forecast; Step S33, sequentially using a linear regression model, a random forest regression model and a physical correction model to correct the initial flood forecast respectively, obtaining correction results of the three correction models, coupling the correction results of the three correction models and optimizing the flood forecast model parameters, inputting the rainfall forecast into the optimized flood forecast model, and simulating to obtain a flood forecast; Step S4, collect flash flood literature, identify initial flash flood risk factors, calculate the sensitivity of initial flash flood risk factors, and screen initial flash flood risk factors with sensitivity greater than a threshold, construct a flash flood risk factor set, use the projection pursuit method based on t-SNE optimization to reduce the dimension of the flash flood risk factors, obtain the rainstorm-flash flood chain comprehensive risk index, and use the XGBoost model to build a flash flood risk assessment model, calculate the rainstorm-flash flood chain comprehensive risk index value based on flood forecasts and input it into the flash flood risk assessment model, and solve the model to obtain the rainstorm-flash flood chain disaster comprehensive risk forecast.

2. The multivariate progressive rainstorm-flash flood chain disaster comprehensive risk assessment method as claimed in claim 1, characterized in that: The step S1 is further as follows: Step S11, collecting multi-band weather radar data of the study area, including: weather radar type, number of weather radars, weather radar distribution, historical measurement data of multi-band weather radars in the study area, and real historical rainfall data; Step S12, constructing a radar rainfall measurement simulation model, extracting historical measurement data of multi-band weather radar in the study area and inputting it into the radar rainfall measurement simulation model, calculating the historical rainfall forecast, and comparing it with the real historical rainfall data to obtain the forecast error of the multi-band weather radar in the study area; Step S13, collecting multi-band weather radar networking data, extracting all multi-band weather radar networking solutions, and screening multi-band weather radar networking solutions based on different forecast periods, and constructing a multi-band weather radar optimization networking solution set corresponding to different forecast periods, wherein the different forecast periods include: short-term forecast, medium-term forecast and long-term forecast.

3. The multivariate progressive rainstorm-flash flood chain disaster comprehensive risk assessment method as claimed in claim 1, characterized in that: The step S33 is further as follows: Step S33a, extracting the flow value predicted by the flood forecasting model as an independent variable, collecting the corresponding actual observed flow value as a dependent variable, using historical data to build a linear regression model, and using the initial flood forecast as an input to the linear regression model to obtain a statistically corrected flow forecast value; Step S33b, extracting the flow value predicted by the flood forecasting model, real-time meteorological data, basin characteristics of the study area and actual observed flow to train the random forest regression model, and using cross-validation to evaluate the model performance, inputting the initial flood forecast into the random forest regression model to obtain the flow forecast value corrected by machine learning; Step S33c, constructing a hydrological model using the SCS-CN method, extracting the flow prediction value as the model input, and comparing it with the actual observed value, dynamically adjusting the hydrological model parameters, and generating the flow prediction value after the physical model correction; Step S33d, respectively weight the flow prediction value after statistical correction, the flow prediction value corrected by machine learning, and the flow prediction value corrected by the physical model, and adjust the weights based on the actual flow to obtain the comprehensive corrected flow prediction value and optimize the flood forecast model parameters, input the rainfall forecast into the optimized flood forecast model, and simulate the flood forecast.

4. The multi-element progressive rainstorm-flash flood chain disaster comprehensive risk assessment method as claimed in claim 1, characterized in that: The step S4 is further as follows: Step S41, collecting flash flood literature, using meta-analysis method to cyclically identify flash flood risk factors, including: disaster-causing factors, disaster-pregnant factors and disaster-bearing factors, and constructing an initial flash flood risk factor set; Step S42, sequentially calculating the sensitivity of each initial flash flood risk factor in the initial flash flood risk factor set, extracting the initial flash flood risk factors with a sensitivity greater than a threshold as flash flood risk factors, and using the projection pursuit method based on t-SNE optimization to reduce the dimension of the flash flood risk factors to obtain a rainstorm-flash flood chain comprehensive risk index; Step S43, constructing a flash flood risk assessment model using an XGBoost model based on the rainstorm-flash flood chain comprehensive risk index; Step S44, calculating the rainstorm-flash flood chain comprehensive risk index value based on the flood forecast and inputting it into the flash flood risk assessment model, solving the model to obtain the rainstorm-flash flood chain disaster comprehensive risk forecast.

5. The multi-element progressive rainstorm-flash flood chain disaster comprehensive risk assessment method as claimed in claim 4, characterized in that: The step S41 is further as follows: Step S41a, collecting flash flood literature, using rainstorm-flash flood chain disaster as a keyword to conduct a meta-analysis of the flash flood literature, and obtaining the flash flood risk factor corresponding to the rainstorm-flash flood chain disaster; Step S41b, using rainstorm, rainstorm-flash flood runoff and flash flood disaster as keywords, respectively, to conduct meta-analysis on flash flood literature, and obtain the flash flood risk factors corresponding to each; Step S41c, classify the flash flood risk factors corresponding to the rainstorm-flash flood chain disaster according to rainstorm, rainstorm-flash flood runoff, flash flood disaster and others to obtain four types of flash flood risk factors, and compare the three types of flash flood factors classified based on rainstorm, rainstorm-flash flood runoff and flash flood disaster with the flash flood risk factors corresponding to rainstorm, rainstorm-flash flood runoff and flash flood disaster, and screen out the same flash flood factors; Step S41d, extract the flash flood risk factors obtained based on other classifications, and use the semantic recognition method to extract keywords. Based on the extracted keywords, the meta-analysis method is used again to identify flash flood literature. The obtained flash flood risk factors and the screened out identical flash flood factors are merged to obtain an initial flash flood risk factor set.

6. The multi-element progressive rainstorm-flash flood chain disaster comprehensive risk assessment method as claimed in claim 4, characterized in that: The step S42 is further as follows: Step S42a, sequentially calculating the sensitivity of each initial flash flood risk factor in the initial flash flood risk factor set, and extracting the initial flash flood risk factor with a sensitivity greater than a threshold as the flash flood risk factor; Step S42b, setting an initial objective function, initializing a projection vector and performing data projection, calculating a projection result, iterating in a loop of multi-dimensional projection directions, and obtaining multiple optimal projection directions; Step S42c, extracting low-dimensional data obtained by the projection pursuit method as the input of t-SNE, and setting the parameters of t-SNE, wherein the parameters include: balancing local and global data, learning rate, and number of iterations; Step S42d, run t-SNE to obtain the flash flood risk factor after dimensionality reduction, that is, the rainstorm-flash flood chain comprehensive risk index.

7. The multi-element progressive rainstorm-flash flood chain disaster comprehensive risk assessment method as claimed in claim 4, characterized in that: The step S43 is further as follows: Step S43a, constructing a flash flood risk assessment model; Step S43b, using the grid search method to determine the hyperparameters of XGBoost, collecting historical flash flood data in the study area, calculating the corresponding rainstorm-flash flood chain comprehensive risk index value, and inputting it into the flash flood risk assessment model for training; Step S43c, based on the rainstorm-flash flood chain comprehensive risk index, a forward cloud generator program is used to draw an evaluation index cloud map, and a comprehensive cloud model of the evaluation object is constructed; Step S43d, calculating the similarity between the comprehensive cloud model parameters of the evaluation object and the evaluation index cloud map, extracting the parameters corresponding to the evaluation level with the highest similarity and configuring the flash flood risk assessment model.

8. A multi-element progressive rainstorm-flash flood chain disaster comprehensive risk assessment system, characterized in that: Includes the following modules: Collection and construction module: Collect multi-band weather radar data in the study area, extract historical measurement data of multi-band weather radar in the study area and input it into the pre-built radar rainfall measurement simulation model, obtain historical rainfall forecast and compare it with the actual historical rainfall to obtain the forecast error of multi-band weather radar in the study area, collect all multi-band weather radar networking schemes, and select multi-band weather radar networking schemes based on different forecast periods, and construct a set of multi-band weather radar optimization networking schemes corresponding to different forecast periods; Solution model module: construct a multi-band weather radar networking multi-objective optimization model, and couple the radar rainfall measurement simulation model to obtain the multi-band rainfall measurement radar optimization networking model, input the multi-band weather radar optimization networking solution set corresponding to different forecast periods into the multi-band rainfall measurement radar optimization networking model, solve the model to obtain the optimal solution, that is, the multi-band rainfall measurement radar optimal networking solution, and obtain the rainfall forecast based on the radar data of the multi-band rainfall measurement radar optimal networking solution; specifically: construct a multi-band weather radar networking multi-objective optimization model, and the objective function is: the highest predictable accuracy and lowest cost for different forecast periods; couple the radar rainfall measurement simulation model and the multi-band weather radar networking multi-objective optimization model to obtain the multi-band rainfall measurement radar optimization networking model; input the multi-band weather radar optimization networking solution set corresponding to different forecast periods into the multi-band rainfall measurement radar optimization networking model, and use the comprehensive multi-objective decision-making and priority evaluation method to solve the model to obtain the multi-band rainfall measurement radar. The optimal networking scheme for rain radars is to determine the weight of each target in the multi-target optimization model of multi-band weather radar networking based on expert opinions, and assign the weights to the corresponding targets, determine the ideal solution and the negative ideal solution, calculate the distance between each multi-band weather radar optimization networking scheme and the ideal solution and the negative ideal solution in turn, and use the TOPSIS method to select the solution closest to the ideal solution and farthest from the negative ideal solution from the multi-band weather radar optimization networking scheme set as the preliminary optimal solution. Based on the preliminary optimal solution, the minimum acceptable standard for each target is set as a constraint condition and applied to each target, and each solution in the multi-band weather radar optimization networking scheme set is re-evaluated, and the solution that meets the acceptable range of all targets is selected as a candidate solution; for each candidate solution, the weighted method is used in turn to calculate the score of each candidate solution, and the candidate solution is sorted according to the score size, and the candidate solution with the highest score is selected as the optimal solution, that is, the optimal networking scheme for multi-band rain radars; Rainfall forecast is obtained based on radar data of the optimal networking solution of multi-band rain radar; Optimization model module: construct a flood forecasting model based on the spatial combination mechanism of AI-coupled runoff patterns, extract rainfall forecast as model input, obtain initial flood forecast, use no less than two correction models to correct the initial flood forecast, couple multiple correction results to optimize flood forecasting model parameters, input rainfall forecast into the optimized flood forecasting model, and simulate flood forecast; specifically: construct a hydrological model based on HEC-HMS, construct a flood prediction model based on convolutional neural network, couple the hydrological model and flood prediction model to obtain a flood forecasting model, including: precipitation-runoff module , namely the HEC-HMS hydrological model, and the flood prediction module, namely the flood prediction model, determine the super-storage combination mode based on the spatial combination mechanism of runoff patterns, and optimize the parameters of the flood forecast model; extract the rainfall forecast and input it into the flood forecast model to obtain the initial flood forecast; use the linear regression model, random forest regression model and physical correction model to correct the initial flood forecast respectively, and obtain the correction results of the three correction models, couple the correction results of the three correction models and optimize the parameters of the flood forecast model, input the rainfall forecast into the optimized flood forecast model, and simulate the flood forecast; Solving the risk forecasting module: collect flash flood literature, identify initial flash flood risk factors, calculate the sensitivity of initial flash flood risk factors, and screen initial flash flood risk factors with sensitivity greater than the threshold, construct a flash flood risk factor set, and use the projection pursuit method based on t-SNE optimization to reduce the dimension of flash flood risk factors to obtain the rainstorm-flash flood chain comprehensive risk index, and use the XGBoost model to build a flash flood risk assessment model. Based on flood forecasts, the rainstorm-flash flood chain comprehensive risk index value is calculated and input into the flash flood risk assessment model, and the model is solved to obtain the comprehensive risk forecast of the rainstorm-flash flood chain disaster.

Images