Flood early warning method based on rainfall multi-layer perception

Abstract

The invention belongs to the technical field of meteorological disaster prediction, and discloses a flood early warning method based on rainfall multilayer perception. Comprising the steps of collecting meteorological parameter data, terrain parameter data, hydrological soil data and human activity data; preprocessing the collected meteorological parameter data, topographic parameter data, hydrological soil data and artificial activity data to obtain a meteorological feature data set, a topographic feature data set, a hydrological soil feature data set and an artificial activity feature data set; fusing the meteorological feature data set, the topographic feature data set, the hydrological soil feature data set and the human activity feature data set to obtain a comprehensive feature data set; training according to the comprehensive feature data set to obtain a flood risk prediction model, and predicting through the flood risk prediction model to obtain a flood risk index; the predicted flood risk index is compared with a preset flood risk index threshold value, and whether flood early warning needs to be carried out or not is judged; and a worker is helped to accurately make flood control measures.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological disaster prediction, and more specifically, to a flood warning method based on multi-layer rainfall perception. Background Art

[0002] The patent with the patent announcement number CN118297442A discloses a flood warning method for urban underpass tunnels based on a multi-layer perception model, which relates to the field of urban management technology. The present invention is based on the existing water environment science research system and uses a multi-layer perception model to overcome the high hardware requirements of the traditional flood diffusion mechanism model; obtains municipal rainfall monitoring data and flood drainage network GIS layers to more accurately depict the spatiotemporal characteristics of rainfall water body diffusion in the pipe streets, and realizes urban rainfall flood warning and quantitative analysis in a small-scale environment; according to rainfall and geographical distance, quickly and accurately locks the location of the underpass tunnel where the flood occurs, helping to realize the refinement, automation and intelligence of flood control scheduling, and avoiding the lag of flood control measures and improving timeliness.

[0003] The existing flood warning methods based on multi-layer rainfall perception have the following main problems: The meteorological time series is not processed through autocorrelation function analysis and adaptive delay adjustment. The dynamic change characteristics in the time series may be ignored, resulting in the model's insufficient understanding of the short-term fluctuations and long-term trends of meteorological parameters, affecting the subsequent prediction performance; meteorological data without extracting autocorrelation characteristics cannot fully reflect the periodicity and trend, resulting in the model's lack of sensitivity to seasonal fluctuations in meteorological parameters; the adaptive delay adjustment method is not considered to adjust the delay between time points, because fixed delays cause distortion of time series characteristics; the geographically weighted regression model is not used to extract local regression coefficients as features, which cannot accurately reflect the spatial heterogeneity of terrain on the response variable; the local regression coefficients are not used, resulting in the neglect of the differences in characteristics of different locations, which in turn limits the model's ability to characterize the complexity of terrain and reduces the model's recognition accuracy of the impact of terrain on the target variable; Undefined optimization hyperparameter space and initial sampling distribution may cause the optimization process to become blind, making it difficult to effectively explore the potential performance of hyperparameter combinations, increasing search time and computing costs; without a mechanism to dynamically adjust the sampling distribution, the optimization stays in a fixed hyperparameter area and cannot fully explore the entire space, making it easy to fall into a local optimal solution and limiting model performance improvement; lacking a mechanism to adjust the sampling probability based on the validation set loss function value, model optimization is difficult to quickly focus on the hyperparameter area with better performance, resulting in slow convergence or suboptimal final performance; without the introduction of dynamic adjustment factors, the model has poor adaptability to validation set feedback, and the optimization process may be too rigid, resulting in an inflexible hyperparameter adjustment strategy that is difficult to cope with complex data characteristics; without the introduction of adjustment parameters and exponential parameters to control the growth rate of the dynamic adjustment factors, the optimization step size is inappropriate, and it is impossible to effectively balance exploration and utilization, affecting the overall optimization efficiency.

[0004] In view of this, the present invention proposes a flood warning method based on multi-layer rainfall perception to solve the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: a flood warning method based on multi-layer rainfall perception, comprising: S1. Collect meteorological parameter data, terrain parameter data, hydrological soil data and human activity data; S2, preprocessing the collected meteorological parameter data, terrain parameter data, hydrological soil data and human activity data to obtain a meteorological characteristic data set, a terrain characteristic data set, a hydrological soil characteristic data set and a human activity characteristic data set; S3, fusing the meteorological characteristic data set, the terrain characteristic data set, the hydrological soil characteristic data set and the human activity characteristic data set to obtain a comprehensive characteristic data set; training a flood risk prediction model based on the comprehensive characteristic data set, and predicting a flood risk index through the flood risk prediction model; S4, comparing the predicted flood risk index with the preset flood risk index threshold to determine whether a flood warning is needed; S5. If flood warning is not required, the flood warning intelligent management terminal enters the continuous monitoring mode; if flood warning is required, the flood warning intelligent management terminal issues warning information and classifies the flood risk level according to the flood risk coefficient; S6. Delineate flood warning areas based on the classified flood risk levels and terrain feature data sets, and generate flood warning maps; send flood warning maps to staff in a timely manner through the flood warning intelligent management terminal.

[0006] Furthermore, the meteorological parameter data include rainfall, rainfall intensity, rainfall duration, air pressure, cloud cover and temperature; the terrain parameter data include ground height, ground slope and terrain undulation; the hydrological soil data include river water level, flow rate, flow, groundwater level, river distribution area, soil moisture content at different depths and soil water absorption capacity; the human activity data include urbanization degree, land use type, population density and drainage equipment status.

[0007] Furthermore, the method of preprocessing the collected meteorological parameter data to obtain a meteorological characteristic data set includes: S31. For meteorological parameter data: sort out the collected meteorological parameter data and arrange them in chronological order to obtain a meteorological parameter time series: ;in, For at time point The collected meteorological parameter data; is the total length of the meteorological parameter time series; is the index of the time point, ; S32, measuring the correlation of meteorological parameter data in the meteorological parameter time series at different time point delays by an autocorrelation function formula; the autocorrelation function formula is: ;in, For delay and time point In the current time point, the meteorological parameter data in the meteorological parameter time series With delay The autocorrelation value between ; is the time point in the meteorological parameter time series Observed value of is the time point in the meteorological parameter time series Observed value of For at time point The average value of all collected data; is the time delay between two data points in the meteorological parameter time series; S33, dynamically adjusting the time delay between two data points in the meteorological parameter time series by an adaptive delay adjustment formula; The adaptive delay adjustment formula is: ; is the time delay between two data points in the adjusted meteorological parameter time series; is the frequency of collecting meteorological parameter data; To control the frequency of meteorological data collection, the time delay The adjustment coefficient of the degree of influence; To control the total length of the meteorological parameter time series Time delay The adjustment coefficient of the degree of influence; To control the logarithmic term The weight factor of S34. Based on the autocorrelation value, an autocorrelation function between the time delay and the time delay is obtained, and an autocorrelation graph is drawn, wherein the horizontal axis of the graph represents the time delay between two data points in the adjusted meteorological parameter time series, and the vertical axis of the graph represents the autocorrelation value; the peak value of the autocorrelation function is identified from the autocorrelation graph, and the attenuation degree of the autocorrelation value with the time delay is calculated, and it is used as a feature to construct a meteorological feature data set.

[0008] Furthermore, the method of preprocessing the terrain parameter data, the hydrological soil data and the human activity data to obtain the terrain characteristic data set, the hydrological soil characteristic data set and the human activity characteristic data set includes: S41. For terrain parameter data: record the data set consisting of terrain parameter data as The data set ;in, For the The ground height of each location point; For the The ground slope at each location; For the The terrain relief at each location; is the index of the location point; S42, constructing a geographically weighted regression model to analyze the relationship between terrain parameter data and response variables; the geographically weighted regression model is: ;in, For location point The response variable; , , and For location The local regression coefficient at represents the relationship between the response variable and terrain parameters at that location in the geographically weighted regression model; For location point The coordinates of ;in, For location point The horizontal axis of For location point The vertical coordinate of is the error term in the geographically weighted regression model; The local regression coefficient of each position point is estimated by a weighting function; the weighting function is: ;in, For location point and location points The weight between For location point and location points The Euclidean distance between is the bandwidth parameter, which controls the speed of weight decay; The bandwidth parameter in the weighting function is limited by a bandwidth limitation formula, and the bandwidth limitation formula is: ;in, is the bandwidth parameter after limitation; is the number of types of terrain parameter data; is the number of location points; A constant factor for adjusting the ratio between the number of terrain parameter types and the number of location points; S43, using the weighted least squares method to estimate the local regression coefficient of each location point according to the weighting function and the bandwidth limitation formula; for each location point, calculating the location point according to the weighting function and the weights of other location points, and use the weights to calculate the regression coefficients according to the weighted least squares method. , , and Estimation is performed until the local regression coefficients of all the position points are calculated, and the local regression coefficients of all the position points constitute a terrain feature data set; S44. For the hydrological soil data and the human activity data: using the DBSCAN algorithm to identify the outliers in the hydrological soil data and the human activity data, and removing the identified outliers to obtain a hydrological soil characteristic data set and a human activity characteristic data set; S45, performing standard deviation normalization processing on the meteorological characteristic data set, the terrain characteristic data set, the hydrological and soil characteristic data set, and the human activity characteristic data set, converting them into a standard normal distribution, and obtaining normalized meteorological characteristic data set, the terrain characteristic data set, the hydrological and soil characteristic data set, and the human activity characteristic data set.

[0009] Furthermore, the training method of the flood risk prediction model includes: The data set is divided into training set, test set and validation set to build a flood risk prediction model; the sample set is a subset of the data set, and each sample set includes a historical comprehensive feature data set and a corresponding flood risk index; The flood risk prediction model includes an input layer, a hidden layer and an output layer; the input layer is used to input a historical comprehensive feature data set, and the output layer is used to output a flood risk index; the hidden layer uses a ReLU activation function, and the number of neurons in the output layer is set to 1 node; the flood risk prediction model is a multi-layer perceptron MLP model; the mean square error is used as a loss function to measure the difference between the predicted value and the actual value of the model; The flood risk prediction model is trained using the training set, and the model parameters are updated through the back propagation algorithm to minimize the loss function; the performance of the flood risk prediction model is evaluated by calculating the accuracy index using the validation set; The SGD optimization algorithm is selected as the optimizer, and the model is tuned according to the performance feedback of the validation set. The model parameters are adjusted until the performance no longer improves or the preset stopping conditions are reached. The test set is used to evaluate the performance of the model in the prediction task, and the trained flood risk prediction model is used to predict the current comprehensive feature data set to obtain the flood risk index.

[0010] Furthermore, the method of adjusting the hyperparameters of the model to optimize the model includes: S61. Define the hyperparameters that need to be optimized, including learning rate, number of hidden layers, number of neurons in each layer, and regularization parameters. The hyperparameters constitute a hyperparameter space. ; S62. In the hyperparameter space Initial sampling probability within , randomly generated Hyperparameter combinations, denoted as ;in, is the number of hyperparameter combinations; is the index of the hyperparameter, ; S63. For each set of hyperparameter combinations , use the training set to train the multilayer perceptron model and calculate the loss function on the validation set: ;in, is the loss function value of the multilayer perceptron model on the validation set; For the validation set samples in the hyperparameter combination The model prediction value under ; For the validation set The true value of samples; For the validation set The input features of samples; is the number of samples in the validation set; is the index of the sample in the validation set; S64, updating the sampling distribution according to the loss function value on the validation set by using a sampling distribution dynamic adjustment formula; the sampling distribution dynamic adjustment formula is: ;in, After the update Hyperparameter combinations in rounds The sampling distribution probability value of ; is a combination of hyperparameters Probability values ​​in the initial sampling distribution; For the The dynamic adjustment factor of the round is used to control the impact of the loss function on the sampling probability; is the number of sampling rounds; S65. Limit the adjustment range of the dynamic adjustment factor and gradually optimize the hyperparameter distribution; use the updated Hyperparameter combinations in rounds The sampling distribution probability value of Resample to generate a new set of hyperparameters; train and verify the new set of hyperparameters, and update the loss function value of the multilayer perceptron model on the verification set; Calculate the change in loss of the multilayer perceptron model on the validation set: ;in, is the loss change of the multilayer perceptron model on the validation set; For the The loss function value of the multilayer perceptron model on the validation set during round sampling; For the The loss function value of the multilayer perceptron model on the validation set during round sampling; S66. When the loss change of the multilayer perceptron model on the validation set is less than the preset loss change threshold or reaches the maximum number of sampling rounds, the iteration is stopped, and the hyperparameter combination with the smallest loss function value on the validation set is selected.

[0011] Furthermore, the method of limiting the adjustment range of the dynamic adjustment factor and gradually optimizing the hyperparameter distribution includes: The adjustment range of the dynamic adjustment factor is limited by the adjustment factor limiting formula, and the adjustment factor limiting formula is: ;in, is the dynamic adjustment factor after restriction; An adjustment parameter for controlling the amplitude of the dynamic adjustment factor; To control the number of rounds Dynamic Adjustment Factor Exponential parameters affecting the speed of growth; is the type of hyperparameter combination; is the smoothing factor.

[0012] Furthermore, the method of comparing the predicted flood risk index with a preset flood risk index threshold to determine whether a flood warning is needed includes: If the predicted flood risk index is greater than or equal to the preset flood risk index threshold, it is determined that a flood warning is needed; If the predicted flood risk index is less than the preset flood risk index threshold, it is judged that no flood warning is needed.

[0013] Furthermore, the method for classifying flood risk levels according to flood risk coefficients includes: Two flood risk level thresholds are preset, namely the first flood risk level threshold and the second flood risk level threshold ; The preset flood risk level coefficient is , the flood risk factor The first flood risk level threshold and the second flood risk level threshold In comparison; like , then the flood risk level is judged to be low; if , then the flood risk level is judged to be medium; if , the flood risk level is judged to be high.

[0014] Furthermore, the method of defining flood warning areas according to the divided flood risk levels and terrain feature data sets and generating flood warning maps includes: Integrate flood risk level data and terrain feature data sets into the same geographic coordinate system, perform spatial analysis using GIS software, and overlay flood risk level data and terrain feature data; based on the spatial analysis results and combined with flood risk levels, define flood warning areas of different risk levels; If the flood risk level is low, the flood warning area will be designated as a low-risk area; if the flood risk level is medium, the flood warning area will be designated as a medium-risk area; if the flood risk level is high, the flood warning area will be designated as a high-risk area; different colors are used to represent different risk levels, and the designated flood warning areas are drawn on the map through GIS software to obtain a flood warning map.

[0015] The technical effects and advantages of the flood warning method based on multi-layer rainfall perception of the present invention are as follows: The present invention uses autocorrelation function analysis and adaptive delay adjustment to more accurately describe the time series characteristics of meteorological parameter data, which helps capture the dynamic change characteristics of meteorological data; through the analysis of autocorrelation function and autocorrelation graph, the autocorrelation characteristics of meteorological parameter data are extracted, and these characteristics can reflect important information such as periodicity and trend of meteorological data; terrain parameter data is analyzed through a geographically weighted regression model, and local regression coefficients related to the response variable are extracted as features, and these features can more accurately reflect the influence of terrain on the response variable; through a bandwidth limitation formula, the size of the bandwidth parameter can be more accurately controlled, thereby avoiding regression parameter estimation deviation or excessive variance caused by excessive or too small bandwidth; it helps to improve the prediction accuracy and stability of the geographically weighted regression model; By defining the hyperparameters that need to be optimized and constructing a hyperparameter space, a foundation is provided for the subsequent optimization process. The setting of the initial sampling distribution probability allows random exploration in the hyperparameter space, which helps to quickly discover potential hyperparameter combinations with better performance. According to the loss function value on the validation set, the sampling distribution probability value is updated through the sampling distribution dynamic adjustment formula, so that the model can sample those hyperparameter combinations with better performance more frequently, thereby accelerating the optimization process. The introduction of dynamic adjustment factors allows the loss function value to have a direct impact on the sampling probability, which helps the model gradually focus on the optimal solution during the iteration process; limiting the adjustment range of the dynamic adjustment factor by adjusting the factor restriction formula can prevent excessive fluctuations in the optimization process and ensure the stability and continuity of the optimization process; the introduction of adjustment parameters and exponential parameters allows fine control of the amplitude and growth rate of the dynamic adjustment factor, which helps to balance the relationship between exploration and utilization and avoid falling into local optimal solutions; the optimized model can more accurately capture the complex relationship between rainfall data and flood warnings, providing a more reliable basis for flood warning and prevention. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of a flood warning method based on multi-layer rainfall perception; Figure 2 This is a schematic diagram of the structure of a flood warning system based on multi-layer rainfall perception. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Example 1 See also Figure 1 As shown, the flood warning method based on multi-layer rainfall perception described in this embodiment includes: S1. Collect meteorological parameter data, terrain parameter data, hydrological soil data and human activity data; S2, preprocessing the collected meteorological parameter data, terrain parameter data, hydrological soil data and human activity data to obtain a meteorological characteristic data set, a terrain characteristic data set, a hydrological soil characteristic data set and a human activity characteristic data set; S3, fusing the meteorological characteristic data set, the terrain characteristic data set, the hydrological soil characteristic data set and the human activity characteristic data set to obtain a comprehensive characteristic data set; training a flood risk prediction model based on the comprehensive characteristic data set, and predicting a flood risk index through the flood risk prediction model; S4, comparing the predicted flood risk index with the preset flood risk index threshold to determine whether a flood warning is needed; S5. If flood warning is not required, the flood warning intelligent management terminal enters the continuous monitoring mode; if flood warning is required, the flood warning intelligent management terminal issues warning information and classifies the flood risk level according to the flood risk coefficient; S6. Delineate flood warning areas based on the classified flood risk levels and terrain feature data sets, and generate flood warning maps; send flood warning maps to staff in a timely manner through the flood warning intelligent management terminal.

[0019] Meteorological parameter data include rainfall, rainfall intensity, rainfall duration, air pressure, cloud cover and temperature; terrain parameter data include ground height, ground slope and terrain undulation; hydrological soil data include river water level, flow velocity, flow, groundwater level, river distribution area, soil moisture content at different depths and soil water absorption capacity; human activity data include urbanization degree, land use type, population density and drainage equipment status.

[0020] The method of preprocessing the collected meteorological parameter data to obtain a meteorological characteristic data set includes: S31. For meteorological parameter data: sort out the collected meteorological parameter data and arrange them in chronological order to obtain a meteorological parameter time series: ;in, For at time point The collected meteorological parameter data; is the total length of the meteorological parameter time series; is the index of the time point, ; S32, measuring the correlation of meteorological parameter data in the meteorological parameter time series at different time point delays by an autocorrelation function formula; the autocorrelation function formula is: ;in, For delay and time point In the current time point, the meteorological parameter data in the meteorological parameter time series With delay The autocorrelation value between ; is the time point in the meteorological parameter time series Observed value of is the time point in the meteorological parameter time series Observed value of For at time point The average value of all collected data; is the time delay between two data points in the meteorological parameter time series; S33, dynamically adjusting the time delay between two data points in the meteorological parameter time series by an adaptive delay adjustment formula; The adaptive delay adjustment formula is: ; is the time delay between two data points in the adjusted meteorological parameter time series; is the frequency of collecting meteorological parameter data; To control the frequency of meteorological data collection, the time delay The adjustment coefficient of the degree of influence; To control the total length of the meteorological parameter time series Time delay The adjustment coefficient of the degree of influence; To control the logarithmic term The weight factor of For example, there is a meteorological parameter dataset with a time span of 1 year (365 days). =365, the frequency of collecting meteorological parameter data is once every hour, that is =24 times / day; controls the frequency of meteorological data collection and the time delay The adjustment coefficient of the degree of influence is 0.5, controlling the total length of the meteorological parameter time series Time delay The adjustment coefficient of the degree of influence is 0.3, controlling the logarithmic term The weight factor is 0.2, then the time delay between two data points in the adjusted meteorological parameter time series is sky; S34. Based on the autocorrelation value, an autocorrelation function between the time delay and the time delay is obtained, and an autocorrelation graph is drawn, wherein the horizontal axis of the graph represents the time delay between two data points in the adjusted meteorological parameter time series, and the vertical axis of the graph represents the autocorrelation value; the peak value of the autocorrelation function is identified from the autocorrelation graph, and the attenuation degree of the autocorrelation value with the time delay is calculated, and it is used as a feature to construct a meteorological feature data set.

[0021] The method of preprocessing terrain parameter data, hydrological soil data and human activity data to obtain terrain characteristic data set, hydrological soil characteristic data set and human activity characteristic data set includes: S41. For terrain parameter data: record the data set consisting of terrain parameter data as The data set ;in, For the The ground height of each location point; For the The ground slope at each location; For the The terrain relief at each location; is the index of the location point; S42, constructing a geographically weighted regression model to analyze the relationship between terrain parameter data and response variables; the geographically weighted regression model is: ;in, For location point The response variable; , , and For location The local regression coefficient at represents the relationship between the response variable and terrain parameters at that location in the geographically weighted regression model; For location point The coordinates of ;in, For location point The horizontal axis of For location point The vertical coordinate of is the error term in the geographically weighted regression model; The local regression coefficient of each position point is estimated by a weighting function; the weighting function is: ;in, For location point and location points The weight between For location point and location points The Euclidean distance between is the bandwidth parameter, which controls the speed of weight decay; The bandwidth parameter in the weighting function is limited by a bandwidth limitation formula, and the bandwidth limitation formula is: ;in, is the bandwidth parameter after limitation; is the number of types of terrain parameter data; is the number of location points; A constant factor for adjusting the ratio between the number of terrain parameter types and the number of location points; S43, using the weighted least squares method to estimate the local regression coefficient of each location point according to the weighting function and the bandwidth limitation formula; for each location point, calculating the location point according to the weighting function and the weights of other location points, and use the weights to calculate the regression coefficients according to the weighted least squares method. , , and Estimation is performed until the local regression coefficients of all the position points are calculated, and the local regression coefficients of all the position points constitute a terrain feature data set; S44. For the hydrological soil data and the human activity data: using the DBSCAN algorithm to identify the outliers in the hydrological soil data and the human activity data, and removing the identified outliers to obtain a hydrological soil characteristic data set and a human activity characteristic data set; S45, performing standard deviation normalization processing on the meteorological characteristic data set, the terrain characteristic data set, the hydrological and soil characteristic data set, and the human activity characteristic data set, converting them into a standard normal distribution, and obtaining normalized meteorological characteristic data set, the terrain characteristic data set, the hydrological and soil characteristic data set, and the human activity characteristic data set.

[0022] Methods for fusing meteorological characteristic datasets, terrain characteristic datasets, hydrological and soil characteristic datasets, and human activity characteristic datasets to obtain a comprehensive characteristic dataset include: The meteorological characteristic dataset, terrain characteristic dataset, hydrological soil characteristic dataset and human activity characteristic dataset are fused through the weighted formula to obtain a comprehensive characteristic dataset; the meteorological characteristic dataset is recorded as , the terrain feature dataset is recorded as , the hydrological and soil characteristic dataset is recorded as , the human activity feature dataset is recorded as ; The weighted model is: ;in, It is a comprehensive feature dataset; is the weight coefficient of the meteorological characteristic dataset; is the weight coefficient of the terrain feature dataset; is the weight coefficient of the hydrological soil characteristic dataset; is the weight coefficient of the human activity feature dataset.

[0023] The training methods of flood risk prediction models include: The data set is divided into training set, test set and validation set to build a flood risk prediction model; the sample set is a subset of the data set, and each sample set includes a historical comprehensive feature data set and a corresponding flood risk index; The flood risk prediction model includes an input layer, a hidden layer and an output layer; the input layer is used to input a historical comprehensive feature data set, and the output layer is used to output a flood risk index; the hidden layer uses a ReLU activation function, and the number of neurons in the output layer is set to 1 node; the flood risk prediction model is a multi-layer perceptron MLP model; The mean square error is used as the loss function to measure the difference between the model's predicted value and the actual value; the mean square error loss function is: ;in, is the number of samples in the data set; For the data set The actual value of samples; For the data set The predicted value of samples; is the index of the sample; The flood risk prediction model is trained using the training set, and the model parameters are updated through the back propagation algorithm to minimize the loss function; the performance of the flood risk prediction model is evaluated by calculating the accuracy index using the validation set; The SGD optimization algorithm is selected as the optimizer, and the model is tuned according to the performance feedback of the validation set. The model parameters are adjusted until the performance no longer improves or the preset stopping conditions are reached. The test set is used to evaluate the performance of the model in the prediction task, and the trained flood risk prediction model is used to predict the current comprehensive feature data set to obtain the flood risk index.

[0024] Methods for adjusting the model's hyperparameters to optimize the model include: S61. Define the hyperparameters that need to be optimized, including learning rate, number of hidden layers, number of neurons in each layer, and regularization parameters. The hyperparameters constitute a hyperparameter space. ; S62. In the hyperparameter space Initial sampling probability within , randomly generated Hyperparameter combinations, denoted as ;in, is the number of hyperparameter combinations; is the index of the hyperparameter, ; S63. For each set of hyperparameter combinations , use the training set to train the multilayer perceptron model and calculate the loss function on the validation set: ;in, is the loss function value of the multilayer perceptron model on the validation set; For the validation set samples in the hyperparameter combination The model prediction value under ; For the validation set The true value of samples; For the validation set The input features of samples; is the number of samples in the validation set; is the index of the sample in the validation set; S64, updating the sampling distribution according to the loss function value on the validation set by using a sampling distribution dynamic adjustment formula; the sampling distribution dynamic adjustment formula is: ;in, After the update Hyperparameter combinations in rounds The sampling distribution probability value of ; is a combination of hyperparameters Probability values ​​in the initial sampling distribution; For the The dynamic adjustment factor of the round is used to control the impact of the loss function on the sampling probability; is the number of sampling rounds; For example, the number of hyperparameter combinations =3, randomly initialize 3 hyperparameter combinations: : The learning rate is 0.01, the number of hidden layers is 2, and the number of neurons in each layer is 64; : The learning rate is 0.1, the number of hidden layers is 3, and the number of neurons in each layer is 128; : The learning rate is 0.001, the number of hidden layers is 4, and the number of neurons in each layer is 256; Initial sampling distribution probability : Assume that the initial sampling probability of each combination is equal: ; Validation set loss function value : Assume that in In this round, the loss value calculated by the validation set is: , ; Assuming dynamic adjustment factor , numerator part: calculate each The weight of is determined by the initial distribution and the exponential function value according to the formula: ; for : ; for : ; for : ; The denominator is the sum of all weights: ; After the first round of updates, the sampling distribution probability of each hyperparameter combination is calculated as: ; ; ; Smaller loss It has obtained a higher weight and is more likely to be selected by subsequent sampling, resulting in a larger loss. The lower the weight, the smaller the probability of subsequent sampling.

[0025] S65. Limit the adjustment range of the dynamic adjustment factor and gradually optimize the hyperparameter distribution; use the updated Hyperparameter combinations in rounds The sampling distribution probability value of Resample to generate a new set of hyperparameters; train and verify the new set of hyperparameters, and update the loss function value of the multilayer perceptron model on the verification set; Calculate the change in loss of the multilayer perceptron model on the validation set: ;in, is the loss change of the multilayer perceptron model on the validation set; For the The loss function value of the multilayer perceptron model on the validation set during round sampling; For the The loss function value of the multilayer perceptron model on the validation set during round sampling; S66. When the loss change of the multilayer perceptron model on the validation set is less than the preset loss change threshold or reaches the maximum number of sampling rounds, the iteration is stopped, and the hyperparameter combination with the smallest loss function value on the validation set is selected.

[0026] Methods for limiting the adjustment range of dynamic adjustment factors and gradually optimizing the distribution of hyperparameters include: The adjustment range of the dynamic adjustment factor is limited by the adjustment factor limiting formula, and the adjustment factor limiting formula is: ;in, is the dynamic adjustment factor after restriction; An adjustment parameter for controlling the amplitude of the dynamic adjustment factor; To control the number of rounds Dynamic Adjustment Factor Exponential parameters affecting the speed of growth; is the type of hyperparameter combination; is a smoothing factor used to prevent the denominator from being too small; For example, the number of hyperparameter combinations is 20, the types of hyperparameter combinations is 4, the number of sampling rounds is 5, which controls the adjustment parameter of the dynamic adjustment factor amplitude is 0.5, the smoothing factor is 0.1, controlling the number of rounds Dynamic Adjustment Factor Exponential parameters affecting the speed of growth is 1.2, then the dynamic adjustment factor after limitation is .

[0027] The method of comparing the predicted flood risk index with the preset flood risk index threshold to determine whether a flood warning is needed includes: If the predicted flood risk index is greater than or equal to the preset flood risk index threshold, it is determined that a flood warning is needed; If the predicted flood risk index is less than the preset flood risk index threshold, it is judged that no flood warning is needed.

[0028] The methods for classifying flood risk levels according to flood risk coefficients include: Two flood risk level thresholds are preset, namely the first flood risk level threshold and the second flood risk level threshold ; The preset flood risk level coefficient is , the flood risk factor The first flood risk level threshold and the second flood risk level threshold In comparison; like , then the flood risk level is judged to be low; if , then the flood risk level is judged to be medium; if , the flood risk level is judged to be high.

[0029] Methods for delineating flood warning areas based on the divided flood risk levels and terrain feature datasets and generating flood warning maps include: Integrate flood risk level data and terrain feature data sets into the same geographic coordinate system, perform spatial analysis using GIS software, and overlay flood risk level data and terrain feature data; based on the spatial analysis results and combined with flood risk levels, define flood warning areas of different risk levels; If the flood risk level is low, the flood warning area will be designated as a low-risk area; if the flood risk level is medium, the flood warning area will be designated as a medium-risk area; if the flood risk level is high, the flood warning area will be designated as a high-risk area; different colors are used to represent different risk levels, and the designated flood warning areas are drawn on the map through GIS software to obtain a flood warning map.

[0030] The preset flood risk index threshold is set by the staff. Different flood risk indices are collected through the flood warning intelligent management terminal, and the average value of multiple flood risk indices is taken as the preset flood risk index threshold.

[0031] Through autocorrelation function analysis and adaptive delay adjustment, the time series characteristics of meteorological parameter data are described more accurately in this embodiment, which helps to capture the dynamic change characteristics of meteorological data; through the analysis of autocorrelation function and autocorrelation graph, the autocorrelation characteristics of meteorological parameter data are extracted, and these characteristics can reflect important information such as periodicity and trend of meteorological data; terrain parameter data are analyzed through geographical weighted regression model, and local regression coefficients related to response variables are extracted as features, which can more accurately reflect the impact of terrain on response variables; through bandwidth limitation formula, the size of bandwidth parameter can be more accurately controlled, so as to avoid regression parameter estimation deviation or excessive variance caused by excessive or too small bandwidth; it helps to improve the prediction accuracy and stability of geographical weighted regression model; By defining the hyperparameters that need to be optimized and constructing a hyperparameter space, a foundation is provided for the subsequent optimization process. The setting of the initial sampling distribution probability allows random exploration in the hyperparameter space, which helps to quickly discover potential hyperparameter combinations with better performance. According to the loss function value on the validation set, the sampling distribution probability value is updated through the sampling distribution dynamic adjustment formula, so that the model can sample those hyperparameter combinations with better performance more frequently, thereby accelerating the optimization process. The introduction of dynamic adjustment factors allows the loss function value to have a direct impact on the sampling probability, which helps the model gradually focus on the optimal solution during the iteration process; limiting the adjustment range of the dynamic adjustment factor by adjusting the factor restriction formula can prevent excessive fluctuations in the optimization process and ensure the stability and continuity of the optimization process; the introduction of adjustment parameters and exponential parameters allows fine control of the amplitude and growth rate of the dynamic adjustment factor, which helps to balance the relationship between exploration and utilization and avoid falling into local optimal solutions; the optimized model can more accurately capture the complex relationship between rainfall data and flood warnings, providing a more reliable basis for flood warning and prevention.

[0032] Example 2 See also Figure 2As shown, the part not described in detail in this embodiment is described in Example 1, which provides a flood warning system based on multi-layer rainfall perception, including: Data collection module, used to collect meteorological parameter data, terrain parameter data, hydrological soil data and human activity data; The data processing module is used to pre-process the collected meteorological parameter data, terrain parameter data, hydrological soil data and human activity data to obtain a meteorological characteristic data set, a terrain characteristic data set, a hydrological soil characteristic data set and a human activity characteristic data set; The flood risk prediction module is used to fuse the meteorological characteristic data set, the terrain characteristic data set, the hydrological soil characteristic data set and the human activity characteristic data set to obtain a comprehensive characteristic data set; the flood risk prediction model is obtained by training the comprehensive characteristic data set, and the flood risk index is obtained by predicting the flood risk prediction model; The flood risk assessment module is used to compare the predicted flood risk index with the preset flood risk index threshold to determine whether a flood warning is needed; Warning risk classification module: if flood warning is not required, the flood warning intelligent management terminal enters the continuous monitoring mode; if flood warning is required, the flood warning intelligent management terminal issues warning information and classifies the flood risk level according to the flood risk coefficient; The early warning risk notification module is used to define the flood warning area according to the divided flood risk level and terrain feature data set, and generate a flood warning map; the flood warning map is sent to the staff in a timely manner through the flood warning intelligent management terminal; and each module is connected by wired and / or wireless means.

[0033] Since the electronic device introduced in this embodiment is an electronic device used to implement a flood warning method based on multi-layer rainfall perception in the embodiment of this application, based on the flood warning method based on multi-layer rainfall perception introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as the technical personnel of this field implement the electronic device used in the flood warning method based on multi-layer rainfall perception in the embodiment of this application, it belongs to the scope of protection of this application.

[0034] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0035] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A flood warning method based on multi-layer rainfall perception, characterized in that: include: S1. Collect meteorological parameter data, terrain parameter data, hydrological soil data and human activity data; S2, preprocessing the collected meteorological parameter data, terrain parameter data, hydrological soil data and human activity data to obtain a meteorological characteristic data set, a terrain characteristic data set, a hydrological soil characteristic data set and a human activity characteristic data set; S3, fusing the meteorological characteristic data set, the terrain characteristic data set, the hydrological and soil characteristic data set and the human activity characteristic data set to obtain a comprehensive characteristic data set; A flood risk prediction model is obtained by training the comprehensive feature data set, and a flood risk index is obtained by predicting the flood risk prediction model; S4, comparing the predicted flood risk index with the preset flood risk index threshold to determine whether a flood warning is needed; S5. If flood warning is not required, the flood warning intelligent management terminal enters the continuous monitoring mode; if flood warning is required, the flood warning intelligent management terminal issues warning information and classifies the flood risk level according to the flood risk coefficient; S6. Delineate flood warning areas based on the classified flood risk levels and terrain feature data sets, and generate flood warning maps; Flood warning maps are sent to staff in a timely manner through the flood warning intelligent management terminal.

2. The flood warning method based on multi-layer rainfall perception according to claim 1 is characterized in that: The meteorological parameter data include rainfall, rainfall intensity, rainfall duration, air pressure, cloud cover and temperature; the terrain parameter data include ground height, ground slope and terrain undulation; the hydrological soil data include river water level, flow rate, flow, groundwater level, river distribution area, soil moisture content at different depths and soil water absorption capacity; the human activity data include urbanization degree, land use type, population density and drainage equipment status.

3. The flood warning method based on multi-layer rainfall perception according to claim 2 is characterized in that: The method of preprocessing the collected meteorological parameter data to obtain a meteorological characteristic data set includes: S31. For meteorological parameter data: sort out the collected meteorological parameter data and arrange them in chronological order to obtain a meteorological parameter time series: ;in, For at time point The collected meteorological parameter data; is the total length of the meteorological parameter time series; is the index of the time point, ; S32, measuring the correlation of meteorological parameter data in the meteorological parameter time series at different time point delays by an autocorrelation function formula; the autocorrelation function formula is: ;in, For delay and time point In the current time point, the meteorological parameter data in the meteorological parameter time series With delay The autocorrelation value between ; is the time point in the meteorological parameter time series Observed value of is the time point in the meteorological parameter time series Observed value of For at time point The average value of all collected data; is the time delay between two data points in the meteorological parameter time series; S33, dynamically adjusting the time delay between two data points in the meteorological parameter time series by an adaptive delay adjustment formula; The adaptive delay adjustment formula is: ; is the time delay between two data points in the adjusted meteorological parameter time series; is the frequency of collecting meteorological parameter data; To control the frequency of meteorological data collection, the time delay The adjustment coefficient of the degree of influence; To control the total length of the meteorological parameter time series Time delay The adjustment coefficient of the degree of influence; To control the logarithmic term The weight factor of S34. Based on the autocorrelation value, an autocorrelation function between the time delay and the time delay is obtained, and an autocorrelation graph is drawn, wherein the horizontal axis of the graph represents the time delay between two data points in the adjusted meteorological parameter time series, and the vertical axis of the graph represents the autocorrelation value; the peak value of the autocorrelation function is identified from the autocorrelation graph, and the attenuation degree of the autocorrelation value with the time delay is calculated, and it is used as a feature to construct a meteorological feature data set.

4. The flood warning method based on multi-layer rainfall perception according to claim 3 is characterized in that: The method of preprocessing terrain parameter data, hydrological soil data and human activity data to obtain terrain characteristic data set, hydrological soil characteristic data set and human activity characteristic data set includes: S41. For terrain parameter data: record the data set consisting of terrain parameter data as The data set ;in, For the The ground height of each location point; For the The ground slope at each location; For the The terrain relief at each location; is the index of the location point; S42, constructing a geographically weighted regression model to analyze the relationship between terrain parameter data and response variables; the geographically weighted regression model is: ;in, For location point The response variable; , , and For location The local regression coefficient at represents the relationship between the response variable and terrain parameters at that location in the geographically weighted regression model; For location point The coordinates of ;in, For location point The horizontal axis of For location point The vertical coordinate of is the error term in the geographically weighted regression model; The local regression coefficient of each position point is estimated by a weighting function; the weighting function is: ;in, For location point and location points The weight between For location point and location points The Euclidean distance between is the bandwidth parameter, which controls the speed of weight decay; The bandwidth parameter in the weighting function is limited by a bandwidth limitation formula, and the bandwidth limitation formula is: ;in, is the bandwidth parameter after limitation; is the number of types of terrain parameter data; is the number of location points; A constant factor for adjusting the ratio between the number of terrain parameter types and the number of location points; S43, using the weighted least squares method to estimate the local regression coefficient of each location point according to the weighting function and the bandwidth limitation formula; for each location point, calculating the location point according to the weighting function and the weights of other location points, and use the weights to calculate the regression coefficients according to the weighted least squares method. , , and Estimation is performed until the local regression coefficients of all the position points are calculated, and the local regression coefficients of all the position points constitute a terrain feature data set; S44. For the hydrological soil data and the human activity data: using the DBSCAN algorithm to identify the outliers in the hydrological soil data and the human activity data, and removing the identified outliers to obtain a hydrological soil characteristic data set and a human activity characteristic data set; S45, performing standard deviation normalization processing on the meteorological characteristic data set, the terrain characteristic data set, the hydrological and soil characteristic data set, and the human activity characteristic data set, converting them into a standard normal distribution, and obtaining normalized meteorological characteristic data set, the terrain characteristic data set, the hydrological and soil characteristic data set, and the human activity characteristic data set.

5. The flood warning method based on multi-layer rainfall perception according to claim 4 is characterized in that: The training method of the flood risk prediction model includes: The data set is divided into training set, test set and validation set to build a flood risk prediction model; the sample set is a subset of the data set, and each sample set includes a historical comprehensive feature data set and a corresponding flood risk index; The flood risk prediction model includes an input layer, a hidden layer and an output layer; the input layer is used to input a historical comprehensive feature data set, and the output layer is used to output a flood risk index; the hidden layer uses a ReLU activation function, and the number of neurons in the output layer is set to 1 node; the flood risk prediction model is a multi-layer perceptron MLP model; the mean square error is used as a loss function to measure the difference between the predicted value and the actual value of the model; The flood risk prediction model is trained using the training set, and the model parameters are updated through the back propagation algorithm to minimize the loss function; the performance of the flood risk prediction model is evaluated by calculating the accuracy index using the validation set; The SGD optimization algorithm is selected as the optimizer, and the model is tuned according to the performance feedback of the validation set. The model parameters are adjusted until the performance no longer improves or the preset stopping conditions are reached. The test set is used to evaluate the performance of the model in the prediction task, and the trained flood risk prediction model is used to predict the current comprehensive feature data set to obtain the flood risk index.

6. The flood warning method based on multi-layer rainfall perception according to claim 5 is characterized in that: The method for adjusting the hyperparameters of the model to optimize the model includes: S61. Define the hyperparameters that need to be optimized, including learning rate, number of hidden layers, number of neurons in each layer, and regularization parameters. The hyperparameters constitute a hyperparameter space. ; S62. In the hyperparameter space Initial sampling probability within , randomly generated Hyperparameter combinations, denoted as ;in, is the number of hyperparameter combinations; is the index of the hyperparameter, ; S63. For each set of hyperparameter combinations , use the training set to train the multilayer perceptron model and calculate the loss function on the validation set: ;in, is the loss function value of the multilayer perceptron model on the validation set; For the validation set samples in the hyperparameter combination The model prediction value under ; For the validation set The true value of samples; For the validation set The input features of samples; is the number of samples in the validation set; is the index of the sample in the validation set; S64, updating the sampling distribution according to the loss function value on the validation set by using a sampling distribution dynamic adjustment formula; the sampling distribution dynamic adjustment formula is: ;in, After the update Hyperparameter combinations in rounds The sampling distribution probability value of ; is a combination of hyperparameters Probability values ​​in the initial sampling distribution; For the The dynamic adjustment factor of the round is used to control the impact of the loss function on the sampling probability; is the number of sampling rounds; S65. Limit the adjustment range of the dynamic adjustment factor and gradually optimize the hyperparameter distribution; use the updated Hyperparameter combinations in rounds The sampling distribution probability value of Resample to generate a new set of hyperparameters; train and verify the new set of hyperparameters, and update the loss function value of the multilayer perceptron model on the verification set; Calculate the change in loss of the multilayer perceptron model on the validation set: ;in, is the loss change of the multilayer perceptron model on the validation set; For the The loss function value of the multilayer perceptron model on the validation set during round sampling; For the The loss function value of the multilayer perceptron model on the validation set during round sampling; S66. When the loss change of the multilayer perceptron model on the validation set is less than the preset loss change threshold or reaches the maximum number of sampling rounds, the iteration is stopped, and the hyperparameter combination with the smallest loss function value on the validation set is selected.

7. The flood warning method based on multi-layer rainfall perception according to claim 6 is characterized in that: The method of limiting the adjustment range of the dynamic adjustment factor and gradually optimizing the hyperparameter distribution includes: The adjustment range of the dynamic adjustment factor is limited by the adjustment factor limiting formula, and the adjustment factor limiting formula is: ;in, is the dynamic adjustment factor after restriction; An adjustment parameter for controlling the amplitude of the dynamic adjustment factor; To control the number of rounds Dynamic Adjustment Factor Exponential parameters affecting the speed of growth; is the type of hyperparameter combination; is the smoothing factor.

8. The flood warning method based on multi-layer rainfall perception according to claim 7 is characterized in that: The method of comparing the predicted flood risk index with a preset flood risk index threshold to determine whether a flood warning is needed includes: If the predicted flood risk index is greater than or equal to the preset flood risk index threshold, it is determined that a flood warning is needed; If the predicted flood risk index is less than the preset flood risk index threshold, it is judged that no flood warning is needed.

9. The flood warning method based on multi-layer rainfall perception according to claim 8 is characterized in that: The method for classifying flood risk levels according to flood risk coefficients includes: Two flood risk level thresholds are preset, namely the first flood risk level threshold and the second flood risk level threshold ; The preset flood risk level coefficient is , the flood risk factor The first flood risk level threshold and the second flood risk level threshold In comparison; like , then the flood risk level is judged to be low; if , then the flood risk level is judged to be medium; if , the flood risk level is judged to be high.

10. The flood warning method based on multi-layer rainfall perception according to claim 9 is characterized in that: The method of demarcating flood warning areas according to the divided flood risk levels and terrain feature data sets and generating flood warning maps includes: Integrate flood risk level data and terrain feature data sets into the same geographic coordinate system, perform spatial analysis using GIS software, and overlay flood risk level data and terrain feature data; based on the results of spatial analysis and combined with flood risk levels, define flood warning areas of different risk levels; If the flood risk level is low, the flood warning area will be designated as a low-risk area; if the flood risk level is medium, the flood warning area will be designated as a medium-risk area; if the flood risk level is high, the flood warning area will be designated as a high-risk area; different colors are used to represent different risk levels, and the designated flood warning areas are drawn on the map through GIS software to obtain a flood warning map.

Images