Method and system for quickly identifying hydrological forecasting factors before flood based on environmental perception

Abstract

The invention discloses a pre-flood hydrological forecasting factor rapid identification method and system based on environmental perception, and the method abandons a conventional forecasting mode which depends on a hysteresis index after a flood occurs, and achieves the rapid and accurate identification of a hydrological forecasting factor through sensing a key environmental factor before the flood occurs. The design system comprises an environment factor sensing module, a data preprocessing module, a factor rapid screening module, a model training optimization module and an identification result output module, and the pre-flood environment data such as the soil water content, the early-stage accumulated rainfall and the vegetation coverage are collected and preprocessed, and then a multi-algorithm fusion screening strategy is adopted. The method can quickly lock the core forecasting factor with strong relevance to flood generation, provides advance support for hydrological forecasting, and solves the technical problems that the traditional hydrological forecasting factor is lagged in identification and poor in timeliness. The hydrological forecasting method remarkably improves the perspectiveness and accuracy of hydrological forecasting, and is suitable for scenes such as basin flood early warning and water resource scheduling.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hydrological forecasting and environmental perception, and in particular to a method for quickly identifying hydrological forecasting factors before floods based on environmental perception, and to a system for quickly identifying hydrological forecasting factors before floods based on environmental perception. BACKGROUND

[0002] Hydrological forecasting is a key technical support for flood control and disaster reduction and optimal scheduling of water resources, and its core lies in accurately identifying key forecasting factors that affect the occurrence of floods. Traditional methods for identifying hydrological forecasting factors mainly rely on lagging indicators after the occurrence of floods, such as flood peak flow and flood hydrograph, which have obvious limitations:

[0003] Firstly, lagging indicators cannot capture early warning signals before the occurrence of floods, resulting in insufficient forecasting lead time and difficulty in meeting the needs of real-time flood control scheduling. Secondly, traditional methods focus on the hydrological process itself and ignore the cumulative effect of the environment before the occurrence of floods, such as the saturation degree of soil moisture and the cumulative amount of rainfall before the occurrence of floods. In addition, traditional methods for identifying factors mostly use a single algorithm, which has low efficiency and poor targeting, making it difficult to quickly lock in core influencing factors and affecting the timeliness and accuracy of forecasting.

[0004] Therefore, there is an urgent need for a system for quickly identifying environmental factors before floods to address the technical problems of strong lagging and insufficient forecasting lead time in traditional methods. SUMMARY

[0005] The present application aims to address the problems of incomplete factor identification, low screening efficiency, and insufficient warning accuracy in the prior art, and provides a method and system for quickly identifying hydrological forecasting factors before floods based on environmental perception, which realizes accurate collection, efficient processing, and rapid screening of core factors of multi-dimensional environmental data, and accurate flood probability early warning.

[0006] To achieve the above functions, the present application designs a method for quickly identifying hydrological forecasting factors before floods based on environmental perception, which performs the following steps S1-S5 to complete flood warning in the target area:

[0007] Step S1: For the target area, collect multi-dimensional environmental raw data before the occurrence of floods, including soil moisture, cumulative rainfall before the occurrence of floods, vegetation coverage, terrain slope, surface temperature, and short-term weather forecast data.

[0008] Step S2: Perform denoising, missing value filling, and standardization processing on the environmental raw data to output standardized environmental data.

[0009] Step S3: Calculate the correlation degree of each standardized environmental data and historical flood occurrence event by using mutual information method, preliminarily screen the standardized environmental data, sort the standardized environmental data based on the random forest model, further screen the standardized environmental data, and screen out core prediction factors;

[0010] Step S4: Based on the historical flood event data and the corresponding core prediction factors, a hydrological prediction factor identification model is constructed, the core prediction factors are taken as the input, and the flood occurrence probability is taken as the output, the hydrological prediction factor identification model is trained and iteratively optimized;

[0011] Step S5: Real-time output of core prediction factors and corresponding flood occurrence probability warning information, complete the flood warning of the target area.

[0012] As a preferred technical solution of the application: the specific steps of step S1 are as follows:

[0013] Step S1.1: For the target area, a distributed soil moisture sensor is used to collect the soil moisture content at a preset depth at a preset frequency;

[0014] Step S1.2: A tipping bucket rain gauge is used to collect the cumulative rainfall in the target area within a preset time period at a preset frequency;

[0015] Step S1.3: Obtain the spatial and temporal distribution data of the vegetation coverage of the target area by remote sensing image interpretation combined with ground vegetation monitoring station data;

[0016] Step S1.4: Based on digital elevation model data, extract the terrain parameters of the target area, including terrain slope, slope direction, and watershed convergence area;

[0017] Step S1.5: Receive the precipitation forecast and air temperature forecast data of the target area within a preset time period released by the meteorological department.

[0018] As a preferred technical solution of the application: the specific steps of step S2 are as follows:

[0019] Step S2.1: The original environmental data is denoised by using the moving average method to remove random interference signals, and the calculation formula is as follows:

[0020] ;

[0021] Wherein, is the environmental original data value at the i-th moment after denoising, N is the size of the sliding window, is the value of the k-th moment of the environmental original data;

[0022] Step S2.2: Fill in the missing data in the original environmental data using the K-neighbor algorithm with K value of 5, and when the single-factor data missing rate exceeds 10%, combine the linear interpolation method with the historical same period data in the past 5 years to supplement;

[0023] ;

[0024] wherein, is the missing data estimation value at the jth moment, are the effective data values adjacent to the missing moment respectively, are the times corresponding to the moments respectively;

[0025] Step S2.3: Normalize the original environmental data processed in step S2.2 using the Z-score standardization method to obtain standardized environmental data conforming to the [0, 1] interval distribution.

[0026] As a preferred technical solution of the present application: the specific steps of step S3 are as follows:

[0027] Step S3.1: Calculate the correlation degree of each standardized environmental data and historical flood occurrence event by mutual information method, respectively take each standardized environmental data as a candidate factor in the mutual information method, retain the candidate factors with correlation degree greater than 0.6, and the calculation formula is as follows:

[0028] The mutual information method calculation formula is as follows:

[0029] ;

[0030] In the formula, indicates the correlation degree of the candidate factor X and the flood occurrence event Y, indicates the information entropy of the candidate factor X, indicates the information entropy of the flood occurrence event Y, indicates the joint information entropy of the candidate factor X and the flood occurrence event Y;

[0031] Wherein, the information entropy calculation formula of the candidate factor X is as follows:

[0032] ;

[0033] Wherein, x is the observation value of the candidate factor X, is the probability that the observation value of the candidate factor X is equal to x, indicates the logarithmic function with base 2;

[0034] The information entropy calculation formula of the flood occurrence event Y is as follows:

[0035] ;

[0036] Wherein, y is the specific state of the historical flood occurrence event Y, including occurrence and non-occurrence; The probability that the specific state of the historical flood occurrence event Y is equal to y;

[0037] The joint information entropy calculation formula of the candidate factor X and the flood occurrence event Y is as follows:

[0038] ;

[0039] Wherein, The probability that the observed value of the candidate factor X is equal to x, and the specific state of the historical flood occurrence event Y is equal to y;

[0040] The correlation degree > 0.6 candidate factor;

[0041] Step S3.2: input the candidate factors obtained in step S3.1 into a random forest model containing 100 decision trees, calculate the importance score of each candidate factor by out-of-bag data error, and the calculation formula is as follows:

[0042] ;

[0043] Wherein, The importance score of the candidate factor X, T=100 is the number of decision trees, The out-of-bag data original error rate of the tth decision tree, The error rate after random permutation of the out-of-bag data of the candidate factor X;

[0044] Step S3.3: set the importance score threshold to 0.7, and select the candidate factors with importance score higher than the threshold as the core prediction factor.

[0045] As a preferred technical solution of the present application: the specific steps of step S4 are as follows:

[0046] Step S4.1: match the nearly 30-year historical environmental factor data with the corresponding flood occurrence results, construct a training set and a validation set, the ratio of which is 7:3, and the data sample size is not less than 1000 groups;

[0047] Step S4.2: adopt gradient boosting decision tree to construct a hydrological prediction factor identification model, take the core prediction factor as input, and take the flood occurrence probability as output, and the initial number of iterations of the model is set to 100 times; The output formula of the model of the mth iteration is as follows:

[0048] ;

[0049] Wherein, The output value of the model after the mth iteration, The model output value of the m-1th iteration, The learning rate, The weak classifier obtained by the mth training;

[0050] The model output value is mapped to the flood occurrence probability by the sigmoid function:

[0051]

[0052] The probability of flood occurrence when the input x is given, M is the total number of iterations; F is the cumulative output value of the gradient boosting decision tree model after M iterations;

[0053] Step S4.3: The Gaussian process is used as the Bayesian optimization algorithm of the probability model to iteratively optimize the hyperparameters of the model, including the learning rate, tree depth, and leaf node number, and the number of iterations is 50 times, so that the prediction accuracy of the model is higher than 85%; The optimization function is as follows:

[0054]

[0055] The combination of hyperparameters, The learning rate, The tree depth, The leaf node number, The search space of the hyperparameters, wherein the search range of the learning rate is 0.01-0.3, the search range of the tree depth is 3-10, and the search range of the leaf node number is 10-100; The model validation set accuracy corresponding to the combination of hyperparameters, and the prediction accuracy of the model is higher than 85% after optimization.

[0056] As a preferred technical solution of the present application: the specific steps of step S5 are as follows:

[0057] Step S5.1: Output the core prediction factor name, correlation value and importance score in list form, and support Excel format export;

[0058] Step S5.2: Based on the numerical combination of the core prediction factor, output the low, medium and high three levels of flood occurrence probability warning, wherein the low risk warning corresponds to the flood occurrence probability <30%, the medium risk warning corresponds to the flood occurrence probability 30%-69%, and the high risk warning corresponds to the flood occurrence probability ≥70%, and the warning information supports the synchronous push of the webpage end and the applet end.

[0059] ​​​​​The application also designs a flood-precursor hydrological forecasting factor rapid identification system based on environmental perception, which comprises an environmental factor perception module, a data preprocessing module, a factor rapid screening module, a model training optimization module, and an identification result output module, so as to realize the flood-precursor hydrological forecasting factor rapid identification method based on environmental perception.

[0060] The environmental factor perception module is used for collecting multi-dimensional environmental original data before flood occurrence in a target area, and the environmental original data comprises soil moisture content, previous cumulative rainfall, vegetation coverage, terrain slope, ground temperature, and short-term weather forecast data.

[0061] The data preprocessing module is in communication connection with the environmental factor perception module, and is used for denoising, missing value filling, and standardization processing of the environmental original data, and outputs standardized environmental data.

[0062] The factor rapid screening module is in communication connection with the data preprocessing module, and is used for calculating the correlation degree of each standardized environmental data and historical flood occurrence events by using a mutual information method, preliminarily screening the standardized environmental data, sorting the standardized environmental data based on a random forest model, further screening the standardized environmental data, and screening core forecasting factors.

[0063] The model training optimization module is in communication connection with the factor rapid screening module, and is used for constructing a hydrological forecasting factor identification model based on historical flood event data and corresponding core forecasting factors, taking the core forecasting factors as input and taking flood occurrence probability as output, training and iteratively optimizing the hydrological forecasting factor identification model.

[0064] The identification result output module is in communication connection with the model training optimization module, and is used for real-time output of core forecasting factors and corresponding flood occurrence probability warning information, and completes flood warning of the target area.

[0065] The application also designs an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and the processor realizes the flood-precursor hydrological forecasting factor rapid identification method based on environmental perception when executing the program.

[0066] The application also designs a computer readable storage medium, which stores a computer program, and the program realizes any one of the flood-precursor hydrological forecasting factor rapid identification methods based on environmental perception when executed by a processor.

[0067] Advantages:

[0068] Compared with the prior art, the present application has obvious advantages, and the core beneficial effects are reflected in four aspects of multi-dimensional perception, efficient screening, accurate prediction and wide application. First, the perception dimension is more comprehensive, the system covers five key environmental factors of soil, rainfall, vegetation, terrain and weather, through multi-layer sensor vertical arrangement and multi-source data cross fusion, breaking the limitation of single dimension perception of traditional technology, realizing all-around and three-dimensional capture of flood influencing factors. Second, the screening efficiency is greatly improved, the fusion algorithm combining mutual information method and random forest importance sorting is adopted, which not only ensures the correlation screening of factors and flood events, but also accurately identifies the contribution of factors to the warning result, and the whole core factor screening process takes no more than 30 minutes, which significantly meets the timeliness demand of flood warning. Third, the prediction accuracy is more guaranteed, the hyperparameters of gradient boosting decision tree (GBDT) model are optimized by Bayesian optimization algorithm, so that the model prediction accuracy is higher than 85%, and the low, medium and high three flood occurrence probability warning levels are divided, which provides quantitative and accurate technical basis for disaster prevention and mitigation decision-making. In addition, the system is very practical, the wireless communication technology is adopted to realize remote transmission of environmental data and real-time push of warning information, the hardware deployment is flexible and adaptive to different terrain conditions, and the system can be widely applied to flood warning scenes of small watershed in hilly area, plain watershed and other types of regions, and provides reliable support for flood prevention and control of different scale watersheds. BRIEF DESCRIPTION OF DRAWINGS

[0069] Figure 1 is a flow chart of a flood pre-hydrological forecasting factor rapid identification method based on environmental perception provided according to an embodiment of the present application;

[0070] Figure 2 is a schematic diagram of an environmental factor perception module according to an embodiment of the present application;

[0071] Figure 3 is a schematic diagram of a factor rapid screening module according to an embodiment of the present application;

[0072] Figure 4 is a schematic diagram of a model training optimization module according to an embodiment of the present application. DETAILED DESCRIPTION

[0073] The present application will be further described below with reference to the drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present application, and cannot be used to limit the protection scope of the present application.

[0074] The flood pre-hydrological forecasting factor rapid identification method based on environmental perception provided by the embodiment of the present application, with reference to Figure 1 , the following steps S1-S5 are executed to complete the flood warning of the target area:

[0075] Step S1: collecting multi-dimensional environmental original data before the occurrence of the flood in the target area, the environmental original data including soil moisture content, cumulative rainfall in the previous period, vegetation coverage, terrain slope, surface temperature, and short-term weather forecast data;

[0076] Referring to Figure 2 The specific steps of step S1 are as follows:

[0077] Step S1.1: for the target area, a distributed soil moisture sensor is used to collect the soil moisture content at a depth of 5-30 cm at a preset frequency; in the embodiment, the soil moisture sensing unit uses a frequency domain reflectometer (FDR) type distributed soil moisture sensor, the sensor measurement range is 0-100%, the accuracy is ±2%, the sensor is vertically arranged in three layers of 5 cm, 15 cm, and 30 cm, and the collection frequency is 1 time / hour;

[0078] Step S1.2: a tipping bucket rain gauge is used to collect the cumulative rainfall in the target area within a preset time period at a preset frequency; in the embodiment, the resolution of the tipping bucket rain gauge is 0.2 mm, the measurement range is 0-200 mm / h, the cumulative rainfall in the basin within nearly 72 hours is collected, and the collection frequency is 1 time / 15 minutes;

[0079] Step S1.3: obtain the spatial and temporal distribution data of the vegetation coverage of the target area by remote sensing image interpretation combined with ground vegetation monitoring station data; in the embodiment, the spatial and temporal distribution data of the vegetation coverage of the basin are obtained by Landsat series satellite remote sensing image interpretation combined with ground vegetation monitoring station data, using the NDVI vegetation index method, and the data update cycle is 1 day;

[0080] Step S1.4: based on digital elevation (DEM) model data, extract the terrain parameters of the target area, including terrain slope, slope direction, and watershed convergence area; in the embodiment, based on 30-meter resolution digital elevation model data, terrain parameters such as terrain slope, slope direction, and watershed convergence area are extracted by ArcGIS software;

[0081] Step S1.5: receive the precipitation forecast and temperature forecast data of the target area within a preset time period issued by the meteorological department; in the embodiment, the 4G full-network communication module is used to receive the precipitation forecast and temperature forecast data of the future 24-72 hours issued by the meteorological department, and the data update frequency is 1 time / 6 hours;

[0082] Step S2: denoising, missing value filling, and standardization processing are performed on the environmental original data, and standardized environmental data are output;

[0083] The specific steps of step S2 are as follows:

[0084] Step S2.1: The sliding average method with a window size of 24 hours is used to denoise the environmental raw data, remove random interference signals, and the calculation formula is as follows:

[0085]

[0086] is the environmental raw data value at the i-th moment after denoising, N is the sliding window size, N = 24, is the value of the k-th moment of the environmental raw data;

[0087] Step S2.2: The K-Nearest Neighbor algorithm with K value of 5 is used to fill in the missing data in the environmental raw data. When the single-factor data missing rate exceeds 10%, linear interpolation method is used to supplement combined with the historical data of the same period in the past 5 years.

[0088]

[0089] is the estimated value of the missing data at the j-th moment, are the effective data values before and after the missing moment, respectively, are the times corresponding to the moments, respectively;

[0090] Step S2.3: The Z-score standardization method is used to normalize the environmental raw data processed in step S2.2, and the standardized environmental data conforming to the [0, 1] interval distribution is obtained, and the formula is as follows:

[0091]

[0092]

[0093] is the data mean, is the data standard deviation, is the standardized data mapped to the [0, 1] interval, are the minimum and maximum values of the Z-score standard score, respectively.

[0094] Step S3: The mutual information method is used to calculate the correlation degree of each standardized environmental data and historical flood occurrence event, and the standardized environmental data is preliminarily screened. Based on the random forest model, the standardized environmental data is sorted by importance, and the standardized environmental data is further screened to select the core prediction factor.

[0095] Referring to Figure 3 , the specific steps of step S3 are as follows: ​​​​​​​​

[0096] Step S3.1: Calculate the correlation degree of each standardized environmental data and historical flood occurrence event by mutual information method, respectively take each standardized environmental data as candidate factor in mutual information method, retain candidate factors with correlation degree greater than 0.6, the calculation formula is as follows:

[0097] The calculation formula of mutual information method is as follows:

[0098] ;

[0099] In the formula, represents the mutual information value of candidate factor X and flood occurrence event Y, that is, the correlation degree, represents the information entropy of candidate factor X, represents the information entropy of flood occurrence event Y, represents the joint information entropy of candidate factor X and flood occurrence event Y;

[0100] Wherein, the information entropy calculation formula of candidate factor X is as follows:

[0101] ;

[0102] Wherein, x is the observation value of candidate factor X, is the probability that the observation value of candidate factor X is equal to x, represents the logarithmic function with 2 as the base;

[0103] The information entropy calculation formula of flood occurrence event Y is as follows:

[0104] ;

[0105] Wherein, y is the specific state of historical flood occurrence event Y, including occurrence and non-occurrence; is the probability that the specific state of historical flood occurrence event Y is equal to y;

[0106] The joint information entropy calculation formula of candidate factor X and flood occurrence event Y is as follows:

[0107] ;

[0108] Wherein, is the probability that the observation value of candidate factor X is equal to x, and the specific state of historical flood occurrence event Y is equal to y;

[0109] Retain candidate factors with correlation degree > 0.6;

[0110] Step S3.2: input the candidate factors obtained in step S3.1 into a random forest model containing 100 decision trees, and calculate the importance score of each candidate factor by the out-of-bag data error, the calculation formula is as follows:

[0111] ;

[0112] Wherein, is the importance score of the candidate factor X, T = 100 is the number of decision trees, is the out-of-bag data original error rate of the tth decision tree, is the error rate after random permutation of the out-of-bag data of the candidate factor X;

[0113] Step S3.3: set the importance score threshold to 0.7, and select the candidate factors with importance score higher than the threshold as the core prediction factors. The whole screening process takes no more than 30 minutes.

[0114] Step S4: based on the historical flood event data and the corresponding core prediction factors, a hydrological prediction factor identification model is constructed, the core prediction factors are input, and the flood occurrence probability is output, and the hydrological prediction factor identification model is trained and iteratively optimized;

[0115] Referring to Figure 4 , the specific steps of step S4 are as follows:

[0116] Step S4.1: match the nearly 30 years of historical environmental factor data with the corresponding flood occurrence results, construct a training set and a validation set, the ratio of which is 7:3, and the data sample size is not less than 1000 groups;

[0117] Step S4.2: adopt gradient boosting decision tree (GBDT) to construct a hydrological prediction factor identification model, input the core prediction factors, and output the flood occurrence probability, the initial number of model iterations is set to 100 times; the model output formula of the mth iteration is as follows:

[0118] ;

[0119] Wherein, is the model output value after the mth iteration, is the model output value of the (m-1) th iteration, is the learning rate, is the weak classifier obtained by the mth training;

[0120] The model output value is mapped to the flood occurrence probability by the sigmoid function:

[0121] ;

[0122] wherein, is the probability of flood occurrence for a given input x, M is the total number of iterations, and F is the cumulative output value of the gradient boosting decision tree model after M rounds of iterations.

[0123] Step S4.3: Using the Bayesian optimization algorithm with Gaussian process as the probability model, the hyperparameters of the model are iteratively optimized, including learning rate, tree depth, leaf node number, and the number of iterations is 50 times, so that the model prediction accuracy is higher than 85%; the optimization function is as follows:

[0124]

[0125] wherein, is the combination of hyperparameters, represents the learning rate, is the tree depth, is the number of leaf nodes, is the search space of hyperparameters, wherein the search range of learning rate is 0.01-0.3, the search range of tree depth is 3-10, and the search range of leaf node number is 10-100; is the model validation set accuracy corresponding to the combination of hyperparameters, and after optimization, the model prediction accuracy is higher than 85%.

[0126] Step S5: Real-time output of core prediction factors and corresponding flood occurrence probability warning information, and completion of flood warning in the target area.

[0127] The specific steps of step S5 are as follows:

[0128] Step S5.1: Output the core prediction factor name, correlation value and importance score in list form, and support Excel format export;

[0129] Step S5.2: Based on the numerical combination of core prediction factors, output low, medium and high level flood occurrence probability warnings, wherein the low risk warning corresponds to flood occurrence probability < 30%, the medium risk warning corresponds to flood occurrence probability 30%-69%, and the high risk warning corresponds to flood occurrence probability ≥ 70%, and the warning information supports web and app end synchronous push.

[0130] The embodiment of the application also provides an environmental perception-based flood-precursor hydrological prediction factor rapid identification system, which comprises an environmental factor perception module, a data preprocessing module, a factor rapid screening module, a model training and optimization module, and an identification result output module, to realize the environmental perception-based flood-precursor hydrological prediction factor rapid identification method.

[0131] Environmental factor perception module: refer to Figure 2 ​, for collecting multi-dimensional environmental original data before the occurrence of floods, including soil moisture, cumulative rainfall, vegetation coverage, terrain slope, surface temperature and short-term weather forecast data for the target area;

[0132] The data preprocessing module is in communication connection with the environmental factor perception module, and performs denoising, missing value filling and standardization processing on the environmental original data, and outputs standardized environmental data.

[0133] The factor rapid screening module is in communication connection with the data preprocessing module, and calculates the correlation degree of each standardized environmental data and historical flood occurrence event by using the mutual information method, performs preliminary screening on the standardized environmental data, performs importance sorting on the standardized environmental data based on the random forest model, performs further screening on the standardized environmental data, and screens out core prediction factors. Figure 3 The model training optimization module is in communication connection with the factor rapid screening module, and constructs a hydrological prediction factor identification model based on historical flood event data and corresponding core prediction factors, takes the core prediction factors as input, takes the flood occurrence probability as output, trains and iteratively optimizes the hydrological prediction factor identification model.

[0134] Figure 4 The identification result output module is in communication connection with the model training optimization module, and outputs the core prediction factors and corresponding flood occurrence probability warning information in real time, and completes the flood warning of the target area.

[0135] The embodiment of the application also provides an electronic device, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the flood pre-hydrological prediction factor rapid identification method based on environmental perception when executing the program.

[0136] The embodiment of the application also provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the flood pre-hydrological prediction factor rapid identification method based on environmental perception.

[0137] The embodiment of the application also provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the flood pre-hydrological prediction factor rapid identification method based on environmental perception.

[0138] The embodiments of the application are described in detail above in combination with the drawings, but the application is not limited to the above-mentioned embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the purpose of the application.​

Claims

1. A rapid identification method for pre-flood hydrological forecasting factors based on environmental perception, characterized in that, Perform the following steps S1-S5 to complete the flood warning for the target area: Step S1: For the target area, collect multi-dimensional raw environmental data before the flood occurs. The raw environmental data includes soil moisture content, previous cumulative rainfall, vegetation coverage, terrain slope, surface temperature and short-term weather forecast data. Step S2: Perform noise reduction, missing value imputation, and standardization on the raw environmental data to output standardized environmental data; Step S3: The correlation between each standardized environmental data and historical flood events is calculated using the mutual information method. The standardized environmental data is initially screened. Based on the random forest model, the importance of the standardized environmental data is ranked. The standardized environmental data is further screened to identify the core forecasting factors. Step S4: Based on historical flood event data and corresponding core forecast factors, construct a hydrological forecast factor identification model. Using the core forecast factors as input and the probability of flood occurrence as output, train and iteratively optimize the hydrological forecast factor identification model. Step S5: Output the core forecast factors and corresponding flood occurrence probability early warning information in real time to complete the flood early warning for the target area.

2. The method for rapid identification of pre-flood hydrological forecasting factors based on environmental perception according to claim 1, characterized in that, The specific steps of step S1 are as follows: Step S1.1: For the target area, use distributed soil moisture sensors to collect soil moisture content at a preset depth at a preset frequency; Step S1.2: Use a tipping bucket rain gauge to collect the cumulative rainfall in the target area within a preset time period at a preset frequency; Step S1.3: Obtain the spatiotemporal distribution data of vegetation cover in the target area by interpreting remote sensing images and combining them with data from ground vegetation monitoring stations; Step S1.4: Based on the digital elevation model data, extract the topographic parameters of the target area, including topographic slope, aspect, and catchment area of ​​the watershed; Step S1.5: Receive precipitation and temperature forecast data for the target area within a preset time period issued by the meteorological department.

3. The method for rapid identification of pre-flood hydrological forecasting factors based on environmental perception according to claim 1, characterized in that, The specific steps of step S2 are as follows: Step S2.1: Denoise the raw environmental data using the moving average method to remove random interference signals. The calculation formula is as follows: ； in, Let N be the original environmental data value at time i after denoising, and N be the sliding window size. This represents the value of the raw environmental data at time k. Step S2.2: Use the K-nearest neighbor algorithm with a K value of 5 to fill the missing data in the original environmental data. When the missing rate of single-factor data exceeds 10%, use linear interpolation to supplement it by combining the historical data of the same period in the past 5 years. ； in, The missing data estimate at time j. These represent the valid data values ​​adjacent to the missing time point. These are the times corresponding to the given moments; Step S2.3: The raw environmental data processed in step S2.2 is normalized using the Z-score normalization method to obtain standardized environmental data that conforms to the distribution in the [0,1] interval.

4. The method for rapid identification of pre-flood hydrological forecasting factors based on environmental perception according to claim 1, characterized in that, The specific steps of step S3 are as follows: Step S3.1: Calculate the correlation between each standardized environmental data point and historical flood events using the mutual information method. Each standardized environmental data point is used as a candidate factor in the mutual information method, and candidate factors with a correlation greater than 0.6 are retained. The calculation formula is as follows: The formula for calculating mutual information is as follows: ； In the formula, This indicates the degree of association between candidate factor X and flood event Y. The information entropy of candidate factor X is represented by... The information entropy represents the event Y in which the flood occurs. The joint information entropy of candidate factor X and flood event Y; The formula for calculating the information entropy of candidate factor X is as follows: ； Where x is the observed value of the candidate factor X, Let x be the probability that the observed value of candidate factor X is equal to x. Represents the logarithmic function with base 2; The formula for calculating the information entropy of flood event Y is as follows: ； Where y represents the specific state of the historical flood event Y, including whether it has occurred or not; Let Y be the probability that the specific state of the historical flood event Y is equal to y. The formula for calculating the joint information entropy of candidate factor X and flood event Y is as follows: ； in, The probability that the observed value of candidate factor X is equal to x, and the specific state of historical flood event Y is equal to y; Preserve correlation Candidate factors > 0.6; Step S3.2: Input the candidate factors obtained in Step S3.1 into a random forest model containing 100 decision trees, and calculate the importance score of each candidate factor using the out-of-bag error. The calculation formula is as follows: ； in, The importance score of candidate factor X, T=100, is the number of decision trees. Let be the raw error rate of the out-of-bag data for the t-th decision tree. The error rate is the result of random permutation of the out-of-bag data for candidate factor X. Step S3.3: Set the importance score threshold to 0.7 and select candidate factors with importance scores higher than the threshold as core forecast factors.

5. The method for rapid identification of pre-flood hydrological forecasting factors based on environmental perception according to claim 1, characterized in that, The specific steps of step S4 are as follows: Step S4.1: Match the historical environmental factor data of nearly 30 years with the corresponding flood occurrence results to construct a training set and a validation set with a ratio of 7:3 and a data sample size of no less than 1000 sets; Step S4.2: Construct a hydrological forecasting factor identification model using a gradient boosting decision tree, with core forecasting factors as input and flood occurrence probability as output. The initial number of model iterations is set to 100. The model output formula for the m-th iteration is as follows: ； in, This represents the model output value after the m-th iteration. This represents the model output value in the (m-1)th iteration. For learning rate, This is the weak classifier obtained in the m-th training round; The sigmoid function is used to output the model value. Mapped to the probability of flood occurrence: ； in, Let x be the probability of a flood occurring given an input x, M be the total number of iterations, and F be the cumulative output value of the gradient boosting decision tree model after M iterations. Step S4.3: Using a Gaussian process as the Bayesian optimization algorithm for the probabilistic model, iteratively optimize the model's hyperparameters, including the learning rate, tree depth, and number of leaf nodes. The iteration count is 50, aiming to achieve a model prediction accuracy higher than 85%. The optimization function is as follows: ； in, For hyperparameter combination, Indicates the learning rate. For tree depth, The number of leaf nodes. The hyperparameter search space is defined as follows: the search range for the learning rate is 0.01-0.3, the search range for the tree depth is 3-10, and the search range for the number of leaf nodes is 10-100. The model validation set accuracy is the result of the hyperparameter combination. After optimization, the model prediction accuracy is higher than 85%.

6. The method for rapid identification of pre-flood hydrological forecasting factors based on environmental perception according to claim 1, characterized in that, The specific steps of step S5 are as follows: Step S5.1: Output the core forecast factor names, correlation values, and importance scores in list format, supporting export to Excel format; Step S5.2: Based on the numerical combination of core forecast factors, output flood probability warnings at three levels: low, medium, and high. The low-risk warning corresponds to a flood probability of <30%, the medium-risk warning corresponds to a flood probability of 30%-69%, and the high-risk warning corresponds to a flood probability of ≥70%. Warning information can be pushed simultaneously to web pages and mini-programs.

7. A rapid identification system for pre-flood hydrological forecasting factors based on environmental perception, characterized in that, It includes an environmental factor perception module, a data preprocessing module, a factor rapid screening module, a model training and optimization module, and an identification result output module, to realize the rapid identification method for pre-flood hydrological forecasting factors based on environmental perception as described in any one of claims 1-6. Environmental Factor Sensing Module: Used to collect multi-dimensional raw environmental data for the target area before the occurrence of floods. The raw environmental data includes soil moisture content, previous cumulative rainfall, vegetation coverage, terrain slope, surface temperature and short-term weather forecast data. Data preprocessing module: Communicatively connected to the environmental factor sensing module, it performs noise reduction, missing value imputation and standardization processing on the raw environmental data, and outputs standardized environmental data; Factor rapid screening module: It is connected to the data preprocessing module and uses the mutual information method to calculate the correlation between each standardized environmental data and historical flood events. It performs preliminary screening of standardized environmental data, ranks the standardized environmental data by importance based on the random forest model, and further screens the standardized environmental data to select core forecasting factors. Model training and optimization module: It is connected to the factor fast screening module and constructs a hydrological forecast factor identification model based on historical flood event data and corresponding core forecast factors. The model is trained and iteratively optimized with the core forecast factors as input and the probability of flood occurrence as output. The identification result output module communicates with the model training and optimization module to output the core forecast factors and corresponding flood occurrence probability early warning information in real time, thus completing the flood early warning for the target area.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the rapid identification method for pre-flood hydrological forecasting factors based on environmental perception as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the rapid identification method for pre-flood hydrological forecasting factors based on environmental perception as described in any one of claims 1 to 5.

Images