A demonstration method for heavy rain types with multiple scenarios and return periods

Abstract

The invention relates to the field of hydrological technology, in particular to a method for demonstrating rainstorm types with multiple return periods, comprising the following steps: division of rain type scenarios; derivation of rainstorm intensity formula; generation of rain type data set; and cluster analysis of rain type curves with multiple scenarios. The invention selects the Gumbel distribution curve for probability distribution model fitting through comparative analysis, and solves the rainstorm intensity formula and the rainstorm intensity formula with a single return period by using the least square method. Based on the rainstorm intensity formula with a single return period and the division of rain type scenarios, a multi-scenario rainfall process database is established by using the Chicago rainfall method and a SMOTE sample expansion algorithm, and a multi-scenario rainfall process cluster analysis is performed by using a K-means clustering algorithm, and a multi-scenario rainfall process simulation function is established to obtain rainfall types with different return periods and different durations in multiple scenarios.

Description

Technical Field

[0001] The invention relates to the field of hydrological technology, and in particular to a method for demonstrating rainstorm types with multiple scenario recurrence periods. Background Art

[0002] The rainstorm pattern refers to the characteristics of the change of rainstorm intensity over time in different rainfall periods. It is a linear expression of the rainfall process in different rainfall durations or different rainfall periods, and provides data support for the calculation of surface runoff, peak flow, and flood evolution. To design a rainstorm pattern, it is necessary to comprehensively consider the geographical characteristics of the study basin and the causes of rainfall. By describing the spatiotemporal changes and spatiotemporal evolution characteristics of the rainstorm, it can truly reflect the actual occurrence process of a rainstorm and describe the change of rainfall intensity over time.

[0003] The traditional rain pattern design model is affected by factors such as insufficient rainfall data, fixed rain peak position, too sharp rain peak, too much reliance on rainfall data, complex reasoning and deduction process, and uniform rainfall pattern. Summary of the invention

[0004] The object of the present invention is to provide a method for demonstrating rainstorm types with multiple scenarios and return periods, so as to solve the problems raised in the above-mentioned background technology.

[0005] The technical solution of the present invention is: a method for demonstrating rainstorm types with multiple scenarios and return periods, comprising the following steps:

[0006] S1. Classification of rainfall scenarios: According to the rainfall sample data of the rainfall station observation data, the rainfall events and rainfall processes are counted, and the rain peak position coefficient corresponding to each actual rainfall process is calculated. The rain peak position coefficient is used to divide the rainfall process into different rainfall scenarios. The calculation formula is:

[0007]

[0008] In the formula, r i is the rain peak position coefficient, t i is the peak rainfall moment, T i is the duration of rainfall;

[0009] S2. Derivation of rainstorm intensity formula: Different theoretical frequency curves such as Pearson III distribution curve, exponential distribution curve and Gumbel distribution curve are used for fitting and solving respectively, and the optimal distribution curve is selected for probability distribution model fitting to obtain the relationship between recurrence period, rainfall intensity and rainfall duration, that is, the relationship value of p, i, t. According to the relationship value of p, i, t, the least squares method is used to obtain the parameters A1, b, C, n, and the rainstorm intensity formula can be calculated. The basic form of the rainstorm intensity formula is:

[0010]

[0011] S3, Rain type data set generation: The classification of rain type scenarios, the rainstorm intensity formulas with different return periods, and the Chicago rain type method are comprehensively considered to construct a basic rainfall process database;

[0012] S4. Cluster analysis of multi-scenario rainfall pattern curves: The K-means clustering algorithm is used to perform cluster analysis on the rainfall pattern data sets with different return periods and durations in multiple scenarios to obtain rainfall patterns with multiple scenarios, different return periods and different durations.

[0013] Preferably, in S3, the specific steps of generating the rainfall type dataset include:

[0014] S31. Chicago Rain Pattern Method generates basic rain patterns: The total rainfall of a rainstorm with a duration of t is calculated based on the empirical relationship between rainfall duration and average rainstorm intensity, and the instantaneous rainfall intensity at time t is calculated by partial derivatives. The position of the rain peak is corrected by introducing the rain peak position coefficient r, and the time series of rainfall duration is divided into two parts: the pre-peak part and the post-peak part. Assume that the instantaneous intensity before the peak is i(t b ), the corresponding duration is t b ; The instantaneous intensity after the peak is i(t a ), the corresponding duration is t a The instantaneous rainfall intensity before and after the rain peak can be calculated by the following formula:

[0015]

[0016] S32. Construction of rainfall process database: Using the Chicago rainfall pattern, the basic rainfall patterns with different durations and return periods under multiple scenarios are deduced, so as to establish a rainfall process database;

[0017] S33. Rainfall type data expansion: Based on the rainfall process database, the improved SMOTE algorithm is used to expand the samples of the rainfall process database to obtain rainfall type data sets with different durations and return periods under multiple scenarios.

[0018] Preferably, in S32, the step of constructing the rainfall process database specifically includes:

[0019] S321. In the process of constructing the rainfall process database, the rainfall type scenarios are: the rain peak is in front, the rain peak is in the middle, and the rain peak is in the back. The rainfall duration is 2h, 6h, 12h, and 24h, a total of 4 durations. The rainfall recurrence period is 5a, 10a, 20a, 30a, 50a, and 100a, a total of 6 recurrence periods. The step length is 5min for 2h rainfall duration, 15min for 6h rainfall duration, 30min for 12h rainfall duration, and 60min for 24h rainfall duration.

[0020] S322. For the rainfall pattern scenario with the rainfall peak being relatively early (0.15 ≤ r ≤ 0.45), according to formula (3), the parameters of the rainfall intensity formula for different return periods (A, b, n), and the range of the rainfall peak position coefficient r (with a step size of 0.03), respectively derive the basic rainfall patterns for different durations and different return periods;

[0021] S323. For the rainfall pattern scenario with the rainfall peak being in the middle (0.45 < r < 0.55), according to formula (3), the parameters of the rainfall intensity formula for different return periods (A, b, n), and the range of the rainfall peak position coefficient r (with a step size of 0.01), respectively derive the basic rainfall patterns for different durations and different return periods;

[0022] S324. For the rainfall pattern scenario with the rainfall peak being relatively late (0.55 ≤ r ≤ 0.85), according to formula (3), the parameters of the rainfall intensity formula for different return periods (A, b, n), and the range of the rainfall peak position coefficient r (with a step size of 0.03), respectively derive the basic rainfall patterns for different durations and different return periods.

[0023] Preferably, in S33, the specific steps of expanding the rainfall pattern data include:

[0024] S331. Take the data sets for each duration and each return period in the rainfall process database as the minority class samples X, traverse each sample, and for each sample x in the minority class samples i , calculate its distances to all samples in the minority class sample set, using the Euclidean distance, to obtain the k nearest neighbors of sample x i ;

[0025] S332. According to the situation of the sample imbalance ratio, determine the sampling magnification N (take N = 1). For each minority class sample x i , randomly select N nearest neighbors from the k nearest neighbors. Assume the selected nearest neighbors are: x (1) , x (2) , ··· x (N) ;

[0026] S333. For each randomly selected nearest neighbor x (j) (j = 1, 2, 3, ··· N), respectively perform linear interpolation with the original sample to construct a new sample set X new ;

[0027] S334. Judge whether the sample set X new overlaps: If there are overlapping samples, perform the de - overlapping operation and output the de - overlapping samples X' new ; If there are no overlapping samples, output the new sample data set X" new ;

[0028] S335. Statistically output the de - overlapping samples X' newOr a new sample dataset X" new The number of samples is compared with the set threshold: if it is less than the threshold, a sampling multiple of N = 1 is used for iterative expansion; if it is greater than or equal to the threshold, the final rainfall type data set is generated.

[0029] Preferably, in S4, the specific steps of cluster analysis of multi-scenario rainfall pattern curves include:

[0030] S41, clustering the data sets of a certain rainfall scenario, different return periods and different durations, k = 1;

[0031] S42, randomly select a sample point in the data set and use it as the center of the cluster, that is, the "cluster line";

[0032] S43, setting the bandwidth limit of path finding and introducing LB_Keogh distance to optimize the dynamic time planning to calculate the distance from all samples of the rain type data set to the "cluster line";

[0033] S44, according to each obtained cluster, find the sample point in each cluster that is closest to all sample points, and use it as a new "cluster line";

[0034] S45. Repeat the above operation, continuously iterate to obtain the optimal "cluster line", and return the cluster center value.

[0035] The present invention provides a method for demonstrating rainstorm types with multiple scenarios and return periods through improvement, which has the following improvements and advantages compared with the prior art:

[0036] The present invention selects the Gumbel distribution curve for probability distribution model fitting through comparative analysis, and uses the least square method to solve the rainstorm intensity formula and the single return period rainstorm intensity formula; based on the single return period rainstorm intensity formula and the division of rain type scenarios, the Chicago rain type method and the SMOTE sample expansion algorithm are used to establish a multi-scenario rainfall process database, and the K-means clustering algorithm is used to perform clustering analysis of multi-scenario rain type curves, and a multi-scenario return period rainfall process simulation function is established to obtain rain types with different return periods and different durations in multiple scenarios; BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The present invention will be further explained below in conjunction with the accompanying drawings and embodiments:

[0038] Figure 1 It is a flowchart of a demonstration method of multi-scenario return period rainstorm type;

[0039] Figure 2 The distribution diagram of the rain peak position coefficient r corresponding to the rainfall events;

[0040] Figure 3 It is a sample set with different duration and return period of rainfall peak deviation;

[0041] Figure 4 It is a sample set with different duration and return period of rain peak center;

[0042] Figure 5 It is a sample set with different durations and return periods after the rain peak;

[0043] Figure 6 The rainfall type clustering result curve for different rainfall scenarios with different return periods (2 / 6 / 12 / 24h) Figure 1 ;

[0044] Figure 7 The rainfall type clustering result curve for different rainfall scenarios with different return periods (2 / 6 / 12 / 24h) Figure 2 ;

[0045] Figure 8 The rainfall type clustering result curve for different rainfall scenarios with different return periods (2 / 6 / 12 / 24h) Figure 3 ;

[0046] Fig. 9 The rainfall type clustering result curve for different rainfall scenarios with different return periods (2 / 6 / 12 / 24h) Figure 4 ;

[0047] Fig.10 The rainfall type clustering result curve for different rainfall scenarios with different return periods (2 / 6 / 12 / 24h) Figure 5 ;

[0048] Fig.11 The rainfall type clustering result curve for different rainfall scenarios with different return periods (2 / 6 / 12 / 24h) Figure 6 . DETAILED DESCRIPTION

[0049] The present invention is described in detail below, and the technical solutions in the embodiments of the present invention are clearly and completely described. 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.

[0050] The present invention provides a method for demonstrating rainstorm types with multiple scenarios and return periods through improvement. The technical solution of the present invention is:

[0051] The specific implementation process is as follows Figure 1 As shown, a method for demonstrating a multi-scenario return period rainstorm type includes the following steps:

[0052] S1. Classification of rain type scenarios: According to the rainfall sample data of the rainfall station observation data, the rainfall events and rainfall processes are counted, and the rain peak position coefficient corresponding to each actual rainfall process is calculated. The rain peak position coefficient is used to divide the rainfall process into different rain type scenarios (rain peak in front, rain peak in the middle, rain peak in the back). The rain peak position coefficient is a parameter that characterizes the rain peak position of the rainstorm intensity process. Its calculation formula is:

[0053]

[0054] In the formula, r i is the rain peak position coefficient, t i is the peak rainfall moment, T i is the rainfall duration. The distribution of the rain peak position coefficient r corresponding to the rainfall sample data is as follows: Figure 2 As shown in the figure, according to the distribution diagram of the rain peak position coefficient r corresponding to the rainfall events, it can be found that most of the rain peak position coefficients are concentrated between 0.15 and 0.85. Based on this, the corresponding rain type scenarios can be divided into: the rain peak is in front (0.15-0.45), the rain peak is in the middle (0.45-0.55), and the rain peak is in the back (0.55-0.85);

[0055] S2. Derivation of rainstorm intensity formula: The rainstorm intensity formula reflects the extreme intensity and average intensity of the rainstorm. The basic form of the rainstorm intensity formula is:

[0056]

[0057] According to the continuous self-recording rainfall records of the rain gauge station in the past 30 years, the annual maximum value method was used to select the rainfall data of each rainfall duration. Different theoretical frequency curves such as Pearson III distribution curve, exponential distribution curve and Gumbel distribution curve were used for fitting and solving respectively. The optimal distribution curve was selected for probability distribution model fitting to obtain the relationship between return period, rainfall intensity and rainfall duration. The least square method was used to solve the rainstorm intensity formula and the single return period rainstorm intensity formula. The average absolute variance value was 0.049 (mm / min), and the average relative variance value was 6.64%, which met the fitting accuracy requirements. The empirical parameters of the single return period rainstorm intensity formula were obtained: A1, b, C, n;

[0058] S3. Rain type data set generation: The process of establishing a rainfall process database is actually a process of diversified rain type design. The division of rainfall type scenarios, the rainstorm intensity formula with different return periods, and the Chicago rainfall type method are comprehensively considered to build a basic rainfall process database. The specific steps of generating a rainfall type data set include:

[0059] S31. Chicago rain pattern method generates basic rain pattern: The Chicago rain pattern is similar to a composite rain pattern, which is composed of the maximum rainfall intensity of different rainfall durations under a certain recurrence period. The determination of its rain pattern is based on the IDF relationship curve under a specific recurrence period. According to the empirical relationship between rainfall duration and average rainstorm intensity (rainstorm intensity formula under a certain recurrence period), the total rainfall of the rainstorm with a duration of t is calculated, and the partial derivative is calculated to calculate the instantaneous rainfall intensity at time t. The position of the rain peak is corrected by introducing the rain peak position coefficient r, and the time series of the rainfall duration is divided into two parts: before the peak and after the peak. Assume that the instantaneous intensity before the peak is i(t b ), the corresponding duration is t b ; The instantaneous intensity after the peak is i(t a ), the corresponding duration is t a The instantaneous rainfall intensity before and after the rain peak can be calculated by the following formula:

[0060]

[0061] S32. Construction of rainfall process database: Using the Chicago rainfall pattern, the basic rainfall patterns with different durations (2h, 6h, 12h, 24h) and different return periods (5a, 10a, 20a, 30a, 50a, 100a) under multiple scenarios are deduced, so as to establish a rainfall process database, such as Figure 3 , Figure 4 , Figure 5 As shown in Figure 1, the steps of constructing the rainfall process database specifically include:

[0062] S321, in the process of constructing the rainfall process database, the rain type scenarios adopt three scenarios: the rain peak is in front, the rain peak is in the middle, and the rain peak is in the back. The rainfall duration adopts 4 durations, namely 2h, 6h, 12h, and 24h. The rainfall recurrence period adopts 6 recurrence periods, namely 5a, 10a, 20a, 30a, 50a, and 100a. In order to make different rainfall durations more reasonably represented, the present invention adopts a step length of 5min to represent a 2h rainfall duration, a step length of 15min to represent a 6h rainfall duration, a step length of 30min to represent a 12h rainfall duration, and a step length of 60min to represent a 24h rainfall duration;

[0063] S322. For the rain type scenario with the rain peak shifted forward (0.15≤r≤0.45), according to formula (3) and the parameters (A, b, n) of the rainstorm intensity formula with different return periods, and the range of the rain peak position coefficient r (with a step size of 0.03), the basic rain types with different durations and return periods are derived respectively;

[0064] S323. For the rainfall pattern scenario with the rainfall peak in the middle (0.45 < r < 0.55), according to formula (3), the parameters of the rainfall intensity formula for different return periods (A, b, n), and the range of the rainfall peak position coefficient r (with a step size of 0.01), the basic rainfall patterns for different durations and different return periods are respectively derived.

[0065] S324. For the rainfall pattern scenario with the rainfall peak at the rear (0.55 ≤ r ≤ 0.85), according to formula (3), the parameters of the rainfall intensity formula for different return periods (A, b, n), and the range of the rainfall peak position coefficient r (with a step size of 0.03), the basic rainfall patterns for different durations and different return periods are respectively derived.

[0066] S33. Rainfall pattern data augmentation: The number of samples for each return period and each duration in the rainfall process database is relatively small, and it is derived under specific conditions, lacking representativeness and universality. Therefore, based on the rainfall process database, the present invention uses an improved SMOTE algorithm to augment the samples of the rainfall process database, obtaining a rainfall pattern dataset with different durations (2h, 6h, 12h, 24h) and different return periods (5a, 10a, 20a, 30a, 50a, 100a) under multiple scenarios. The specific steps of the rainfall pattern data augmentation include:

[0067] S331. Take the datasets of each duration and each return period in the rainfall process database as the minority class samples X, and traverse each sample. For each sample x in the minority class samples i , calculate its distances to all samples in the minority class sample set, using the Euclidean distance, to obtain the k nearest neighbors of sample x i .

[0068] S332. According to the sample imbalance ratio situation, determine the sampling magnification N (take N = 1). For each minority class sample x i , randomly select N nearest neighbors from the k nearest neighbors. Suppose the selected nearest neighbors are: x (1) , x (2) , ··· x (N) ;

[0069] S333. For each randomly selected nearest neighbor x (j) (j = 1, 2, 3, ··· N), perform linear interpolation with the original sample respectively to construct a new sample set X new ;

[0070] S334. Judge whether there is overlap in the sample set X new . If there is sample overlap, perform the de - overlapping operation and output the de - overlapping samples X' new ; if there is no sample overlap, output the new sample dataset X'' new ;

[0071] S335, statistically outputting overlapping samples X' new Or a new sample dataset X" new The number of samples is compared with the set threshold. If it is less than the threshold, the sampling rate is N=1 and iterative expansion is performed; if it is greater than or equal to the threshold, the final rain type data set is generated;

[0072] S4. Cluster analysis of multi-scenario rainfall pattern curves: The K-means clustering algorithm is used to perform cluster analysis on the rainfall pattern data sets with different return periods and durations under multiple scenarios. That is, the curves with the same changes are fitted by clustering to obtain rainfall patterns with multiple scenarios, different return periods and different durations, such as Figure 6-Figure 11 As shown, the specific steps of the multi-scenario rainfall pattern curve cluster analysis include:

[0073] S41. Since the rain type data set has been determined to which category it belongs after operations such as before and after expansion, when clustering data sets of a certain rain type scenario, different return periods and different durations, the cluster category is determined, that is, clustering a certain category of data sets, k=1;

[0074] S42, randomly select a sample point in the data set. For a time series, this sample point is actually a randomly selected time series, which is used as the center of the cluster, i.e., the "cluster line";

[0075] S43, setting the bandwidth limit (search area) of path finding and introducing LB_Keogh distance to optimize the dynamic time planning (DTW) to calculate the distance from all samples of the rain type data set to the "cluster line";

[0076] S44, according to each obtained cluster, find the sample point (line) in each cluster that is closest to all sample points, and use it as a new "cluster line";

[0077] S45. Repeat the above operation, continuously iterate to obtain the optimal "cluster line", and return the cluster center value. For time series clustering, the cluster center value is a time series.

[0078] The above description enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for demonstrating rainstorm types with multiple scenarios and return periods, characterized by: The following steps are involved: S1. Classification of rainfall scenarios: According to the rainfall sample data of the rainfall station observation data, the rainfall events and rainfall processes are counted, and the rain peak position coefficient corresponding to each actual rainfall process is calculated. The rain peak position coefficient is used to divide the rainfall process into different rainfall scenarios. The calculation formula is: In the formula, r i is the rain peak position coefficient, t i is the peak rainfall moment, T i is the duration of rainfall; S2. Derivation of rainstorm intensity formula: Using the Pearson III distribution curve, exponential distribution curve and Gumbel distribution curve, different theoretical frequency curves are fitted and solved respectively, and the optimal distribution curve is selected for probability distribution model fitting to obtain the relationship between the return period, rainfall intensity and rainfall duration, that is, the relationship value of p, i, t. According to the relationship value of p, i, t, the least squares method is used to obtain the parameters A1, b, C, n, and the rainstorm intensity formula can be calculated. The basic form of the rainstorm intensity formula is: S3, Rain type data set generation: The classification of rain type scenarios, the rainstorm intensity formulas with different return periods, and the Chicago rain type method are comprehensively considered to construct a basic rainfall process database; S4. Cluster analysis of multi-scenario rainfall pattern curves: K-means clustering algorithm is used to perform cluster analysis on rainfall pattern data sets with different return periods and durations in multiple scenarios. The specific steps include: S41, clustering the data sets of a certain rainfall scenario, different return periods and different durations, k = 1; S42, randomly select a sample point in the data set and use it as the center of the cluster, that is, the "cluster line"; S43, setting the bandwidth limit of path finding and introducing LB_Keogh distance to optimize the dynamic time planning to calculate the distance from all samples of the rain type data set to the "cluster line"; S44, according to each obtained cluster, find the sample point in each cluster that is closest to all sample points, and use it as a new "cluster line"; S45, repeat the above operation, continuously iterate to obtain the optimal "cluster line", and return the cluster center value; Obtain rainfall patterns with multiple scenarios, different return periods, and different durations.

2. The method for demonstrating a multi-scenario return period rainstorm type according to claim 1, characterized in that: In S3, the specific steps of generating the rainfall type dataset include: S31. Chicago Rain Pattern Method generates basic rain patterns: The total rainfall of a rainstorm with a duration of t is calculated based on the empirical relationship between rainfall duration and average rainstorm intensity, and the instantaneous rainfall intensity at time t is calculated by partial derivatives. The position of the rain peak is corrected by introducing the rain peak position coefficient r, and the time series of rainfall duration is divided into two parts: the pre-peak part and the post-peak part. Assume that the instantaneous intensity before the peak is i(t b ), the corresponding duration is t b ; The instantaneous intensity after the peak is i(t a ), the corresponding duration is t a , then the instantaneous rainfall intensity before and after the rain peak can be calculated by the following formula: S32. Construction of rainfall process database: Using the Chicago rainfall pattern, the basic rainfall patterns with different durations and return periods under multiple scenarios are deduced, so as to establish a rainfall process database; S33. Rainfall type data expansion: Based on the rainfall process database, the improved SMOTE algorithm is used to expand the samples of the rainfall process database to obtain rainfall type data sets with different durations and return periods under multiple scenarios.

3. The method for demonstrating a multi-scenario return period rainstorm type according to claim 2, characterized in that: In S32, the step of constructing the rainfall process database specifically includes: S321. In the process of constructing the rainfall process database, the rainfall type scenarios are: the rain peak is in front, the rain peak is in the middle, and the rain peak is in the back. The rainfall duration is 2h, 6h, 12h, and 24h, a total of 4 durations. The rainfall recurrence period is 5a, 10a, 20a, 30a, 50a, and 100a, a total of 6 recurrence periods. The step length is 5min for 2h rainfall duration, 15min for 6h rainfall duration, 30min for 12h rainfall duration, and 60min for 24h rainfall duration. S322. For the rainfall pattern scenario where the rainfall peak is 0.15 ≤ r ≤ 0.45 ahead, based on the formula, the parameters A, b, n of the rainfall intensity formula for different return periods, and the range of the rainfall peak position coefficient r, with a step size of 0.03, respectively derive the basic rainfall patterns for different durations and different return periods; S323. For the rainfall pattern scenario where the rainfall peak is centered at 0.45 < r < 0.55, based on the formula, the parameters A, b, n of the rainfall intensity formula for different return periods, and the range of the rainfall peak position coefficient r, with a step size of 0.01, respectively derive the basic rainfall patterns for different durations and different return periods; S324. For the rainfall pattern scenario where the rainfall peak is 0.55 ≤ r ≤ 0.85 behind, based on the formula, the parameters A, b, n of the rainfall intensity formula for different return periods, and the range of the rainfall peak position coefficient r, with a step size of 0.03, respectively derive the basic rainfall patterns for different durations and different return periods.

4. The method for demonstrating a multi-scenario return period rainstorm type according to claim 2, characterized in that: In S33, the specific steps for expanding the rainfall pattern data include: S331, taking the datasets of each duration and return period in the rainfall process database as minority class samples X, traversing each sample, for each sample x in the minority class samples i , calculate its distance to all samples in the minority class sample set, using Euclidean distance, and get sample x i The k nearest neighbors of ; S332, according to the sample imbalance ratio, determine the sampling ratio M, take M = 1, for each minority class sample x i , randomly select N neighbors from the k nearest neighbors, assuming that the selected neighbors are: x (1) , x (2) ,···x (N) ; S333, for each randomly selected neighbor x (j) (j=1, 2, 3, ···N), respectively, perform linear interpolation with the original samples to construct a new sample set X new ; S334, judging sample set X new Whether overlap occurs: If there is sample overlap, perform a de-overlapping operation and output the de-overlapping sample X' new ; If there is no sample overlap, then output a new sample data set X' n ' ew ; S335, statistically outputting overlapping samples X' new Or a new sample dataset X' n ' ew The number of samples is compared with the set threshold: if it is less than the threshold, a sampling multiple of M = 1 is used for iterative expansion; if it is greater than or equal to the threshold, the final rain type data set is generated.

Images