JP2023166977A - Adaptive division method for rainfall runoff frequency in period - Google Patents

Abstract

To provide an adaptive division method for the rainfall runoff frequency in a period.SOLUTION: A method includes: a step S1 of collecting continuous rainfall runoff data in a known region; a step S2 of automatically calculating and recognizing peaks and valleys; a step S3 of automatically dividing the frequency of rainfall, recognizing the start and end times of rainfall, and calculating the cumulative rainfall amount every time; a step S4 of automatically dividing the frequency of rainfall runoff, and recognizing the start and end times of corresponding frequency of floods based on the start and end times of rainfall; a step S5 of recognizing the number of peaks in the single flood, and specifying the number of peaks in the flood based on the difference between the peaks and valleys in the single flood; a step S6 of calculating the correlation coefficient between the cumulative rainfall amount and the maximum increase range of flow rate in the single rainfall runoff; a step S7 of optimizing a parameter; and a step S8 of selecting a rainfall runoff process satisfying requirements.SELECTED DRAWING: Figure 1

Description

The present invention relates to the field of hydrological forecasting, and more particularly to a method for adaptively dividing the number of rainfall runoffs within a period.

If a certain amount of rainfall changes the cross-sectional flow of a river, rainfall does not necessarily change the flow. Conversely, fluctuations in river flow may be due to rainfall, and may be due to other causes such as upstream flow fluctuations or dam openings. Currently, it is relatively easy to extract a single flood from continuous hydrological data, but the extraction of complete rainfall runoff is mainly artificial and empirically recognized, with low efficiency or It is not possible to strictly extract the rainfall that corresponds to the floods, and the accuracy of the extraction results is low.

The invention patent application with publication number CN109785979A discloses a method for defining a flood rainfall runoff process, which provides a mathematical recognition method for flood peak discharge, but for both flood and rainfall start and end times. No specific recognition method is provided, and it is necessary to make an artificial determination. The invention patent application with publication number CN112561214A discloses a method and system for automatically recognizing the number of floods, but the invention does not provide a method to specifically search for the start and end position of the rainfall process; Since the rainfall process and the flood process are not associated, in the third embodiment, the durations of the third and fifth floods as a result of division are too long. The invention patent application with publication number CN110929956B discloses a preferred method for timely forecasting of floods based on machine learning, and the invention describes the following: The method of relating rainfall and flooding is that if the center of mass of the rainfall event belongs to a predetermined time window related to the flooding event, then the rainfall event and the flooding event are considered to be correlated events, but if the rainfall event is a concavity deformation, It is difficult for the center of mass to fit into the time window, and the patent does not describe how to set the time length of the time window.

The number of rainfall runoffs is the main data for studying rainfall runoff in a basin and is the basic data for hydrological forecasting, so we researched methods that can rationally and accurately extract the number of rainfall runoffs, and summarized artificial experience into scientific laws. , can be automatically processed in large quantities using computer language, and the number of rainfall runoff can be divided quickly and efficiently.

The technical problem to be solved by the present invention is to provide an adaptive dividing method for the number of rain runoffs within a period, so that related floods and rainfall form one complete rain runoff, and divide the number of rain runoffs. It improves the accuracy rate, has high adaptability, can realize complete automation, and can quickly and efficiently divide the number of rainfall runoffs.

In order to solve the above technical problem, the technical solutions used by the present invention are as follows. In order to solve the above technical problem, the technical solutions used by the present invention are as follows. An adaptive dividing method for the number of rainfall runoff within a period, the method comprising:
Step S1 of collecting continuous rainfall runoff data in a known area;
Step S2 of automatically calculating and recognizing peaks and valleys;
step S3 of automatically dividing the number of rainfalls, setting a rainfall intensity threshold and a rainfall interval time threshold, recognizing the start and end times of rainfall, and calculating the cumulative rainfall amount each time;
Step S4: automatically dividing the number of rain runoffs, setting a time window threshold, and recognizing the start and end times of the corresponding number of floods based on the start and end times of the rain;
Step S5 of recognizing the number of peaks of one flood and identifying the number of peaks of flood based on the difference between the peak and the valley in one flood;
Step S6 of calculating a correlation coefficient between the cumulative rainfall amount p _n ^sum of one rain runoff and the maximum increase width Q _n ^up of the flow rate;
In step S7 of optimizing the parameters, the optimization evaluation index is such that the correlation coefficient between the cumulative rainfall p _n ^sum and the maximum increase width Q _n ^up of the flow rate is the maximum in step S6, and the peak of the flood is step S7, which satisfies that the correlation coefficient decreases as the number increases;
Step S8 of selecting rain runoff processes that meet the requirements.

The step S1 is as follows:
Step S11 of collecting flow rate data and time sequence that need to be divided and are continuous and uninterrupted, which are included in the known area;
Acquire the area precipitation amount and time sequence of the corresponding water catchment area in step S11, request that the time step size is the same and that the data is continuous and uninterrupted, and create a buffer before and after the data that needs to be divided. Step S12 of setting a period is included.

In a preferred solution, the step S2 includes:
Step S21 of performing smoothing processing on a flow rate process with obvious sawtooth characteristics and directly using the smooth flow rate process;
Mark the peak and valley on the flow rate data, record the peak flow rate and peak time, and the valley flow rate and valley time, determine the peak flow rate and peak time according to the formula below,
q' _t-1 <q' _t and q' _t >q' _t+1
In the formula, q' is the smoothed flow rate, q' _t-1 , q' _t , q' _t+1 are the smoothed flow rates at t-1, t, and t+1, and t is the peak time, The original flow rate corresponding to that point in time is the peak flow rate;
Determine the valley flow rate and valley point according to the formula below,
q' _t-1 >q' _t and q' _t <q' _t+1
a step S22 in which the time point t is a valley time point and the original flow rate corresponding to the time point is a valley flow rate;
The method includes a step S23 of recording and encoding the peak time and the corresponding peak flow rate, and the trough time and the corresponding valley flow rate in time order.

Ｎａｓｈ－Ｓｕｔｃｌｉｆｆｅ効率係数（ＮＳＥ）の計算式は以下のとおりであり、

式では、ｙ_ｉは観測値であり、

は観測値の平均値であり、

はフィッティング値であり、ｙ_ｉ ^ｐｒｅｄはフィッティング値である。 In a preferred method, in step S21, the smoothing process uses a Savitzky-Golay filter, and the calculation formula is as follows:
_Assuming that the width of the smoothing window _of the filter is n= _2m +1, each measurement data Fitting the data, x _i can be fitted using a polynomial,
p(x _i )=a ₀ +a ₁ x _i +a ₂ x _i ² +. ．． ．． +a _k-1 x _i ^k-1
The determination evaluation index for the parameters m and k includes the determinable coefficient R ² or the Nash-Sutcliffe efficiency coefficient (NSE), and the larger R ² or NSE, the more suitable the parameter is, and the calculation formula for the determinable coefficient R ² is It is as follows,

The formula for calculating the Nash-Sutcliffe efficiency coefficient (NSE) is as follows:

In the formula, y _i is the observed value,

is the average value of the observed values,

is the fitting value and y _i ^pred is the fitting value.

In a preferred solution, the step S3 includes:
Step S31 of performing a filling process of rainfall data within the period and filling the rainfall amount of 0 mm at the time of no rainfall;
step 32 of recognizing the start and end times of rainfall, setting a rainfall threshold parameter P _min and a maximum interval time threshold parameter T _stop of rainfall within a period, T _stop times of consecutive rainfall within the period P; If it is smaller than _{the min} value, it is identified as two rainfall events; if the continuous rainfall is smaller than the P _min value and the time is less than the T _stop value, it is identified as one rainfall event, and the calculation formula is as follows: That's right,
diff=t _i -t _i-1
In the formula, t is the time point when the rainfall is greater than P _min , t _i and t _i-1 represent this time point and the previous adjacent time point, and if diff is greater than T _stop , t _i-1 is 1 step S32, where t i is the end point of the previous rain, and t _i is the start point of the current rain;
The cumulative amount of rainfall each time is calculated, and the amount of rainfall from the start of one rain to the end of the rain is added, and the amount of rainfall at the start and end is included.The calculation formula is as follows,
P _sum =Σp _t
In the formula, step S33 where p _t is the amount of rainfall within the same period;
The start time, end time, and cumulative rainfall amount of each rainfall are recorded and encoded in time order, and P _n , p _n ^strt , p _n ^endt , p _n ^sum are the nth rainfall and its corresponding start time and end. and step S34 representing time and cumulative rainfall amount, respectively.

In a preferred solution, the step S4 includes:
Set the time threshold parameter T _pro , and if there is a flow valley within T _pro time after the start time of a certain rainfall for a natural watershed, the last one flow valley is set as the flood start time corresponding to the current rainfall. , for a basin with water conservancy construction or with external influencing factors, if there is no flow valley at the T _pro time after the start time of some rainfall, the flood start time is the time of the nearest valley before the start time of rainfall. Step S41 where the time is set as the time;
The time threshold parameter T _bre is set, and the first flow valley after T _bre time from the rain end time is the flood end time corresponding to the current rainfall, and the flow rate process between the flood start time and end time is A step S42 in which a flow rate change process corresponds to the current rainfall, and one rainfall and the corresponding number of floods are set as one rain runoff process;
Calculate the maximum increase in flow rate in one rainfall runoff process, that is, subtract the flow rate value corresponding to the flood start time from the maximum flow value, and input the following calculation formula,
Q _i ^up =Q _i ^max -Q _i ^str
In the formula, Q _i ^max is the maximum value in one rainfall runoff, Q _i ^str is the starting flow rate of one rainfall runoff, step S43;
Rain start time p _n ^strt , rain end time p _n ^endt , cumulative rainfall amount p _n ^sum , flood start time Q _i ^strt and its flow rate Q _i ^str , flood end time Q _i ^endt and its flow rate Q of each rain runoff process _i ^end , the maximum flow rate Q _n ^max within the period, the maximum rise Q _n ^up of the flow rate, all peaks Q _{n_i} ^peak , and a step S44 of recording the valley Q _{n_i} ^valley .

In a preferred solution, the step S5 of determining the number of flood peaks comprises:
a step S51 of recognizing a single peak flood, where in one rainfall runoff process, if there is only one peak, identifying it as a single peak flood;
Step S52 of recognizing floods with double or more peaks, where there are multiple peaks in one rainfall runoff process, a proportion threshold parameter Q _diff is set, and the difference between the valley and the adjacent peaks on both sides is set. If the proportion of each of the peak values on both sides exceeds Q _diff , it is a multi-peak flood peak, otherwise it is a single-peak flood, and the formula is as follows,
(Q _i−1 ^peak −Q _i ^valley )/Q _i−1 ^peak ＞=Q _diff and (Q _i+1 ^peak −Q _i ^valley )/Q _i+1 ^peak ＞=Q _diff
In the formula, Q _i ^valley is the valley flow rate of one rainfall runoff, Q _i-1 ^peak and Q _i+1 ^peak are the adjacent peak flow rates on both sides, respectively, and if the above relationship is satisfied, it is recorded as an effective valley, When the number of effective valleys is 1, it is a double peak flood, when the number of effective valleys is 2, it is a triple peak flood, and when the number of effective valleys is 3, it is a quadruple peak flood. , and includes step S52 which is analogized in this way.

式では、ｒ（Ｘ，Ｙ）は２組の同じ数のＸ、Ｙデータの相関係数であり、Ｖａｒ［Ｘ］はＸデータの分散であり、Ｖａｒ［Ｙ］はＹデータの分散であり、Ｃｏｖ［Ｘ，Ｙ］は、ＸとＹの共分散であり、
非線形相関の方法はＳｐｅａｒｍａｎランク相関を用い、式は以下のとおりであり、

式では、ｐ_ｓは２組の同じ数のＸ、ＹデータのＳｐｅａｒｍａｎランク相関係数であり、ｄ_ｉは２組のデータのランクの差、すなわちｘ_ｉ、ｙ_ｉを大きさ順に並べる番号の差であり、ｎはデータの数である。 In a preferred method, the correlation coefficient between the cumulative rainfall amount and the maximum increase in flow rate is calculated in step S6, and a linear correlation or nonlinear correlation method is used, and the calculation formula for the linear correlation is as follows:

In the formula, r(X, Y) is the correlation coefficient of two sets of the same number of X, Y data, Var[X] is the variance of the X data, and Var[Y] is the variance of the Y data. , Cov[X,Y] is the covariance of X and Y,
The method of nonlinear correlation uses Spearman rank correlation, and the formula is as follows,

In the formula, p _s is the Spearman rank correlation coefficient of two sets of the same number of X, Y data, and d _i is the difference in rank between the two sets of data, that is, the number of numbers that arrange x _i and y _i in order of magnitude. is the difference, and n is the number of data.

（２）線形投影を行い、
異なる角度からデータを観察し、データ特徴を最も掘削できる最適な投影方向を探し、複数の初期投影方向ａ（ａ_１，ａ_２、…_，ａ_ｍ）をランダムに抽出し、その投影指標の大きさを計算し、最大指標の投影の解をその最適な投影方向として決定し、
サンプルｉの１次元空間の投影方向である（ａ_１，ａ_２，…_，ａ_ｍ）における投影特徴値を以下のように定義し、
Ｚ_ｉ＝Σ_ｊ＝１ ^ｍａ_ｊｘ_ｉｊ
（３）目標関数を探し、
目標関数をＱ（ａ）、クラス間距離をｓ（ａ）、クラス内密度をｄ（ａ）として定義し、
Ｑ（ａ）＝ｓ（ａ）ｄ（ａ）
クラス間距離はサンプル配列の投影特徴値分散で計算し、

ここでｉ＝１_，２…ｎであり、
投影特徴値間の距離は、ｒ_ｉｋ｜ｚ_ｉ－ｚ_ｋ｜（ｉ、ｋ＝１_，２_，…ｎ）であり、それでは、
ｄ（ａ）＝Σ_ｉ＝１ ^ｎΣ_ｋ＝１ ^ｎ（Ｒ－ｒ_ｉｋ）ｆ（Ｒ－ｒ_ｉｋ）
ここでｆ（ｔ）はステップ信号であり、

Ｒは局所のプロット密度を推定する窓幅パラメータであり、幅内に少なくとも１つのプロットを含む原則に従って、その値はサンプルデータ構造に関連し、
ｍａｘ（ｒ_ｉｋ）＜Ｒ＜２ｍ
ｉ，ｋ＝１，２，．．ｎ
（４）投影方向を最適化させ、
最適な投影方向を探す問題を以下の最適化問題に変換し、
ｍａｘＱ（ａ）＝ｓ（ａ）ｄ（ａ）
Σ_ｊ＝１ ^ｍａ_ｊ ^２＝１
（５）統合評価及びクラスタリング分析を行い、
ｚ_ｉの差異レベルでサンプル群に対してクラスタリング分析を行い、最適な投影方向に基づいて、各評価指標の統合情報を反応する投影特徴値ｚ_ｉの差異レベルを計算し、最適な投影係数ａ＝ａ（ａ_１，ａ_２，…ａ_ｍ）を求めることができる。 In a preferred solution, in step S7, the parameter optimization method includes a grid search optimization method or a projection tracking optimization method;
The grid search optimization method searches for optimal parameters by performing an exhaustive search in the parameter array and training each situation;
The projection tracking optimization method projects high-dimensional data into a low-dimensional subspace, searches for a projection that can reflect the structure or characteristics of the original high-dimensional data, and achieves the purpose of studying and analyzing high-dimensional data. and includes the following steps (1) to (5),
(1) Normalize the sample data,

(2) Perform linear projection,
Observe the data from different angles, search for the optimal projection direction that can best excavate the data features, randomly extract multiple initial projection directions a (a _1, a ₂ , ... _, a _m ), and calculate the size of the projection index. and determine the solution of the projection of the maximum index as its optimal projection direction,
The projection feature value in (a _1, a _2, ... _, a _m ), which is the projection direction of the one-dimensional space of sample i, is defined as follows,
Z _i =Σ _j=1 ^m a _j x _ij
(3) Find the objective function,
Define the objective function as Q(a), the interclass distance as s(a), and the intraclass density as d(a),
Q(a)=s(a)d(a)
The interclass distance is calculated using the projected feature value variance of the sample array,

Here, i=1 _, 2...n,
The distance between projected feature values is r _ik |z _i -z _k |(i, k=1 _, 2 _, ...n), then,
d(a)=Σ _i=1 ⁿ Σ _k=1 ⁿ (R-r _ik ) f(R-r _ik )
Here f(t) is a step signal,

R is a window width parameter estimating the local plot density, whose value is related to the sample data structure according to the principle of including at least one plot within the width;
max(r _ik )<R<2m
i, k=1, 2, . ．． n
(4) Optimize the projection direction,
Convert the problem of finding the optimal projection direction to the following optimization problem,
maxQ(a)=s(a)d(a)
Σ _j=1 ^m a _j ² =1
(5) Perform integrated evaluation and clustering analysis,
Clustering analysis is performed on the sample group at the difference level of z _i , and based on the optimal projection direction, the difference level of the projection feature value z _i that responds to the integrated information of each evaluation index is calculated, and the optimal projection coefficient a =a(a _1, a _2, ... _am ) can be obtained.

In a preferred method, in step S8, a rain runoff process that meets the requirements is selected, and a selection index and an index threshold are set, and the selection index is the cumulative rainfall amount, the maximum flood peak flow rate, the flood duration, or the maximum increase in flow rate. Use width.

The present invention provides an adaptive dividing method for the number of rainfall runoff within a period, which accurately divides the rainfall corresponding to the corresponding number of runoff processes, improves the accuracy rate of dividing the number of rainfall runoff, and has a high degree of adaptation. , the automation can be fully realized and the rainfall runoff frequency can be divided quickly and effectively. The segmentation results can be widely used in hydrological research, and when used to determine hydrological model parameters, the efficiency and accuracy of determining hydrological model parameters can be greatly improved.

The present invention will be further described below with reference to drawings and examples.

3 is a method flowchart of the present invention. Flow rate process of the water catchment area selected in the example and surface rainfall process of the basin. FIG. 2 is a diagram of the rain runoff process numbered 1 in Table 1 in the example. FIG. 2 is a diagram of the rainfall runoff process number 2 in Table 1 in the example. It is a rainfall runoff process diagram of number 3 in Table 1 in an example. FIG. 4 is a diagram of the rainfall runoff process of number 4 in Table 1 in the example. FIG. 5 is a diagram of the rainfall runoff process numbered 5 in Table 1 in the example. It is a rainfall runoff process diagram of number 6 in Table 1 in an example. FIG. 7 is a diagram of the rain runoff process numbered 7 in Table 1 in the example. FIG. 8 is a diagram of the rain runoff process numbered 8 in Table 1 in the example. FIG. 3 is a diagram of correlation coefficient calculation data performed by employing a grid search optimization method.

Embodiment 1: A water catchment area will be described as an example, and a method for adaptively dividing the number of rainfall and runoff within a period will be described, as shown in FIG. 1, including the following steps.

Step S1 of collecting continuous rainfall runoff data in a known area includes step S11 and step S12.

In S11, the flow rate data and time sequence that need to be divided and are included in a certain known area and are continuous and uninterrupted are collected, and the time (h) that needs to be divided is set as a scale for the water catchment area. There is continuous runoff data from June 2nd to July 30th, 2016, and large dams are distributed within the catchment area.

In S12, obtain the areal precipitation amount and time sequence of the corresponding water catchment area in step S11, request that the time step size is the same and that the data is continuous and uninterrupted, and the time scale is set to time (h ).

In steps S11 and S12, buffer periods are set before and after the data that needs to be divided in order to maintain the accuracy of the division results. Since the length of the buffer period is one day before and after the data that needs to be divided, the time of rainfall and runoff data is divided into June 1, 2016 to July 31, 2016, as shown in FIG.

Step S2 of automatically calculating and recognizing peaks and valleys specifically includes the following steps.

In S21, a smoothing process is performed on the flow rate process with obvious sawtooth characteristics, and the smooth flow rate process is directly used.

Ｎａｓｈ－Ｓｕｔｃｌｉｆｆｅ効率係数（ＮＳＥ）の計算式は以下のとおりであり、

式では、ｙ_ｉは観測値であり、

は観測値の平均値であり、

はフィッティング値であり、ｙ_ｉ ^ｐｒｅｄはフィッティング値であり、
Ｒ^２又はＮＳＥ値が高いほど、パラメータｍ、ｋの値の信頼性の高いことを表す。本実施例においてパラメータの最終値は、４５、１０、Ｒ^２で、ＮＳＥ値は、それぞれ０．９９、０．９９である。 In this embodiment, the smoothing process employs a Savitzky-Golay filter, and the method is essentially polynomial smoothing filtering based on the least squares method, and the calculation formula is as follows:
_Assuming that the width of the smoothing window _of the filter is n= _2m +1, each measurement data Fitting the data, x _i can be fitted using a polynomial,
p(x _i )=a ₀ +a ₁ x _i +a ₂ x _i ² +. ．． ．． +a _k-1 x _i ^k-1
The determination evaluation index for the parameters m and k includes the determinable coefficient R ² or the Nash-Sutcliffe efficiency coefficient (NSE), and the larger R ² or NSE, the more suitable the parameter is, and the calculation formula for the determinable coefficient R ² is It is as follows,

The formula for calculating the Nash-Sutcliffe efficiency coefficient (NSE) is as follows:

In the formula, y _i is the observed value,

is the average value of the observed values,

is the fitting value, y _i ^pred is the fitting value,
The higher the R ² or NSE value, the more reliable the values of parameters m and k are. In this example, the final values of the parameters are 45, 10, and ^R2 , and the NSE values are 0.99 and 0.99, respectively.

In S22, mark the peak and valley on the flow rate data, record the peak flow rate and peak time, and the valley flow rate and valley time, and determine the peak flow rate and peak time according to the following formula,
q' _t-1 <q' _t and q' _t >q' _t+1
In the formula, q' is the smoothed flow rate, q' _t-1 , q' _t , q' _t+1 are the smoothed flow rates at t-1, t, and t+1, and t is the peak time, The original flow rate corresponding to that point in time is the peak flow rate;
Determine the valley flow rate and valley point according to the formula below,
q' _t-1 >q' _t and q' _t <q' _t+1
Time t is the valley time, and the original flow rate corresponding to the time point is the valley flow rate.

In S23, the peak time and the corresponding peak flow rate, and the trough time and the corresponding valley flow rate are recorded in time order and encoded.

In S3, the number of rainfalls is automatically divided, a rainfall intensity threshold and a rainfall interval time threshold are set, the start and end times of rainfall are recognized, and the cumulative rainfall amount each time is calculated. Step S3 is
Step S31 of performing a filling process of rainfall data within the period and filling the rainfall amount of 0 mm at the time of no rainfall;
step 32 of recognizing the start and end times of rainfall, setting a rainfall threshold parameter P _min and a maximum interval time threshold parameter T _stop of rainfall within a period, T _stop times of consecutive rainfall within the period P; If it is smaller than _{the min} value, it is identified as two rainfall events; if the continuous rainfall is smaller than the P _min value and the time is less than the T _stop value, it is identified as one rainfall event, and the calculation formula is as follows: That's right,
diff=t _i -t _i-1
In the formula, t is the time point when the rainfall is greater than P _min , t _i and t _i-1 represent this time point and the previous adjacent time point, and if diff is greater than T _stop , t _i-1 is 1 step S32, where t i is the end point of the previous rain, and t _i is the start point of the current rain;
The cumulative amount of rainfall each time is calculated, and the amount of rainfall from the start of one rain to the end of the rain is added, and the amount of rainfall at the start and end is included.The calculation formula is as follows,
P _sum =Σp _t
In the formula, step S33 where p _t is the amount of rainfall within the same period;
The start time, end time, and cumulative rainfall amount of each rainfall are recorded and encoded in time order, and P _n , p _n ^strt , p _n ^endt , p _n ^sum are the nth rainfall and its corresponding start time and end. and step S34 representing time and cumulative rainfall amount, respectively.

In S4, automatically divide the number of rain runoffs, set a time window threshold, and recognize the start and end times of the corresponding number of floods based on the start and end times of the rain. The step S4 includes the following steps.

In S41, a time threshold parameter T _pro is set, and if there is a flow valley within the T _pro time after the start time of a certain rainfall for a natural watershed, the last flow valley is set as the flood corresponding to the current rainfall. start time, and if there is no flow trough at the T _pro time after the start time of some rainfall for a basin with water conservancy construction or a basin with external influence factors, the nearest trough time before the start time of the rainfall. Let be the flood start time.

In this example, in a water catchment area affected by a dam, if there is a flow valley within T _pro time after the start time of a certain rainfall, the last flow valley is taken as the flood start time corresponding to the current rainfall. , if there is no flow trough at T _pro time after the start time of some rainfall, then let the nearest trough time before the start time of the rain be the flood start time.

In S42, a time threshold parameter T _bre is set, and the first flow valley after T _bre time from the rain end time is the flood end time corresponding to the current rainfall, and the time between the flood start time and end time is Let the flow rate process be a flow rate change process corresponding to the current rainfall, and let one rainfall and the corresponding number of floods be one rain runoff process.

In S43, calculate the maximum increase width of the flow rate in one rainfall runoff process, that is, subtract the flow value corresponding to the flood start time from the maximum flow value, input the following calculation formula,
Q _i ^up =Q _i ^max -Q _i ^str
In the equation, Q _i ^max is the maximum value in one rain runoff and Q _i ^str is the starting flow rate in one rain runoff.

In S44, the rain start time p _n ^strt , the rain end time p _n ^endt , the cumulative rainfall amount p _n ^sum , the flood start time Q i ^strt and its flow rate _{Q i str} ^, the flood end time Q _i ^endt and The flow rate Q _i ^end , the maximum flow rate Q _n ^max within the period, the maximum increase width Q _n ^up of the flow rate, all peaks Q _{n_i} ^peak , and valleys Q _{n_i} ^valley are recorded.

In S5, the number of peaks in one flood is recognized, and the number of peaks in flood is specified based on the difference between the peak and the valley in one flood.

Recognize single peak flood: If there is only one peak in one rainfall runoff process, identify it as a single peak flood.

Recognize double or more peak flooding: In one rainfall runoff process, if there are multiple peaks, set the proportion threshold parameter Q _diff , and the difference between the valley and the adjacent peaks on both sides is the peak value on both sides. If the proportion of each exceeds Q _diff , it is a multi-peak flood peak, otherwise it is a single-peak flood, the formula is
(Q _i−1 ^peak −Q _i ^valley )/Q _i−1 ^peak ＞=Q _diff and (Q _i+1 ^peak −Q _i ^valley )/Q _i+1 ^peak ＞=Q _diff
In the formula, Q _i ^valley is the valley flow rate of one rainfall runoff, Q _i-1 ^peak and Q _i+1 ^peak are the adjacent peak flow rates on both sides, respectively, and if the above relationship is satisfied, it is recorded as an effective valley, When the number of effective valleys is 1, it is a double peak flood, when the number of effective valleys is 2, it is a triple peak flood, and when the number of effective valleys is 3, it is a quadruple peak flood. , and the analogy is as follows.

In S6, a correlation coefficient is calculated between the cumulative rainfall amount p _n ^sum of one rain runoff and the maximum increase width Q _n ^up of the flow rate.

式では、ｒ（Ｘ，Ｙ）は２組の同じ数のＸ、Ｙデータの相関係数であり、、Ｖａｒ［Ｘ］はＸデータの分散であり、Ｖａｒ［Ｙ］はＹデータの分散であり、Ｃｏｖ［Ｘ，Ｙ］は、ＸとＹの共分散であり、
非線形相関の方法はＳｐｅａｒｍａｎランク相関を用い、式は以下のとおりであり、

式では、ｐ_ｓは２組の同じ数のＸ、ＹデータのＳｐｅａｒｍａｎランク相関係数であり、ｄ_ｉは２組のデータのランクの差、すなわちｘ_ｉ、ｙ_ｉを大きさ順に並べる番号の差であり、ｎはデータの数である。 In a preferred method, in step S6, a correlation coefficient between the cumulative rainfall amount and the maximum increase in flow rate is calculated, and a linear correlation or nonlinear correlation method is used, and the calculation formula for the linear correlation is as follows:

In the formula, r(X, Y) is the correlation coefficient between two sets of the same number of X, Y data, Var[X] is the variance of the X data, and Var[Y] is the variance of the Y data. , Cov[X, Y] is the covariance of X and Y,
The method of nonlinear correlation uses Spearman rank correlation, and the formula is as follows,

In the formula, p _s is the Spearman rank correlation coefficient of two sets of the same number of X, Y data, and d _i is the difference in rank between the two sets of data, that is, the number of numbers that arrange x _i and y _i in order of magnitude. is the difference, and n is the number of data.

In step S7, the parameters are optimized, and the optimization evaluation index is determined in step S6 when the correlation coefficient between the cumulative rainfall amount p _n ^sum and the maximum increase width Q _n ^up of the flow rate is the maximum, and the number of flood peaks is the maximum. The requirement is that the correlation coefficient decreases as the correlation coefficient increases.

In step S7, the parameter optimization method includes a grid search optimization method, a projection tracking optimization method, or other methods.

（２）線形投影を行い、
異なる角度からデータを観察し、データ特徴を最も掘削できる最適な投影方向を探し、複数の初期投影方向ａ（ａ_１，ａ_２、…_，ａ_ｍ）をランダムに抽出し、その投影指標の大きさを計算し、最大指標の投影の解をその最適な投影方向として決定し、
サンプルｉの１次元空間の投影方向である（ａ_１，ａ_２，…_，ａ_ｍ）における投影特徴値を以下のように定義し、
Ｚ_ｉ＝Σ_ｊ＝１ ^ｍａ_ｊｘ_ｉｊ
（３）目標関数を探し、
目標関数をＱ（ａ）、クラス間距離をｓ（ａ）、クラス内密度をｄ（ａ）として定義し、
Ｑ（ａ）＝ｓ（ａ）ｄ（ａ）
クラス間距離はサンプル配列の投影特徴値分散で計算し、

ここでｉ＝１_，２…ｎであり、
投影特徴値間の距離は、ｒ_ｉｋ｜ｚ_ｉ－ｚ_ｋ｜（ｉ、ｋ＝１_，２_，…ｎ）であり、それでは、
ｄ（ａ）＝Σ_ｉ＝１ ^ｎΣ_ｋ＝１ ^ｎ（Ｒ－ｒ_ｉｋ）ｆ（Ｒ－ｒ_ｉｋ）
ここでｆ（ｔ）はステップ信号であり、

Ｒは局所のプロット密度を推定する窓幅パラメータであり、幅内に少なくとも１つのプロットを含む原則に従って、その値はサンプルデータ構造に関連し、
ｍａｘ（ｒ_ｉｋ）＜Ｒ＜２ｍ
ｉ，ｋ＝１，２，．．ｎ
（４）投影方向を最適化させ、
最適な投影方向を探す問題を以下の最適化問題に変換し、
ｍａｘＱ（ａ）＝ｓ（ａ）ｄ（ａ）
Σ_ｊ＝１ ^ｍａ_ｊ ^２＝１
（５）統合評価及びクラスタリング分析を行い、
ｚ_ｉの差異レベルでサンプル群に対してクラスタリング分析を行い、最適な投影方向に基づいて、各評価指標の統合情報を反応する投影特徴値ｚ_ｉの差異レベルを計算し、最適な投影係数ａ＝ａ（ａ_１，ａ_２，…ａ_ｍ）を求めることができる。最後期間内の降雨閾値パラメータＰ_ｍｉｎ、降雨最大間隔時間閾値パラメータＴ_ｓｔｏｐ、時間閾値パラメータＴ_ｐｒｏ及びＴ_ｂｒｅ、割合閾値パラメータＱ_ｄｉｆｆを０．２ｍｍ、１５ｈ、１４ｈ、１０ｈ、１／３として決定する。この時、降雨流出回数の分布精度が最適である。 In this example, the projection tracking optimization method is used to study the high-dimensional data by projecting the high-dimensional data into a low-dimensional subspace, searching for a projection that can reflect the structure or characteristics of the original high-dimensional data. Achieve the purpose of analysis and include the following steps (1) to (5),
(1) Normalize the sample data,

(2) Perform linear projection,
Observe the data from different angles, search for the optimal projection direction that can best excavate the data features, randomly extract multiple initial projection directions a (a _1, a ₂ , ... _, a _m ), and calculate the size of the projection index. and determine the solution of the projection of the maximum index as its optimal projection direction,
The projection feature value in (a _1, a _2, ... _, a _m ), which is the projection direction of the one-dimensional space of sample i, is defined as follows,
Z _i =Σ _j=1 ^m a _j x _ij
(3) Find the objective function,
Define the objective function as Q(a), the interclass distance as s(a), and the intraclass density as d(a),
Q(a)=s(a)d(a)
The interclass distance is calculated using the projected feature value variance of the sample array,

Here, i=1 _, 2...n,
The distance between projected feature values is r _ik |z _i -z _k |(i, k=1 _, 2 _, ...n), then,
d(a)=Σ _i=1 ⁿ Σ _k=1 ⁿ (R-r _ik ) f(R-r _ik )
Here f(t) is a step signal,

R is a window width parameter estimating the local plot density, whose value is related to the sample data structure according to the principle of including at least one plot within the width;
max(r _ik )<R<2m
i, k=1, 2, . ．． n
(4) Optimize the projection direction,
Convert the problem of finding the optimal projection direction to the following optimization problem,
maxQ(a)=s(a)d(a)
Σ _j=1 ^m a _j ² =1
(5) Perform integrated evaluation and clustering analysis,
Clustering analysis is performed on the sample group at the difference level of z _i , and based on the optimal projection direction, the difference level of the projection feature value z _i that responds to the integrated information of each evaluation index is calculated, and the optimal projection coefficient a =a(a _1, a _2, ... _am ) can be obtained. The rainfall threshold parameter P _min in the last period, the maximum rainfall interval time threshold parameter T _stop , the time threshold parameters T _pro and T _bre , and the proportion threshold parameter Q _diff are determined as 0.2 mm, 15 h, 14 h, 10 h, and 1/3. . At this time, the distribution accuracy of the number of rainfall runoff is optimal.

In S8, rain runoff processes that meet the requirements are selected. A screening index and an index threshold are set, and the cumulative rainfall amount, the maximum flood peak flow rate, the flood duration, or the maximum rise width of the flow rate is used as the selection index.

In this example, according to the size of the rainfall process, we screen the corresponding number of events that meet the requirements, and if the minimum threshold of cumulative rainfall is 5 mm, the screening result is 8 rain runoff processes, and Table 1 The screening results are shown in FIGS. 3 to 10.

Table 1 Statistical table of division results of rainfall runoff frequency

Embodiment 2: The difference from Embodiment 1 is that in step S7, the parameter optimization method is a grid search optimization method that searches for optimal parameters by performing a thorough search in the parameter array and training each situation. It is to adopt the method of

In order to perform quick calculations and accelerate convergence, based on the hydrological characteristics of the watershed, the intra-period rainfall threshold parameter P _min , the maximum rainfall interval time threshold parameter T _stop , the time threshold parameters T _pro and T _bre , the rate The value ranges of the five parameters of the threshold parameter Q _diff are 0.1 to 0.3, 14 to 17, 13 to 15, 9 to 11, and 1/2 to 1/4, respectively, and the change step size is 0.1, respectively. , 1, 1, 1, 1/n (the value of n is 2, 3, 4), and random search is adopted as the search method. Therefore, there are a total of 324 types of parameter combinations. The objective function is set so that the total correlation coefficient between the cumulative rainfall amount p _n ^sum and the maximum increase width Q _n ^up of the flow rate is maximum, and the correlation decreases as the number of flood peaks increases.

Objective function: R (P _n ^sum , Q _n ^up ) is maximum, R _{single peak} (P _n ^sum , Q _n ^up ) ≧ R _{double peak} (P _n ^sum , Q _n ^up ), R correlation The formula for calculating gender is as follows.

図１１に示すように、３２４組のパラメータに対して相関性計算を行い、目標値を満たす対応するパラメータを戻す。２２２組目のパラメータの場合、累積降雨量ｐ_ｎ ^ｓｕｍと流量の最大上昇幅Ｑ_ｎ ^ｕｐの相関係数は、最大約０．８９であり、対応するする単一ピークの洪水の累積降雨量と流量の最大上昇幅の相関係数、及び二重ピークの洪水の累積降雨量と流量の最大上昇幅の相関係数は、それぞれ０．９１及び０．８５であり、単一ピークの相関係数が二重ピークの相関係数よりも大きいことを満たす。従って、５つのパラメータの値が０．２ｍｍ、１５ｈ、１４ｈ、１０ｈ、１／３である場合、降雨流出回数の分布精度が最適である。

As shown in FIG. 11, correlation calculations are performed on 324 sets of parameters, and corresponding parameters that satisfy the target value are returned. In the case of the 222nd set of parameters, the correlation coefficient between the cumulative rainfall amount p _n ^sum and the maximum increase width Q _n ^up of the flow rate is approximately 0.89 at maximum, which is similar to the corresponding cumulative rainfall amount of a single peak flood. The correlation coefficient between the maximum rise in flow rate and the correlation coefficient between the cumulative rainfall amount and the maximum rise in flow rate for double-peak floods are 0.91 and 0.85, respectively, and the correlation coefficient for a single peak is 0.91 and 0.85, respectively. is larger than the double peak correlation coefficient. Therefore, when the values of the five parameters are 0.2 mm, 15 h, 14 h, 10 h, and 1/3, the distribution accuracy of the number of rainfall runoff is optimal.

(Additional note)
(Additional note 1)
An adaptive dividing method for the number of rainfall runoff within a period, the method comprising:
Step S1 of collecting continuous rainfall runoff data in a known area;
Step S2 of automatically calculating and recognizing peaks and valleys;
step S3 of automatically dividing the number of rainfalls, setting a rainfall intensity threshold and a rainfall interval time threshold, recognizing the start and end times of rainfall, and calculating the cumulative rainfall amount each time;
Step S4: automatically dividing the number of rain runoffs, setting a time window threshold, and recognizing the start and end times of the corresponding number of floods based on the start and end times of the rain;
Step S5 of recognizing the number of peaks of one flood and identifying the number of peaks of flood based on the difference between the peak and the valley in one flood;
Step S6 of calculating a correlation coefficient between the cumulative rainfall amount p _n ^sum of one rainfall runoff and the maximum increase width Q _n ^up of the flow rate;
In step S7 of optimizing the parameters, the optimization evaluation index is such that the correlation coefficient between the cumulative rainfall p _n ^sum and the maximum increase width Q _n ^up of the flow rate is the maximum in step S6, and the peak of the flood is step S7, which satisfies that the correlation coefficient decreases as the number increases;
A method for adaptively dividing the number of rainfall runoffs within a period, comprising step S8 of selecting rainwater runoff processes that meet requirements.

(Additional note 2)
The step S1 is as follows:
Step S11 of collecting flow rate data and time sequence that need to be divided and are continuous and uninterrupted, which are included in the known area;
Acquire the area precipitation amount and time sequence of the corresponding water catchment area in step S11, request that the time step size is the same and that the data is continuous and uninterrupted, and create a buffer before and after the data that needs to be divided. The method for adaptively dividing the number of rainfall and runoff within a period according to appendix 1, characterized in that the method includes step S12 of setting a period.

(Additional note 3)
The step S2 is
Step S21 of performing smoothing processing on a flow rate process with obvious sawtooth characteristics and directly using the smooth flow rate process;
Mark the peak and valley on the flow rate data, record the peak flow rate and peak time, and the valley flow rate and valley time, determine the peak flow rate and peak time according to the formula below,
q' _t-1 <q' _t and q' _t >q' _t+1
In the formula, q' is the smoothed flow rate, q' _t-1 , q' _t , q' _t+1 are the smoothed flow rates at t-1, t, and t+1, and t is the peak time, The original flow rate corresponding to that point in time is the peak flow rate;
Determine the valley flow rate and valley point according to the formula below,
q' _t-1 >q' _t and q' _t <q' _t+1
a step S22 in which the time point t is a valley time point and the original flow rate corresponding to the time point is a valley flow rate;
A step S23 of recording and encoding the peak time and the corresponding peak flow rate, and the trough time and the corresponding valley flow rate in time order, the method of calculating the number of rainfall runoff times within the period described in Supplementary Note 1. Adaptive partitioning method.

Ｎａｓｈ－Ｓｕｔｃｌｉｆｆｅ効率係数（ＮＳＥ）の計算式は以下のとおりであり、

式では、ｙ_ｉは観測値であり、

は観測値の平均値であり、

はフィッティング値であり、ｙ_ｉ ^ｐｒｅｄはフィッティング値である、ことを特徴とする付記３に記載の期間内の降雨流出回数の適応分割方法。 (Additional note 4)
In step S21, the smoothing process uses a Savitzky-Golay filter, and the calculation formula is as follows:
_Assuming that the width of the smoothing window _of the filter is n= _2m +1, each measurement data Fitting the data, x _i can be fitted using a polynomial,
p(x _i )=a ₀ +a ₁ x _i +a ₂ x _i ² +. ．． ．． +a _k-1 x _i ^k-1
The determination evaluation index for the parameters m and k includes the determinable coefficient R ² or the Nash-Sutcliffe efficiency coefficient (NSE), and the larger R ² or NSE, the more suitable the parameter is, and the calculation formula for the determinable coefficient R ² is It is as follows,

The formula for calculating the Nash-Sutcliffe efficiency coefficient (NSE) is as follows:

In the formula, y _i is the observed value,

is the average value of the observed values,

is a fitting value, and y _i ^pred is a fitting value.

(Appendix 5)
The step S3 is
Step S31 of performing a filling process of rainfall data within the period and filling the rainfall amount of 0 mm at the time of no rainfall;
step 32 of recognizing the start and end times of rainfall, setting a rainfall threshold parameter P _min and a maximum interval time threshold parameter T _stop of rainfall within a period, T _stop times of consecutive rainfall within the period P; If it is smaller than _{the min} value, it is identified as two rainfall events; if the continuous rainfall is smaller than the P _min value and the time is less than the T _stop value, it is identified as one rainfall event, and the calculation formula is as follows: That's right,
diff=t _i -t _i-1
In the formula, t is the time point when the rainfall is greater than P _min , t _i and t _i-1 represent this time point and the previous adjacent time point, and if diff is greater than T _stop , t _i-1 is 1 step S32, where t i is the end point of the previous rain, and t _i is the start point of the current rain;
The cumulative amount of rainfall each time is calculated, and the amount of rainfall from the start of one rain to the end of the rain is added, and the amount of rainfall at the start and end is included.The calculation formula is as follows,
P _sum =Σp _t
In the formula, step S33 where p _t is the amount of rainfall within the same period;
The start time, end time, and cumulative rainfall amount of each rainfall are recorded and encoded in time order, and P _n , p _n ^strt , p _n ^endt , p _n ^sum are the nth rainfall and its corresponding start time and end. The method for adaptively dividing the number of times of rainfall and runoff within a period according to Supplementary Note 1, characterized in that the method includes step S34 representing time and cumulative rainfall amount, respectively.

(Appendix 6)
The step S4 is
Set the time threshold parameter T _pro , and if there is a flow valley within T _pro time after the start time of a certain rainfall for a natural watershed, the last one flow valley is set as the flood start time corresponding to the current rainfall. , for a basin with water conservancy construction or with external influencing factors, if there is no flow valley at the T _pro time after the start time of some rainfall, the flood start time is the time of the nearest valley before the start time of rainfall. Step S41 where the time is set as the time;
The time threshold parameter T _bre is set, and the first flow valley after T _bre time from the rain end time is the flood end time corresponding to the current rainfall, and the flow rate process between the flood start time and end time is A step S42 in which a flow rate change process corresponds to the current rainfall, and one rainfall and the corresponding number of floods are set as one rain runoff process;
Calculate the maximum increase in flow rate in one rainfall runoff process, that is, subtract the flow rate value corresponding to the flood start time from the maximum flow value, and input the following calculation formula,
Q _i ^up =Q _i ^max -Q _i ^str
In the formula, Q _i ^max is the maximum value in one rainfall runoff, Q _i ^str is the starting flow rate of one rainfall runoff, step S43;
Rain start time p _n ^strt , rain end time p _n ^endt , cumulative rainfall amount p _n ^sum , flood start time Q _i ^strt and its flow rate Q _i ^str , flood end time Q _i ^endt and its flow rate Q of each rain runoff process Supplementary note 1 characterized in that it includes step S44 of recording _i ^end , the maximum flow rate Q _n ^max within the period, the maximum increase width Q _n ^up of the flow rate, all the peaks Q _{n_i} ^peak , and the valley Q _{n_i} ^valley . Adaptive division method for the number of rainfall runoffs within the period described in .

(Appendix 7)
The step S5 is
a step S51 of recognizing a single peak flood, where in one rainfall runoff process, if there is only one peak, identifying it as a single peak flood;
Step S52 of recognizing floods with double or more peaks, where there are multiple peaks in one rainfall runoff process, a proportion threshold parameter Q _diff is set, and the difference between the valley and the adjacent peaks on both sides is set. If the proportion of each of the peak values on both sides exceeds Q _diff , it is a multi-peak flood peak, otherwise it is a single-peak flood, and the formula is as follows,
(Q _i−1 ^peak −Q _i ^valley )/Q _i−1 ^peak ＞=Q _diff and (Q _i+1 ^peak −Q _i ^valley )/Q _i+1 ^peak ＞=Q _diff
In the formula, Q _i ^valley is the valley flow rate of one rainfall runoff, Q _i-1 ^peak and Q _i+1 ^peak are the adjacent peak flow rates on both sides, respectively, and if the above relationship is satisfied, it is recorded as an effective valley, When the number of effective valleys is 1, it is a double peak flood, when the number of effective valleys is 2, it is a triple peak flood, and when the number of effective valleys is 3, it is a quadruple peak flood. , and step S52 of making an analogy in this way.

式では、ｒ（Ｘ，Ｙ）は２組の同じ数のＸ、Ｙデータの相関係数であり、Ｖａｒ［Ｘ］はＸデータの分散であり、Ｖａｒ［Ｙ］はＹデータの分散であり、Ｃｏｖ［Ｘ，Ｙ］は、ＸとＹの共分散であり、
非線形相関の方法はＳｐｅａｒｍａｎランク相関を用い、式は以下のとおりであり、

式では、ｐ_ｓは２組の同じ数のＸ、ＹデータのＳｐｅａｒｍａｎランク相関係数であり、ｄ_ｉは２組のデータのランクの差、すなわちｘ_ｉ、ｙ_ｉを大きさ順に並べる番号の差であり、ｎはデータの数である、ことを特徴とする付記１に記載の期間内の降雨流出回数の適応分割方法。 (Appendix 8)
In step S6, the correlation coefficient between the cumulative rainfall and the maximum increase in flow rate is calculated using a linear correlation or nonlinear correlation method, and the linear correlation calculation formula is as follows:

In the formula, r(X, Y) is the correlation coefficient of two sets of the same number of X, Y data, Var[X] is the variance of the X data, and Var[Y] is the variance of the Y data. , Cov[X,Y] is the covariance of X and Y,
The method of nonlinear correlation uses Spearman rank correlation, and the formula is as follows,

In the formula, p _s is the Spearman rank correlation coefficient of two sets of the same number of X, Y data, and d _i is the difference in rank between the two sets of data, that is, the number of numbers that arrange x _i and y _i in order of magnitude. The adaptive dividing method for the number of rainfall runoffs within a period according to appendix 1, wherein n is the difference, and n is the number of data.

（２）線形投影を行い、
異なる角度からデータを観察し、データ特徴を最も掘削できる最適な投影方向を探し、複数の初期投影方向ａ（ａ_１，ａ_２、…_，ａ_ｍ）をランダムに抽出し、その投影指標の大きさを計算し、最大指標の投影の解をその最適な投影方向として決定し、
サンプルｉの１次元空間の投影方向である（ａ_１，ａ_２，…_，ａ_ｍ）における投影特徴値を以下のように定義し、
Ｚ_ｉ＝Σ_ｊ＝１ ^ｍａ_ｊｘ_ｉｊ
（３）目標関数を探し、
目標関数をＱ（ａ）、クラス間距離をｓ（ａ）、クラス内密度をｄ（ａ）として定義し、
Ｑ（ａ）＝ｓ（ａ）ｄ（ａ）
クラス間距離はサンプル配列の投影特徴値分散で計算し、

ここでｉ＝１_，２…ｎであり、
投影特徴値間の距離は、ｒ_ｉｋ｜ｚ_ｉ－ｚ_ｋ｜（ｉ、ｋ＝１_，２_，…ｎ）であり、それでは、
ｄ（ａ）＝Σ_ｉ＝１ ^ｎΣ_ｋ＝１ ^ｎ（Ｒ－ｒ_ｉｋ）ｆ（Ｒ－ｒ_ｉｋ）
ここでｆ（ｔ）はステップ信号であり、

Ｒは局所のプロット密度を推定する窓幅パラメータであり、幅内に少なくとも１つのプロットを含む原則に従って、その値はサンプルデータ構造に関連し、
ｍａｘ（ｒ_ｉｋ）＜Ｒ＜２ｍ
ｉ，ｋ＝１，２，．．ｎ
（４）投影方向を最適化させ、
最適な投影方向を探す問題を以下の最適化問題に変換し、
ｍａｘＱ（ａ）＝ｓ（ａ）ｄ（ａ）
Σ_ｊ＝１ ^ｍａ_ｊ ^２＝１
（５）統合評価及びクラスタリング分析を行い、
ｚ_ｉの差異レベルでサンプル群に対してクラスタリング分析を行い、最適な投影方向に基づいて、各評価指標の統合情報を反応する投影特徴値ｚ_ｉの差異レベルを計算し、最適な投影係数ａ＝ａ（ａ_１，ａ_２，…ａ_ｍ）を求めることができる、ことを特徴とする付記１に記載の期間内の降雨流出回数の適応分割方法。 (Appendix 9)
In step S7, the parameter optimization method includes a grid search optimization method or a projection tracking optimization method,
The grid search optimization method searches for optimal parameters by performing an exhaustive search in the parameter array and training each situation;
The projection tracking optimization method projects high-dimensional data into a low-dimensional subspace, searches for a projection that can reflect the structure or characteristics of the original high-dimensional data, and achieves the purpose of studying and analyzing high-dimensional data. and includes the following steps (1) to (5),
(1) Normalize the sample data,

(2) Perform linear projection,
Observe the data from different angles, search for the optimal projection direction that can best excavate the data features, randomly extract multiple initial projection directions a (a _1, a ₂ , ... _, a _m ), and calculate the size of the projection index. and determine the solution of the projection of the maximum index as its optimal projection direction,
The projection feature value in (a _1, a _2, ... _, a _m ), which is the projection direction of the one-dimensional space of sample i, is defined as follows,
Z _i =Σ _j=1 ^m a _j x _ij
(3) Find the objective function,
Define the objective function as Q(a), the interclass distance as s(a), and the intraclass density as d(a),
Q(a)=s(a)d(a)
The interclass distance is calculated using the projected feature value variance of the sample array,

Here, i=1 _, 2...n,
The distance between projected feature values is r _ik |z _i -z _k |(i, k=1 _, 2 _, ...n), then,
d(a)=Σ _i=1 ⁿ Σ _k=1 ⁿ (R-r _ik ) f(R-r _ik )
Here f(t) is a step signal,

R is a window width parameter estimating the local plot density, whose value is related to the sample data structure according to the principle of including at least one plot within the width;
max(r _ik )<R<2m
i, k=1, 2, . ．． n
(4) Optimize the projection direction,
Convert the problem of finding the optimal projection direction to the following optimization problem,
maxQ(a)=s(a)d(a)
Σ _j=1 ^m a _j ² =1
(5) Perform integrated evaluation and clustering analysis,
Clustering analysis is performed on the sample group at the difference level of z _i , and based on the optimal projection direction, the difference level of the projection feature value z _i that responds to the integrated information of each evaluation index is calculated, and the optimal projection coefficient a = a (a _{1 ,} a _{2 ,} ... a _m ).

(Appendix 10)
In the step S8, a rain runoff process that meets the requirements is selected, and a selection index and an index threshold are set, and the selection index uses the cumulative rainfall amount, the maximum flood peak flow rate, the flood duration, or the maximum rise width of the flow rate. An adaptive dividing method for the number of rainfall and runoff within the period described in Appendix 1, characterized by:

Claims (10)

An adaptive dividing method for the number of rainfall runoff within a period, the method comprising:
Step S1 of collecting continuous rainfall runoff data in a known area;
Step S2 of automatically calculating and recognizing peaks and valleys;
step S3 of automatically dividing the number of rainfalls, setting a rainfall intensity threshold and a rainfall interval time threshold, recognizing the start and end times of rainfall, and calculating the cumulative rainfall amount each time;
Step S4: automatically dividing the number of rain runoffs, setting a time window threshold, and recognizing the start and end times of the corresponding number of floods based on the start and end times of the rain;
Step S5 of recognizing the number of peaks of one flood and identifying the number of peaks of flood based on the difference between the peak and the valley in one flood;
Step S6 of calculating a correlation coefficient between the cumulative rainfall amount p _n ^sum of one rain runoff and the maximum increase width Q _n ^up of the flow rate;
In step S7 of optimizing the parameters, the optimization evaluation index is such that the correlation coefficient between the cumulative rainfall p _n ^sum and the maximum increase width Q _n ^up of the flow rate is the maximum in step S6, and the peak of the flood is step S7, which satisfies that the correlation coefficient decreases as the number increases;
A method for adaptively dividing the number of rainfall runoffs within a period, comprising step S8 of selecting rainwater runoff processes that meet requirements. The step S1 is as follows:
Step S11 of collecting flow rate data and time sequence that need to be divided and are continuous and uninterrupted, which are included in the known area;
Acquire the area precipitation amount and time sequence of the corresponding water catchment area in step S11, request that the time step size is the same and that the data is continuous and uninterrupted, and create a buffer before and after the data that needs to be divided. The method of adaptively dividing the number of rainfall and runoff within a period according to claim 1, further comprising step S12 of setting a period. The step S2 is
Step S21 of performing smoothing processing on a flow rate process with obvious sawtooth characteristics and directly using the smooth flow rate process;
Mark the peak and valley on the flow rate data, record the peak flow rate and peak time, and the valley flow rate and valley time, determine the peak flow rate and peak time according to the formula below,
q' _t-1 <q' _t and q' _t >q' _t+1
In the formula, q' is the smoothed flow rate, q' _t-1 , q' _t , q' _t+1 are the smoothed flow rates at t-1, t, and t+1, and t is the peak time, The original flow rate corresponding to that point in time is the peak flow rate;
Determine the valley flow rate and valley point according to the formula below,
q' _t-1 >q' _t and q' _t <q' _t+1
a step S22 in which the time point t is a valley time point and the original flow rate corresponding to the time point is a valley flow rate;
The number of rainfall runoff times within a period according to claim 1, further comprising: recording and encoding the peak time and the corresponding peak flow rate, and the trough time and the corresponding valley flow rate in time order. adaptive partitioning method. In step S21, the smoothing process uses a Savitzky-Golay filter, and the calculation formula is as follows:
_Assuming that the width of the smoothing window _of the filter is n= _2m +1, each measurement data Fitting the data, x _i can be fitted using a polynomial,
p(x _i )=a ₀ +a ₁ x _i +a ₂ x _i ² +. ．． ．． +a _k-1 x _i ^k-1
The determination evaluation index for the parameters m and k includes the determinable coefficient R ² or the Nash-Sutcliffe efficiency coefficient (NSE), and the larger R ² or NSE, the more suitable the parameter is, and the calculation formula for the determinable coefficient R ² is It is as follows,

The formula for calculating the Nash-Sutcliffe efficiency coefficient (NSE) is as follows:

In the formula, y _i is the observed value,

is the average value of the observed values,

The method of adaptively dividing the number of rainfall runoffs within a period according to claim 3, wherein is a fitting value, and y _i ^pred is a fitting value. The step S3 is
Step S31 of performing a filling process of rainfall data within the period and filling the rainfall amount of 0 mm at the time of no rainfall;
step 32 of recognizing the start and end times of rainfall, setting a rainfall threshold parameter P _min and a maximum interval time threshold parameter T _stop of rainfall within a period, T _stop times of consecutive rainfall within the period P; If it is smaller than _{the min} value, it is identified as two rainfall events; if the continuous rainfall is smaller than the P _min value and the time is less than the T _stop value, it is identified as one rainfall event, and the calculation formula is as follows: That's right,
diff=t _i -t _i-1
In the formula, t is the time point when the rainfall is greater than P _min , t _i and t _i-1 represent this time point and the previous adjacent time point, and if diff is greater than T _stop , t _i-1 is 1 step S32, where t i is the end point of the previous rain, and t _i is the start point of the current rain;
The cumulative amount of rainfall each time is calculated, and the amount of rainfall from the start of one rain to the end of the rain is added, and the amount of rainfall at the start and end is included.The calculation formula is as follows,
P _sum =Σp _t
In the formula, step S33 where p _t is the amount of rainfall within the same period;
The start time, end time, and cumulative rainfall amount of each rainfall are recorded and encoded in time order, and P _n , p _n ^strt , p _n ^endt , p _n ^sum are the nth rainfall and its corresponding start time and end. 2. The method of claim 1, further comprising step S34 representing time and cumulative rainfall, respectively. The step S4 is
Set the time threshold parameter T _pro , and if there is a flow valley within T _pro time after the start time of a certain rainfall for a natural watershed, the last one flow valley is set as the flood start time corresponding to the current rainfall. , for a basin with water conservancy construction or with external influencing factors, if there is no flow valley at the T _pro time after the start time of some rainfall, the flood start time is the time of the nearest valley before the start time of rainfall. Step S41 where the time is set as the time;
The time threshold parameter T _bre is set, and the first flow valley after T _bre time from the rain end time is the flood end time corresponding to the current rainfall, and the flow rate process between the flood start time and end time is A step S42 in which a flow rate change process corresponds to the current rainfall, and one rainfall and the corresponding number of floods are set as one rain runoff process;
Calculate the maximum increase in flow rate in one rainfall runoff process, that is, subtract the flow rate value corresponding to the flood start time from the maximum flow value, and input the following calculation formula,
Q _i ^up =Q _i ^max -Q _i ^str
In the formula, Q _i ^max is the maximum value in one rainfall runoff, Q _i ^str is the starting flow rate of one rainfall runoff, step S43;
Rain start time p _n ^strt , rain end time p _n ^endt , cumulative rainfall amount p _n ^sum , flood start time Q _i ^strt and its flow rate Q _i ^str , flood end time Q _i ^endt and its flow rate Q of each rain runoff process _i ^end , a maximum flow rate Q _n ^max within the period, a maximum rise Q _n ^up of the flow rate, all peaks Q _{n_i} ^peak , and a step S44 of recording the valley Q _{n_i} ^valley . Adaptive division method of the number of rainfall runoff within the period described in 1. The step S5 is
a step S51 of recognizing a single peak flood, where in one rainfall runoff process, if there is only one peak, identifying it as a single peak flood;
Step S52 of recognizing floods with double or more peaks, where there are multiple peaks in one rainfall runoff process, a proportion threshold parameter Q _diff is set, and the difference between the valley and the adjacent peaks on both sides is set. If the proportion of each of the peak values on both sides exceeds Q _diff , it is a multi-peak flood peak, otherwise it is a single-peak flood, and the formula is as follows,
(Q _i−1 ^peak −Q _i ^valley )/Q _i−1 ^peak ＞=Q _diff and (Q _i+1 ^peak −Q _i ^valley )/Q _i+1 ^peak ＞=Q _diff
In the formula, Q _i ^valley is the valley flow rate of one rainfall runoff, Q _i-1 ^peak and Q _i+1 ^peak are the adjacent peak flow rates on both sides, respectively, and if the above relationship is satisfied, it is recorded as an effective valley, When the number of effective valleys is 1, it is a double peak flood, when the number of effective valleys is 2, it is a triple peak flood, and when the number of effective valleys is 3, it is a quadruple peak flood. The method of adaptively dividing the number of rainfall and runoff within a period according to claim 1, further comprising step S52 of making an analogy in this way. In step S6, the correlation coefficient between the cumulative rainfall and the maximum increase in flow rate is calculated using a linear correlation or nonlinear correlation method, and the linear correlation calculation formula is as follows:

In the formula, r(X, Y) is the correlation coefficient of two sets of the same number of X, Y data, Var[X] is the variance of the X data, and Var[Y] is the variance of the Y data. , Cov[X,Y] is the covariance of X and Y,
The method of nonlinear correlation uses Spearman rank correlation, and the formula is as follows,

In the formula, p _s is the Spearman rank correlation coefficient of two sets of the same number of X, Y data, and d _i is the difference in rank between the two sets of data, that is, the number of numbers that arrange x _i and y _i in order of size. 2. The method of adaptively dividing the number of rainfall runoff within a period according to claim 1, wherein n is the difference and n is the number of data. In step S7, the parameter optimization method includes a grid search optimization method or a projection tracking optimization method,
The grid search optimization method searches for optimal parameters by performing an exhaustive search in the parameter array and training each situation;
The projection tracking optimization method projects high-dimensional data into a low-dimensional subspace, searches for a projection that can reflect the structure or characteristics of the original high-dimensional data, and achieves the purpose of studying and analyzing high-dimensional data. and includes the following steps (1) to (5),
(1) Normalize the sample data,

(2) Perform linear projection,
Observe the data from different angles, search for the optimal projection direction that can best excavate the data features, randomly extract multiple initial projection directions a (a _1, a ₂ , ... _, a _m ), and calculate the size of the projection index. and determine the solution of the projection of the maximum index as its optimal projection direction,
The projection feature value in (a _1, a _2, ... _, a _m ), which is the projection direction of the one-dimensional space of sample i, is defined as follows,
Z _i =Σ _j=1 ^m a _j x _ij
(3) Find the objective function,
Define the objective function as Q(a), the interclass distance as s(a), and the intraclass density as d(a),
Q(a)=s(a)d(a)
The interclass distance is calculated using the projected feature value variance of the sample array,

Here, i=1 _, 2...n,
The distance between projected feature values is r _ik |z _i -z _k |(i, k=1 _, 2 _, ...n), then,
d(a)=Σ _i=1 ⁿ Σ _k=1 ⁿ (R-r _ik ) f(R-r _ik )
Here f(t) is a step signal,

R is a window width parameter estimating the local plot density, whose value is related to the sample data structure according to the principle of including at least one plot within the width;
max(r _ik )<R<2m
i, k=1, 2, . ．． n
(4) Optimize the projection direction,
Convert the problem of finding the optimal projection direction to the following optimization problem,
maxQ(a)=s(a)d(a)
Σ _j=1 ^m a _j ² =1
(5) Perform integrated evaluation and clustering analysis,
Clustering analysis is performed on the sample group at the difference level of z _i , and based on the optimal projection direction, the difference level of the projection feature value z _i that responds to the integrated information of each evaluation index is calculated, and the optimal projection coefficient a The adaptive dividing method of the number of rainfall runoffs within a period according to claim 1, characterized in that it is possible to obtain = a (a _1, a _2, ... _am ). In the step S8, a rain runoff process that meets the requirements is selected, and a selection index and an index threshold are set, and the selection index uses the cumulative rainfall amount, the maximum flood peak flow rate, the flood duration, or the maximum rise width of the flow rate. The method of adaptively dividing the number of rainfall and runoff within a period according to claim 1.