A daily-scale runoff reconstruction method and system based on "prediction-decomposition"

Abstract

The present invention discloses a method and system for reconstructing daily-scale runoff based on "prediction-decomposition", the method comprising: runoff prediction, using integrated empirical mode decomposition technology to spatially decompose monthly-scale runoff, predicting each component through the nearest neighbor method of coupled similarity index, and then realizing monthly-scale runoff prediction after component superposition; runoff decomposition, using nearest neighbor sampling estimation method, while improving monthly runoff similarity judgment strategy, proposing improved non-parametric solution set model, and decomposing predicted monthly-scale runoff to days, and finally reconstructing long-sequence daily-scale runoff sequence. The present invention reconstructs long-sequence daily runoff through two steps: monthly runoff simulation prediction and predicted monthly runoff decomposition to days, avoiding the difficulty of runoff reconstruction by directly taking daily-scale runoff as the research object, providing a new method and technology for the reconstruction of small-scale runoff data, and also laying a good foundation for the subsequent water conservancy planning.

Description

Technical Field

[0001] The invention relates to a runoff reconstruction technology, and in particular to a daily-scale runoff reconstruction method based on "prediction-decomposition". Background Art

[0002] Long-sequence daily-scale runoff data is the basis for water conservancy planning calculations. However, due to the imperfection of some river hydrological monitoring facilities and inappropriate data storage methods, the actual runoff data is short-lived and inconsistent in scale within the year. Some months have daily-scale data, while some months have only monthly-scale data, and the data missing problem is more serious. The complex river hydrological system and the obvious seasonal characteristics of river runoff, coupled with the combined influence of climate change and human activities, make the runoff time series highly nonlinear and non-stationary. Traditional linear methods are mainly aimed at stationary runoff sequences, so the accuracy of the reconstructed sequence is not high. In addition, the scale of the runoff sequence currently studied is large, mostly on the monthly and annual scales, and the runoff sequence at the daily scale is rarely studied. Since the runoff sequence at the daily scale is more random and volatile, and the internal relationship is more complex, directly taking the runoff sequence at the daily scale as the research object will increase the difficulty of runoff sequence reconstruction, and also affect the accuracy of the reconstructed sequence. How to reconstruct a daily-scale runoff series with a long sequence and high accuracy has important practical significance for subsequent water conservancy planning calculations. Summary of the invention

[0003] Purpose of the invention: One purpose of the present invention is to provide a daily-scale runoff reconstruction method based on "prediction-decomposition", which provides a feasible method and technology for small-scale runoff reconstruction.

[0004] Another object of the present invention is to provide a daily-scale runoff reconstruction system based on "prediction-decomposition".

[0005] Technical solution: A daily-scale runoff reconstruction method based on "prediction-decomposition" of the present invention comprises the following steps:

[0006] S1. Obtain the monthly historical runoff sequence Z and the monthly and daily historical runoff process Y based on the daily historical runoff data, and form the historical sample points [Z, Y] of the monthly and daily runoff;

[0007] S2. The non-stationary monthly historical runoff sequence Z is spatially decomposed using the integrated empirical mode decomposition technology and converted into a series of relatively stable IMF component time series signals; the nearest neighbor method with coupled similarity indicators is used to predict the time series signals of each IMF component, and the prediction results of each IMF component are superimposed to obtain m predicted monthly runoff values;

[0008] S3. Rotate and transform the historical runoff process Y on each monthly and daily scale to obtain the transformed historical runoff process X on each monthly and daily scale. Use the improved non-parametric solution set model to decompose the m predicted monthly runoff values ​​respectively: First, find k predicted monthly runoff values ​​Z that need to be decomposed in the transformed historical runoff process X on each monthly and daily scale. * Similar months, and construct the cumulative weight distribution function g(m). Secondly, a roulette wheel is used to select one of the k similar months to estimate the rotation transformation X of the daily runoff process in that month. * , and then through inverse transformation, we can get the actual runoff process Y on the monthly and daily scale * ; Then the remaining m-1 predicted monthly runoff values ​​are decomposed, and finally m decomposed monthly and daily runoff processes are obtained to achieve the reconstruction of long-sequence daily runoff.

[0009] Furthermore, in step S2, the method of spatially decomposing the non-stationary monthly historical runoff series Z using the integrated empirical mode decomposition technique is as follows:

[0010] (1) Add Gaussian white noise n to the monthly historical runoff series Z i (t), to obtain the new runoff sequence y i (t):

[0011] y i (t) = Z + n i (t) (1)

[0012] Where i is the number of times Gaussian white noise is added;

[0013] (2) Using the empirical mode decomposition technique to analyze the new runoff series y i (t) is decomposed to obtain n IMF components imf i,j (t) and a residual term r i,n (t):

[0014]

[0015] Among them, IMF i,j (t) is to add y i The jth IMF component of (t);

[0016] (3) Repeat steps (1) to (2) N times, integrate and average the IMF components and residual terms obtained each time and use them as the final decomposition result:

[0017]

[0018] (4) The final monthly historical runoff series Z is expressed by n IMF components and 1 residual term:

[0019]

[0020] Furthermore, in step S2, the method for predicting each IMF component time series signal using the nearest neighbor method of the coupling similarity index is:

[0021] Determine the maximum lag order p of the IMF component series and mark the historical value x of the IMF component series t And the corresponding eigenvector D t ={x t-p ,x t-p+1 ,...,x t-1}, t=1,2,...,n, through the Euclidean distance between vectors, from D t Find k features that are consistent with the eigenvector D n+1 ={x n-p+1 ,x n-p+2 ,...,x n} feature vectors with good similarity, and the subsequent value x' corresponding to these k feature vectors j , j = 1, 2, ..., k, predict x n+1 :

[0022]

[0023] Among them, w j is the weight function, k is the number of nearest neighbors, j is the sorting position of the k eigenvectors of the nearest neighbors, and w j The calculation formulas for and k are as follows:

[0024]

[0025]

[0026] The prediction results of each IMF component series are superimposed to obtain the predicted monthly runoff value.

[0027] Furthermore, the historical runoff process Y = (y1, y2, ..., y d ) T , the transformed monthly and daily historical runoff process X=(x1,x2,...,x d ) T The transformation formula is:

[0028] X=RY (8)

[0029] Among them, R is the unit orthogonal matrix, and the dth component x in X after rotation transformation is d The following conditions are met:

[0030]

[0031] Among them, d is the number of days corresponding to each month.

[0032] Furthermore, in step S3, the specific steps of using the improved non-parametric solution set model to decompose the m predicted monthly runoff values ​​are as follows:

[0033] First, according to the rotation transformation formula, calculate X * The dth component in

[0034] Then, calculate the first d-1 coordinates If the roulette wheel method is used to select the rotation transformation X of the daily runoff process in the lth month, * , take X l The first d-1 components U l As X * The first d-1 coordinates U * ;

[0035] Combined with calculation Estimate the daily runoff of the month after rotation transformation Then, through the inverse rotation transformation, the actual runoff process Y on the monthly and daily scale is obtained. * , and so on, the remaining m-1 predicted monthly runoff values ​​are decomposed.

[0036] A daily-scale runoff reconstruction system based on "prediction-decomposition" of the present invention comprises:

[0037] The data collection and processing module is used to collect daily historical runoff data, obtain the monthly historical runoff sequence Z and the monthly and daily historical runoff process Y, and form the historical sample points [Z, Y] of the monthly and daily runoff;

[0038] The runoff prediction module uses the integrated empirical mode decomposition technology to spatially decompose the monthly runoff sequence, and uses the nearest neighbor method of coupled similarity indicators to predict the time series signals of each component, superimposes the prediction results of the time series signals of each component, and obtains the predicted monthly scale runoff;

[0039] The runoff decomposition module uses an improved non-parametric solution set model to decompose the predicted monthly runoff and reconstruct the daily runoff.

[0040] Optionally, an improved nonparametric solution set model is used to decompose the predicted monthly runoff as follows:

[0041] Rotate the historical runoff process Y on each month and day scale to obtain the transformed historical runoff process X on each month and day scale; find k predicted monthly runoff values ​​Z that need to be decomposed in the transformed historical runoff process X on each month and day scale * similar months, and construct the cumulative weight distribution function g(m); select one from the k similar months by roulette to estimate the rotation transformation X of the daily scale runoff process in that month * , and then through the inverse rotation transformation, the decomposed actual runoff process Y * ; Then the remaining m-1 predicted monthly runoff values ​​are decomposed, and finally m decomposed monthly daily runoff processes are obtained to achieve long-sequence daily runoff reconstruction.

[0042] A device of the present invention includes a memory and a processor, wherein:

[0043] A memory for storing computer programs that can be run on the processor;

[0044] The processor is used to execute the steps of the above-mentioned daily-scale runoff reconstruction method based on "prediction-decomposition" when running the computer program.

[0045] A storage medium of the present invention stores a computer program, which, when executed by at least one processor, implements the steps of the above-mentioned daily-scale runoff reconstruction method based on "prediction-decomposition".

[0046] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0047] (1) Through the two steps of simulating and predicting monthly runoff and decomposing the predicted monthly runoff into daily data, the reconstruction of daily runoff is achieved. The operation is simple, flexible and easy to implement, avoiding the difficulty of runoff reconstruction caused by directly taking daily runoff as the research object, and providing a feasible method and technology for the reconstruction of small-scale runoff data.

[0048] (2) Through the spatial decomposition of monthly runoff, component simulation and prediction, and then superposition to obtain the predicted monthly runoff results, the non-stationary time series can be converted into a stationary time series, and then the stationary time series can be analyzed and processed. In addition, the nearest neighbor method with coupled similarity indicators is used to model each component, avoiding the prediction bias caused by assuming the distribution of the sequence, and effectively improving the prediction accuracy of runoff.

[0049] (3) The kernel density estimation method used in the nonparametric solution set model requires the number of parameters to be estimated to increase rapidly with the increase of components. However, the number of components within the day of the monthly runoff is relatively large, and it is quite difficult to use the kernel density estimation method. The present invention proposes to use the nearest neighbor sampling estimation method instead of the kernel density estimation method, which effectively improves the solution efficiency. In addition, in the monthly runoff similarity judgment strategy, the deviation of the runoff and the head-to-tail correlation of the daily runoff are taken into account, which improves the accuracy of the reconstructed sequence. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of the method of the present invention;

[0051] Figure 2 A schematic diagram of the nearest neighbor method of the coupling similarity index used in the technical solution of the present invention;

[0052] Figure 3 The runoff process of Liujiaping River from 1962 to 1981 in the embodiment of the present invention;

[0053] Figure 4 It is a distribution diagram of various statistical parameters of the measured runoff and reconstructed runoff of Liujiaping River in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The technical solution of the present invention is described in detail below, but the protection scope of the present invention is not limited to the embodiments.

[0055] like Figure 1 As shown, a daily-scale runoff reconstruction method based on "prediction-decomposition" of the present invention comprises the following steps:

[0056] S1. Obtain the monthly historical runoff sequence Z and the monthly and daily historical runoff process Y based on the daily historical runoff data, and form the historical sample points [Z, Y] of the monthly and daily runoff;

[0057] S2. The non-stationary monthly historical runoff sequence Z is spatially decomposed using the integrated empirical mode decomposition technology and converted into a series of relatively stable IMF component time series signals; the nearest neighbor method with coupled similarity indicators is used to predict the time series signals of each IMF component, and the prediction results of each IMF component are superimposed to obtain m predicted monthly runoff values;

[0058] The method of spatially decomposing the non-stationary monthly historical runoff series Z using the integrated empirical mode decomposition technology is as follows:

[0059] (1) Add Gaussian white noise n to the monthly historical runoff series Z i (t), to obtain the new runoff sequence y i (t):

[0060] yi (t) = Z + n i (t) (1)

[0061] Where i is the number of times Gaussian white noise is added;

[0062] (2) Using the empirical mode decomposition technique to analyze the new runoff series y i (t) is decomposed to obtain n IMF components imf i,j (t) and a residual term r i,n (t):

[0063]

[0064] Among them, IMF i,j (t) is to add y i The jth IMF component of (t);

[0065] (3) Repeat steps (1) to (2) N times, integrate and average the IMF components and residual terms obtained each time and use them as the final decomposition result:

[0066]

[0067] (4) The final monthly historical runoff series Z is expressed by n IMF components and 1 residual term:

[0068]

[0069] like Figure 2 As shown in the figure, the method of using the nearest neighbor method of coupled similarity index to predict the time series signals of each IMF component is:

[0070] Determine the maximum lag order p of the IMF component series and mark the historical value x of the IMF component series t And the corresponding eigenvector D t ={x t-p ,x t-p+1 ,...,x t-1}, t=1,2,...,n, through the Euclidean distance between vectors, from D t Find k features that are consistent with the eigenvector D n+1 ={x n-p+1 ,x n-p+2 ,...,x n} feature vectors with good similarity, and the subsequent value x' corresponding to these k feature vectors j , j = 1, 2, ..., k, predict x n+1 :

[0071]

[0072] Among them, w j is the weight function, and the eigenvector D n+1 The higher the similarity, the greater the weight, and the position inverse weighting method can be used for assignment; k is the number of nearest neighbors, j is the sorting position of the k feature vectors of the nearest neighbors; w j The calculation formulas for and k are as follows:

[0073]

[0074]

[0075] The prediction results of each IMF component series are superimposed to obtain the predicted monthly runoff value.

[0076] S3. Rotate and transform the historical runoff process Y on each monthly and daily scale to obtain the transformed historical runoff process X on each monthly and daily scale. Use the improved non-parametric solution set model to decompose the m predicted monthly runoff values ​​respectively: First, find k predicted monthly runoff values ​​Z that need to be decomposed in the transformed historical runoff process X on each monthly and daily scale. * Similar months, and construct the cumulative weight distribution function g(m). Secondly, a roulette wheel is used to select one of the k similar months to estimate the rotation transformation X of the daily runoff process in that month. * , and then through inverse transformation, we can get the actual runoff process Y on the monthly and daily scale * ; Then the remaining m-1 predicted monthly runoff values ​​are decomposed, and finally m decomposed monthly and daily runoff processes are obtained to achieve the reconstruction of long-sequence daily runoff.

[0077] Among them, the historical runoff process Y = (y1, y2, ..., y d ) T , the transformed monthly and daily historical runoff process X=(x1,x2,...,x d ) T The transformation formula is:

[0078] X=RY (8)

[0079] Among them, R is the unit orthogonal matrix, and the dth component x in X after rotation transformation is d The following conditions are met:

[0080]

[0081] Among them, d is the number of days corresponding to each month.

[0082] The specific steps of using the improved non-parametric solution set model to decompose the m predicted monthly runoff values ​​are as follows:

[0083] First, according to the rotation transformation formula, calculate X * The dth component in

[0084] Then, calculate the first d-1 coordinates If the roulette wheel method is used to select the rotation transformation X of the daily runoff process in the lth month, * , take X l The first d-1 components U l As X * The first d-1 coordinates U * ;

[0085] Combined with calculation Estimate the daily runoff of the month after rotation transformation Then, through the inverse rotation transformation, the actual runoff process Y on the monthly and daily scale is obtained. * , and so on, the remaining m-1 predicted monthly runoff values ​​are decomposed.

[0086] A daily-scale runoff reconstruction system based on "prediction-decomposition" of the present invention comprises:

[0087] The data collection and processing module is used to collect daily historical runoff data, obtain the monthly historical runoff sequence Z and the monthly and daily historical runoff process Y, and form the historical sample points [Z, Y] of the monthly and daily runoff;

[0088] The runoff prediction module uses the integrated empirical mode decomposition technology to spatially decompose the monthly runoff sequence, and uses the nearest neighbor method of coupled similarity indicators to predict the time series signals of each component, superimposes the prediction results of the time series signals of each component, and obtains the predicted monthly scale runoff;

[0089] The runoff decomposition module uses an improved non-parametric solution set model to decompose the predicted monthly runoff and reconstruct the daily runoff. The specific method of using the improved non-parametric solution set model to decompose the predicted monthly runoff is as follows:

[0090] Rotate the historical runoff process Y on each month and day scale to obtain the transformed historical runoff process X on each month and day scale; find k predicted monthly runoff values ​​Z that need to be decomposed in the transformed historical runoff process X on each month and day scale * similar months, and construct the cumulative weight distribution function g(m); select one from the k similar months by roulette to estimate the rotation transformation X of the daily scale runoff process in that month * , and then through the inverse rotation transformation, the decomposed actual runoff process Y *; Then the remaining m-1 predicted monthly runoff values ​​are decomposed, and finally m decomposed monthly daily runoff processes are obtained to achieve long-sequence daily runoff reconstruction.

[0091] A device of the present invention includes a memory and a processor, wherein:

[0092] A memory for storing computer programs that can be run on the processor;

[0093] The processor is used to execute the steps of the above-mentioned daily-scale runoff reconstruction method based on "prediction-decomposition" when running the computer program, and achieve the technical effect consistent with the above-mentioned method.

[0094] A storage medium of the present invention stores a computer program, which, when executed by at least one processor, implements the steps of the above-mentioned daily-scale runoff reconstruction method based on "prediction-decomposition" and achieves the same technical effect as the above-mentioned method.

[0095] Example:

[0096] This embodiment takes the Liujiaping River in Hunan Province as an example to illustrate, and uses the daily-scale runoff reconstruction method of the present invention to reconstruct a long sequence of daily-scale runoff to illustrate the effectiveness and rationality of the present invention.

[0097] Liujiaping River is located in Xupu County, Hunan Province, at 110°39'~110°43' east longitude and 27°27'~27°30' north latitude. It is a first-level tributary of Erdu River of Xushui. It originates from Liangfengjie, Hunan Province, and flows from southeast to northwest in a zigzag shape. It flows through Longzhuang Bay and Liujiaping in Xupu County and flows into Erdu River in Gezhuping Town. The total basin area is 72km 2 The river is 30km long, with a large slope and an average flow of 1.51m over many years. 3 / s. At present, only daily runoff data for Liujiaping River from 1962 to 1976, a total of 15 years, have been collected. The period is relatively short. In order to eliminate the influence of interannual climate change and other factors on the calculation results of hydrological indicators, a hydrological time series of more than 20 years is usually required.

[0098] To this end, the monthly runoff from 1962 to 1976, a total of 15 years, was first used as a data sample. The monthly runoff was decomposed based on the integrated empirical mode decomposition technology, and the nearest neighbor method with coupled similarity indicators was used to predict and simulate the sequence number of each component. After superposition of the components, the monthly runoff from 1977 to 1981 was obtained. The monthly runoff prediction results are as follows:

[0099] In order to verify the reliability of the monthly runoff prediction model in the present invention, the runoff series of 180 months from 1962 to 1975 was decomposed into two parts, of which the runoff of 168 months in the first 14 years was used as training data, and the runoff of the last 12 months was used as test data. The training data was decomposed into several sub-series using the integrated empirical mode decomposition technology EEMD, and the ANN, ARIMA, SARIMA and the nearest neighbor method of coupled similarity index (hereinafter referred to as KNN) were used to perform time series analysis on each sub-series, denoted as EEMD-ANN, EEMD-ARIMA, EEMD-SARIMA and EEMD-KNN, and compared with the separate ANN, ARIMA and SARIMA and KNN models. The commonly used Nash coefficient NSE, mean relative error MARE and minimum root mean square error RMSE three indicators were used to evaluate the simulation effect of the model, and the results are shown in Table 1.

[0100] Table 1 Comparison of runoff simulation results

[0101]

[0102] As can be seen from the table, the EEMD-KNN model has significantly superior simulation effects compared to other models, both in the training period and the test period. The independent KNN model also has good simulation effects, indicating that compared with conventional time series processing methods, the nearest neighbor method of coupled similarity indicators performs better in sequence simulation prediction. In addition, the simulation effect of the monthly runoff prediction model based on the integrated empirical mode decomposition technology is also better than that of a single prediction model, indicating that modal decomposition of the original sequence before prediction can significantly improve the accuracy of the prediction model.

[0103] The EEMD-KNN model was used to predict and simulate monthly runoff from 1977 to 1981. The results are shown in Table 2.

[0104] Table 2 Predicted monthly runoff from 1977 to 1981

[0105]

[0106] Then, the improved nonparametric solution set model was used to decompose the predicted monthly runoff and reconstruct the daily runoff from 1977 to 1981. The daily runoff process from 1962 to 1981, a total of 20 years, was plotted ( Figure 3 ). The dark line in the figure corresponds to the measured daily runoff, and the light line corresponds to the reconstructed daily runoff. It can be seen that the reconstructed runoff shows a fluctuation characteristic similar to the measured runoff, retaining the inter-annual and intra-annual distribution characteristics of the measured runoff, and the reconstruction effect is good.

[0107] In order to quantitatively describe the effect of the reconstructed sequence and verify the performance of the improved nonparametric solution set model, the paper statistically analyzed the relevant parameters of the measured runoff and the reconstructed runoff (see Table 3), including the mean Q, mean square error S, and coefficient of variation C on the monthly and annual scales. v , skewness coefficient C s , minimum value Q min , maximum value Q max And the first-order correlation coefficient R 1 .

[0108] Table 3 Statistical parameters of measured runoff and reconstructed runoff

[0109]

[0110]

[0111] Note: ε is the relative error.

[0112] Draw box plots of 7 statistical parameters ( Figure 4 (a) to (g), where the boxes represent the distribution of the 15-year runoff statistical parameters, the middle red line represents the measured mean, the upper and lower box lines represent the 25th and 75th quantiles of the parameters, and the upper and lower box lines represent the 10th and 90th quantiles of the parameters. The solid triangle represents the reconstructed 5-year runoff statistical parameters. The horizontal axis 1 to 12 represents the statistical period from January to December, and 13 represents the statistical period of the year.

[0113] As can be seen from the figure, there is a large difference between the reconstructed runoff and the measured runoff in the parameter of runoff mean Q. Only in the three statistical periods of April, November and December, the mean of the reconstructed runoff can be within the upper and lower box lines (between the 10th and 90th percentiles of the measured runoff), while in the other statistical periods, the reconstructed runoff mean exceeds the upper and lower box lines. Taking the year as the statistical period, in the reconstructed 5-year runoff, only 1 year's runoff mean is within the upper and lower box lines, and the runoff mean parameter has a large difference. Since the improved non-parametric solution set model decomposes the predicted monthly scale runoff value into days, the runoff mean is only affected by the runoff prediction model, but has nothing to do with the non-parametric solution set model. The reconstructed runoff mean is quite different from the measured runoff, which shows that the monthly runoff process has a trend and is difficult to meet the distribution characteristics of the historical measured runoff.

[0114] For other statistical parameters, the parameter values ​​of reconstructed runoff are almost all within the upper and lower box lines, and there are more values ​​within the upper and lower box lines (between the 25th and 75th percentiles of the measured runoff). For example, the standard deviation S, of the 65 statistical parameters of the reconstructed 5-year runoff in 13 statistical periods, only the standard deviation of September 1978 exceeds the upper and lower box lines; for the coefficient of variation C v and the skewness coefficient C s, there is only one parameter value beyond the upper and lower box lines; for the maximum value Q max , minimum value Q min And the first-order correlation coefficient R 1 , the number of parameters exceeding the upper and lower box lines is relatively large, but all are less than 5. Therefore, the daily-scale runoff reconstructed by “forecast-decomposition” can reproduce the relevant characteristics of the measured historical runoff, and the simulation effect is good.

Claims

1. A daily-scale runoff reconstruction method based on "prediction-decomposition", characterized in that: The following steps are involved: S1. Obtain the monthly historical runoff sequence Z and the monthly and daily historical runoff process Y based on the daily historical runoff data, and form the historical sample points [Z, Y] of the monthly and daily runoff; S2. The non-stationary monthly historical runoff sequence Z is spatially decomposed using the integrated empirical mode decomposition technology and converted into a series of relatively stable IMF component time series signals; the nearest neighbor method with coupled similarity indicators is used to predict the time series signals of each IMF component, and the prediction results of each IMF component are superimposed to obtain m predicted monthly runoff values; S3. Rotate and transform the historical runoff process Y on each monthly and daily scale to obtain the transformed historical runoff process X on each monthly and daily scale. Use the improved non-parametric solution set model to decompose the m predicted monthly runoff values ​​respectively: First, find k predicted monthly runoff values ​​Z that need to be decomposed in the transformed historical runoff process X on each monthly and daily scale. * Similar months, and construct the cumulative weight distribution function g(m). Secondly, a roulette wheel is used to select one of the k similar months to estimate the rotation transformation X of the daily runoff process in that month. * , and then through inverse transformation, we can get the actual runoff process Y on the monthly and daily scale * ; Then the remaining m-1 predicted monthly runoff values ​​are decomposed, and finally m decomposed monthly and daily runoff processes are obtained to achieve the reconstruction of long-sequence daily runoff.

2. A daily-scale runoff reconstruction method based on "prediction-decomposition" according to claim 1, characterized in that: In step S2, the method of spatially decomposing the non-stationary monthly historical runoff series Z using the integrated empirical mode decomposition technique is as follows: (1) Add Gaussian white noise n to the monthly historical runoff series Z i (t), to obtain the new runoff sequence y i (t): y i (t)=Z+n i (t) (1) Where i is the number of times Gaussian white noise is added; (2) Using the empirical mode decomposition technique to analyze the new runoff series y i (t) is decomposed to obtain n IMF components imf i,j (t) and a residual term r i,n (t): Among them, IMF i,j (t) is to add y i The jth IMF component of (t); (3) Repeat steps (1) to (2) N times, integrate and average the IMF components and residual terms obtained each time and use them as the final decomposition result: (4) The final monthly historical runoff series Z is expressed by n IMF components and 1 residual term:

3. A daily-scale runoff reconstruction method based on "prediction-decomposition" according to claim 1, characterized in that: The method for predicting the time series signals of each IMF component using the nearest neighbor method of the coupling similarity index in step S2 is: Determine the maximum lag order p of the IMF component series and mark the historical value x of the IMF component series t And the corresponding eigenvector D t ={x t-p ,x t-p+1 ,...,x t-1 }, t=1,2,...,n, through the Euclidean distance between vectors, from D t Find k features that are consistent with the eigenvector D n+1 ={x n-p+1 ,x n-p+2 ,...,x n } feature vectors with good similarity, and the subsequent value x' corresponding to these k feature vectors j , j = 1, 2, ..., k, predict x n+1 : Among them, w j is the weight function, k is the number of nearest neighbors, j is the sorting position of the k eigenvectors of the nearest neighbors, and w j The calculation formulas for and k are as follows: The prediction results of each IMF component series are superimposed to obtain the predicted monthly runoff value.

4. A daily-scale runoff reconstruction method based on "prediction-decomposition" according to claim 1, characterized in that: The historical runoff process Y = (y1, y2, ..., y d ) T , the transformed monthly and daily historical runoff process X=(x1,x2,...,x d ) T The transformation formula is: X=RY (8) Among them, R is the unit orthogonal matrix, and the dth component x in X after rotation transformation is d The following conditions are met: Among them, d is the number of days corresponding to each month.

5. The daily-scale runoff reconstruction method based on "prediction-decomposition" according to claim 1 is characterized in that: The specific steps of using the improved non-parametric solution set model to decompose the m predicted monthly runoff values ​​in step S3 are as follows: First, according to the rotation transformation formula, calculate X * The dth component in Then, calculate the first d-1 coordinates If the roulette wheel method is used to estimate the rotation transformation X of the daily runoff process in the lth month, * , take X l The first d-1 components U l As X * The first d-1 coordinates U * ; Combined with calculation Estimate the daily runoff of the month after rotation transformation Then, through the inverse rotation transformation, the actual runoff process Y on the monthly and daily scale is obtained. * , and so on, the remaining m-1 predicted monthly runoff values ​​are decomposed.

6. A daily-scale runoff reconstruction system based on "prediction-decomposition", characterized in that: include: The data collection and processing module is used to collect daily historical runoff data, obtain the monthly historical runoff sequence Z and the monthly and daily historical runoff process Y, and form the historical sample points [Z, Y] of the monthly and daily runoff; The runoff prediction module uses the integrated empirical mode decomposition technology to spatially decompose the monthly runoff sequence, and uses the nearest neighbor method of coupled similarity indicators to predict the time series signals of each component, superimposes the prediction results of the time series signals of each component, and obtains the predicted monthly scale runoff; The runoff decomposition module uses an improved non-parametric solution set model to decompose the predicted monthly runoff and reconstruct the daily runoff. Specifically: Rotate the historical runoff process Y on each month and day scale to obtain the transformed historical runoff process X on each month and day scale; find k predicted monthly runoff values ​​Z that need to be decomposed in the transformed historical runoff process X on each month and day scale * similar months, and construct the cumulative weight distribution function g(m); select one from the k similar months by roulette to estimate the rotation transformation X of the daily scale runoff process in that month * , and then through the inverse rotation transformation, the decomposed actual runoff process Y * ; Then the remaining m-1 predicted monthly runoff values ​​are decomposed, and finally m decomposed monthly daily runoff processes are obtained to achieve long-sequence daily runoff reconstruction.

7. A device, characterized in that: comprising a memory and a processor, wherein: A memory for storing computer programs that can be run on the processor; A processor, for executing the steps of a daily-scale runoff reconstruction method based on "prediction-decomposition" as described in any one of claims 1 to 5 when running the computer program.

8. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by at least one processor, implements the steps of a daily-scale runoff reconstruction method based on "prediction-decomposition" as described in any one of claims 1 to 5.

Images