A method and system for estimating reservoir daily runoff history series by integrating machine learning correction

Abstract

The present invention discloses a method and system for estimating a historical series of daily runoff of a reservoir that integrates machine learning correction. The method comprises: S100: estimating the historical long series of ten-day runoff data of the target reservoir station using the runoff correlation method and the hydrological analogy method, respectively, based on the short series of measured runoff data of the target reservoir station and the long series of measured runoff data of the reference station; S200: correcting the estimated historical long series of ten-day runoff data using a machine learning model to obtain the corrected historical long series of ten-day runoff data of the target reservoir station; S300: temporally downscaling the corrected historical long series of ten-day runoff data to obtain the long series of daily runoff data of the target reservoir station. When the measured runoff data of the reservoir basin is insufficient, the present invention can be used to estimate the historical long series of daily runoff data of the target reservoir, and the estimation accuracy and time resolution are significantly improved, and the estimated runoff series results are verified to be reasonable.

Description

Technical Field

[0001] The present application belongs to the technical field of reservoir runoff estimation, and specifically relates to a method and system for estimating a historical series of reservoir daily runoff that integrates machine learning correction. Background Art

[0002] The design runoff series (usually required to be at least 30 years) is a crucial basis for determining reservoir size, operating water levels, and developing reservoir operation regulations. However, many newly constructed reservoirs lack sufficient measured runoff data. This often requires selecting similar basins upstream and downstream of the reservoir, or in adjacent basins. Hydrological stations with long-term measured runoff data in these similar basins are used as reference stations. Methods such as runoff correlation and hydrological analogy are then used to estimate the reservoir's design runoff series. These methods for estimating the reservoir's design runoff series are documented in SL / T-278, "Specifications for Hydrological Calculations for Water Conservancy and Hydropower Projects."

[0003] The technical approach of the runoff correlation method is to establish a statistical correlation between a short series of measured runoff at a reservoir station and a short series of measured runoff at a reference station over the same period. This statistical correlation is then combined with the long series of measured runoff data from the reference station to estimate long-term runoff data at the reservoir station over a ten-day period. The technical approach of the hydrological analogy method is to establish a proportional relationship between the catchment area and surface rainfall of the reservoir station and the reference station. Based on this proportional relationship, the long series of measured runoff data from the reference station is used to estimate long-term runoff data at the reservoir station over a ten-day period.

[0004] However, neither the runoff correlation method nor the hydrological analogy method considers the complexity of the basin's runoff generation and confluence processes. Both methods assume that the rainfall-runoff similarity between the reservoir basin (the basin where the reservoir is located) and the reference basin (the basin where the reference station is located) follows a linear relationship. This deviates from the nonlinear characteristics of actual runoff generation and confluence processes, potentially making it difficult to achieve high accuracy in the resulting long-series reservoir runoff. Furthermore, the rainfall-runoff similarity between the reservoir basin and the reference basin is generally only achieved over longer timescales, such as ten days or longer. This also results in the difficulty of achieving daily temporal resolution in the resulting long-series reservoir runoff. As reservoir operation becomes increasingly sophisticated, the requirements for the accuracy and temporal resolution of the designed runoff series are also increasing. Therefore, the designed runoff series derived from methods such as the runoff correlation method and the hydrological analogy method are unable to support these refined reservoir operations. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for estimating the daily runoff history series of a reservoir that integrates machine learning correction. The method and system of this application can significantly improve the estimation accuracy and time resolution of the long series of reservoir runoff data. The estimated long series of reservoir runoff data can meet the refined scheduling and operation of the reservoir.

[0006] To achieve the above objectives, on the one hand, this application provides a method for estimating reservoir daily runoff history series by integrating machine learning correction, including:

[0007] S100: Based on the short series of measured runoff data of the target reservoir station and the long series of measured runoff data of the reference station, the runoff correlation method and the hydrological analogy method are used to estimate the historical long series of ten-day runoff data of the target reservoir station, which are recorded as the first long series estimated ten-day runoff data and the second long series estimated ten-day runoff data respectively;

[0008] S200: Use the machine learning model to correct the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data to obtain the corrected historical long series of ten-day runoff data for the target reservoir station;

[0009] The machine learning model is first trained using training samples, wherein the training samples include a short series of measured ten-day runoff data of the target reservoir station and a first long series of estimated ten-day runoff data and a second long series of estimated ten-day runoff data for the same period; the machine learning model is trained using the short series of measured ten-day runoff data of the target reservoir station as output and the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data for the same period as input;

[0010] S300: Temporally downscaling the corrected historical long series of ten-day runoff data to obtain the long series of daily runoff data of the target reservoir station.

[0011] In some specific embodiments, the machine learning model is first trained using training samples, further comprising:

[0012] S210: constructing a sample data set, wherein the sample data set includes a short series of measured ten-day runoff data of the target reservoir station and a first long series of estimated ten-day runoff data and a second long series of estimated ten-day runoff data for the same period; dividing the sample data set into a training sample set and a test sample set;

[0013] S220: Training the machine learning model using the training sample set;

[0014] S230: Testing the trained machine learning model using a test sample set;

[0015] Select multiple different machine learning models for training and testing respectively, and use the machine learning model with the highest accuracy for calibration.

[0016] The above-mentioned machine learning model can be selected from one or more of the support vector machine model, random forest model, gradient boosting regression tree model, and long short-term memory network model.

[0017] In some specific implementations, step S200 further includes:

[0018] The corrected historical long series of ten-day runoff data are corrected according to the relative deviation of water volume:

[0019]

[0020] Among them, Q' ml,旬 represents the corrected historical long-series ten-day runoff data; n represents the ten-day number of the short-series measured runoff data of the target reservoir station; Indicates the short series of measured ten-day runoff at the target reservoir station the sum of represents the measured ten-day runoff in the i-th short series; Indicates the corrected historical long series of ten-day runoff and The sum of runoff during the same period.

[0021] In some specific implementations, step S300 further includes:

[0022] The following are performed on the runoff of each ten-day period in the corrected historical long series of ten-day runoff data:

[0023] Determine whether there is rainfall in the current ten-day period. If there is no rainfall, the current ten-day runoff is underground runoff. If there is rainfall, the current ten-day runoff is divided into underground runoff and surface runoff in proportion according to the rainfall in the current ten-day period.

[0024] For underground runoff, the daily underground runoff is obtained by temporal downscaling based on the equal runoff volume every day within ten days;

[0025] For surface runoff, Perform time downscaling to obtain daily surface runoff Among them, m represents the total number of days in the current decade, represents the surface runoff in the current decade, P 目标,j represents the rainfall in the reservoir basin on the jth day of the current ten-day period, Indicates the total rainfall in the reservoir basin in the current ten days;

[0026] For the current ten-day period without rainfall, the daily groundrunoff is the daily runoff data; for the current ten-day period with rainfall, the daily groundrunoff and daily surface runoff on the same day are added together to obtain the daily runoff data.

[0027] On the other hand, the present application also provides a reservoir daily runoff history series estimation system integrated with machine learning correction, including:

[0028] The first module is used to estimate the historical long-series ten-day runoff data of the target reservoir station based on the short-series measured runoff data of the target reservoir station and the long-series measured runoff data of the reference station using the runoff correlation method and the hydrological analogy method, respectively. These are recorded as the first long-series estimated ten-day runoff data and the second long-series estimated ten-day runoff data;

[0029] The second module is used to use the machine learning model to correct the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data to obtain the corrected historical long series of ten-day runoff data for the target reservoir station;

[0030] The machine learning model is first trained using training samples, wherein the training samples include a short series of measured ten-day runoff data of the target reservoir station and a first long series of estimated ten-day runoff data and a second long series of estimated ten-day runoff data for the same period; the machine learning model is trained using the short series of measured ten-day runoff data of the target reservoir station as output and the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data for the same period as input;

[0031] The third module is used to time downscale the corrected historical long series of ten-day runoff data to obtain the long series of daily runoff data of the target reservoir station.

[0032] In some specific embodiments, the second module further includes a correction submodule for correcting the corrected historical long series of ten-day runoff data according to the relative deviation of water volume:

[0033]

[0034] Among them, Q' ml,旬 represents the corrected historical long-series ten-day runoff data; n represents the ten-day number of the short-series measured runoff data of the target reservoir station; Indicates the short series of measured ten-day runoff at the target reservoir station the sum of represents the measured ten-day runoff in the i-th short series; Indicates the corrected historical long series of ten-day runoff and The sum of runoff during the same period.

[0035] In some embodiments, the third module further includes:

[0036] Submodule 1 is used to determine whether there is rainfall in the current ten-day period. If there is no rainfall, the current ten-day runoff is underground runoff. If there is rainfall, the current ten-day runoff is divided into underground runoff and surface runoff in proportion according to the rainfall in the current ten-day period.

[0037] Submodule 2 is used to perform time downscaling on the underground runoff, assuming that the runoff volume is equal every day within ten days, to obtain the daily underground runoff;

[0038] Submodule 3 is used to analyze surface runoff according to Perform time downscaling to obtain daily surface runoff Among them, m represents the total number of days in the current decade, represents the surface runoff in the current decade, P 目标,jrepresents the rainfall in the reservoir basin on the jth day of the current ten-day period, Indicates the total rainfall in the reservoir basin in the current ten days;

[0039] Submodule 4 is used to obtain the daily underground runoff data for the current ten-day period without rainfall; and to obtain the daily runoff data by adding the daily underground runoff and the daily surface runoff for the current ten-day period with rainfall.

[0040] Compared with the prior art, this application has the following advantages and beneficial effects:

[0041] When the measured runoff data in the reservoir basin is insufficient, the present invention can be used to infer the long series of daily runoff data of the target reservoir, with the time resolution down to the day. The inference accuracy and time resolution are significantly improved, and the calculated runoff series results are verified to be reasonable. The calculated long series of reservoir runoff data can support the refined scheduling and operation of the reservoir. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic diagram of the process of this application method;

[0043] Figure 2 The comparison between the measured ten-day runoff data and the estimated ten-day runoff data of the target reservoir station in the embodiment;

[0044] Figure 3 The comparison between the measured ten-day runoff data and the corrected ten-day runoff data of the target reservoir station during the training period in the embodiment;

[0045] Figure 4 The comparison between the measured ten-day runoff data and the corrected ten-day runoff data of the target reservoir station during the inspection period in the embodiment;

[0046] Figure 5 The comparison between the estimated long series daily runoff data and the short series measured daily runoff data of the target reservoir station in the embodiment;

[0047] Figure 6 It is the variation trend curve of the annual average runoff and annual rainfall in the reservoir basin in the embodiment. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions, and advantages of this application more clear, the specific embodiments of this application will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the specific embodiments described are only part of the specific embodiments of this application, not all of the specific embodiments. Based on the specific embodiments of this application, all other specific embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0049] The technical idea behind this application's method for estimating reservoir daily runoff history series using machine learning correction is:

[0050] First, the historical long-term runoff data of the target reservoir station is estimated using existing standard methods, including the runoff correlation method and the hydrological analogy method.

[0051] Next, considering the nonlinear characteristics of the actual runoff generation and runoff process, a machine learning model that can handle nonlinear correspondence is introduced to correct the estimated historical long-series runoff data to improve the estimation accuracy.

[0052] Finally, the corrected historical long series of ten-day runoff data were time-downscaled to obtain the long series of daily runoff data of the target reservoir station.

[0053] See Figure 1 , shown is a flow chart of the reservoir daily runoff history series estimation method integrated with machine learning correction in this application. Figure 1 The specific implementation of this application is described in detail, and the steps are as follows:

[0054] S100: Based on the short series of measured runoff data of the target reservoir station and the long series of measured runoff data of the reference station, the runoff correlation method and the hydrological analogy method are used to estimate the historical long series of ten-day runoff data of the target reservoir station, which are recorded as the first long series estimated ten-day runoff data and the second long series estimated ten-day runoff data respectively;

[0055] Reference stations refer to the reference stations of the target reservoir. Selection of reference stations follows conventional industry practice and must meet the following two criteria: ① The reference station's basin is upstream, downstream, or adjacent to the reservoir basin; and ② The reference station's basin has a long series of measured runoff data. Here, the reservoir basin refers to the basin of the target reservoir.

[0056] In this specific implementation, the method for selecting the reference station is:

[0057] First, obtain the upstream, downstream, and adjacent basins with long-term measured runoff data and record them as potential reference basins; upstream, downstream, and adjacent basins here refer to the upstream, downstream, or adjacent basins of the reservoir basin;

[0058] Then, from the potential reference basins, basins with similar rainfall and underlying surface characteristics to the reservoir basin are selected as reference basins. The hydrological stations in the reference basins are referred to as reference stations. The underlying surface characteristics include but are not limited to the basin's topography and vegetation.

[0059] S200: Use the machine learning model to correct the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data to obtain the corrected historical long series of ten-day runoff data for the target reservoir station;

[0060] This step mainly includes two parts: using training samples to train the machine learning model and using the trained machine learning model to correct the estimated historical long series of ten-day runoff data.

[0061] In this specific embodiment, using training samples to train a machine learning model further includes:

[0062] S210: Constructing a sample data set, wherein the sample data set includes a short series of measured ten-day runoff data of the target reservoir station and a first long series of estimated ten-day runoff data and a second long series of estimated ten-day runoff data for the same period; dividing the sample data set into a training sample set and a test sample set, wherein the sample data set is used to train the machine learning model and the test sample set is used to test the machine learning model; the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data for the same period, wherein "the same period" refers to the same period as the short series of measured ten-day runoff data;

[0063] S220: Training the machine learning model using the training sample set; specifically, using the short series of measured ten-day runoff data of the target reservoir station as output, and the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data of the same period as input, to train and calibrate the structure and parameters of the machine learning model;

[0064] The machine learning model is represented as follows:

[0065] Q ml,旬 =f(Q 1,旬 +Q 2,旬 ) (1)

[0066] In formula (1):

[0067] f() represents the machine learning model function;

[0068] Q ml,旬 Represents the historical long series of ten-day runoff data after correction by the machine learning model;

[0069] Q 1,旬 It represents the first long series of estimated ten-day runoff data estimated using the runoff correlation method;

[0070] Q 2,旬 Represents the second longest series of estimated ten-day runoff data estimated using the hydrological analogy method.

[0071] S230: Testing the trained machine learning model using a test sample set;

[0072] In this specific embodiment, at least two of the support vector machine model, random forest model, gradient boosting regression tree model, long short-term memory network model, etc. are selected for training respectively, and then the test sample set is used to test the trained multiple machine learning models, and the machine learning model with the highest accuracy is selected.

[0073] In this specific embodiment, the method of using the trained machine learning model to correct the estimated historical long series of ten-day runoff data further includes: inputting the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data into the trained machine learning model, and outputting the corrected historical long series of ten-day runoff data Q ml,旬 .

[0074] Furthermore, the corrected historical long series of ten-day runoff data Q is analyzed based on the relative deviation of water volume. ml,旬 Make corrections to further improve accuracy.

[0075] Specifically, the historical long series of ten-day runoff data Q ml,旬 The correction formula is as follows:

[0076]

[0077] In formula (2):

[0078] Q' ml,旬 It represents the revised historical long series of ten-day runoff data;

[0079] n represents the number of ten days of short series measured runoff data at the target reservoir station;

[0080] Indicates the short series of measured ten-day runoff at the target reservoir station the sum of represents the measured ten-day runoff in the i-th short series;

[0081] Indicates the corrected historical long series of ten-day runoff and The sum of runoff during the same period.

[0082] S300: Temporally downscaling the corrected historical long series of ten-day runoff data to obtain the long series of daily runoff data of the target reservoir station;

[0083] Determine whether there has been rainfall in the current ten-day period. If there has been no rainfall, all runoff in the current ten-day period is groundwater runoff. If there has been rainfall, the runoff is divided into groundwater runoff and surface runoff in proportion to the current ten-day rainfall. Groundwater runoff is temporally downscaled to equalize the runoff volume every day within the ten-day period to obtain daily groundwater runoff. Surface runoff is temporally downscaled to obtain daily surface runoff based on the variation of other hydrometeorological factors with daily resolution (such as the daily runoff process at the reference station and the daily rainfall process in the reservoir basin). Finally, the daily groundwater runoff and daily surface runoff for the same day are added together to obtain the daily runoff data.

[0084] Specifically, the time downscaling formula is as follows:

[0085]

[0086] In formulas (3) to (6):

[0087] and represent ten-day underground runoff and ten-day surface runoff respectively;

[0088] P 目标,旬 Indicates the ten-day rainfall in the reservoir basin;

[0089] k represents the distribution coefficient of groundwater runoff. This value is an empirical value determined by referring to the ratio of surface water resources to groundwater resources in the reference basin over many years. For the humid areas in the south, the distribution coefficient k is generally taken as 0.1.

[0090] and They represent the daily underground runoff and daily surface runoff on the jth day of the current ten-day period respectively;

[0091] P 目标,j represents the daily rainfall in the reservoir basin on the jth day of the current ten-day period;

[0092] Q' ml,j represents the historical long series of daily runoff data after time downscaling;

[0093] m represents the total number of days in the current decade.

[0094] The technical effects of this application will be further explained below with reference to embodiments.

[0095] In this example, hydrological station A is located on a river adjacent to the target reservoir. The rainfall and underlying surface characteristics of the basin where hydrological station A is located are similar to those of the target reservoir. Hydrological station A has a long series of continuous measured runoff data from 1983 to 2023. Therefore, hydrological station A was selected as the reference station for the target reservoir.

[0096] Based on the long series of measured runoff data of the reference station and the short series of measured runoff data of the target reservoir station, the runoff correlation method and the hydrological analogy method are used to estimate the historical long series of ten-day runoff data of the target reservoir station, which are recorded as the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data respectively. The short series of measured runoff data of the target reservoir station is compared with the runoff data of the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data during the same period. Figure 2 In the figure, "measured" represents the short series of measured runoff data of the target reservoir station, "series one" and "series two" represent the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data respectively; and the accuracy of the runoff correlation method and the hydrological analogy method for the same period is evaluated. The accuracy data are shown in Table 1.

[0097] Table 1 Accuracy of runoff data estimated by runoff correlation method and hydrological analogy method during the same period

[0098] Evaluation indicators Runoff correlation method Hydrological analogy method Nash efficiency coefficient 0.965 0.958 <![CDATA[Average absolute flow deviation (unit: m 3 / s)]]> 3.54 5.32 Average relative flow deviation (unit: %) 14.22 33.48 Relative deviation of total water volume (unit: %) 6.19 19.18

[0099] A sample data set was constructed using the short series of measured ten-day runoff data of the target reservoir station and the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data of the same period. 2020-2022 was selected as the training period, and 2019 was selected as the test period. The ten-day runoff data estimated during the training period was used as the input of the machine learning model, and the short series of measured ten-day runoff data of the target reservoir station during the training period was used as the output of the machine learning model to train the machine learning model. In this embodiment, four different machine learning models were selected for training, including a support vector machine model, a random forest model, a gradient boosting regression tree model, and a long short-term memory network model.

[0100] The trained machine learning model was used to calibrate and correct the estimated ten-day runoff data during the training and test periods, and the corrected ten-day runoff data were compared with the short series of measured ten-day runoff data of the target reservoir station during the test period. For comparison, see Figures 3-4 In the figure, "measured" represents the short series of measured ten-day runoff data at the target reservoir station, "support vector machine" represents the ten-day runoff data corrected by the support vector machine model, "random forest" represents the ten-day runoff data corrected by the random forest model, "gradient boosting regression tree" represents the ten-day runoff data corrected by the gradient boosting regression tree model, and "long short-term memory" represents the ten-day runoff data corrected by the long short-term memory network model; and the accuracy of the runoff forecast after the machine learning model is evaluated. The accuracy data of different machine learning models are shown in Table 2. Figures 3-4 As shown in Table 2, the accuracy of the ten-day runoff data after correction by the gradient boosting regression tree model is the highest.

[0101] Table 2 Accuracy of runoff estimation during the same period after correction by four machine learning models

[0102]

[0103] The historical long series of ten-day runoff data after the gradient boosting regression tree model correction was downscaled to calculate the historical long series of daily runoff data of the target reservoir station, and compared with the short series of measured daily runoff data of the target reservoir station. Figure 5 As shown. A frequency analysis was performed on the estimated long-series daily runoff data from the target reservoir station and the long-series measured daily runoff data from the reference station. The results are shown in Table 3. The reservoir basin is located upstream of the reference basin, and the ratio of the reservoir basin area to the reference basin area is 43.62%. Comparisons show that the ratios of the long-series runoff mean and the average flow at each frequency range from 46.21% to 47.83% for the reservoir basin and the reference basin, respectively. This is greater than the area ratio of the reservoir basin to the reference basin, consistent with the trend that the rainfall center of the basin is close to the upstream reservoir area, with rainfall decreasing from upstream to downstream. Furthermore, the coefficient of variation (Cv) of the long-series runoff for the target reservoir station and the reference station are 0.23 and 0.24, respectively, consistent with the basin similarity assessment.

[0104] Table 3 Frequency analysis results of long series daily runoff data of target reservoir station and reference station

[0105]

[0106] The annual average runoff of the reservoir basin can be calculated using the estimated long series daily runoff data of the target reservoir station. According to the MK test, the annual rainfall in the reservoir basin has a continuous upward trend since 1991, and the annual average runoff has also a continuous upward trend since 1991. The annual average runoff of the reservoir basin from 1983 to 2022 is consistent with the change trend of annual rainfall. Figure 6 .

[0107] In combination with the above embodiments, it can be concluded that: when the reservoir basin runoff series is short, the use of the method of the present application can significantly improve the calculation accuracy and time resolution of the long series of reservoir runoff data, and the calculated runoff series results are reasonable.

[0108] Note that the above are only preferred embodiments of the present application and the technical principles employed. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and that various obvious changes, readjustments, and substitutions can be made by those skilled in the art without departing from the scope of protection of the present application. Therefore, although the present application has been described in more detail through the above embodiments, the present application is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the scope of protection of the present application, all of which fall within the scope of protection of the present application.

Claims

1. A method for estimating reservoir daily runoff history series by integrating machine learning correction, characterized by: include: S100: Based on the short series of measured runoff data of the target reservoir station and the long series of measured runoff data of the reference station, the runoff correlation method and the hydrological analogy method are used to estimate the historical long series of ten-day runoff data of the target reservoir station, which are recorded as the first long series estimated ten-day runoff data and the second long series estimated ten-day runoff data respectively; S200: Use the machine learning model to correct the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data to obtain the corrected historical long series of ten-day runoff data Q of the target reservoir station. ml,旬 ; The machine learning model is first trained using training samples, wherein the training samples include a short series of measured ten-day runoff data of the target reservoir station and a first long series of estimated ten-day runoff data and a second long series of estimated ten-day runoff data for the same period; The machine learning model is trained using the short series of measured ten-day runoff data at the target reservoir station as output and the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data over the same period as input. S300: Temporally downscaling the corrected historical long series of ten-day runoff data to obtain the long series of daily runoff data of the target reservoir station; Step S200 also includes: The corrected historical long series of ten-day runoff data are corrected according to the relative deviation of water volume: Among them, Q ' ml,旬 represents the corrected historical long-series ten-day runoff data; n represents the ten-day number of the short-series measured runoff data of the target reservoir station; Indicates the short series of measured ten-day runoff at the target reservoir station The sum of represents the measured ten-day runoff in the i-th short series; Indicates the corrected historical long series of ten-day runoff and The sum of runoff during the same period.

2. The method for estimating reservoir daily runoff history series by integrating machine learning correction as claimed in claim 1 is characterized by: The machine learning model is first trained using training samples, further comprising: S210: constructing a sample data set, wherein the sample data set includes a short series of measured ten-day runoff data of the target reservoir station and a first long series of estimated ten-day runoff data and a second long series of estimated ten-day runoff data for the same period; dividing the sample data set into a training sample set and a test sample set; S220: Training the machine learning model using the training sample set; S230: Testing the trained machine learning model using a test sample set; Select multiple different machine learning models for training and testing respectively, and use the machine learning model with the highest accuracy for calibration.

3. The method for estimating reservoir daily runoff history series by integrating machine learning correction as claimed in claim 1 or 2, characterized in that: The machine learning model is one or more of a support vector machine model, a random forest model, a gradient boosting regression tree model, and a long short-term memory network model.

4. The method for estimating reservoir daily runoff history series by integrating machine learning correction as claimed in claim 1 is characterized by: Step S300 further includes: The following are performed on the runoff of each ten-day period in the corrected historical long series of ten-day runoff data: Determine whether there is rainfall in the current ten-day period. If there is no rainfall, the current ten-day runoff is underground runoff. If there is rainfall, the current ten-day runoff is divided into underground runoff and surface runoff in proportion according to the rainfall in the current ten-day period. For underground runoff, the daily underground runoff is obtained by temporal downscaling based on the equal runoff volume every day within ten days; For surface runoff, Perform time downscaling to obtain daily surface runoff Among them, m represents the total number of days in the current decade, represents the surface runoff in the current decade, P 目标,j represents the rainfall in the reservoir basin on the jth day of the current ten-day period, It represents the total rainfall in the reservoir basin in the current ten days; For the current ten-day period without rainfall, the daily groundwater runoff is the daily runoff data; For the current decade with rainfall, the daily groundwater runoff and daily surface runoff on the same day are added together to obtain the daily runoff data.

5. A reservoir daily runoff history series estimation system integrated with machine learning correction, characterized by: include: The first module is used to estimate the historical long-series ten-day runoff data of the target reservoir station based on the short-series measured runoff data of the target reservoir station and the long-series measured runoff data of the reference station using the runoff correlation method and the hydrological analogy method, respectively. These are recorded as the first long-series estimated ten-day runoff data and the second long-series estimated ten-day runoff data; The second module is used to use the machine learning model to correct the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data to obtain the corrected historical long series of ten-day runoff data Q of the target reservoir station. ml,旬 ; The machine learning model is first trained using training samples, wherein the training samples include a short series of measured ten-day runoff data of the target reservoir station and a first long series of estimated ten-day runoff data and a second long series of estimated ten-day runoff data for the same period; The machine learning model is trained using the short series of measured ten-day runoff data at the target reservoir station as output and the first long series of estimated ten-day runoff data and the second long series of estimated ten-day runoff data over the same period as input. The third module is used to temporally downscale the corrected historical long-series ten-day runoff data to obtain the long-series daily runoff data of the target reservoir station; The second module also includes a correction submodule for correcting the corrected historical long series of ten-day runoff data according to the relative deviation of water volume: Among them, Q ' ml,旬 represents the corrected historical long-series ten-day runoff data; n represents the ten-day number of the short-series measured runoff data of the target reservoir station; Indicates the short series of measured ten-day runoff at the target reservoir station The sum of represents the measured ten-day runoff in the i-th short series; Indicates the corrected historical long series of ten-day runoff and The sum of runoff during the same period.

6. The reservoir daily runoff history series estimation system integrated with machine learning correction as claimed in claim 5 is characterized by: The machine learning model is a support vector machine model, a random forest model, a gradient boosting regression tree model or a long short-term memory network model.

7. The reservoir daily runoff history series estimation system integrated with machine learning correction as claimed in claim 5 is characterized by: The third module further includes: Submodule 1 is used to determine whether there is rainfall in the current ten-day period. If there is no rainfall, the current ten-day runoff is underground runoff. If there is rainfall, the current ten-day runoff is divided into underground runoff and surface runoff in proportion according to the rainfall in the current ten-day period. Submodule 2 is used to perform time downscaling on the underground runoff, assuming that the runoff volume is equal every day within ten days, to obtain the daily underground runoff; Submodule 3 is used to analyze surface runoff according to Perform time downscaling to obtain daily surface runoff Among them, m represents the total number of days in the current decade, represents the surface runoff in the current decade, P 目标,j represents the rainfall in the reservoir basin on the jth day of the current ten-day period, It represents the total rainfall in the reservoir basin in the current ten days; Submodule 4 is used to obtain the daily underground runoff data for the current ten-day period without rainfall; and to obtain the daily runoff data by adding the daily underground runoff and the daily surface runoff for the current ten-day period with rainfall.

Images