A method and system for interpolating and extending daily flow series in mountainous rivers based on machine learning

Abstract

The present invention discloses a method and system for interpolating and extending a daily flow series of a mountain river based on machine learning. The method comprises: collecting and collating data: including collecting a daily rainfall series of a rainfall observation station in a basin above a mountain river station, collating to obtain a daily surface rainfall series in the basin above the mountain river station, collecting a daily water surface evaporation series, a daily water level series, and a daily flow series; naming the data of a mountain river station having a daily surface rainfall series, a daily flow series, a daily water level series, and a daily water surface evaporation series as a short series, naming the data of a mountain river station having a daily surface rainfall series, a daily water level series, and a daily water surface evaporation series but not having a daily flow series as a long series; using a machine learning model to interpolate and extend the daily flow series of a mountain river station; and conducting a reasonableness test on the interpolation and extension results. The present invention improves the interpolation and extension accuracy of the daily flow series of a mountain river station.

Description

Technical Field

[0001] The present invention relates to the technical field of interpolation and extension of mountain river flow series, and in particular to a method and system for interpolation and extension of mountain river daily flow series based on machine learning. Background Art

[0002] High-quality, long-series flow series are important bases for water resources planning, flood and drought prevention, water conservancy project design and other businesses. However, due to various factors such as changes in the nature of the measuring station, equipment failure, and insufficient funds, many mountain river stations lack long-series continuous observation flow data. It is usually necessary to compile a daily flow table based on the measured water level and flow series of mountain river stations, using the water level flow relationship legal line and interpolation extension.

[0003] Normally, when compiling the flow series of mountainous rivers, the idea of ​​the water level-flow relationship method is to use the water level and flow data of the synchronous observation of the year to determine the water level-flow relationship of the year, and then deduce the flow series based on the water level series of continuous observation according to the water level-flow relationship curve. However, due to the limitations of the actual situation, many stations have changed from the original synchronous observation of water level and flow to only observing water level for a period of time. In order to interpolate and extend the daily flow series of mountainous river stations, the water level and flow series of the short series of synchronous observations before the missing years can be used to construct a comprehensive line of multi-year water level-flow relationship. However, when using the water level-flow relationship method to interpolate and extend the daily flow series of mountainous river stations, it is believed that the water level-flow relationship curve of mountainous rivers is a stable single line, and the changes in the water level-flow relationship curve of mountainous rivers over a long period of time are not considered. In addition, a single water level-flow relationship curve may be difficult to simultaneously guarantee the high-precision fitting of different magnitudes of flow in the dry season and flood season, resulting in the inferred long series of flow results of mountainous rivers being difficult to support the high-precision requirements of hydrological data compilation. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes a method and system for interpolating and extending the daily flow series of mountainous rivers based on machine learning, which can improve the estimation accuracy of long series daily flow data of mountainous river stations.

[0005] In order to achieve the above object, the present invention is implemented by the following technical solutions:

[0006] The method for interpolating and extending the daily flow series of a mountain river based on machine learning of the present invention comprises the following operations:

[0007] Collect and organize data: including naming mountain river stations and upstream catchment areas as the basins above mountain river stations, collecting daily rainfall series of rainfall observation stations in the basins above mountain river stations, daily water surface evaporation series of representative stations in the basins where mountain river stations are located, daily water level series of mountain river stations and unifying the base surface, daily flow series of mountain river stations, and statistically organizing the daily rainfall series of rainfall observation stations in the basins above mountain river stations to obtain the daily surface rainfall series of the basins above mountain river stations;

[0008] The data of mountain river stations that have daily surface rainfall series, daily flow series, daily water level series, and daily water surface evaporation series are named short series, and the data of mountain river stations that have daily surface rainfall series, daily water level series, and daily water surface evaporation series but do not have daily flow series are named long series;

[0009] Based on the short series and the long series, the daily flow series of the mountain river station is interpolated and extended using a machine learning model, including constructing a sample data set, using the sample data set to train and test multiple machine learning models, comparing and analyzing to select the optimal machine learning model, and using the optimal machine learning model to interpolate and extend the daily flow series;

[0010] The rationality of the results of the daily flow series obtained by interpolation and extension using the optimal machine learning model was tested.

[0011] A further improvement of the present invention is to construct a sample data set, which specifically includes:

[0012] Based on the daily surface rainfall series of the basin above the mountainous river station, the short series and long series are divided into short series low flow data set, short series high flow data set, long series low flow data set and long series high flow data set according to whether the daily rainfall in the basin reaches the set level for several consecutive days. The daily flow series of the short series low flow data set and the short series high flow data set are statistically analyzed to verify the rationality of the data set division. The data set division method is as follows:

[0013] ;

[0014] ;

[0015] ;

[0016] ;

[0017] in, , are short series and long series respectively. , , , They are short series low flow data set, short series high flow data set, long series low flow data set and long series high flow data set. , , , Short series of moments The daily surface rainfall series, daily surface evaporation series, daily water level series and daily flow series of the basin above the mountainous river station are , , , The moments in a short series The daily surface rainfall series, daily surface evaporation series, daily water level series and daily flow series of the basin above the mountainous river station are , , The moments in the long series The daily surface rainfall series, daily surface evaporation series and daily water level series of the basin above the mountainous river station are , , The moments in the long series The daily surface rainfall series, daily surface evaporation series and daily water level series of the basin above the mountainous river station are To divide the rainfall level thresholds into low-flow dataset and high-flow dataset, The number of days with rainfall levels for dividing the low-flow dataset and the high-flow dataset.

[0018] A further improvement of the present invention is to use sample data sets to train and test multiple machine learning models, and to compare and analyze to select the optimal machine learning model, specifically including:

[0019] The short series low flow data set and the short series high flow data set are divided into a short series low flow training sample data set, a short series low flow test sample data set and a short series high flow training sample data set, and a short series high flow test sample data set. The daily surface rainfall series, daily water surface evaporation series and daily water level series in the basin above the short series medium mountain river station are used as the input of the machine learning model, and the short series medium and daily flow series are used as the output of the machine learning model. Multiple machine learning models are trained on the short series low flow training sample data set and the short series high flow training sample data set respectively; among them, the daily flow series simulated by any machine learning model in the short series low flow training sample data set and the short series high flow training sample data set is:

[0020] ;

[0021] in, The daily traffic simulated by any machine learning model on a short series of low-flow training sample data sets and a short series of high-flow training sample data sets, , They are short series low-flow training sample data sets and short series high-flow training sample data sets, respectively. , They are the fitting functions of the machine learning model after the model structure and parameter optimization and calibration in a short series of low-flow training sample data sets and a short series of high-flow training sample data sets;

[0022] The machine learning model trained by the short series low flow training sample data set and the short series high flow training sample data set is tested by using the short series low flow test sample data set and the short series high flow training sample data set, and the machine learning model with the highest accuracy is screened out by comparative analysis. The method for screening out the machine learning model with the highest accuracy includes: evaluating the accuracy of the machine learning model based on the relative deviation of the total water volume of different magnitudes of flow, and the expression is:

[0023] ;

[0024] ;

[0025] in, , They are the daily flow series in the short series and the daily flow series simulated by the machine learning model. is the number of days of the short series, is the low traffic statistics threshold, is the high traffic statistics threshold, The traffic is less than the low traffic statistics threshold The relative deviation of the total water volume at low flow rate, The traffic is greater than the high traffic statistics threshold Relative deviation of total water volume at high flow rate;

[0026] A further improvement of the present invention is to use the optimal machine learning model to interpolate and extend the daily flow series, specifically including taking the daily surface rainfall series, daily water level series, and daily water surface evaporation series within a long series of several days as the input of the optimal machine learning model, and obtaining a long series of mountain river station daily flow series. The expression of the obtained long series of mountain river station daily flow series is:

[0027] ;

[0028] in, Interpolate the extended long series of daily traffic for the selected machine learning model, For a long series of low-flow data sets, For a long series of high-traffic datasets, , The fitting functions of the machine learning model are selected after training and testing using a short series of low-flow training sample data sets, a short series of high-flow training sample data sets, a short series of low-flow test sample data sets, and a short series of high-flow test sample data sets.

[0029] A further improvement of the present invention is to perform a result rationality test on the daily flow series obtained by interpolation and extension using the optimal machine learning model, specifically including:

[0030] The frequency statistical analysis of the estimated series and the reference series was carried out respectively to obtain the mean, coefficient of variation and statistical characteristic values ​​of different frequencies of the estimated series and the reference series, and the rationality was compared and analyzed. Among them, the estimated series is a long series of daily flow series of mountain river stations obtained by interpolation and extension using the optimal machine learning model, and the reference series is the daily flow data of the long series measured at the nearby reference stations;

[0031] Statistical tests on the changing trends of the station series of mountainous rivers and the rainfall data in their basins were carried out, and the consistency of the changing trends of flow and rainfall was compared and analyzed.

[0032] The system for interpolating and extending the daily flow series of a mountain river based on machine learning of the present invention comprises:

[0033] The acquisition module is used to collect and organize data and divide the data into short series and long series;

[0034] Model screening module, used to interpolate and extend the daily flow series of mountain river stations based on short series and long series using machine learning models, including constructing sample data sets; using sample data sets to train and test multiple machine learning models, and comparing and analyzing to select the optimal machine learning model; and using the optimal machine learning model to interpolate and extend the daily flow series;

[0035] The verification module is used to verify the rationality of the daily flow series obtained by interpolation and extension using the optimal machine learning model.

[0036] A further improvement of the present invention is that the operations performed by the acquisition module include:

[0037] Collect the daily rainfall series of the rainfall observation stations in the basin above the mountainous river station, the daily water surface evaporation series of the representative stations in the basin where the mountainous river station is located, the daily water level series of the mountainous river station and unify the base surface, the daily flow series of the mountainous river station, and statistically organize the daily rainfall series of the rainfall observation stations in the basin above the mountainous river station to obtain the daily surface rainfall series of the basin above the mountainous river station, among which the mountainous river station and the upstream catchment area are named as the basin above the mountainous river station;

[0038] The data of mountainous river stations that have daily surface rainfall series, daily flow series, daily water level series, and daily water surface evaporation series are named short series, and the data of mountainous river stations that have daily surface rainfall series, daily water level series, daily water surface evaporation series but do not have daily flow series are named long series.

[0039] A further improvement of the present invention is that the operations performed by the model screening module include:

[0040] Based on the daily surface rainfall series of the basin above the mountainous river station, the short series and long series are divided into short series low flow data set, short series high flow data set, long series low flow data set and long series high flow data set according to whether the daily rainfall in the basin reaches the set level for several consecutive days. The daily flow series of the short series low flow data set and the short series high flow data set are statistically analyzed to verify the rationality of the data set division. The data set division method is as follows:

[0041] ;

[0042] ;

[0043] ;

[0044] ;

[0045] in, , are short series and long series respectively. , , , They are short series low flow data set, short series high flow data set, long series low flow data set and long series high flow data set. , , , Short series of moments The daily surface rainfall series, daily surface evaporation series, daily water level series and daily flow series of the basin above the mountainous river station are , , , Short series of moments The daily surface rainfall series, daily surface evaporation series, daily water level series and daily flow series of the basin above the mountainous river station are , , The moments in the long series The daily surface rainfall series, daily surface evaporation series and daily water level series of the basin above the mountainous river station are , , The moments in the long series The daily surface rainfall series, daily surface evaporation series and daily water level series of the basin above the mountainous river station are To divide the rainfall level thresholds into low-flow dataset and high-flow dataset, The number of days with rainfall levels for dividing low-flow datasets and high-flow datasets;

[0046] The short series low flow data set and the short series high flow data set are divided into a short series low flow training sample data set, a short series low flow test sample data set and a short series high flow training sample data set, and a short series high flow test sample data set. The daily surface rainfall series, daily water surface evaporation series and daily water level series in the basin above the short series medium mountain river station are used as the input of the machine learning model, and the short series medium and daily flow series are used as the output of the machine learning model. Multiple machine learning models are trained on the short series low flow training sample data set and the short series high flow training sample data set respectively; among them, the daily flow series simulated by any machine learning model in the short series low flow training sample data set and the short series high flow training sample data set is:

[0047] ;

[0048] in, The daily traffic simulated by any machine learning model on a short series of low-flow training sample data sets and a short series of high-flow training sample data sets, , They are short series low-flow training sample data sets and short series high-flow training sample data sets, respectively. , They are the fitting functions of the machine learning model after the model structure and parameter optimization and calibration in a short series of low-flow training sample data sets and a short series of high-flow training sample data sets;

[0049] The machine learning model trained by the short series low flow training sample data set and the short series high flow training sample data set is tested by using the short series low flow test sample data set and the short series high flow training sample data set, and the machine learning model with the highest accuracy is screened out by comparative analysis. The method for screening out the machine learning model with the highest accuracy includes: evaluating the accuracy of the machine learning model based on the relative deviation of the total water volume of different magnitudes of flow, and the expression is:

[0050] ;

[0051] ;

[0052] in, , They are the daily flow series in the short series and the daily flow series simulated by the machine learning model. is the number of days of the short series, is the low traffic statistics threshold, is the high traffic statistics threshold, The traffic is less than the low traffic statistics threshold The relative deviation of the total water volume at low flow rate, The traffic is greater than the high traffic statistics threshold Relative deviation of total water volume at high flow rate;

[0053] The optimal machine learning model is used to interpolate and extend the daily flow series. Specifically, the daily surface rainfall series, daily water level series, and daily water surface evaporation series within a long series of days are used as the input of the optimal machine learning model to obtain a long series of daily flow series of mountain river stations. The expression of the long series of daily flow series of mountain river stations is:

[0054] ;

[0055] in, Interpolate the extended long series of daily traffic for the selected machine learning model, For a long series of low-flow data sets, For a long series of high-traffic datasets, , The fitting functions of the machine learning model are selected after training and testing using a short series of low-flow training sample data sets, a short series of high-flow training sample data sets, a short series of low-flow test sample data sets, and a short series of high-flow test sample data sets.

[0056] A further improvement of the present invention is that the operations performed by the inspection module include:

[0057] The frequency statistical analysis of the estimated series and the reference series was carried out respectively to obtain the mean, coefficient of variation and statistical characteristic values ​​of different frequencies of the estimated series and the reference series, and the rationality was compared and analyzed. Among them, the estimated series is a long series of daily flow series of mountain river stations obtained by interpolation and extension using the optimal machine learning model, and the reference series is the daily flow data of the long series measured at the nearby reference stations;

[0058] Statistical tests on the changing trends of the station series of mountainous rivers and the rainfall data in their basins were carried out, and the consistency of the changing trends of flow and rainfall was compared and analyzed.

[0059] The beneficial effects of the present invention are as follows: the present invention utilizes a variety of machine learning models to interpolate and extend the daily flow series of mountainous rivers, and selects the optimal machine learning model to interpolate and extend the daily flow series of mountainous river stations, taking into account the changes in the water level-flow relationship curve of mountainous rivers over a long period of time, and significantly improving the calculation accuracy of long series daily flow data of mountainous river stations. The calculated long series flow data can meet the high-precision requirements of hydrological data compilation. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a method flow chart of an embodiment of the present invention;

[0061] Figure 2 It is a comparison diagram of the measured water level-flow relationship and the multi-year water level-flow relationship curve of mountain river station A from 2020 to 2022 in the embodiment of the present invention;

[0062] Figure 3 It is a comparison chart of the short series daily flow rate during the same period from 1983 to 1988 simulated by the three machine learning models in the embodiment of the present invention and the measured flow rate;

[0063] Figure 4 It is a comparison chart of the short series daily flow and the measured flow in the same period of 2020-2022 simulated by the three machine learning models in the embodiment of the present invention;

[0064] Figure 5 It is a long series daily flow process diagram of a mountain river station A with interpolation and extension in an embodiment of the present invention;

[0065] Figure 6 It is a relationship diagram between the annual average daily flow of a mountain river station from 1983 to 2022 and the annual total rainfall in the basin after interpolation and extension in an embodiment of the present invention. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0067] like Figure 1 As shown, the method for interpolating and extending the daily flow series of a mountain river based on machine learning in this embodiment includes the following steps:

[0068] Step 1, collect and organize data (meteorological and hydrological series);

[0069] The mountain river station and its upstream catchment area are named as the basin above the mountain river station, and the daily rainfall series of the rainfall observation stations in the basin above the mountain river station are collected to obtain the daily surface rainfall series in the basin above the mountain river station.

[0070] Collect the daily water surface evaporation series of representative stations in the basin where the mountain river station is located;

[0071] Collect daily water level series from mountain river stations and unify the base surface;

[0072] Collect daily flow series from mountain river stations;

[0073] The data of mountainous river stations that have daily surface rainfall series, daily flow series, daily water level series, and daily water surface evaporation series are named short series, and the data of mountainous river stations that have daily surface rainfall series, daily water level series, daily water surface evaporation series but do not have daily flow series are named long series.

[0074] Step 2: Use machine learning models to interpolate and extend the daily flow series of mountain river stations;

[0075] It includes the selection of machine learning model input factors, sample data set construction, training and testing of machine learning models, comparative analysis to select the optimal machine learning model, and using the optimal machine learning model to interpolate and extend the daily flow series, including:

[0076] Step 2.1, build a sample data set for the machine learning model;

[0077] Based on the daily surface rainfall series of the basin above the mountainous river station, the short series and long series are divided into short series low flow data set, short series high flow data set, long series low flow data set and long series high flow data set according to whether the daily rainfall in the basin reaches the set level for several consecutive days. The daily flow series of the short series low flow data set and the short series high flow data set are statistically analyzed to verify the rationality of the data set division. The data set division method is as follows:

[0078] ;

[0079] ;

[0080] ;

[0081] ;

[0082] in, , are short series and long series respectively. , , , They are short series low flow data set, short series high flow data set, long series low flow data set and long series high flow data set. , , , The moments in a short series The daily surface rainfall series, daily surface evaporation series, daily water level series and daily flow series of the basin above the mountainous river station are , , , The moments in a short series The daily surface rainfall series, daily surface evaporation series, daily water level series and daily flow series of the basin above the mountainous river station are , , The moments in the long series The daily surface rainfall series, daily surface evaporation series and daily water level series of the basin above the mountainous river station are , , The moments in the long series The daily surface rainfall series, daily surface evaporation series and daily water level series of the basin above the mountainous river station are To divide the rainfall level thresholds into low-flow dataset and high-flow dataset, The number of days with rainfall levels for dividing low-flow datasets and high-flow datasets;

[0083] The indicators for verifying the rationality of data set division include: minimum daily flow, maximum daily flow, average daily flow, daily flow at different percentiles (such as the 25th percentile, 50th percentile, 75th percentile, etc.), daily flow at different magnitudes (such as less than 10m 3 / s, greater than 100m 3 / s, etc.) proportion, etc.

[0084] Step 2.2, divide the data set into a training sample data set and a test sample data set, and train the machine learning model based on the training sample data set, specifically including:

[0085] The short series low flow data set and the short series high flow data set are divided into a short series low flow training sample data set, a short series low flow test sample data set and a short series high flow training sample data set, and a short series high flow test sample data set. The daily surface rainfall series, daily water surface evaporation series and daily water level series in the basin above the short series medium mountain river station are used as the input of the machine learning model, and the short series medium and daily flow series are used as the output of the machine learning model. Multiple machine learning models are trained on the short series low flow training sample data set and the short series high flow training sample data set respectively.

[0086] The daily traffic series simulated by any machine learning model in the short series of low-flow training sample data sets and the short series of high-flow training sample data sets are:

[0087] ;

[0088] in, The daily traffic simulated by any machine learning model on a short series of low-flow training sample data sets and a short series of high-flow training sample data sets, , They are short series low-flow training sample data sets and short series high-flow training sample data sets, respectively. , They are the fitting functions of the machine learning model after model structure and parameter optimization and calibration on a short series of low-flow training sample data sets and a short series of high-flow training sample data sets.

[0089] Step 2.3, machine learning model accuracy evaluation and comparison;

[0090] Select at least 2 different machine learning models (such as support vector machines, gradient boosting regression trees, long short-term memory networks, etc.) for training respectively, and use test sample data sets to test the trained machine learning models to select the machine learning model with the highest accuracy.

[0091] In addition to the evaluation indicators such as Nash efficiency coefficient, average absolute deviation, average relative deviation, and total water volume relative deviation, in view of the large range of flow variation in mountainous rivers and the difficulty of a single water level-flow relationship curve to take into account the accuracy of high and low flows of different magnitudes, the total water volume relative deviation of flows of different magnitudes (such as low flow and high flow) is additionally considered to evaluate the model accuracy. The expression is:

[0092] ;

[0093] ;

[0094] in, , They are the daily flow series in the short series and the daily flow series simulated by the machine learning model. is the number of days of the short series, is the low traffic statistics threshold, is the high traffic statistics threshold, The traffic is less than the low traffic statistics threshold The relative deviation of the total water volume at low flow rate, The traffic is greater than the high traffic statistics threshold Relative deviation of total water volume at high flow rate;

[0095] Step 2.4, using the optimal machine learning model to interpolate and extend the daily flow series, specifically includes:

[0096] Taking the daily surface rainfall series, daily water level series, and daily water surface evaporation series in a long series of several days as the input of the optimal machine learning model, a long series of daily flow series of mountain river stations is obtained. The expression of the long series of daily flow series of mountain river stations is:

[0097] ;

[0098] in, Interpolate the extended long series of daily traffic for the selected machine learning model, For a long series of low-flow data sets, For a long series of high-traffic datasets, , The fitting functions of the machine learning model are selected after training and testing using a short series of low-flow training sample data sets, a short series of high-flow training sample data sets, a short series of low-flow test sample data sets, and a short series of high-flow test sample data sets.

[0099] Step 3: Check the rationality of the daily flow series obtained by interpolation and extension using the optimal machine learning model, including:

[0100] Step 3.1, conduct frequency statistical analysis on the estimated series and the reference series respectively, obtain the mean, coefficient of variation and statistical characteristic values ​​of different frequencies of the estimated series and the reference series, and compare and analyze the rationality. Among them, the estimated series is a long series of daily flow series of mountain river stations obtained by interpolation and extension using the optimal machine learning model, and the reference series is the daily flow data of the long series measured at the nearby reference station;

[0101] Step 3.2, conduct statistical tests on the changing trends of the mountain river station series and the rainfall data in the basin where they are located, and compare and analyze the consistency of the flow and rainfall changing trends.

[0102] This embodiment uses the daily flow interpolation extension of a mountain river station A to further illustrate the effect of this embodiment. Mountain river station A has measured water level and flow data from 1983 to 1988 and from 2020 to 2022. The station only observed water level from 1989 to 2019, and the flow data was interrupted. From 1983 to 2022, there were multiple rainfall stations in the basin where the station was located with continuous daily rainfall observations.

[0103] Based on the short series of mountain river station A from 1983 to 1988, the traditional water level flow relationship method was used to calculate the daily flow series of the same period according to the measured water level from 2020 to 2022, and the accuracy of the traditional water level flow relationship method was evaluated by comparing the measured flow data. 3 / s, greater than 100m 3 / s and greater than 500m 3 / s total water volume relative error of different magnitudes of flow. When the water level-flow relationship curve from many years ago is used to deduce the flow series from 2020 to 2022, the error is less than 10m 3 / s flow rate and above 500m 3 The flow rate accuracy of / s is low, and the results are as follows Figure 2 As shown in Table 1, the reason is that the changes in erosion and siltation in the river channel between 1989 and 2019 caused the water level-flow relationship to change.

[0104] Table 1 Accuracy evaluation of daily flow estimated by water level-flow relationship method from 2020 to 2022

[0105]

[0106] According to the calculated daily surface rainfall in the basin above the mountainous river station A from 1983 to 1988 and 2020 to 2022, the data with daily surface rainfall in the basin that did not reach 25mm for 5 consecutive days were divided into a short series low-flow data set, and the data with surface rainfall in the basin reaching 25mm on any day within 5 consecutive days were divided into a short series high-flow data set. According to statistical analysis, in the short series low-flow data set, the daily flow of mountainous river station A is significantly lower than the daily flow in the short series high-flow data set, and the daily flow in the short series low-flow data set is less than 100m 3 / s, while 28.04% of the daily flow in the short series high flow data set is greater than 100m 3 / s, as shown in Table 2, the data set is divided reasonably.

[0107] Table 2 Comparison of daily flow series statistics of short series low flow data set and short series high flow data set

[0108]

[0109] Three different machine learning models, namely support vector machine model, gradient boosting regression tree model and long short-term memory network model, were selected respectively. The data from 1983 to 1988 were selected as the training set, and the data from 2020 to 2022 were selected as the test set. The daily rainfall, water surface evaporation and water level in the previous 5 days were used as the input of the machine learning model. The daily flow process was simulated in the low-flow data set and the high-flow data set respectively. The results are as follows Figure 3 , Figure 4 As shown. From the simulation results, the short series of daily flow series simulated by support vector machine and long short-term memory network will have negative flow, which is inconsistent with common sense. It is believed that the two machine learning models of support vector machine and long short-term memory network are not suitable for daily flow interpolation extension in this case. The gradient boosting regression tree model is a more applicable machine learning model in this embodiment. The accuracy evaluation of the short-term series of daily flow series derived by the gradient boosting regression tree model is shown in Table 3.

[0110] Table 3 Accuracy evaluation of short series of daily flow using gradient boosting regression tree model

[0111]

[0112] By selecting the optimal machine learning model (gradient boosting regression tree model) after comparison, based on the daily surface rainfall, water surface evaporation, and water level in the basin, the long series of daily flow of the mountain river station A that is finally interpolated and extended can be obtained, such as Figure 5 shown.

[0113] There is hydrological station B downstream of mountain river station A. The rainfall and underlying surface characteristics of the two stations are similar. Hydrological station B has a long series of continuous measured daily flow data from 1983 to 2022. Therefore, hydrological station B is selected as the reference station for mountain river station A. Based on the long series of daily calculation results of mountain river station A and the long series of daily flow measured data of reference station B, frequency analysis is performed, and the results are shown in Table 4.

[0114] Table 4 Results of long series flow frequency analysis

[0115]

[0116] After comparison, the ratio of the mean long series flow rate and the average flow rate at each frequency of mountain river station A to the reference station was relatively stable between 29.74% and 30.77%, and the coefficients (Cv) of the two long series flow series were 0.24 and 0.23, respectively, which was consistent with the basin similarity judgment.

[0117] The annual average daily flow can be calculated by using the interpolated and extended mountain river station daily flow series. After interpolation and extension, the annual average daily flow of mountain river station A from 1983 to 2022 is positively correlated with the annual total rainfall in the basin (p<0.05), and the flow and rainfall are in good agreement. Figure 6 shown.

[0118] In summary, by adopting the method provided by the present invention, when the flow series of a river station in a mountainous area is relatively short, the interpolation and extension accuracy of a long series of flow data can be significantly improved, and the calculated flow series results are reasonable.

[0119] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have the same meaning as in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless defined similarly here.

[0120] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for interpolating and extending the daily flow series of mountainous rivers based on machine learning, characterized by: The following operations are included: Collect and organize data: including naming mountain river stations and upstream catchment areas as the basins above mountain river stations, collecting daily rainfall series of rainfall observation stations in the basins above mountain river stations, daily water surface evaporation series of representative stations in the basins where mountain river stations are located, daily water level series of mountain river stations and unifying the base surface, daily flow series of mountain river stations, and statistically organizing the daily rainfall series of rainfall observation stations in the basins above mountain river stations to obtain the daily surface rainfall series of the basins above mountain river stations; The data of mountain river stations that have daily surface rainfall series, daily flow series, daily water level series, and daily water surface evaporation series are named short series, and the data of mountain river stations that have daily surface rainfall series, daily water level series, and daily water surface evaporation series but do not have daily flow series are named long series; Based on the short series and the long series, the machine learning model is used to interpolate and extend the daily flow series of mountain river stations, including constructing a sample data set based on the short series, using the sample data set to train and test multiple machine learning models, and comparing and analyzing to select the optimal machine learning model. The optimal machine learning model is then used to interpolate and extend the daily flow series based on the long series. The rationality of the results of the daily flow series obtained by interpolation and extension using the optimal machine learning model is tested; The method of using the optimal machine learning model to interpolate and extend the daily flow series specifically includes taking the daily surface rainfall series, daily water level series, and daily water surface evaporation series within a long series of several days as the input of the optimal machine learning model to obtain a long series of daily flow series of mountain river stations. The expression of the obtained long series of daily flow series of mountain river stations is: ； in, Interpolate the extended long series of daily traffic for the selected machine learning model, For a long series of low-flow data sets, For a long series of high-traffic datasets, , The fitting functions of the machine learning model are selected after training and testing using a short series of low-flow training sample data sets, a short series of high-flow training sample data sets, a short series of low-flow test sample data sets, and a short series of high-flow test sample data sets.

2. The method for interpolating and extending the daily flow series of a mountain river based on machine learning according to claim 1 is characterized in that: The constructing of the sample data set specifically includes: Based on the daily surface rainfall series of the basin above the mountainous river station, the short series and long series are divided into short series low flow data set, short series high flow data set, long series low flow data set and long series high flow data set according to whether the daily rainfall in the basin reaches the set level for several consecutive days. The daily flow series of the short series low flow data set and the short series high flow data set are statistically analyzed to verify the rationality of the data set division. The data set division method is as follows: ； ； ； ； in, , are short series and long series respectively. , , , They are short series low flow data set, short series high flow data set, long series low flow data set and long series high flow data set. , , , Short series of moments The daily surface rainfall series, daily surface evaporation series, daily water level series and daily flow series of the basin above the mountainous river station are , , , The moments in a short series The daily surface rainfall series, daily surface evaporation series, daily water level series and daily flow series of the basin above the mountainous river station are , , The moments in the long series The daily surface rainfall series, daily surface evaporation series and daily water level series of the basin above the mountainous river station are , , The moments in the long series The daily surface rainfall series, daily surface evaporation series and daily water level series of the basin above the mountainous river station are To divide the rainfall level thresholds into low-flow dataset and high-flow dataset, The number of days with rainfall levels for dividing the low-flow dataset and the high-flow dataset.

3. The method for interpolating and extending the daily flow series of a mountain river based on machine learning according to claim 2 is characterized in that: The method of using sample data sets to train and test multiple machine learning models and compare and analyze to select the optimal machine learning model specifically includes: The short series low flow data set and the short series high flow data set are divided into a short series low flow training sample data set, a short series low flow test sample data set and a short series high flow training sample data set, and a short series high flow test sample data set. The daily surface rainfall series, daily water surface evaporation series and daily water level series in the basin above the short series medium mountain river station are used as the input of the machine learning model, and the short series medium and daily flow series are used as the output of the machine learning model. Multiple machine learning models are trained on the short series low flow training sample data set and the short series high flow training sample data set respectively; among them, the daily flow series simulated by any machine learning model in the short series low flow training sample data set and the short series high flow training sample data set is: ； in, The daily traffic simulated by any machine learning model on a short series of low-flow training sample data sets and a short series of high-flow training sample data sets, , They are short series low-flow training sample data sets and short series high-flow training sample data sets, respectively. , They are the fitting functions of the machine learning model after the model structure and parameter optimization and calibration in a short series of low-flow training sample data sets and a short series of high-flow training sample data sets; The short series of low-flow inspection sample data sets and the short series of high-flow inspection sample data sets are used to test the machine learning models trained by the short series of low-flow training sample data sets and the short series of high-flow training sample data sets, and the machine learning model with the highest accuracy is selected through comparative analysis.

4. The method for interpolating and extending the daily flow series of a mountain river based on machine learning according to claim 3 is characterized in that: The method of selecting the machine learning model with the highest accuracy includes: evaluating the accuracy of the machine learning model based on the relative deviation of the total water volume of different magnitudes of flow, and the expression is: ； ； in, , They are the daily flow series in the short series and the daily flow series simulated by the machine learning model. is the number of days of the short series, is the low traffic statistics threshold, is the high traffic statistics threshold, The traffic is less than the low traffic statistics threshold The relative deviation of the total water volume at low flow rate, The traffic is greater than the high traffic statistics threshold Relative deviation of total water volume at high flow rate.

5. The method for interpolating and extending the daily flow series of a mountain river based on machine learning according to claim 1 is characterized in that: The rationality test of the results of the daily flow series obtained by interpolation and extension using the optimal machine learning model specifically includes: The frequency statistical analysis of the estimated series and the reference series was carried out respectively to obtain the mean, coefficient of variation and statistical characteristic values ​​of different frequencies of the estimated series and the reference series, and the rationality was compared and analyzed. Among them, the estimated series is a long series of daily flow series of mountain river stations obtained by interpolation and extension using the optimal machine learning model, and the reference series is the daily flow data of the long series measured at the nearby reference stations; Statistical tests on the changing trends of the station series of mountainous rivers and the rainfall data in their basins were carried out, and the consistency of the changing trends of flow and rainfall was compared and analyzed.

6. A system for interpolating and extending the daily flow of mountainous rivers based on machine learning, characterized by: The system comprises: The acquisition module is used to collect and organize data and divide the data into short series and long series; Model screening module, used to interpolate and extend the daily flow series of mountain river stations based on short series and long series using machine learning models, including building a sample data set based on the short series, using the sample data set to train and test multiple machine learning models, and comparing and analyzing to select the optimal machine learning model, and using the optimal machine learning model based on the long series to interpolate and extend the daily flow series; The test module is used to test the rationality of the daily flow series obtained by interpolation and extension using the optimal machine learning model; Collect the daily rainfall series of the rainfall observation stations in the basin above the mountainous river station, the daily water surface evaporation series of the representative stations in the basin where the mountainous river station is located, the daily water level series of the mountainous river station and unify the base surface, the daily flow series of the mountainous river station, and statistically organize the daily rainfall series of the rainfall observation stations in the basin above the mountainous river station to obtain the daily surface rainfall series of the basin above the mountainous river station, among which the mountainous river station and the upstream catchment area are named as the basin above the mountainous river station; The data of mountain river stations that have daily surface rainfall series, daily flow series, daily water level series, and daily water surface evaporation series are named short series, and the data of mountain river stations that have daily surface rainfall series, daily water level series, and daily water surface evaporation series but do not have daily flow series are named long series; The operations performed by the model screening module include: Based on the daily surface rainfall series of the basin above the mountainous river station, the short series and long series are divided into short series low flow data set, short series high flow data set, long series low flow data set and long series high flow data set according to whether the daily rainfall in the basin reaches the set level for several consecutive days. The daily flow series of the short series low flow data set and the short series high flow data set are statistically analyzed to verify the rationality of the data set division. The data set division method is as follows: ； ； ； ； in, , are short series and long series respectively. , , , They are short series low flow data set, short series high flow data set, long series low flow data set and long series high flow data set. , , , The moments in a short series The daily surface rainfall series, daily surface evaporation series, daily water level series and daily flow series of the basin above the mountainous river station are , , , The moments in a short series The daily surface rainfall series, daily surface evaporation series, daily water level series and daily flow series of the basin above the mountainous river station are , , The moments in the long series The daily surface rainfall series, daily surface evaporation series and daily water level series of the basin above the mountainous river station are , , The moments in the long series The daily surface rainfall series, daily surface evaporation series and daily water level series of the basin above the mountainous river station are To divide the rainfall level thresholds into low-flow dataset and high-flow dataset, The number of days with rainfall levels for dividing low-flow datasets and high-flow datasets; The short series low flow data set and the short series high flow data set are divided into a short series low flow training sample data set, a short series low flow test sample data set and a short series high flow training sample data set, and a short series high flow test sample data set. The daily surface rainfall series, daily water surface evaporation series and daily water level series in the basin above the short series medium mountain river station are used as the input of the machine learning model, and the short series medium and daily flow series are used as the output of the machine learning model. Multiple machine learning models are trained on the short series low flow training sample data set and the short series high flow training sample data set respectively; among them, the daily flow series simulated by any machine learning model in the short series low flow training sample data set and the short series high flow training sample data set is: ； in, The daily traffic simulated by any machine learning model on a short series of low-flow training sample data sets and a short series of high-flow training sample data sets, , They are short series low-flow training sample data sets and short series high-flow training sample data sets, respectively. , They are the fitting functions of the machine learning model after the model structure and parameter optimization and calibration in a short series of low-flow training sample data sets and a short series of high-flow training sample data sets; The machine learning model trained by the short series low flow training sample data set and the short series high flow training sample data set is tested by using the short series low flow test sample data set and the short series high flow training sample data set, and the machine learning model with the highest accuracy is screened out by comparative analysis. The method for screening out the machine learning model with the highest accuracy includes: evaluating the accuracy of the machine learning model based on the relative deviation of the total water volume of different magnitudes of flow, and the expression is: ； ； in, , They are the daily flow series in the short series and the daily flow series simulated by the machine learning model. is the number of days of the short series, is the low traffic statistics threshold, is the high traffic statistics threshold, The traffic is less than the low traffic statistics threshold The relative deviation of the total water volume at low flow rate, The traffic is greater than the high traffic statistics threshold Relative deviation of total water volume at high flow rate; The optimal machine learning model is used to interpolate and extend the daily flow series. Specifically, the daily surface rainfall series, daily water level series, and daily water surface evaporation series within a long series of days are used as the input of the optimal machine learning model to obtain a long series of daily flow series of mountain river stations. The expression of the long series of daily flow series of mountain river stations is: ； in, Interpolate the extended long series of daily traffic for the selected machine learning model, For a long series of low-flow data sets, For a long series of high-traffic datasets, , The fitting functions of the machine learning model are selected after training and testing using a short series of low-flow training sample data sets, a short series of high-flow training sample data sets, a short series of low-flow test sample data sets, and a short series of high-flow test sample data sets.

7. The system for interpolating and extending the daily flow series of mountainous rivers based on machine learning according to claim 6 is characterized by: The operations performed by the inspection module include: The frequency statistical analysis of the estimated series and the reference series was carried out respectively to obtain the mean, coefficient of variation and statistical characteristic values ​​of different frequencies of the estimated series and the reference series, and the rationality was compared and analyzed. Among them, the estimated series is a long series of daily flow series of mountain river stations obtained by interpolation and extension using the optimal machine learning model, and the reference series is the daily flow data of the long series measured at the nearby reference stations; Statistical tests on the changing trends of the station series of mountainous rivers and the rainfall data in their basins were carried out, and the consistency of the changing trends of flow and rainfall was compared and analyzed.

Images