CN115293037A - Hydrodynamic modeling method of river-lake composite system based on machine learning - Google Patents

Abstract

The embodiment of the application provides a hydrodynamic modeling method of a river and lake composite system based on machine learning, wherein the boundary condition of a one-dimensional hydrodynamic model provides input for the machine learning; the output of the machine learning provides complex boundary conditions and downstream boundary conditions at river and lake junctions for the one-dimensional hydrodynamic model. The method improves the operating efficiency of hydrodynamic simulation of the river and lake composite system, can quickly calculate the water level and the flow of the important section position of the river and lake composite system, simultaneously reduces the requirement on terrain data, does not need to input complex lake terrain elevation survey data, and can effectively improve the water level simulation precision of the section near the downstream boundary.

Description

Hydrodynamic modeling method of river-lake composite system based on machine learning

technical field

This application relates to the field of hydrodynamic modeling, and can also be used in the field of machine learning, and specifically relates to a method for hydrodynamic modeling of river-lake composite systems based on machine learning.

Background technique

Machine learning is an important tool for current research on watershed hydrological simulation. When using machine learning for hydrological modeling, the rainfall or upstream flow in the watershed is generally used as input, and the downstream water level or flow is used as output. The parameters of the machine learning model are trained according to the measured data, and the model after parameter training can simulate and predict the water level and flow of rivers or lakes in the watershed. Machine learning is a hydrological and hydrodynamic simulation method with high speed and high precision. However, it can only simulate a single site sequentially, and it is difficult to model the hydrological and hydrodynamic elements of each section of a large-scale river section.

One-dimensional, two-dimensional or three-dimensional physical hydrodynamic models can be established for the hydrodynamic numerical simulation of the composite system of rivers and lakes. Choosing different model dimensions is essentially a trade-off between solution accuracy and computational cost, and is also limited by data availability. For river simulations with long stretches and long periods of time, the use of a one-dimensional model along the flow direction of the river can help greatly reduce the amount of calculation. Due to the relatively large length and width in the horizontal direction, the lake area cannot be generalized by a one-dimensional model, and a two-dimensional or three-dimensional model averaged along the water depth direction is often used. However, for large lakes, it is difficult to obtain the high-resolution bathymetry data or digital elevation data required to build a 2D or 3D model. At the same time, there are also problems such as low model operation efficiency, repeated iterations of water level and flow at the coupling boundary, and easy calculation instability. question. On the premise of ensuring the calculation accuracy, it is a major challenge to simulate the hydrological and hydrodynamic process of the river-lake complex system by rationally generalizing the lake model to reduce the calculation cost. Furthermore, the downstream boundary conditions of river-lake complex systems are unknown in a changing environment without hydraulic engineering regulation. The commonly used water level-discharge relationship curve method is often not stable due to the interference of various factors such as terrain and vegetation, and it is easy to affect the accuracy of hydrodynamic simulation of river-lake complex systems.

Machine learning has the advantage of exploring implicit relationships in nonlinear systems. Coupling machine learning and hydrodynamic methods may provide new ideas for solving complex problems in hydrodynamic modeling of river-lake complex systems. Practical applications and in-depth research have shown that there are some shortcomings in the current modeling methods of river-lake complex systems related to machine learning and hydrodynamics: (1) The calculation sites of pure machine learning methods are limited, and it is difficult to model each unit of the entire study area at the same time; (2) The pure hydrodynamic method takes a long time to calculate and has high requirements for terrain data, especially for lakes; (3) The hydrodynamic model-driven machine learning method for alternative modeling is limited, and parameters need to be updated frequently according to terrain changes, etc. , the scalability is not strong, and the advantages of machine learning and hydrodynamic methods are not fully combined.

Contents of the invention

Aiming at the problems in the prior art, the present application provides a machine learning-based hydrodynamic modeling method of the river-lake composite system, which can effectively improve the operating efficiency of the river-lake composite system hydrodynamic simulation.

In order to solve at least one of the above problems, the application provides the following technical solutions:

This application provides a machine learning-based hydrodynamic modeling method for river-lake composite systems, including:

Collect flow data at the upstream boundary, lateral boundary, and junctions of rivers and lakes in the study area, as well as water level data at downstream boundary stations;

According to the flow of the upstream boundary site and the flow of the lateral tributary site that may affect the water exchange between the river and the lake, the machine learning method is used to train and predict the flow of the river and lake intersection;

Using machine learning methods to train and predict water levels at downstream boundaries based on flows at upstream and lateral boundary stations, as well as flows exchanged between rivers and lakes;

Discretize the researched river section according to the topographic data of the river channel, divide a section at intervals on the river channel, and sort out the bank start distance and elevation data on the section;

The initial flow of each section is assigned as the flow of the first day of the upstream boundary, and the initial water level is calculated by linear interpolation according to the water level of the first day of the downstream boundary and the slope of the river;

Discrete Saint-Venant's equations, and input the initial and boundary conditions of each channel section for one-dimensional hydrodynamic calculation of the channel.

Further, the use of machine learning methods to train and predict the flow at the intersection of rivers and lakes includes:

The specific calculation paradigm adopted is:

是第j个湖泊与河流交换的水量；Q_in是上游来水量；Q_B是河流旁侧支流流量，且第j个湖泊与河流交换的水量一共受到i条河流支流的影响；

是第j个湖泊自身承接的支流流量，且一共有k条湖泊支流；f(g)代表预设机器学习方法。in,

is the amount of water exchanged between the j-th lake and the river; Q _in is the inflow of upstream water; Q _B is the flow rate of the side tributaries of the river, and the amount of water exchanged between the j-th lake and the river is affected by the tributaries of the i river;

is the tributary flow undertaken by the jth lake itself, and there are k lake tributaries in total; f(g) represents the default machine learning method.

Further, said adopting machine learning method to train and predict the water level of the downstream boundary includes:

The specific calculation paradigm of the downstream boundary water level adopted is as follows:

Among them, Z is the water level of the downstream boundary; Q _in is the inflow of upstream water; Q _B is the flow of tributaries beside the river; Q _L is the water exchange volume between lakes and rivers; m and n are the number of tributaries connected to the river and the number of lakes connected to the river ; f(g) represents the preset machine learning method.

It can be seen from the above technical solutions that this application provides a machine learning-based hydrodynamic modeling method for river-lake composite systems, which can improve the operation efficiency of hydrodynamic simulation of river-lake composite systems, and quickly calculate the water level and At the same time, it reduces the requirements for topographic data, does not need to input complex lake topographic elevation survey data, and can effectively improve the accuracy of water level simulation of sections near the downstream boundary.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are For some embodiments of the present application, those of ordinary skill in the art can also obtain other drawings based on these drawings without creative effort.

Fig. 1 is the schematic flow chart of the hydrodynamic modeling method of river-lake composite system based on machine learning in the embodiment of the present application;

Fig. 2 is the measured and simulated water level, flow comparison schematic diagram of main site in a specific embodiment of the present application;

Fig. 3 is a schematic diagram of comparison of measured and simulated downstream boundary water levels in a specific embodiment of the present application.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments It is a part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of this application.

The acquisition, storage, use, and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations.

In order to effectively improve the operating efficiency of the hydrodynamic simulation of the river-lake composite system, the present application provides an embodiment of a machine learning-based hydrodynamic modeling method for the river-lake composite system, see Figure 1, the machine-learning-based river-lake The hydrodynamic modeling method of composite system specifically includes the following contents:

Step 1, Hydrological data collection: collect the flow data of the upstream boundary, lateral boundary and the intersection of rivers and lakes in the study area, and the water level data of the downstream boundary stations.

Specifically, according to the "Hydrological Yearbook" and other data sources, the upstream boundary flow (Yichang station flow), river tributary flow (Hanjiang Xiantao station flow ), lake tributary flow (Shimen Station, Taojiang Station, Taoyuan Station and Xiangtan Station flow in Dongting Lake, Waizhou Station, Lijiadu Station, Meigang Station, Hushan Station, Qiujin Station, Wanjiabu Station and Ferry Station in Poyang Lake Fengkeng station flow), river and lake confluence point flow (Dongting Lake’s Chenglingji station flow, Poyang Lake’s Hukou station flow) and downstream boundary water level (Datong station water level), all flow or hydrological data are daily-scale hydrological data .

Step 2. Calculation of water exchange between rivers and lakes: According to the flow of upstream boundary sites and the flow of lateral tributary sites that may affect the water exchange between rivers and lakes, machine learning methods are used to train and predict the flow at the intersection of rivers and lakes. The specific calculation paradigm is as follows:

是第j个湖泊与河流交换的水量；Q_in是上游来水量,即宜昌站流量；Q_B是河流旁侧支流流量，且第j个湖泊与河流交换的水量一共受到i条河流支流的影响；

是第j个湖泊自身承接的支流流量，且一共有k条湖泊支流；RNN(g)代表一种称为循环神经网络的机器学习方法。in,

is the amount of water exchanged between the jth lake and the river; Q _in is the amount of upstream water, that is, the flow of Yichang station; Q _B is the flow rate of the tributaries beside the river, and the amount of water exchanged between the jth lake and the river is affected by the tributaries of the i river ;

is the tributary flow undertaken by the jth lake itself, and there are k lake tributaries in total; RNN(g) represents a machine learning method called a recurrent neural network.

Taking the flow at the confluence of rivers and lakes from 2007 to 2013 as the output of machine learning, the upstream boundary flow (Yichang Station flow), the tributary flow of the corresponding lake (Dongting Lake’s Shimen Station, Taojiang Station, Taoyuan Station and Xiangtan Station flow, Poyang Lake The flow of Waizhou Station, Lijiadu Station, Meigang Station, Hushan Station, Qiujin Station, Wanjiabu Station and Dufengkeng Station) and the flow of upstream river tributaries that can affect the confluence of the river and lake (for example, Poyang Lake and The Hukou Station at the confluence of the main stream of the Yangtze River is affected by the flow of the Han River) as the input of machine learning to train the parameters of the machine learning model. Use the RNN model with trained parameters to predict the flow at the confluence of rivers and lakes from 2014 to 2016 (the flow at Chenglingji Station in Dongting Lake, the flow at Hukou Station in Poyang Lake).

Step 3, Downstream Boundary Water Level Calculation: According to the flow of upstream and lateral boundary stations, and the flow exchanged between rivers and lakes, machine learning methods are used to train and predict the water level of the downstream boundary. The specific calculation paradigm is as follows:

Among them, Z is the downstream boundary water level, that is, the water level of Datong Station; Q _in is the upstream water flow, that is, the Yichang flow; Q _B is the flow of the tributaries beside the river, that is, the Hanjiang Xiantao Station flow; Q _L is the water exchange volume between the lake and the river; m and n are the number of tributaries connected to the river and the number of lakes connected to the river, which are 1 and 2 respectively; RNN(g) represents a machine learning method called a recurrent neural network.

Taking the water level of the downstream boundary station (Datong station water level) from 2007 to 2013 as the target variable of the machine learning model, the upstream boundary flow (Yichang station flow), all upstream river tributary flows (Hanjiang Xiantao station flow) and the lake flow into and out of the river ( Chenglingji station flow and Hukou station flow) are used as the input of the machine learning model to train the parameters of the machine learning model. Use the RNN model with trained parameters to predict the water level at the downstream boundary station (the water level at Datong Station) from 2014 to 2016.

Step 4. Divide the river section: discretize the research section according to the topographic data of the river, and divide a section at intervals on the river. A total of 535 sections are divided, and the interval between sections is mostly 1-5 kilometers. Organize the shore start distance and elevation data on these sections.

Step 5, initial condition setting: Assign the initial flow of 535 sections as the flow of the first day (January 1, 2014) at the Yichang Station on the upstream boundary, and the initial water level is based on the first day of the Datong Station on the downstream boundary (January 1, 2014). Day) of water level and river slope for linear interpolation calculation;

Step 6, discretizing the Saint-Venant equations in the following form (Saint-Venant equations, the basic equations of the one-dimensional hydrodynamic model):

In the formula: A is the area, m; t is the time, s; Q is the flow rate, m ³ /s; x is the flow, m; q is the lateral flow on the unit flow, a positive value means inflow, m ² /s; g is the acceleration of gravity, taken as 9.81m/s ² ; B is the width of the water surface, in m; Z is the water level, in m; n is the roughness, dimensionless; R is the wetted circumference, in m.

According to the boundary conditions obtained in steps 1-3 and the initial conditions obtained in step 5, the one-dimensional hydrodynamic model Saint-Venant equations are solved. The water level and discharge simulation results of major river and lake stations are shown in Fig. 2. The simulated and measured values of the downstream boundary water level are shown in Fig. 3. This method simulates the water level and discharge of the river-lake complex system well, and is suitable for hydrodynamic modeling of the river-lake complex system. In addition, the method can simulate the water level and flow of the study area for a period of 1 year in 9.4 minutes, which is 38 times higher than that of the traditional one-dimensional coupled hydrodynamic model in the same area, which takes 6 hours to simulate at the same time. Operation efficiency of hydrodynamic model (Xijun Lai, Jiahu Jiang, Qiuhua Liang, Qun Huang, 2013, Large-scale hydrodynamic modeling of the middle Yangtze River Basin with complexriver–lake interactions, Journal of Hydrology, 492, pp.228-243.) .

From the above description, it can be seen that the machine learning-based hydrodynamic modeling method of the river-lake composite system provided by the embodiment of the present application can improve the operation efficiency of the hydrodynamic simulation of the river-lake composite system through flexible modeling, and quickly calculate the important sections of the river-lake composite system At the same time, it reduces the requirements for topographic data, does not need to input complex lake topographic elevation survey data, and can effectively improve the accuracy of water level simulation of sections near the downstream boundary.

In the present invention, specific examples have been applied to explain the principles and implementation methods of the present invention, and the descriptions of the above examples are only used to help understand the method of the present invention and its core idea; meanwhile, for those of ordinary skill in the art, according to this The idea of the invention will have changes in the specific implementation and scope of application. To sum up, the contents of this specification should not be construed as limiting the present invention.

Claims (3)

1. A method for modeling hydrodynamics of a composite system of rivers and lakes based on machine learning, characterized in that the method comprises: Collect flow data at the upstream boundary, lateral boundary, and junctions of rivers and lakes in the study area, as well as water level data at downstream boundary stations; According to the flow of the upstream boundary site and the flow of the lateral tributary site that may affect the water exchange between the river and the lake, the machine learning method is used to train and predict the flow of the river and lake intersection; Using machine learning methods to train and predict water levels at downstream boundaries based on flows at upstream and lateral boundary stations, as well as flows exchanged between rivers and lakes; Discretize the researched river section according to the topographic data of the river channel, divide a section at intervals on the river channel, and sort out the bank start distance and elevation data on the section; The initial flow of each section is assigned as the flow of the first day of the upstream boundary, and the initial water level is calculated by linear interpolation according to the water level of the first day of the downstream boundary and the slope of the river; Discrete Saint-Venant's equations, and input the initial and boundary conditions of each channel section for one-dimensional hydrodynamic calculation of the channel. 2. The hydrodynamic modeling method of river-lake complex system based on machine learning according to claim 1, characterized in that, said adopting machine learning method to train and predict the flow at the intersection of river and lake includes: The specific calculation paradigm adopted is:

in,

is the amount of water exchanged between the j-th lake and the river; Q _in is the inflow of upstream water; Q _B is the flow rate of the side tributaries of the river, and the amount of water exchanged between the j-th lake and the river is affected by the tributaries of the i river;

is the tributary flow undertaken by the jth lake itself, and there are k lake tributaries in total; f(g) represents the default machine learning method. 3. the hydrodynamic modeling method of river-lake composite system based on machine learning according to claim 1, is characterized in that, described adopting machine learning method to train and predict the water level of downstream boundary, comprising: The specific calculation paradigm of the downstream boundary water level adopted is as follows:

Among them, Z is the water level of the downstream boundary; Q _in is the inflow of upstream water; Q _B is the flow of tributaries beside the river; Q _L is the water exchange volume between lakes and rivers; m and n are the number of tributaries connected to the river and the number of lakes connected to the river ; f(g) represents the preset machine learning method.

Images