Index method for fitting relation between complex water level and flow of river channel section in middle and downstream regions of drainage basin

Abstract

The invention discloses an index method for fitting a complex water level and flow relationship of a river channel section in a middle and downstream region of a drainage basin, and the method comprises the following steps: collecting long-time sequence data of a target river channel section, and carrying out the preprocessing of the long-time sequence data; calculating the change rate of water level to time and the change rate of flow to time; based on a one-dimensional river hydrodynamic model, solving a flow velocity v and a water depth h through a Saint-Venant equation set, and calculating a Froude number Fr; the water level Z, the change rate dZ / dt of the water level to the time, the change rate dQ / dt of the flow to the time and the Froude number Fr serve as four input neurons of the neural network, and the flow Q serves as an output neuron; setting a hidden layer activation function, the number of neurons, a learning rate and a momentum factor parameter, and training a neural network through a back propagation algorithm; the trained neural network is used for simulating the flow of the target section, and a complex water level flow relation curve is output; the simulation precision is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of hydrological technology, and in particular to an index method for fitting the complex relationship between water level and flow in a river section in the middle and lower reaches of a river basin. Background Art

[0002] The relationship between water level and flow at river sections plays a crucial role in flood control, integrated water resources management, and other tasks. Typically, in the upper reaches of a river basin, the water level-flow relationship presents a single, easily defined curve. However, in the middle and lower reaches of a river basin, flow varies not only with water level but also with numerous factors, such as backwater support, tidal fluctuations, and riverbed erosion and deposition. Consequently, the water level-flow relationship often appears as a composite loop curve or other more complex curves, which cannot accurately represent the relationship.

[0003] Machine learning methods have the ability to handle complex nonlinear problems, learning from a large number of samples and approximating nonlinear systems. However, water level and flow series as learning samples are insufficient to describe the complex relationship between water level and flow at river sections in the middle and lower reaches of a basin, resulting in poor simulation results. Therefore, finding indicators that better reflect the changing patterns of the water level and flow relationship at river sections is a key issue in solving this problem. Summary of the Invention

[0004] Purpose of the invention: The purpose of the present invention is to provide an indicator method for fitting the complex water level and flow relationship of river sections in the middle and lower reaches of a basin, which can fit the indicator of the complex water level and flow relationship of river sections in the middle and lower reaches of a basin, and can better fit the water level and flow relationship of river sections in the middle and lower reaches of different basins through machine learning methods.

[0005] Technical solution: The present invention provides an index method for fitting the complex relationship between water level and flow in a river section in the middle and lower reaches of a river basin, comprising the following steps:

[0006] (1) Collect long-term series data of the target river section, including water level Z, flow Q and measured large-section data; and perform preprocessing.

[0007] (2) Calculate the time-varying rate of change of water level and flow rate; based on the one-dimensional river hydrodynamic model, solve the flow velocity v and water depth h by using the Saint-Venant equations and calculate the Froude number Fr.

[0008] (3) The water level Z, the rate of change of the water level with time dZ / dt, the rate of change of the flow with time dQ / dt, and the Froude number Fr are used as the four input neurons of the neural network, and the flow Q is used as the output neuron; the hidden layer activation function, the number of neurons, the learning rate and the momentum factor parameters are set, and the neural network is trained using the back propagation algorithm.

[0009] (4) Use the trained neural network to simulate the flow of the target section and output a complex water level-flow relationship curve.

[0010] Furthermore, in step (1), the preprocessing includes: performing interpolation processing on the water level and flow sequence with non-equal time intervals to convert them into a sequence with equal time intervals.

[0011] Furthermore, in step (2), the calculation of the Froude number Fr is achieved by the following steps: in a one-dimensional river hydrodynamic model, a river with a very short length is generalized, and the upper and lower sections use the large section data measured at the site; the upper boundary condition is set as the flow boundary condition, and the lower boundary condition is set as the water level boundary condition; the flow velocity v and water depth h are obtained by solving the Saint-Venant equations, which are:

[0012]

[0013] Among them, Q, A, B, and Z are the average flow of the river section (m 3 / s), water flow area (m 2 ), river width (m) and water level (m). K is the flow modulus; m 3 / (s·km 2 ), α is the momentum correction coefficient.

[0014] Furthermore, in step (3), the hidden layer activation function of the neural network is a Gaussian function, the learning rate is set to 0.5-0.9, the momentum factor is set to 0.3-0.7, and the simulation time step Δt is 3600 seconds.

[0015] Furthermore, in step (3), the neural network model is a BP neural network.

[0016] The present invention provides an index system for fitting the complex relationship between water level and flow in a river section in the middle and lower reaches of a river basin, comprising:

[0017] Acquisition module: used to collect long-term series data of the target river section, including water level Z, flow Q and measured large-section data; and perform preprocessing.

[0018] Calculation module: used to calculate the time rate of change of water level dZ / dt and the time rate of change of flow dQ / dt; based on the one-dimensional river hydrodynamic model, solve the flow velocity v and water depth h through the Saint-Venant equations and calculate the Froude number Fr.

[0019] Training module: used to use the water level Z, the rate of change of water level with time dZ / dt, the rate of change of flow with time dQ / dt, and the Froude number Fr as the four input neurons of the neural network, and the flow Q as the output neuron; set the hidden layer activation function, number of neurons, learning rate and momentum factor parameters, and train the neural network through the back propagation algorithm.

[0020] Output module: used to simulate the flow of the target section using the trained neural network and output a complex water level-flow relationship curve.

[0021] Furthermore, in the acquisition module, preprocessing includes: performing interpolation processing on water level and flow sequences with unequal time intervals to convert them into sequences with equal time intervals.

[0022] Furthermore, in the calculation module, the Froude number Fr is calculated by the following steps: in a one-dimensional river hydrodynamic model, a very short river is generalized, and the upper and lower sections use the large-section data measured at the site; the upper boundary condition is set as the flow boundary condition, and the lower boundary condition is set as the water level boundary condition; the flow velocity v and water depth h are obtained by solving the Saint-Venant equations, which are:

[0023]

[0024] Among them, Q, A, B, and Z are the average flow of the river section (m 3 / s), water flow area (m 2 ), river width (m) and water level (m). K is the flow modulus; m 3 / (s·km 2 ), α is the momentum correction coefficient.

[0025] Furthermore, in the training module, the Froude number Fr is introduced as the input neuron to train the neural network to fit the relationship between water level and flow.

[0026] Furthermore, in the training module, the hidden layer activation function of the neural network is a Gaussian function; and the neural network model is a BP neural network.

[0027] Beneficial effects: Compared with the existing technology, the present invention has the following significant advantages: the simulation results of the present invention show that in the river sections in the middle and lower reaches of most river basins, when Z and Fr are used as input neurons, the simulation accuracy is greatly improved, and the accuracy is still slightly improved when dZ / d and dQ / dt are further added; when only Z, dZ / d and dQ / dt are used as input neurons in the river sections of individual river basins, the simulation accuracy is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 The measured water level and flow relationship curve of Huayuankou Station of the present invention;

[0029] Figure 2 This is a comparison chart of the flow simulation of different groups at Huayuankou Station of the present invention. DETAILED DESCRIPTION

[0030] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0031] like Figure 1 As shown, an embodiment of the present invention provides an index method for fitting the complex relationship between water level and flow in a river section in the middle and lower reaches of a river basin, comprising the following steps:

[0032] S1 collects long-term series data of sites (sections), including water level, flow, and measured large sections. It interpolates the discontinuous water level and flow series and processes them into equal time interval series. It calculates the rate of change of water level over time (dZ / dt) and the rate of change of flow over time (dQ / dt). It uses the one-dimensional river hydrodynamic method to calculate the Froude number Fr series at different times of the site.

[0033] S2 uses the water level Z, the rate of change of water level with time (dZ / dt), the rate of change of flow with time (dQ / dt), and the Froude number Fr as the four input neurons of the neural network, and the flow Q as the one output neuron of the neural network. It sets appropriate hidden layer activation functions and parameters such as the number of hidden layer neurons, learning rate, momentum factor, and simulates the flow Q after training and learning.

[0034] The specific process is as follows:

[0035] A. Indicator calculation

[0036] The present invention uses water level Z, the rate of change of water level over time (dZ / dt), the rate of change of flow over time (dQ / dt), and the Froude number Fr as input neurons in machine learning, and flow Q as the output neuron. dZ / dt and dQ / dt can be calculated, and the Froude number Fr is calculated as follows:

[0037]

[0038] Where v is the velocity, m / s; g is the acceleration due to gravity, m / s 2 ; h is the water depth, m.

[0039] It is usually difficult to obtain flow velocity data, which makes it difficult to calculate the Froude number. Therefore, the hydrodynamic method is used for approximate calculation. In the hydrodynamic model, a river with a very short length is generalized. The upper and lower sections use the large cross-section data measured at the station. The upper boundary condition uses the flow boundary condition of the station, and the lower boundary condition uses the water level boundary condition of the station. The Saint-Venant equations are used to solve the calculation, and the flow velocity and water depth are output to calculate the Froude number. Since the distance between the two sections is extremely small, the Froude number can be approximately considered to be the Froude number of the station. The calculation formula of the Saint-Venant equations is as follows:

[0040]

[0041] Where: Q, A, B, Z are the average flow of the river section (m3 / s), water flow area (m 2 ), river width (m) and water level (m). K is the flow modulus; m 3 / (s·km 2 ), α is the momentum correction coefficient.

[0042] B. BP neural network training and traffic simulation

[0043] The BP neural network consists of an input layer, a hidden layer, and an output layer. The input layer data is calculated using the hidden layer activation function to generate a simulated value for the output layer. The difference between the simulated value and the measured value is the loss function. By continuously adjusting the network weights to minimize the loss function, the neural network's learning and training process is complete. The water level Z, the rate of change of the water level over time (dZ / dt), the rate of change of the flow over time (dQ / dt), and the Froude number Fr are used as the four input neurons of the neural network, and the flow rate Q is used as the one output neuron. Appropriate hidden layer activation functions, as well as parameters such as the number of hidden layer neurons, learning rate, and momentum factor are set. After training and learning, the flow rate is simulated.

[0044] Using the above method, the flow simulation of Huayuankou Station in the Yellow River Basin was carried out. The measured water level and flow relationship curve of Huayuankou Station is shown in the attached figure. Figure 1 , it can be seen that this station is greatly affected by river scouring and silting, and the relationship between water level and flow is complex, showing a composite loop curve. According to the different input neurons, the simulation results are divided into two groups, A and B. Among them, the input neurons of group A are water level Z; the input neurons of group B are Z, dZ / dt, dQ / dt, Fr. Gaussian activation function is used, the simulation time step Δt = 3600s, the learning rate is 0.7, and the momentum factor is 0.5. After BP neural network training, the flow is simulated, and the results are shown in the attached. Figure 2 .

Claims

1. An index method for fitting the complex relationship between water level and flow in river sections in the middle and lower reaches of a river basin, characterized by: The following steps are involved: (1) Collect long-term series data of the target river section, including water level Z, flow Q and measured large-section data; and perform preprocessing. (2) Calculate the time rate of change of water level dZ / dt and the time rate of change of flow dQ / dt; based on the one-dimensional river hydrodynamic model, solve the flow velocity v and water depth h through the Saint-Venant equations and calculate the Froude number Fr. (3) The water level Z, the rate of change of the water level with time dZ / dt, the rate of change of the flow with time dQ / dt, and the Froude number Fr are used as the four input neurons of the neural network, and the flow Q is used as the output neuron; the hidden layer activation function, the number of neurons, the learning rate and the momentum factor parameters are set, and the neural network is trained using the back propagation algorithm. (4) Use the trained neural network to simulate the flow of the target section and output a complex water level-flow relationship curve.

2. The index method for fitting the complex relationship between water level and flow in the river section of the middle and lower reaches of the basin according to claim 1 is characterized in that: In step (1), the preprocessing includes: interpolating the water level and flow sequence with non-equal time intervals to convert them into equal time interval sequences.

3. The index method for fitting the complex relationship between water level and flow in the river section of the middle and lower reaches of the basin according to claim 1 is characterized in that: In step (2), the calculation of the Froude number Fr is achieved by the following steps: in a one-dimensional river hydrodynamic model, a river with a very short length is generalized, and the upper and lower sections use the large section data measured at the site; the upper boundary condition is set as the flow boundary condition, and the lower boundary condition is set as the water level boundary condition; the flow velocity v and water depth h are obtained by solving the Saint-Venant equations, which are: Among them, Q, A, B, and Z are the average flow of the river section (m 3 / s), water flow area (m 2 ), river width (m) and water level (m). K is the flow modulus; m 3 / (s·km 2 ), α is the momentum correction coefficient.

4. The index method for fitting the complex relationship between water level and flow in the river section of the middle and lower reaches of the basin according to claim 1 is characterized in that: In step (3), the Froude number Fr is introduced as the input neuron to train the neural network to fit the relationship between water level and flow.

5. The index method for fitting the complex relationship between water level and flow in the river section of the middle and lower reaches of the basin according to claim 1 is characterized in that: In step (3), the hidden layer activation function of the neural network is a Gaussian function; and the neural network model is a BP neural network.

6. An index system for fitting the complex relationship between water level and flow in river sections in the middle and lower reaches of a river basin, characterized by: include: Acquisition module: used to collect long-term series data of the target river section, including water level Z, flow Q and measured large-section data; And preprocessing. Calculation module: used to calculate the time rate of change of water level dZ / dt and the time rate of change of flow dQ / dt; based on the one-dimensional river hydrodynamic model, solve the flow velocity v and water depth h through the Saint-Venant equations and calculate the Froude number Fr. Training module: used to use the water level Z, the rate of change of water level with time dZ / dt, the rate of change of flow with time dQ / dt, and the Froude number Fr as the four input neurons of the neural network, and the flow Q as the output neuron; set the hidden layer activation function, number of neurons, learning rate and momentum factor parameters, and train the neural network through the back propagation algorithm. Output module: used to simulate the flow of the target section using the trained neural network and output a complex water level-flow relationship curve.

7. The index system for fitting the complex relationship between water level and flow in river sections in the middle and lower reaches of a river basin according to claim 6, characterized in that: In the acquisition module, preprocessing includes: interpolating the water level and flow sequences with unequal time intervals and converting them into sequences with equal time intervals.

8. The index system for fitting the complex relationship between water level and flow in river sections in the middle and lower reaches of a river basin according to claim 6, characterized in that: In the calculation module, the Froude number Fr is calculated through the following steps: In a one-dimensional river hydrodynamic model, a very short river channel is generalized, and the upper and lower sections use the measured large-section data of the site; the upper boundary condition is set as the flow boundary condition, and the lower boundary condition is set as the water level boundary condition; the flow velocity v and water depth h are obtained by solving the Saint-Venant equations, which are: Among them, Q, A, B, and Z are the average flow of the river section (m 3 / s), water flow area (m 2 ), river width (m) and water level (m). K is the flow modulus; m 3 / (s·km 2 ), α is the momentum correction coefficient.

9. The index system for fitting the complex relationship between water level and flow in river sections in the middle and lower reaches of a river basin according to claim 6, characterized in that: In the training module, the Froude number Fr is introduced as the input neuron to train the neural network to fit the relationship between water level and flow.

10. The index system for fitting the complex relationship between water level and flow in river sections in the middle and lower reaches of a river basin according to claim 6, characterized in that: In the training module, the hidden layer activation function of the neural network is a Gaussian function; the neural network model is a BP neural network.

Images