Method for predicting river water level under front water situation of river water level and flow relationship inverse calculation and database construction by coupling recurrent neural network
By coupling a recurrent neural network with the relationship between river water level and flow rate, a method for predicting river water level under the scenario of inflow before reservoir construction is developed. This method solves the problems of large data requirements and prediction distortion in existing technologies, and achieves high-precision water level prediction and rational allocation of water resources.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHANGJIANG RIVER SCI RES INST CHANGJIANG WATER RESOURCES COMMISSION
- Filing Date
- 2023-01-06
- Publication Date
- 2026-06-05
AI Technical Summary
Existing technologies for river level prediction suffer from problems such as high data requirements for model building, high costs, and inability to effectively isolate the influence of water conservancy project construction factors, resulting in inaccurate predictions for long-term river inflow scenarios.
A coupled recurrent neural network method was used to construct a flow simulation and prediction model for key cross-section hydrological stations in the downstream river channel of the reservoir. By utilizing the long-sequence flow sample relationship between the target hydrological station and the representative hydrological station of the upstream water conservancy project, and combining the functional expression of water level and flow, the river water level process before the reservoir was built was calculated.
It enables high-precision prediction of river water levels with low modeling costs, providing key parameters to support scheduling decisions for upstream reservoirs and downstream lakes, and rationally allocating water resources.
Smart Images
Figure CN115983483B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of reservoir scheduling technology, specifically a method for predicting river water levels under the scenario of inflow before reservoir construction, which involves inversely calculating the relationship between river water level and flow rate using a coupled recurrent neural network. Background Technology
[0002] Riverbed morphology is constantly changing due to factors such as water conservancy project construction, riverbank management, and sedimentation. Historical water level-discharge relationships reflect past riverbed environmental conditions, thus simulations using historical data under current riverbed conditions are somewhat distorted. However, accurate river level prediction is of great practical significance for upstream water conservancy projects, downstream lake management decisions, and ensuring water supply security for riverside water intake projects.
[0003] Currently, research on simulating and predicting downstream river cross-section water levels under historical inflow scenarios influenced by factors such as the construction of water conservancy projects generally employs two approaches to constructing simulation models: First, collecting data on river cross-section morphology and parameters reflecting riverbed morphology such as roughness, and constructing a river evolution model using traditional one-dimensional or two-dimensional hydraulic models; second, constructing a simulation model based on collected long-sequence water level data from target hydrological stations using stochastic hydrology or data-driven methods. Then, inputting historical inflow scenarios influenced by factors such as the construction of water conservancy projects, the constructed simulation model is used to predict and calculate the water level process at the target cross-section (usually represented by hydrological station data).
[0004] In summary, the existing technologies have the following problems: (1) Traditional hydraulic models can better reflect the riverbed morphology. If the parameters input to the model are accurate and reasonable, the prediction effect is relatively realistic and can better reflect the water level of each section of the river under the current riverbed morphology and the influence of factors such as the construction of water conservancy projects on the historical inflow of the river. However, the data sample required for building the model is large and the actual collection process is difficult and the model construction cost is high. (2) The research idea of using stochastic hydrology or data-driven methods to build a river water level prediction simulation model has a lower model construction cost, but it cannot distinguish or separate the influence of factors such as the construction of water conservancy projects on the river evolution relationship. The water level prediction simulation effect for the river inflow scenario before the construction of water conservancy projects is relatively poor and even easily distorted. Summary of the Invention
[0005] This invention addresses the aforementioned problems by providing a method for predicting river levels under pre-reservoir inflow scenarios using a coupled recurrent neural network and the relationship between river level and flow rate. This method effectively captures decision-making information from the target hydrological station and representative hydrological stations of upstream water conservancy projects, as well as the correlation between long-sequence flow samples between adjacent upstream hydrological stations. It also utilizes the target station's own level-flow rate relationship to predict the target hydrological station's level with lower modeling costs and higher simulation accuracy. This provides crucial parameter support for upstream reservoirs or downstream lakes with gate control to make scheduling decisions and for the normal water supply of riverside water intake projects, thereby enabling a more rational allocation of water resources utilization in different spaces upstream and downstream of the river.
[0006] This invention provides a method for joint water supply scheduling of reservoirs and water diversion projects to meet the water needs of multiple parties, comprising the following steps:
[0007] Step 1: Construct a flow simulation and prediction model for key cross-section hydrological stations in the downstream river channel after the reservoir is built using the recurrent neural network method.
[0008] Step 2: Using long-term flow series of hydrological stations after the reservoir is officially built and put into operation as data samples, the flow simulation and prediction model of key cross-section hydrological stations in the downstream river channel constructed in Step 1 is used for parameter training and accuracy verification to obtain the trained flow simulation and prediction model.
[0009] Step 3: Using the flow simulation and prediction model obtained after training in Step 2, input the historical flow of the downstream key section hydrological station before the reservoir is officially built and put into operation, and simulate and predict the flow process value of the target hydrological station within the forecast period.
[0010] Step 4: Using the flow process value of the target hydrological station within the forecast period obtained in Step 3 as input, and combining it with the latest fitted relationship between the water level and flow of the target hydrological station, the water level value within the forecast period corresponding to the historical flow process of the target hydrological station before the reservoir is built is predicted under the current river channel and riverbed characteristics. This realizes the calculation of the water level corresponding to the historical flow process of the river hydrological station before the reservoir is built.
[0011] Furthermore, step 1 specifically includes:
[0012] Step 1-1: Taking the river channel where the reservoir is located as the research object, select multiple key cross-section hydrological stations in the downstream river channel of the reservoir as the target hydrological station set for simulation and prediction. The hydrological station set includes the hydrological station at the reservoir outlet section, the target hydrological station, and the upstream hydrological station adjacent to the target hydrological station.
[0013] Steps 1-2 involve constructing a simulation and prediction model for the relationship between the outflow from the reservoir's hydrological station and the flow at key downstream river sections using a recurrent neural network. The prediction time period is denoted as t1, t2, ..., tF , where t F To determine the forecast period, the input parameters of the simulation prediction model include the historical flow sequences of the reservoir outflow section hydrological station and the downstream target hydrological station, as well as the historical flow sequences of the adjacent upstream hydrological stations of the target hydrological station, with each historical time being t. 0-H ,…,t 0-1 ,t0, where t0 is the current time, t 0-H This represents the time point H days prior in history. The value of H is not fixed and is determined based on the fitting results.
[0014] Furthermore, step 2 specifically includes:
[0015] Step 2-1: Using the time when the reservoir is officially completed and put into operation as the node, divide the long-term flow series data of the river hydrological station into two periods: before the reservoir is built and after the reservoir is built, which are denoted as N1 and N2 respectively.
[0016] Step 2-2: Divide the long-term flow sequence N2 of the river hydrological station after reservoir construction into a training set and a validation set in a 2:8 ratio, and perform parameter calibration and simulation accuracy verification on the flow simulation prediction model constructed in Step 1.
[0017] Furthermore, the feature is that step 3 specifically includes:
[0018] Step 3-1: Using the flow sequence N1 of the river hydrological station before reservoir construction as the historical flow process sample, select a historical flow process of a certain year as an example, and use it as the input data for the calculation of the target hydrological station's water level under the historical flow process.
[0019] Step 3-2: Using the flow simulation prediction model obtained after training in Step 2, predict the flow process value of the target hydrological station within the forecast period.
[0020] Furthermore, the feature is that step 4 specifically includes:
[0021] Step 4-1: Fit the latest annual water level and flow data collected from the target hydrological station, and use a function expression to describe the relationship between the water level and flow at the target hydrological station, thereby reflecting the hydrological and flow relationship of the target hydrological station section under the current riverbed condition after the reservoir is built;
[0022] Step 4-2: Using the flow process of the target hydrological station within the forecast period predicted in Step 3 as input, and combining it with the function expression describing the relationship between water level and flow of the target hydrological station obtained by fitting in Step 4-1, the water level value process within the forecast period corresponding to the historical flow process of the target hydrological station before the reservoir is built is back-calculated and predicted under the current river channel and riverbed characteristics.
[0023] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0024] This invention proposes a method for predicting river levels under the scenario of inflow before reservoir construction, which involves coupling a recurrent neural network with the relationship between river level and flow rate. This method can fully capture the decision-making information of the target hydrological station and the representative hydrological station of the upstream water conservancy project, as well as the correlation of long-sequence flow samples between adjacent upstream hydrological stations. It also uses the water level and flow rate relationship of the target station itself to predict the water level value of the target hydrological station with low modeling cost and high simulation accuracy. This provides key parameter support for the scheduling decisions of upstream reservoirs or downstream lakes with gate control, and for the normal water supply of water intake projects along the river, thereby enabling a more rational allocation of water resources utilization in different spaces upstream and downstream of the river. Attached Figure Description
[0025] Figure 1 This is a flowchart illustrating a method for predicting river levels under the scenario of inverse calculation of the relationship between river water level and flow rate in a reservoir construction, provided by an embodiment of the present invention.
[0026] Figure 2 This is a fitted curve of the water level-discharge relationship at a certain cross section before reservoir construction in an embodiment of the present invention;
[0027] Figure 3 This is a fitted curve of the water level-discharge relationship at a certain cross section after the reservoir was built in an embodiment of the present invention;
[0028] Figure 4 This is a schematic diagram comparing the simulation effects of the method proposed in this invention, which uses a coupled recurrent neural network to calculate the inflow of water before reservoir construction, and the method that directly simulates and predicts the water level of a certain section of a river using a recurrent neural network. Detailed Implementation
[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0030] like Figure 1 As shown in Embodiment 1, this method provides a prediction method for river level under the scenario of inflow before reservoir construction, which involves inverse calculation of the relationship between river level and flow rate using a coupled recurrent neural network:
[0031] Step 1: Construct a flow simulation and prediction model for key cross-section hydrological stations in the downstream river channel after the reservoir is built using the recurrent neural network method.
[0032] Step 2: Using long-term flow series of hydrological stations after the reservoir is officially built and put into operation as data samples, the flow simulation and prediction model of key cross-section hydrological stations in the downstream river channel constructed in Step 1 is used for parameter training and accuracy verification to obtain the trained flow simulation and prediction model.
[0033] Step 3: Using the flow simulation and prediction model obtained after training in Step 2, input the historical flow of the downstream key section hydrological station before the reservoir is officially built and put into operation, and simulate and predict the flow process value of the target hydrological station within the forecast period.
[0034] Step 4: Using the flow process value of the target hydrological station within the forecast period obtained in Step 3 as input, and combining it with the latest fitted relationship between the water level and flow of the target hydrological station, the water level value within the forecast period corresponding to the historical flow process of the target hydrological station before the reservoir is built is predicted under the current river channel and riverbed characteristics. This realizes the calculation of the water level corresponding to the historical flow process of the river hydrological station before the reservoir is built.
[0035] Furthermore, step 1 specifically includes:
[0036] Step 1-1: Taking the river channel where the reservoir is located as the research object, select multiple key cross-section hydrological stations in the downstream river channel of the reservoir as the target hydrological station set for simulation and prediction. The hydrological station set includes three stations: the hydrological station at the reservoir outlet section, the target hydrological station, and the upstream hydrological station adjacent to the target hydrological station.
[0037] Steps 1-2 involve constructing a simulation and prediction model for the relationship between the outflow from the reservoir's hydrological station and the flow at key downstream river sections using a recurrent neural network method. The prediction time periods are denoted as t1, t2, ..., t F (t F (For the forecast period length), input parameters include, but are not limited to: historical flow sequences of the reservoir outflow section hydrological station and the downstream target hydrological station, and historical flow sequences of the adjacent upstream hydrological stations of the target hydrological station, with each historical time being t. 0-H ,…,t 0-1 ,t0,,where t0 is the current time, t 0-H The time point corresponding to H days ago in history is not fixed. The specific setting of H depends on the fitting effect and can be 10 days, 20 days or longer.
[0038] Furthermore, step 2 specifically includes:
[0039] Step 2-1: Using the official completion and operation time of the reservoir as the node, divide the long-term flow series data of the river hydrological station into two periods: before the reservoir construction and after the reservoir construction, denoted as N1 and N2 respectively. For example, if a reservoir was officially completed and put into operation in 2003, the hydrological stations are divided into the N1 period before the reservoir construction (1960-2002) and the N2 period after the reservoir construction (2003-2021) based on this time node.
[0040] Step 2-2: Divide the long-term flow sequence N2 of the river hydrological station after reservoir construction into a training set and a validation set in a 2:8 ratio, and perform parameter calibration and simulation accuracy verification on the flow simulation prediction model constructed in Step 1.
[0041] Furthermore, step 3 specifically includes:
[0042] Step 3-1: Using the pre-reservoir flow sequence N1 of the river hydrological station as a historical flow process sample, select a historical flow process of a certain year as an example, and use it as input data for the water level restoration calculation of the target hydrological station under the historical flow process; for example, based on the analysis of the low water frequency of the historical flow process, select 1979 in the N1 period (1960-2002) as a typical year input.
[0043] Step 3-2: Using the flow simulation prediction model obtained after training in Step 2, predict the flow process value of the target hydrological station within the forecast period.
[0044] Furthermore, step 4 specifically includes:
[0045] Step 4-1: Fit the latest annual water level and discharge data collected from the target hydrological station, and use a reasonable function expression to describe the relationship between the water level and discharge at the target hydrological station, thereby reflecting the hydrological-discharge relationship at the cross section of the target hydrological station under the current riverbed conditions after reservoir construction. For example, if the latest year for the N2 period is 2021, then use the water level and discharge series of the target hydrological station from 2021 or 2003 to 2021 after reservoir construction to fit the function relationship, such as... Figure 3 The expression can be written as Z = f(Q), where Z is the water level, Q is the flow rate, and f is the function expression. Figure 2 The fitting results of the water level and flow rate sequence function relationship of the target hydrological station before the reservoir was built were compared. Figure 2 and Figure 3 It is known that the water level-discharge relationship in older historical data reflects the past state of the riverbed environment. Therefore, simulation studies using older historical data under the current riverbed environment conditions will have some distortion.
[0046] Step 4-2: Using the flow process of the target hydrological station within the forecast period predicted in Step 3 as input, and combining it with the function expression describing the relationship between water level and flow at the target hydrological station obtained in Step 4-1, the water level value process within the forecast period corresponding to the historical flow process before reservoir construction at the target hydrological station is back-calculated and predicted under the current riverbed characteristics. For example... Figure 4 As shown, the water level prediction model used to calculate the water level process of the target water station within the forecast period under historical flow scenarios can only reflect the past riverbed environment. The simulated values are close to the distant historical data, but they are inconsistent with the current riverbed environment and are distorted. The calculation results of the method proposed in this invention are consistent with... Figure 3 The water level-discharge relationship of the target hydrological station after the reservoir is built is closer, which can better reflect the water level process of the target hydrological station under the current riverbed environment conditions and if the historical flow process occurs.
[0047] This invention can fully capture the decision-making information of the target hydrological station and the representative hydrological station of the upstream water conservancy project, as well as the correlation of long-sequence flow samples between adjacent upstream hydrological stations. It can also use the water level and flow relationship of the target station itself to predict the water level value of the target hydrological station under the current riverbed conditions with low modeling cost and high simulation accuracy. This invention enables the reproduction calculation of the water level process of the river hydrological station under the historical inflow scenario before the construction of the water conservancy project. This provides key parameter support for the scheduling decisions of upstream reservoirs or downstream lakes with gate control, and for the normal water supply of water intake projects along the river, thereby enabling a more rational allocation of water resources utilization in different spaces in the upstream and downstream of the river.
[0048] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A method for predicting river water level under a pre-reservoir inflow scenario, using a coupled recurrent neural network and the relationship between river water level and flow rate, characterized in that... Includes the following steps: Step 1: Construct a flow simulation and prediction model for key cross-section hydrological stations in the downstream river channel after the reservoir is built using the recurrent neural network method. Step 2: Using long-term flow series of hydrological stations after the reservoir is officially built and put into operation as data samples, the flow simulation and prediction model of key cross-section hydrological stations in the downstream river channel constructed in Step 1 is used for parameter training and accuracy verification to obtain the trained flow simulation and prediction model. Step 3: Using the flow simulation and prediction model obtained after training in Step 2, input the historical flow of the downstream key section hydrological station before the reservoir is officially built and put into operation, and simulate and predict the flow process value of the target hydrological station within the forecast period. Step 4: Using the flow process values within the forecast period obtained in Step 3 as input, and combining them with the latest fitted relationship between water level and flow at the target hydrological station, the water level value within the forecast period corresponding to the historical flow process before reservoir construction at the target hydrological station is back-calculated and predicted under the current riverbed characteristics. This achieves the reconstruction calculation of the water level corresponding to the historical flow process before reservoir construction at the river hydrological station. Step 1 specifically includes: Step 1-1: Taking the river channel where the reservoir is located as the research object, select multiple key cross-section hydrological stations in the downstream river channel of the reservoir as the target hydrological station set for simulation and prediction. The hydrological station set includes the hydrological station at the reservoir outlet section, the target hydrological station, and the upstream hydrological station adjacent to the target hydrological station. Steps 1-2 involve constructing a flow simulation and prediction model for key cross-sections of the downstream river channel after reservoir construction using a recurrent neural network method. The prediction period is denoted as... ,in To determine the forecast period, the input parameters of the flow simulation prediction model include the historical flow sequences of the reservoir outflow section hydrological station and the downstream target hydrological station, as well as the historical flow sequences of the adjacent upstream hydrological stations of the target hydrological station. The corresponding historical times are all... ,in For the current moment, For history The corresponding time 2 days ago The settings are not fixed and should be adjusted based on the fitting results.
2. The method for predicting river level under the scenario of pre-reservoir inflow under the inverse calculation of the relationship between coupled recurrent neural network and river water level-discharge as described in claim 1, characterized in that: Step 2 specifically includes: Step 2-1: Using the time when the reservoir is officially completed and put into operation as the node, divide the long-term flow series data of the river hydrological station into two periods: before the reservoir is built and after the reservoir is built, which are denoted as N1 and N2 respectively. Step 2-2: Divide the long-term flow sequence N2 of the river hydrological station after reservoir construction into a training set and a validation set in a 2:8 ratio, and perform parameter calibration and simulation accuracy verification on the flow simulation prediction model constructed in Step 1.
3. The method for predicting river level under the scenario of pre-reservoir inflow under the inverse calculation of the relationship between coupled recurrent neural network and river water level-discharge as described in claim 2, characterized in that: Step 3 specifically includes: Step 3-1: Using the flow sequence N1 of the river hydrological station before reservoir construction as the historical flow process sample, select a historical flow process of a certain year as an example, and use it as the input data for the calculation of the target hydrological station's water level under the historical flow process. Step 3-2: Using the flow simulation prediction model obtained after training in Step 2, predict the flow process value of the target hydrological station within the forecast period.
4. The method for predicting river level under the scenario of pre-reservoir inflow under the inverse calculation of the relationship between coupled recurrent neural network and river water level-discharge as described in claim 1, characterized in that: Step 4 specifically includes: Step 4-1: Fit the latest annual water level and flow data collected from the target hydrological station, and use a function expression to describe the relationship between the water level and flow at the target hydrological station, thereby reflecting the hydrological and flow relationship of the target hydrological station section under the current riverbed condition after the reservoir is built. Step 4-2: Using the flow process of the target hydrological station within the forecast period predicted in Step 3 as input, and combining it with the function expression describing the relationship between water level and flow of the target hydrological station obtained by fitting in Step 4-1, the water level value process within the forecast period corresponding to the historical flow process of the target hydrological station before the reservoir is built is back-calculated and predicted under the current river channel and riverbed characteristics.