Deep learning method and device for evaluating watershed middle and downstream water level flow change

Abstract

The invention discloses a deep learning method and device for evaluating the water level flow change of the middle and lower streams of a drainage basin. The method comprises the following steps: collecting and preprocessing hydrological section data; carrying out flow reduction calculation and evolution on the upstream control station of the drainage basin; constructing a water level flow change response agent model, and constructing an agent model of upstream reservoir regulation and storage and middle and downstream water level flow change response of the watershed based on deep learning; and data set division and agent model test application: dividing the time sequence data into a training set, a verification set and a test set, and finally obtaining a flow change value of the river basin middle and downstream control stations influenced by regulation and storage of the upstream reservoir. According to the method, the actually measured hydraulic regimen data and the flood routing model are combined, and the agent model of drainage basin upstream reservoir regulation and storage and middle and downstream water level flow change response is constructed based on deep learning, so that immediate response of middle and downstream hydrological regimes in different regulation and storage scenes of the drainage basin upstream reservoir can be realized, and the requirements of real-time performance and quick response during research and judgment of a scheduling strategy are met.

Description

Technical Field

[0001] The present invention relates to the field of hydrology and water resources, and more specifically, to a deep learning method and device for evaluating the water level and flow changes in the middle and lower reaches of a basin. Background Art

[0002] In the context of global climate change and the increasing scarcity of water resources, reservoirs, as the core components of water conservancy projects, play an irreplaceable role in flood control, power generation, irrigation, water supply, and ecological restoration. However, with the construction and operation of large reservoir groups, especially the giant reservoir groups composed of multiple series-parallel reservoirs, as the key controlling projects in the upper reaches of the basin, their dispatching operations directly change the hydrological situation in the middle and lower reaches of the basin.

[0003] Traditional research methods, such as the hydrodynamic model based on the river flood routing process, although significant progress has been made in simulating the hydrodynamic process, the model calibration process is complex and the running time is long, and there are still limitations in quickly judging the impact of different storage conditions of upstream reservoirs in the basin on the hydrological situation in the middle and lower reaches. Facing the complex reservoir regulation system of multiple series-parallel connections and the hydrological response characteristics of the complex river network in the basin, traditional methods are difficult to quickly and accurately capture and simulate the changes in the hydrological situation in the middle and lower reaches of the basin, thus restricting the judgment of different dispatching schemes for the cascade reservoir group in the upstream of the basin.

[0004] Therefore, there is an urgent need for a faster and more accurate evaluation method to explore the response law between the storage regulation of upstream reservoirs in the basin and the water level and flow changes in the middle and lower reaches. The limitations of the current methods are mainly reflected in two aspects: (1) Insufficient timeliness and flexibility of the model: Although significant progress has been made in the traditional hydrodynamic model based on the river flood routing process, its model calibration process is complex, relying on expert experience for parameter adjustment, and the model operation time is long, resulting in difficulties in meeting the real-time and rapid response requirements when quickly judging the impact of different storage conditions of upstream reservoirs in the basin on the hydrological situation in the middle and lower reaches. Therefore, the timeliness and flexibility of the traditional model are limited when dealing with emergency dispatching decisions or evaluating rapidly changing hydrological situations. (2) Lack of efficient data fusion and processing capabilities: When evaluating the response law between the storage regulation of upstream reservoirs in the basin and the water level and flow changes in the middle and lower reaches, it is necessary to comprehensively consider various data sources, including but not limited to measured hydrological and water regime data, meteorological data, flood routing models, etc. However, traditional methods have deficiencies in data fusion and processing, and it is difficult to seamlessly integrate and deeply apply multi-source heterogeneous data. Summary of the Invention

[0005] In view of the deficiencies in the existing technologies, it is necessary to propose a method that can accurately predict the changes in water level and flow in the middle and lower reaches of a river basin, and further propose an evaluation method that takes into account the storage and regulation of upstream reservoirs in the river basin and the response law of water level and flow changes in the middle and lower reaches, so as to quickly judge the impact of different storage and regulation situations of upstream reservoirs in the river basin on the hydrological situation in the middle and lower reaches when analyzing the relationship between water level and flow at hydrological stations.

[0006] To achieve the above object, a first aspect of the present invention provides a deep learning method for evaluating the changes in water level and flow in the middle and lower reaches of a river basin, including: Step S1: Collect and preprocess hydrological section data; Step S2: Use the flow restoration calculation method to restore the measured flow at the upstream control station in the river basin to the natural flow, further calculate the storage and regulation change amount of the upstream reservoir in the river basin, and evolve the measured flow and natural flow at the upstream control station in the river basin to the middle and lower reaches control station through a one-dimensional hydrodynamic model to obtain the flow change value affected by the upstream reservoir storage and regulation at the middle and lower reaches control station; Step S3: Based on the deep learning method, construct a proxy model for the storage and regulation of upstream reservoirs in the river basin and the response of water level and flow changes in the middle and lower reaches. The inputs of the model are the natural flow in the upper reaches of the main stream of the river basin, the measured flows of other tributaries, the measured water levels at the middle and lower reaches control stations considering the time lag, and the storage and regulation change amount of the upstream reservoir in the river basin, and the output of the model is the flow change value affected by the upstream reservoir storage and regulation at the middle and lower reaches control station; Step S4: Train the proxy model for the storage and regulation of upstream reservoirs in the river basin and the response of water level and flow changes in the middle and lower reaches, and use the trained model to predict the changes in water level and flow in the middle and lower reaches of the river basin.

[0007] In one implementation, step S1 includes: Collect the measured water levels and flow data of the flow measurement sections of each hydrological station in the upper and lower reaches of the river basin, including the measured flow in the upper reaches of the main stream of the river basin, the measured flows of other tributaries, the measured flow and water level at the middle and lower reaches control station; Clean the collected data and information to form time series data.

[0008] In one implementation, step S2 includes: Use the flow restoration calculation method to calculate the reservoir inflow before storage and regulation respectively according to the water balance equation for the storage and regulation amount of the upstream reservoir in the river basin; According to the reservoir inflow before storage and regulation, calculate the evolved flow at the outlet section of the upstream control station in the river basin according to the Changjiang Water Resources Commission confluence curve formula to complete the river flood evolution from the dam site of the upstream reservoir in the river basin to the control section; Superimpose the evolved flow of each reservoir at the control section and the measured flow at the control section to obtain the restored natural flow at the upstream control station section in the river basin, and obtain the storage and regulation change amount of the upstream reservoir in the river basin according to the difference between the natural flow and the measured flow; The flow obtained by restoring the natural flow evolution is the flow before regulation, and the flow obtained by measuring the flow evolution is the flow after regulation. According to the difference between the flow before regulation and the flow after regulation, the flow change value affected by the upstream reservoir regulation of the control stations in the middle and lower reaches of the basin is obtained.

[0009] In one implementation, the flow reduction calculation method is used to calculate the reservoir inflow before regulation according to the water balance equation for the regulation volume of the reservoir, including: based on the reservoir water level, storage capacity curve, and outflow, the inflow is inversely calculated based on the water balance method. Specifically:

[0010] In the formula: is the average inflow in the time period, is the average outflow in the time period, is the water loss in the reservoir, is the change value of the reservoir storage volume at the beginning and end of the time period; is the calculation time period; The Changjiang Water Resources Commission confluence curve formula is:

[0011] In the formula: is the ordinate value of the confluence parameter of the river section, called the confluence parameter; is the relative time, is the time; is the average confluence time of a river section; is the number of river sections passed through; is the relative inflow termination time, is the inflow termination time. Starting from the beginning of the inflow as the time origin, then ; is the base of the natural logarithm; is the relative time counted from the inflow termination time.

[0012] In one implementation, in step S3, when constructing a surrogate model for the response of upstream reservoir regulation in the basin to water level and flow changes in the middle and lower reaches based on the deep learning method, the model used is a fitting type of recurrent neural network.

[0013] In one implementation, the mapping relationship constructed for the surrogate model of the response of upstream reservoir regulation in the basin to water level and flow changes in the middle and lower reaches in step S3 is:

[0014] Among them, , , , is the flow change of the downstream control station in the middle and lower reaches of the basin under the influence of upstream reservoir regulation; is the natural flow of the upstream control station in the basin; is the measured flow of the upstream control station in the basin; is the storage regulation volume of the upstream reservoir in the basin; is the measured flow of other tributaries; is the measured water level of the downstream control station in the middle and lower reaches of the basin considering the time lag; is the mapping function of the storage regulation volume of the upstream reservoir in the basin and the response of the water level and flow change in the middle and lower reaches; is the hydrodynamic model for the flow evolution from the upstream control station in the basin to the downstream control station in the middle and lower reaches; is the hydrodynamic model for the evolution of the storage variable of the reservoir to the upstream control station in the basin, where, is the total number of reservoirs above the control station, is the serial number of the reservoir, is the reservoir 's storage variable.

[0015] In one implementation, the method further includes: using the prediction result obtained in step S4 to evaluate the response law of the water level and flow change in the middle and lower reaches of the basin under different storage regulation scenarios of the upstream reservoir in the basin.

[0016] Based on the same inventive concept, the second aspect of the present invention provides a deep learning device for evaluating the water level and flow change in the middle and lower reaches of the basin, including: A data collection and preprocessing module for collecting and preprocessing hydrological section data; A flow restoration calculation and evolution module for restoring the measured flow of the upstream control station in the basin to the natural flow using the flow restoration calculation method, further calculating the storage regulation change amount of the upstream reservoir in the basin, and evolving the measured flow and natural flow of the upstream control station in the basin to the downstream control station in the middle and lower reaches through a one-dimensional hydrodynamic model to obtain the flow change value of the downstream control station in the middle and lower reaches affected by the upstream reservoir regulation; A model construction module for constructing a surrogate model of the storage regulation of the upstream reservoir in the basin and the response of the water level and flow change in the middle and lower reaches based on the deep learning method, where the input of the model is the natural flow of the upstream main stream in the basin, the measured flow of other tributaries, the measured water level of the downstream control station in the middle and lower reaches considering the time lag, and the storage regulation change amount of the upstream reservoir in the basin, and the output of the model is the flow change value of the downstream control station in the middle and lower reaches affected by the upstream reservoir regulation; A training and testing module for training the surrogate model of the storage regulation of the upstream reservoir in the basin and the response of the water level and flow change in the middle and lower reaches, and using the trained model to predict the water level and flow change in the middle and lower reaches of the basin.

[0017] In one embodiment, the model construction module is further configured to determine the mapping relationship between the input and output variables of the model, where the inputs of the model are the natural flow of the upper reaches of the main stream of the basin, the measured flows of other tributaries, the measured water levels of the control stations in the middle and lower reaches of the basin considering the time lag, and the regulated storage change of the upstream reservoirs in the basin, and the output of the model is the change value of the flow affected by the upstream reservoir regulation at the control stations in the middle and lower reaches of the basin; Based on the same inventive concept, the third aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the deep learning method for evaluating the water level and flow changes in the middle and lower reaches of the basin described in the first aspect.

[0018] Based on the same inventive concept, the fourth aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, it implements the deep learning method for evaluating the water level and flow changes in the middle and lower reaches of the basin described in the first aspect.

[0019] Compared with the prior art, the advantages and beneficial technical effects of the present invention are as follows: 1. The core of the present invention lies in using deep learning technology, combining measured hydrological and hydraulic data and flood routing models, to accurately simulate the regulation effect of the upstream reservoir group in the basin, accurately predict the regulation of the upstream reservoirs in the basin and the water level and flow changes in the middle and lower reaches, and evaluate its impact on the hydrological situation of the main flood control nodes in the middle and lower reaches of the basin; 2. The present invention constructs a proxy model for the response of the upstream reservoir regulation in the basin and the water level and flow changes in the middle and lower reaches based on deep learning, which can achieve the instant response of the hydrological situation in the middle and lower reaches under different regulation scenarios of the upstream reservoirs in the basin, and meet the real-time and rapid response requirements during the study of dispatching strategies; 3. The present invention can be used to integrate and apply multi-source heterogeneous data, and by combining the real-time monitoring data of hydrology and hydraulics and the water level and flow data of key control stations simulated by hydrodynamic models, it can achieve the efficient integration and full utilization of diverse data. Description of the Drawings

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0021] Figure 1 is the flowchart of the deep learning method for evaluating the water level and flow changes in the middle and lower reaches of the basin provided by the embodiment of the present invention; Figure 2It is a comparison chart of the simulated response flow change and the actual response flow change of the control station in the middle and lower reaches of the basin provided by the embodiment of the present invention; Figure 3 It is a schematic structural diagram of a deep learning device for evaluating the water level and flow changes in the middle and lower reaches of the basin in the embodiment of the present invention. Detailed implementation manners

[0022] The present invention discloses a deep learning method and device for evaluating the water level and flow changes in the middle and lower reaches of the basin. The method includes: collecting and preprocessing hydrological section data; calculating and evolving the flow restoration of the control station in the upper reaches of the basin, restoring the measured flows of the control stations in the upper and middle and lower reaches of the basin to natural flows, and calculating the regulation change amount of the reservoir in the upper reaches of the basin and the flow change value of the control stations in the middle and lower reaches of the basin affected by the regulation of the upstream reservoir by a one-dimensional hydrodynamic model; constructing a proxy model for the response of water level and flow changes, and constructing a proxy model for the response of the regulation of the upstream reservoir in the basin and the water level and flow changes in the middle and lower reaches based on deep learning; dividing the data set and testing and applying the proxy model, dividing the time series data into a training set, a validation set, and a test set, and finally obtaining the flow change value of the control stations in the middle and lower reaches of the basin affected by the regulation of the upstream reservoir.

[0023] Combining the measured hydraulic situation data and the flood routing model, the present invention constructs a proxy model for the response of the regulation of the upstream reservoir in the basin and the water level and flow changes in the middle and lower reaches based on deep learning, which can realize the instant response of the hydrological situation in the middle and lower reaches under different regulation scenarios of the upstream reservoir in the basin and meet the real-time and rapid response requirements during the study of dispatching strategies.

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0025] Embodiment 1 This embodiment discloses a deep learning method for evaluating the water level and flow changes in the middle and lower reaches of the basin. Please refer to Figure 1 , including: Step S1: Collect and preprocess hydrological section data; Step S2: Use the flow restoration calculation method to restore the measured flow of the control station in the upper reaches of the basin to the natural flow, further calculate the regulation change amount of the reservoir in the upper reaches of the basin, and evolve the measured flow and natural flow of the control station in the upper reaches of the basin to the control stations in the middle and lower reaches of the basin through a one-dimensional hydrodynamic model to obtain the flow change value of the control stations in the middle and lower reaches of the basin affected by the regulation of the upstream reservoir; Step S3: Based on the deep learning method, construct a surrogate model for the response of reservoir regulation in the upper reaches of the basin and the water level and flow changes in the middle and lower reaches. The inputs of the model are the natural flow in the upper reaches of the main stream of the basin, the measured flows of other tributaries, the measured water levels at the control stations in the middle and lower reaches of the basin considering the time lag, and the change in reservoir regulation in the upper reaches of the basin. The output of the model is the change in flow at the control stations in the middle and lower reaches of the basin affected by the reservoir regulation in the upper reaches. Step S4: Train the surrogate model for the response of reservoir regulation in the upper reaches of the basin and the water level and flow changes in the middle and lower reaches, and use the trained model to predict the water level and flow changes in the middle and lower reaches of the basin.

[0026] In one implementation, step S1 includes: Collect the measured water levels and flow data of the flow measurement sections of each hydrological station in the upper and lower reaches of the basin, including the measured flow in the upper reaches of the main stream of the basin, the measured flows of other tributaries, the measured flows and water levels at the control stations in the middle and lower reaches of the basin. Clean the collected data and information to form time series data.

[0027] In the specific implementation process, cleaning the collected data and information includes processing missing values and outliers.

[0028] In one implementation, step S2 includes: Adopt the flow restoration calculation method to calculate the reservoir inflow before regulation for the regulation volume of the reservoirs in the upper reaches of the basin according to the water balance equation respectively. Based on the reservoir inflow before regulation, calculate the evolved flow at the outlet section of the control station in the upper reaches of the basin according to the Changjiang Water Resources Commission's confluence curve formula to complete the river flood routing from the reservoir dam site in the upper reaches of the basin to the control section. Superimpose the evolved flow of each reservoir at the control section and the measured flow at the control section to obtain the restored natural flow at the control section of the upper reaches of the basin. According to the difference between the natural flow and the measured flow, obtain the change in reservoir regulation in the upper reaches of the basin. The flow obtained by restoring the natural flow evolution is the flow before regulation, and the flow obtained by the measured flow evolution is the flow after regulation. According to the difference between the flow before regulation and the flow after regulation, obtain the change in flow at the control stations in the middle and lower reaches of the basin affected by the reservoir regulation in the upper reaches of the basin.

[0029] Specifically, to maintain the water balance of floods in each sub-region, the annual storage variable of each reservoir upstream of the hydrological station is calculated to the hydrological station year by year and the annual restored flow of the hydrological station is calculated in a superimposed manner.

[0030] In this implementation process, taking the Yangtze River Basin and the Three Gorges and its upstream reservoirs as examples, the flow restoration calculation method and the evolution process of the one-dimensional hydrodynamic model are described. The flow restoration calculation method is only carried out for the upstream control stations of the main stream of the Yangtze River. It calculates the storage volume of the Three Gorges Reservoir and the reservoirs upstream of the Three Gorges respectively according to the water balance equation to obtain the reservoir inflow before storage regulation, and then calculates the evolved flow at the outlet section of the upstream control stations of the main stream of the Yangtze River according to the Changjiang Water Conservancy Commission's confluence curve formula, thus completing the river flood evolution from the dam site to the control section.

[0031] Adding the evolved flow of each reservoir at the control section to the measured flow at the control section, the natural flow restored at the upstream control station section of the main stream of the Yangtze River is obtained. Further, subtracting the measured flow from the natural flow gives the storage regulation change amount of the Three Gorges and its upstream reservoirs.

[0032] Then, based on the one-dimensional hydrodynamic model, the restored natural flow and the measured flow at the upstream control station section of the main stream of the Yangtze River are respectively evolved to the control stations in the middle and lower reaches of the Yangtze River, and the flow change value affected by the storage regulation of the Three Gorges and its upstream reservoirs at the control stations in the middle and lower reaches of the Yangtze River can be obtained. That is, the evolved restored natural flow is the flow before storage regulation, and the evolved measured flow is the flow after storage regulation. Subtracting the latter from the former gives the flow change value affected by the storage regulation of the reservoir group.

[0033] In one implementation manner, the storage volume of the upstream reservoirs in the basin is calculated according to the water balance equation by using the flow restoration calculation method to obtain the reservoir inflow before storage regulation, including: based on the reservoir water level, storage capacity curve and outflow, the inflow is calculated by inverse deduction using the water balance method. Specifically:

[0034] In the formula: is the average inflow in the time period, with the unit of m 3 / s; : the average outflow in the time period, which is obtained by adding the power generation flow, idling flow, ship lock passing flow and gate waste water flow to get the outflow, with the unit of m 3 / s; : the water loss of the reservoir, including the water surface evaporation and the leakage loss in the reservoir area, etc., with the unit of m 3 ; : the change value of the reservoir water storage at the beginning and end of the time period, with the unit of m 3 ; : the calculation time period, with the unit of s; The Changjiang Water Conservancy Commission's confluence curve formula is:

[0035] In the formula: is the ordinate value of the confluence parameter of the river reach, called the confluence parameter; is the relative time, is the time; is the average confluence time of a river reach; is the number of river reaches passed through; is the relative inflow termination time, is the inflow termination time. Taking the start of inflow as the time origin, then ; is the base of the natural logarithm; is the relative time counted from the inflow termination time.

[0036] Specifically, above the upstream control station of the main stream of the Yangtze River in this embodiment, the storage variable evolution needs to be carried out for each reservoir step by step, and the specific calculation is as follows: According to the reservoir water level, storage capacity curve and outflow discharge, the inflow discharge is inversely calculated based on the water balance method.

[0037] The Changjiang Water Resources Commission confluence curve formula is a river confluence formula. It describes the outflow process generated at the outflow section after a unit inflow of a certain duration is injected into a certain section of the river and undergoes the regulation of several river reaches. The specific formula is as above. In the river flood routing, the Changjiang Water Resources Commission confluence parameters divide the calculation period into equal value river reaches, which is convenient for considering the position of the tributary (interval) inflow point within the calculation river reach, and the routing result is more in line with the objective situation; , is determined by an optimization method to make the calculated discharge fit well with the measured discharge . It is measured by the following three indicators: Flood process deviation : ; Maximum flood peak discharge deviation : ; Peak appearance time deviation : .

[0038] In the formula: is the discharge at the outflow section of the th river reach calculated by the Changjiang Water Resources Commission confluence curve formula, with the unit of m 3 / s; is the measured discharge at the outflow section of the th river reach, with the unit of m 3 / s; is the number of river reaches passed through; is the calculated value of the maximum flood peak discharge, with the unit of m 3 / s; is the measured value of the maximum flood peak discharge, with the unit of m 3 / s; For calculating the occurrence time of the maximum flood peak flow rate; Is the occurrence time of the measured maximum flood peak flow rate.

[0039] In one implementation, when constructing a surrogate model for the regulation of upstream reservoirs in the basin and the response of water levels and flow rates in the middle and lower reaches based on a deep learning method in step S3, the model used is a fitting-type recurrent neural network.

[0040] In the specific implementation process, the fitting-type recurrent neural networks include long short-term memory networks (LSTM), gated recurrent units (GRU), bidirectional long short-term memory networks (Bi-LSTM), deep long short-term memory networks (Deep LSTM), Transformer, etc.

[0041] In one implementation, for the surrogate model of the regulation of upstream reservoirs in the basin and the response of water levels and flow rates in the middle and lower reaches constructed in step S3, the mapping relationship is:

[0042] Among them, , , , Is the flow rate change at the control station in the middle and lower reaches of the basin under the influence of upstream reservoir regulation, with the unit m 3 / s; Is the natural flow rate at the upstream control station of the basin, with the unit m 3 / s; Is the measured flow rate at the upstream control station of the basin, with the unit m 3 / s; Is the regulation volume of the upstream reservoir in the basin, with the unit m 3 / s; Is the measured flow rate of other tributaries, with the unit m 3 / s; Is the measured water level at the control station in the middle and lower reaches of the basin considering the lag time, with the unit m; Is the mapping function of the regulation volume of the upstream reservoir in the basin and the response of water levels and flow rates in the middle and lower reaches; Is the hydrodynamic model for the flow rate evolution from the upstream control station of the basin to the control station in the middle and lower reaches; Is the hydrodynamic model for the evolution of the storage variable of the upstream reservoir in the basin to the upstream control station of the basin, where Is the total number of reservoirs above the control station, including the Three Gorges and its upstream reservoirs in the specific implementation, Is the serial number of the reservoir, Is the reservoir 's storage variable, with the unit m 3 .

[0043] In one embodiment, the method further includes: using the prediction result obtained in step S4 to evaluate the response law of the water level and flow rate changes in the middle and lower reaches of the basin under different regulation scenarios of the upstream reservoir of the basin.

[0044] In the specific implementation process, for the division of the data set and the test application of the surrogate model, the input and output data sets formed are subjected to time series normalization processing, and the time series data is divided into a training set, a validation set, and a test set. Based on the validation set and combined with the hyperparameter calibration method, the hyperparameters of the deep learning model are optimized. The model after hyperparameter calibration is applied to the training set for training within the number of iterations of the model. Then, the test set is used to test the model. At this time, the backpropagation of the hyperparameters and network parameters of the deep learning model is no longer adjusted. Finally, the flow rate change value affected by the upstream reservoir regulation at the control station in the middle and lower reaches of the basin is obtained, and the response law of the water level and flow rate changes in the middle and lower reaches of the basin under different regulation scenarios of the upstream reservoir of the basin is evaluated.

[0045] The model loss function uses the root mean square error between the simulated response flow rate change and the actual response flow rate change at the control station in the middle and lower reaches of the basin; the Nash-Sutcliffe efficiency coefficient NSE and the root mean square error RMSE are used to evaluate the surrogate model of the upstream reservoir regulation of the basin and the response of the water level and flow rate changes in the middle and lower reaches.

[0046] To facilitate those skilled in the art to better understand the technical solution of the present invention, the following are specific embodiments of the present invention (taking the Yangtze River Basin and the Three Gorges Reservoir as examples): The embodiment of the present invention specifically provides a deep learning method and device for evaluating the water level and flow rate changes in the middle and lower reaches of the basin, including the following steps: (1) Collect and analyze the measured water level and flow rate data of the flow measurement sections of each hydrological station in the upper and lower reaches of the Yangtze River from 2008 to 2021, including the measured flow rate of the upper reaches of the main stream of the Yangtze River, the measured flow rate of other tributaries, the measured flow rate and water level of the control stations in the middle and lower reaches of the Yangtze River, specifically, the measured flow rate of the upstream control station (Yichang Station) of the main stream of the Yangtze River, the measured flow rate of other tributaries (taking the measured flow rate of Gaobazhou Station and the measured flow rate of the four Dongting Rivers as examples), the measured flow rate of the control station in the middle and lower reaches of the main stream of the Yangtze River (taking Luoshan Station as an example), and the measured water level of the control station in the middle and lower reaches of the main stream of the Yangtze River (taking Lianhuatang as an example); process the missing values and outliers of the above data and form time series data; (2) According to the flow reduction calculation method, restore the measured flow rate of Yichang Station to the natural flow rate, and calculate the regulation change amount of the Three Gorges and its upstream reservoirs; further, from the measured flow rate and natural flow rate of Yichang Station, use DHI MIKE11 software to construct a flow calculation model for the middle and lower reaches of the Yangtze River. According to the river channel morphology and the measured data of water level and flow rate, calibrate the roughness coefficient of the one-dimensional hydrodynamic model in three sections for different water levels, and finally obtain the flow rate change value affected by the regulation of the Three Gorges and its upstream reservoirs at Luoshan Station; (3) Based on the deep learning method, a surrogate model for the regulation of the Three Gorges and its upstream reservoirs and the response of water levels and flows in the middle and lower reaches is constructed. The model inputs are the natural flow of Yichang Station, the control station in the upper reaches of the main stream of the Yangtze River, the regulation change of the Three Gorges and its upstream reservoirs, the measured flow of Gaobazhou Station, the measured flow of the four rivers in Dongting Lake, and the measured water levels at two previous moments in front of Lianhuatang. The model output is the change value of the flow at Luoshan Station affected by the regulation of the Three Gorges and its upstream reservoirs. The long short-term memory network (LSTM) model is used for the research. (4) Normalize the input and output data sets in time series, divide the time series data into training set, validation set, and test set. Based on the validation set and combined with the hyperparameter calibration method, optimize the hyperparameters of the deep learning model. The hyperparameters are the number of hidden layers, the number of nodes in the hidden layer, the time step, the training batch size, the number of iterations, and the learning rate. Use the NSE as the calibration function of the hyperparameters. The final obtained hyperparameters are shown in Table 1. Table 1 Hyperparameters of the surrogate model for the regulation of the Three Gorges and its upstream reservoirs and the response of water levels and flows in the middle and lower reaches

[0047] Apply the model with calibrated hyperparameters to the training set and train the model within the number of iterations. Then, use the test set to test the model, and at this time, do not adjust the hyperparameters and the backpropagation of the network parameters of the deep learning model. Finally, obtain the change value of the flow at Luoshan Station affected by the regulation of the Three Gorges and its upstream reservoirs. Compare the simulated response flow change and the actual response flow change at Luoshan Station. The model simulation accuracy indicators of the training set, validation set, and test set are shown in Table 2. It can be seen that in the simulation of the flow change at Luoshan Station affected by the regulation of the Three Gorges and its upstream reservoirs, the accuracy indicators NSE and RMSE perform well, indicating that the deep learning model can better fit the flow change at Luoshan Station under the influence of the regulation of the Three Gorges and its upstream reservoirs.

[0048] Table 2 Accuracy of the surrogate model for water level and flow change response

[0049] (6) Compare the simulated response flow change and the actual response flow change at Luoshan Station obtained by the surrogate model for water level and flow change response. The results are as Figure 2 shown. It can be seen that the surrogate model can better simulate the flow change at Luoshan Station under the influence of the regulation of the Three Gorges and its upstream reservoirs.

[0050] Embodiment 2 Based on the same inventive concept, this embodiment discloses a deep learning device for evaluating water level and flow changes in the middle and lower reaches of a basin. Please refer to Figure 3 , including: A data collection and preprocessing module 101 for collecting and preprocessing hydrological section data. The flow restoration calculation and evolution module 102 is used to restore the measured flow of the upstream control station in the basin to the natural flow by using the flow restoration calculation method, further calculate the regulation and storage change of the Three Gorges and its upstream reservoirs, and evolve the measured flow and natural flow of the upstream control station in the basin to the middle and lower reaches control stations in the basin through a one-dimensional hydrodynamic model to obtain the flow change value affected by the upstream reservoir regulation and storage at the middle and lower reaches control stations in the basin; The model construction module 103 is used to construct a proxy model for the response of the upstream reservoir regulation and storage in the basin to the water level and flow changes in the middle and lower reaches based on the deep learning method. The inputs of the model are the natural flow of the upstream main stream in the basin, the measured flows of other tributaries, the measured water levels of the middle and lower reaches control stations in the basin considering the time lag, and the regulation and storage change of the upstream reservoirs in the basin, and the output of the model is the flow change value affected by the upstream reservoir regulation and storage at the middle and lower reaches control stations in the basin; The training and testing module 104 is used to train the proxy model for the response of the upstream reservoir regulation and storage in the basin to the water level and flow changes in the middle and lower reaches, and use the trained model to predict the water level and flow changes in the middle and lower reaches of the Yangtze River.

[0051] Since the device introduced in the second embodiment of the present invention is the device used for the deep learning method for evaluating the water level and flow changes in the middle and lower reaches of the basin in the first embodiment of the present invention, based on the method introduced in the first embodiment of the present invention, those skilled in the art can understand the specific structure and deformation of the device, so it will not be elaborated here. Any device used for the method in the first embodiment of the present invention belongs to the scope of protection of the present invention.

[0052] Embodiment Three Based on the same inventive concept, the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the method described in Embodiment One.

[0053] Since the computer-readable storage medium introduced in the third embodiment of the present invention is the computer-readable storage medium used for the deep learning method for evaluating the water level and flow changes in the middle and lower reaches of the basin in the first embodiment of the present invention, based on the method introduced in the first embodiment of the present invention, those skilled in the art can understand the specific structure and deformation of the computer-readable storage medium, so it will not be elaborated here. Any computer-readable storage medium used for the method in the first embodiment of the present invention belongs to the scope of protection of the present invention.

[0054] Embodiment Four The present invention also provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the method described in Embodiment One.

[0055] Since the computer device introduced in the fourth embodiment of the present invention is the computer device used in the deep learning method for evaluating the water level and flow rate changes in the middle and lower reaches of the basin in the first embodiment of the present invention, based on the method introduced in the first embodiment of the present invention, those skilled in the art can understand the specific structure and variations of this computer device, so it will not be elaborated here. Any computer device used in the method of the first embodiment of the present invention falls within the scope of protection of the present invention.

[0056] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0057] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for realizing the specified functions in Figure 1 one or more of these processes or multiple processes and / or blocks Figure 1 one or more of these blocks or multiple blocks.

[0058] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present invention. Obviously, those skilled in the art can make various changes and variations to the embodiments of the present invention without departing from the spirit and scope of the embodiments of the present invention. Thus, if these modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these changes and variations.

Claims

1. A deep learning method for evaluating changes in water level and flow in the middle and lower reaches of a river basin, characterized in that: include: Step S1: Collect and preprocess hydrological section data; Step S2: Use the flow restoration calculation method to restore the measured flow of the upstream control station of the basin to the natural flow, further calculate the change in the regulation and storage of the reservoir, and use the one-dimensional hydrodynamic model to evolve the measured flow and natural flow of the upstream control station of the basin to the mid- and downstream control stations of the basin, and obtain the flow change value of the mid- and downstream control stations affected by the regulation and storage of the upstream reservoir of the basin; Step S3: Based on the deep learning method, a proxy model of the upstream reservoir regulation and the mid- and downstream water level and flow change response is constructed. The input of the model is the natural flow of the upstream of the main stream of the basin, the measured flow of other tributaries, the measured water level of the mid- and downstream control stations of the basin considering the time lag, and the change of the upstream reservoir regulation. The output of the model is the flow change value of the mid- and downstream control stations of the basin affected by the regulation of the upstream reservoir; Step S4: Train the proxy model for the regulation of reservoirs in the upper reaches of the basin and the response of water level and flow changes in the middle and lower reaches, and use the trained model to predict the changes in water level and flow in the middle and lower reaches.

2. The deep learning method for evaluating changes in water level and flow in the middle and lower reaches of a river basin as claimed in claim 1, characterized in that: Step S1 includes: Collect the measured water level and flow data of the flow measurement sections of each hydrological station upstream and downstream of the basin, including the measured flow of the upstream of the main stream of the basin, the measured flow of other tributaries, and the measured flow and water level of the control stations in the middle and lower reaches of the basin; Clean the collected data and information to form time series data.

3. The deep learning method for evaluating changes in water level and flow in the middle and lower reaches of a river basin as claimed in claim 1, characterized in that: Step S2 includes: The flow restoration calculation method is used to calculate the reservoir inflow before regulation by using the water balance equation to calculate the regulation and storage capacity of the Three Gorges Reservoir and the reservoir upstream of the Three Gorges. Calculate the evolution flow at the outlet section of the upstream control station according to the Changban confluence curve formula, and complete the river flood evolution from the dam site to the control section; The evolved flow of each reservoir at the control section is superimposed with the measured flow of the control section to obtain the natural flow restored at the section of the upstream control station of the basin. The storage change of the reservoir is obtained based on the difference between the natural flow and the measured flow. The flow before regulation is obtained by restoring the evolution of natural flow, and the flow after regulation is obtained by the evolution of measured flow. According to the difference between the flow before regulation and the flow after regulation, the flow change value of the control station in the middle and downstream of the basin affected by the regulation of the upstream reservoir is obtained.

4. The deep learning method for evaluating changes in water level and flow in the middle and lower reaches of a river basin as claimed in claim 3, characterized in that: The flow restoration calculation method is used to calculate the reservoir inflow before regulation based on the water balance equation for the storage capacity of the Three Gorges Reservoir and the reservoir upstream of the Three Gorges. This includes: according to the reservoir water level, storage capacity curve and outflow, the inflow is calculated based on the water balance method, specifically: Where: is the average inflow flow during the period, is the average outbound flow during the period, is the water loss in the reservoir, is the change in reservoir water storage at the beginning and end of the period; is the calculation period; The formula of the long-distance convergence curve is: Where: For the The ordinate value of the confluence parameter of a river section is called the confluence parameter; is the relative time, For time; is the average confluence time of a river section; is the number of river sections it flows through; is the relative inflow termination time, is the end time of inflow, and the start time of inflow is taken as the starting time. ; is the base of natural logarithms; It is the relative time measured from the time when the inflow stops.

5. The deep learning method for evaluating changes in water level and flow in the middle and lower reaches of a river basin as claimed in claim 1, characterized in that: In step S3, when constructing a proxy model for the regulation of reservoirs in the upper reaches of the basin and the response of water level and flow changes in the middle and lower reaches based on the deep learning method, the model used is a recursive neural network of the fitting class.

6. The deep learning method for evaluating changes in water level and flow in the middle and lower reaches of a river basin as claimed in claim 1, characterized in that: In step S3, the proxy model of the upstream reservoir regulation and the mid- and downstream water level and flow change response in the basin has a mapping relationship constructed as follows: in, , , , The flow changes of the control stations in the middle and lower reaches of the basin under the influence of the regulation of the upstream reservoir; The natural flow at the upstream control station of the basin; is the measured flow at the upstream control station in the basin; To regulate the storage capacity of reservoirs in the upper reaches of the basin; Measure flows for other tributaries; To take into account the time lag, the measured water level at the control station in the middle and downstream of the basin; It is the mapping function between the storage capacity of the upstream reservoir and the response of the water level and flow change in the middle and lower reaches of the basin; A hydrodynamic model for the evolution of flow from the upstream control station to the mid- and downstream control stations in the basin; is the hydrodynamic model of the evolution of the reservoir storage variable to the upstream control station of the basin, where: is the total number of reservoirs above the control station, is the reservoir number, For the reservoir The storage variable.

7. The deep learning method for evaluating changes in water level and flow in the middle and lower reaches of a river basin as claimed in claim 1, characterized in that: The method further comprises: using the prediction result obtained in step S4 to evaluate the response law of the water level and flow change in the middle and lower reaches of the basin under different regulation and storage scenarios of the upstream reservoir.

8. A deep learning device for evaluating changes in water level and flow in the middle and lower reaches of a river basin, characterized in that: include: Data collection and preprocessing module, used to collect and preprocess hydrological section data; The flow restoration calculation and evolution module is used to restore the measured flow of the upstream control station of the basin to the natural flow by using the flow restoration calculation method, further calculate the change in the regulation and storage of the reservoir, and evolve the measured flow and natural flow of the upstream control station of the basin to the mid- and downstream control stations of the basin through a one-dimensional hydrodynamic model to obtain the flow change value of the mid- and downstream control stations of the basin affected by the regulation and storage of the upstream reservoir; The model building module is used to build a proxy model for the regulation of upstream reservoirs and the response of water level and flow changes in the middle and lower reaches of the basin based on deep learning methods. The input of the model is the natural flow of the upstream of the main stream of the basin, the measured flow of other tributaries, the measured water level of the control station in the middle and lower reaches of the basin considering the time lag, and the change in regulation of the upstream reservoir in the basin. The output of the model is the flow change value of the control station in the middle and lower reaches of the basin affected by the regulation of the upstream reservoir. The training and testing module is used to train the proxy model for the regulation of reservoirs in the upper reaches of the basin and the response of water level and flow changes in the middle and lower reaches, and use the trained model to predict the changes in water level and flow in the middle and lower reaches.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the deep learning method for evaluating changes in downstream water level and flow in a river basin as described in any one of claims 1 to 7 is implemented.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the deep learning method for evaluating changes in downstream water level and flow in a river basin as described in any one of claims 1 to 7 is implemented.

Images