Drainage basin flood control scheduling scheme optimization method based on neural network

Abstract

The invention discloses a drainage basin flood control scheduling scheme optimization method based on a neural network. The method specifically comprises the following steps: S1, constructing a hydrological and hydrodynamic coupling model of a drainage basin; s2, performing simulation calculation by using the hydrological and hydrodynamic coupling model constructed in the step S1, and preparing training data required by a neural network model; s3, constructing a neural network model containing scheduling information, and performing neural network model training and verification by using the training data in the step S2; s4, based on the neural network model trained in the step S3, according to a preset target water level, selecting an optimal flood control scheduling scheme by using an optimization algorithm; according to the method, the neural network technology is introduced, the computational efficiency bottleneck of a traditional hydrological hydrodynamic model in flood control scheduling is overcome, the scheduling scheme optimization speed and accuracy are remarkably improved, and a new solution is provided for coping with extreme flood events.

Description

Technical Field

[0001] The invention relates to the technical field of river basin flood control scheduling, and in particular to a method for optimizing river basin flood control scheduling scheme based on a neural network. Background Art

[0002] At present, with the intensification of global climate change, the frequency and intensity of rainstorm events have increased, and extreme flood events have posed a serious threat to social and economic development and the safety of people's lives and property. Floods, as a natural disaster, not only affect the ecological environment of the river basin, but may also trigger secondary disasters and cause huge social and economic losses. Therefore, how to effectively respond to flood risks and scientifically and rationally formulate flood control and dispatching plans has become an important issue in current disaster prevention and mitigation work.

[0003] In watershed management, constructing a hydrological and hydrodynamic model based on basic information of the watershed is an important means of flood forecasting and scheduling. By simulating the hydrological cycle process in the watershed, the model can accurately predict the occurrence and development of floods and their impact on the water level of each section in the watershed, thereby providing a reliable decision-making basis for disaster prevention and mitigation. However, the current hydrological and hydrodynamic models still face some challenges in practical applications, especially in terms of computational efficiency. Due to their complexity and high precision requirements, traditional hydrological and hydrodynamic models often require a lot of computing resources and time, especially when multiple gate scheduling schemes are repeatedly tested under the preset target water level to evaluate their impact on the water level of the downstream section, the calculation time increases significantly. This inefficient calculation process seriously affects the timeliness of flood control scheduling decisions, making it difficult to respond quickly when floods come, increasing the flood control risk in the watershed, and urgently needs to be improved. Summary of the invention

[0004] The problem to be solved by the present invention is to provide a method for optimizing flood control scheduling schemes in a river basin based on a neural network. This method utilizes the powerful learning and prediction capabilities of neural network technology, and trains a neural network model based on historical precipitation data and scheduling scheme parameters to predict the water level change process of the downstream section under different scheduling schemes. Compared with traditional methods, the neural network model can quickly predict the water level change process in a short time, thereby greatly improving the efficiency of the scheduling scheme trial calculation. Through this method, the optimal flood control scheduling scheme can be quickly evaluated and screened under the set target water level conditions, which not only improves the calculation efficiency and reduces the consumption of computing resources, but also significantly improves the timeliness of decision-making, making the river basin flood control scheduling more scientific and efficient, and providing strong technical support for river basin disaster prevention and mitigation work.

[0005] A technical solution adopted by the present invention to solve the above problem is: a method for optimizing a flood control scheduling scheme for a river basin based on a neural network, which specifically includes the following steps: S1. Construct a hydrological and hydrodynamic coupling model of the basin; S2, using the hydrological and hydrodynamic coupling model constructed in step S1 to perform simulation calculations and prepare training data required for the neural network model; S3, constructing a neural network model including scheduling information, and using the training data of step S2 to train and verify the model; S4, based on the neural network model trained in step S3, according to the preset target water level, use the optimization algorithm to select the optimal flood control scheduling plan; The preparation of training data in step S2 specifically includes: Abstraction and generalization of gate joint dispatching rules: generalize the classic dispatching scheme of a single gate, discretize the gate opening, and propose a formula for gate joint dispatching rules for N gates, N≥1; Scaling of historical rainfall events: Collect typical precipitation events in the basin over a long period of time, intercept the n-day precipitation duration in the hourly precipitation process, form a T-hour time series, fill zeros at the end for precipitation processes that are less than T hours long, and scale precipitation events according to different recurrence periods. ; Obtain the water level change process data of important downstream sections under different meteorological conditions and gate joint dispatching rules: input the outflow calculation results of the runoff generation zone under different precipitation scenarios into the one- and two-dimensional hydrodynamic model as boundary conditions to perform hydrological and hydrodynamic coupling calculations; in the hydrodynamic model, for N gates, apply different gate joint dispatching schemes to obtain the water level process of the downstream section under different scenarios; Step S4 is specifically as follows: the gate opening combination in the gate joint scheduling scheme is regarded as a decision variable, and the neural network model trained in step S3 is embedded into the target optimization iteration process of the optimization algorithm by using the optimization algorithm, and the downstream control section water level under different gate joint scheduling rules is iteratively optimized; based on the objective function and constraints, the optimal gate joint scheduling scheme is determined as the optimal flood control scheduling scheme.

[0006] Compared with the prior art, the advantages of the present invention are: (1) the neural network model is used to construct an alternative model of the hydrological and hydrodynamic model, which overcomes the problem of time-consuming calculation of the hydrological and hydrodynamic model, realizes rapid prediction of water level change process in a short time, and effectively improves calculation efficiency.

[0007] (2) Based on the neural network model, efficient reverse calculation is realized, which can quickly evaluate and screen out the optimal flood control scheduling plan under the set target water level conditions; it overcomes the computational efficiency bottleneck of traditional hydrological and hydrodynamic models in flood control scheduling, significantly improves the speed and accuracy of scheduling scheme optimization, and provides strong technical support for basin disaster prevention and mitigation work.

[0008] (3) In preparing the training data for the neural network model, the gate opening is discretized to reflect the response behavior of artificial buildings under different scheduling schemes, which greatly reduces the computational complexity of the neural network model. The historical rainfall events are scaled to cover most of the precipitation processes in reality, thereby improving the accuracy of the neural network model.

[0009] As a preferred method, the abstract generalization of the gate joint dispatching rule is as follows: the gate opening is used as a dispatching parameter, and is discretized in the interval [0,1] with a step size of is 0.2, which reflects the response behavior of artificial buildings under different scheduling schemes; Among them, the gate joint dispatching rule formula for N gates is as follows: when When the gate opening of each station And the gate opening combination in the gate joint dispatching rule is expressed as follows: In the formula, represents the i-th joint scheduling combination in the scheduling scheme group; represents the gate opening of gate No. j, Represents a scheduling plan The gate opening of gate No. j in the middle; represents the calculated water level at gate No. j; represents the control water level threshold at the jth gate, and G is the gate opening combination in the gate joint scheduling rule.

[0010] As a preferred method, the scaling of historical rainfall events is as follows: typical precipitation events in the basin over a long period of time are collected, and the precipitation duration of n days in the hourly precipitation process is intercepted to form a time series of T hours. For precipitation processes with a duration of less than T hours, zero is added at the end. The specific formula is as follows: in, is the processed precipitation time series; is the original precipitation time series; is the time length of the original precipitation series; The precipitation events are scaled according to different return periods to cover most precipitation processes in reality; the formula is as follows: In the formula, is the scaled precipitation time series; is the scaling factor.

[0011] As a preferred embodiment, the water level process in step S2 The simulation calculation formula is as follows: Where M0 is the hydrological and hydrodynamic coupling model constructed in step S1, B is the remaining boundary conditions, and C is the neural network model parameters and initial conditions.

[0012] Preferably, step S3 comprises the following steps: S31. Clarify the input and output matrix of the neural network, the input includes: the outflow Q of the runoff zone obtained by the precipitation-driven hydrological model, the gate opening combination G in the gate joint scheduling rule; the output is the cross-section water level process H calculated by the hydrological and hydrodynamic coupling model; The input matrix X can be expressed as: in, represents the outflow matrix of the runoff-producing partition, represents the gate combination rule matrix, M1 is the number of flow-generating partitions, and M2 is the number of gates for joint scheduling; the input matrix X can be expressed in detail as follows: The output matrix Y can be expressed as: In the formula, The target section water level at time t calculated by the hydrological and hydrodynamic coupling model; S32. Abstract a general framework suitable for neural network models, which can be flexibly applied to long short-term memory networks LSTM, gated recurrent units GRU and fully connected neural networks FCNN; The hidden layer state update formula is: The output generation formula is: The loss function calculation formula is: In the formula, It is represented as the hidden layer state at time step t, F is the feature extraction and state update function, g is the mapping function from the hidden state sequence to the target output, are the parameters of the neural network model, Mapping parameters to control output; S33. Divide the constructed data set into training set and validation set in proportion, of which 70% is used to train the neural network model and 30% is used to verify the performance of the neural network model; in the process of training and verifying the neural network model, the hyperparameters such as the number of hidden layers, the number of neurons in the hidden layer, the number of iterations, the batch size, and the learning rate are optimized and selected on the training set using the leave-one-out cross-validation method to ensure the rationality of the neural network model parameters and the stability of the performance.

[0013] Preferably, in step S4: the optimization algorithm uses the NSGA-II algorithm, the objective function is to minimize the downstream water level deviation, and the constraint condition is that the highest water level of the downstream section shall not exceed the downstream section control water level; Among them, the objective function calculation formula is: In the formula, X represents the gate opening combination, is the objective function, is the trained neural network model, Indicates the controlled water level of the downstream section; Constraints: Optimal solution: In the formula, P represents the set of all possible solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 The present invention is a flowchart of a method for optimizing a watershed flood control scheduling scheme based on a neural network.

[0015] Figure 2 This is a generalized diagram of the modeling of the hydrological and hydrodynamic coupling model of a certain river basin in a certain city in an embodiment of the present invention.

[0016] Figure 3 This is a diagram showing the forecast effect of the water level of an important downstream section during the training period and the verification period under three different valve openings when the precipitation event is typhoon Mina scaled by 1 times in an embodiment of the present invention.

[0017] Figure 4 This is a diagram showing the forecast effect of the water level of an important downstream section during the training period and the verification period under three different valve openings when the precipitation event is 0.25 times the typhoon Chan-hom in an embodiment of the present invention.

[0018] Figure 5This is a diagram showing the forecast effect of the water level of an important downstream section during the training period and the verification period under three different valve openings when the precipitation event is 0.5 times the typhoon Xuanlannuo in an embodiment of the present invention.

[0019] Figure 6 This is a target section water level sequence under multiple different scheduling schemes when the precipitation event is Typhoon Chan-hom scaled 0.25 times in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0021] This preferred embodiment is for example Figure 1 The figure shows a method for optimizing flood control scheduling schemes in a river basin based on a neural network, which specifically includes: S1. Constructing a hydrological and hydrodynamic coupling model: For complex river basins with water conservancy dispatching needs, we collect basic geographic information data, basic water conservancy engineering data, and meteorological and hydrological data of the river basins, generalize the real river basins based on object-oriented thinking, and construct a hydrological and hydrodynamic coupling model. Figure 2 As shown in the figure, it is a generalized diagram of the hydrological and hydrodynamic coupling model of a certain river basin in a certain city, where HmRiver is the hydrological river channel; Node1D is a one-dimensional node; and Reach1D is a one-dimensional river channel.

[0022] Step S1 includes the following steps: S11. Based on the object-oriented concept, the real river basin is generalized to construct the basic unit of the hydrological and hydrodynamic coupling model. The hydrological model components include flow-generating zones, reservoirs, hydrological sections and hydrological channels, and the hydrodynamic model components include one-dimensional nodes, one-dimensional channels, one-dimensional sections, zero-dimensional reservoirs and two-dimensional zones. At the same time, based on the coupling between the hydrological model and the hydrodynamic model, and the coupling between the one-dimensional and two-dimensional components in the hydrodynamic model, the hydrological and hydrodynamic coupling model components are formed.

[0023] S12. Collect basic geographic information of the basin and basic data of key water conservancy facilities such as gates in the basin to complete the configuration of model attributes; at the same time, collect the initial state and boundary condition data required for the calculation of the hydrodynamic coupling model to complete the simulation configuration of the hydrodynamic coupling model. For the generalized model components, establish a water conservancy professional model library, including hydrological model algorithms covering runoff calculation, reservoir joint regulation and river channel calculation, as well as hydrodynamic model algorithms based on numerical methods such as finite difference and finite volume.

[0024] S13. In the process of model calculation, parallel solution technology is adopted. Through multi-level parallel strategies at component level, scenario level and scheme level, the multi-core computing power of hardware servers is fully utilized to achieve efficient solution of large-scale water conservancy models.

[0025] S2. Prepare neural network model training data: Use the hydrological and hydrodynamic coupling model to perform simulation calculations and prepare the training data required for the neural network model (also known as the basin flood forecasting neural network model), including the scaling of historical rainfall events and the abstract generalization of gate joint scheduling rules, as well as the water level change process data of important downstream sections under different meteorological conditions and joint scheduling rules.

[0026] Step S2 includes the following steps: S21. Generalize the classic scheduling scheme of a single gate. When the control object reaches or exceeds a certain control value (for example, the upstream water level is greater than 2m), configure the corresponding gate opening. The gate opening is used as a scheduling parameter and discretized in the interval [0,1] with a step size of is 0.2, which reflects the response behavior of artificial buildings under different scheduling schemes.

[0027] Specifically, the gate joint dispatch rule formula for N gates is as follows (N≥1): when When the gate opening of each station And the scheduling scheme combination is expressed as follows: In the formula, represents the i-th joint scheduling combination in the scheduling scheme group; represents the gate opening of gate No. j, Represents a scheduling plan The gate opening of gate No. j in the middle; represents the calculated water level at gate No. j; It represents the control water level threshold at the jth gate. Therefore, a gate joint dispatching scheme is formulated, the dispatching rules are standardized, and they are abstracted into a dispatching parameter combination.

[0028] S22. Collect typical precipitation events in the basin over a long period of time, intercept the n-day precipitation duration in the hourly precipitation process, form a T-hour time series, and fill in zeros at the end for precipitation processes that are less than T hours long. The specific formula is as follows: in, is the processed precipitation time series; is the original precipitation time series; is the time length of the original precipitation series.

[0029] The precipitation events are scaled according to different return periods to cover most precipitation processes in reality, and the meteorological boundary input conditions of the hydrological and hydrodynamic coupling model are generated in batches. The formula is as follows: In the formula, is the scaled precipitation time series; is the scaling factor.

[0030] S23. The outflow calculation results of the runoff generating zones calculated under different precipitation scenarios are input as boundary conditions into the one- and two-dimensional hydrodynamic model to perform hydrological and hydrodynamic coupling calculations. Then, in the hydrodynamic model, different gate joint scheduling schemes are applied to N gates to obtain the water level process of the downstream section under different scenarios.

[0031] Water level process The simulation calculation formula is as follows: Where M0 is the hydrological and hydrodynamic coupling model constructed in step S1, is the precipitation time series prepared in step S22, G is the gate opening combination in the gate joint scheduling rule, B is the other boundary conditions (such as evaporation, etc.), and C is the model parameters and initial conditions, etc.

[0032] In this embodiment, the scaling range is 0.25 to 2.0, the step size is 0.25, and 7 days of precipitation events are generated in batches as the boundary input conditions of the hydrological and hydrodynamic coupling model; in the hydrological model, the flow generation methods such as the Xin'anjiang model are used to calculate the flow generation, and the linear reservoir, unit line and other methods are used to calculate the confluence. The simple genetic algorithm (SGA) is used to calibrate the model parameters to obtain a set of optimal model parameters. The basic control equations of the hydrodynamic model are the one-dimensional Saint-Venant equation and the two-dimensional shallow water equation. The finite volume numerical solution method is used to perform one-dimensional and two-dimensional hydrodynamic calculations.

[0033] S3. Construct a neural network model with (including) scheduling information: take the outflow of runoff zones driven by meteorological boundary conditions and the joint scheduling parameters of gates as input, and the water level of important downstream sections calculated by the hydrological and hydrodynamic coupling model as the target output, conduct model training and verification, and realize high-precision flood control forecasting in artificially scheduled basins.

[0034] Step S3 includes the following steps: S31. Clarify the input and output matrices of the neural network. The input includes: the outflow Q of the runoff zone obtained by the precipitation-driven hydrological model, and the gate opening combination G in the gate joint scheduling rule; the output is the cross-section water level process H calculated by the hydrological and hydrodynamic model.

[0035] The input matrix X can be expressed as: in, represents the outflow matrix of the runoff-producing partition, represents the gate combination rule matrix, M1 is the number of flow-generating partitions, and M2 is the number of gates for joint scheduling. The input matrix X can be expressed in detail as follows: The output matrix Y can be expressed as: In the formula, It is the target section water level at time t calculated by the hydrological and hydrodynamic coupling model in step S23.

[0036] S32. Abstract a general framework suitable for neural network models, which can be flexibly applied to long short-term memory networks LSTM, gated recurrent units GRU and fully connected neural networks FCNN, etc.

[0037] The hidden layer state update formula is: The output generation formula is: The loss function calculation formula is: In the formula, It is represented as the hidden layer state at time step t, F is the feature extraction and state update function, g is the mapping function from the hidden state sequence to the target output, are the parameters of the neural network model, Mapping parameters to control output.

[0038] S33. Divide the constructed data set into training set and validation set in proportion, of which 70% is used to train the neural network model and 30% is used to validate the performance of the neural network model. During the training and validation process of the neural network model, the number of hidden layers (layers), number of neurons per hidden layer (number of neurons per hidden layer), number of iterations (n_epochs), batch size (batch size), learning rate (learning rate) and other hyperparameters are optimized and selected on the training set using the leave-one-out cross-validation method to ensure the rationality of the neural network model parameters and the stability of performance.

[0039] In this embodiment, the number of hidden layers is 3, the number of neurons in the hidden layer is set to 256, 128, and 64, the number of iterations is 3000, the batch size is 10, and the learning rate is 0.0001.

[0040] Figure 3 The forecast effects of water levels at important downstream sections during the training and validation periods are shown respectively under three different valve openings when the precipitation event is scaled 1 times based on Typhoon Mina. Figure 4 The forecast effects of water levels at important downstream sections during the training and validation periods are shown respectively under three different valve openings when the precipitation event is 0.5 times that of Typhoon Xuanlanno. Figure 5 The forecast effects of the downstream important section water levels under three different valve openings during the training and validation periods are shown respectively, when the precipitation event is scaled 0.5 times based on Typhoon Chan-hom. Taking the header "Chan-hom_0.25_0.8" as an example, it means that the precipitation event is scaled 0.25 times based on Typhoon Chan-hom, and the gate opening ratio is 0.8. The water level comparison process between the hydrological and hydrodynamic model calculation results and the neural network model prediction value. The results show that the forecast effect is good, and the constructed neural network model can be used as an alternative model to the hydrological and hydrodynamic coupling model. The constructed hydrological and hydrodynamic coupling model takes an average of 100 seconds to calculate once, while the neural network model only takes 1 second, which greatly improves the calculation efficiency.

[0041] S4. Optimization of river basin dispatching scheme based on neural network model: Based on the trained neural network model, the water level change process of the downstream section under different gate joint dispatching schemes is quickly calculated. According to the preset target water level, the optimization algorithm is used to select the optimal flood control dispatching scheme.

[0042] Step S4 specifically includes the following steps: The gate opening combination in the gate joint scheduling scheme is regarded as a decision variable. The trained neural network model is embedded into the target optimization iteration process of the optimization algorithm using an optimization algorithm (such as the NSGA-Ⅱ algorithm), and the downstream control section water level under different scheduling rules is iteratively optimized. The optimal gate joint scheduling scheme is determined based on the objective function and constraints. For example, by constraining the maximum water level of the target section not to exceed the control water level and taking the minimization of the downstream water level deviation as the objective function, the best gate joint scheduling scheme is selected as the optimal flood control scheduling scheme.

[0043] Objective function calculation formula: Where X represents the gate opening combination (decision variable), is the objective function, such as minimizing the downstream water level deviation, is the trained neural network model, Indicates the controlled water level of the downstream section.

[0044] Constraints: Optimal solution: In the formula, P represents the set of all possible solutions.

[0045] In this embodiment, if Figure 6 The display takes the typhoon Chan-hom rainfall event scaled 0.25 times as an example, and shows the downstream target section water level process under different gate scheduling opening conditions. If the downstream target is set to no more than 3.5m, the gate opening should be at least 0.6.

[0046] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for optimizing a flood control scheduling scheme for a river basin based on a neural network, comprising the following steps: S1. Construct a hydrological and hydrodynamic coupling model of the basin; It is characterized in that S2, using the hydrological and hydrodynamic coupling model constructed in step S1 to perform simulation calculations and prepare training data required for the neural network model; S3, constructing a neural network model including scheduling information, and using the training data of step S2 to train and verify the neural network model; S4, based on the neural network model trained in step S3, according to the preset target water level, use the optimization algorithm to select the optimal flood control scheduling plan; The preparation of training data in step S2 specifically includes: Abstraction and generalization of gate joint dispatching rules: generalize the classic dispatching scheme of a single gate, discretize the gate opening, and propose a formula for gate joint dispatching rules for N gates, N≥1; Scaling of historical rainfall events: Collect typical precipitation events in the basin over a long period of time, intercept the n-day precipitation duration in the hourly precipitation process, form a T-hour time series, fill zeros at the end for precipitation processes that are less than T hours long, and scale precipitation events according to different recurrence periods. ; Obtain the water level change process data of important downstream sections under different meteorological conditions and gate joint dispatching rules: input the outflow calculation results of the runoff generation zone under different precipitation scenarios into the one- and two-dimensional hydrodynamic model as boundary conditions to perform hydrological and hydrodynamic coupling calculations; in the hydrodynamic model, for N gates, apply different gate joint dispatching schemes to obtain the water level process of the downstream section under different scenarios; Step S4 is specifically as follows: the gate opening combination in the gate joint scheduling scheme is regarded as a decision variable, and the neural network model trained in step S3 is embedded into the target optimization iteration process of the optimization algorithm by using the optimization algorithm, and the downstream control section water level under different gate joint scheduling rules is iteratively optimized; based on the objective function and constraints, the optimal gate joint scheduling scheme is determined as the optimal flood control scheduling scheme.

2. The method for optimizing a flood control scheduling scheme for a river basin based on a neural network according to claim 1, characterized in that: The abstract generalization of the gate joint dispatching rule is as follows: the gate opening is used as the dispatching parameter, which is discretized in the interval [0,1] with a step size of is 0.2, which reflects the response behavior of artificial buildings under different scheduling schemes; Among them, the gate joint dispatching rule formula for N gates is as follows: when When the gate opening of each station And the gate opening combination in the gate joint dispatching rule is expressed as follows: ； ； In the formula, represents the i-th joint scheduling combination in the scheduling scheme group; represents the gate opening of gate No. j, Represents a scheduling plan The gate opening of gate No. j in the middle; represents the calculated water level at gate No. j; represents the control water level threshold at the jth gate, and G is the gate opening combination in the gate joint scheduling rule.

3. The neural network-based method for optimizing flood control scheduling schemes in a river basin according to claim 2, characterized in that: The scaling of historical rainfall events is as follows: typical precipitation events in the basin over a long period of time are collected, and the precipitation duration of n days in the hourly precipitation process is intercepted to form a time series of T hours. For precipitation processes that are less than T hours in length, zero is added at the end. The specific formula is as follows: ； in, is the processed precipitation time series; is the original precipitation time series; is the time length of the original precipitation series; The precipitation events are scaled according to different return periods to cover most precipitation processes in reality; the formula is as follows: ； In the formula, is the scaled precipitation time series; is the scaling factor.

4. The method for optimizing a flood control scheduling scheme for a river basin based on a neural network according to claim 3, characterized in that: Water level process in step S2 The simulation calculation formula is as follows: ； Where M0 is the hydrological and hydrodynamic coupling model constructed in step S1, B is the remaining boundary conditions, and C is the neural network model parameters and initial conditions.

5. A method for optimizing a flood control scheduling scheme for a river basin based on a neural network according to claim 4, characterized in that: Step S3 includes the following steps: S31. Clarify the input and output matrix of the neural network, the input includes: the outflow Q of the runoff zone obtained by the precipitation-driven hydrological model, the gate opening combination G in the gate joint scheduling rule, and the output is the cross-section water level process H calculated by the hydrological and hydrodynamic model; The input matrix X can be expressed as: ； ； in, represents the outflow matrix of the runoff-producing partition, represents the gate combination rule matrix, M1 is the number of flow-generating partitions, and M2 is the number of gates for joint scheduling; the input matrix X can be expressed in detail as follows: ； The output matrix Y can be expressed as: ； In the formula, The target section water level at time t calculated by the hydrological and hydrodynamic coupling model; S32. Abstract a general framework suitable for neural network models, which can be flexibly applied to long short-term memory networks LSTM, gated recurrent units GRU and fully connected neural networks FCNN; The hidden layer state update formula is: ； ； The output generation formula is: ； The loss function calculation formula is: ； In the formula, It is represented as the hidden layer state at time step t, F is the feature extraction and state update function, g is the mapping function from the hidden state sequence to the target output, are the parameters of the neural network model, Mapping parameters to control output; S33. Divide the constructed data set into training set and validation set in proportion, of which 70% is used to train the neural network model and 30% is used to verify the performance of the neural network model; in the process of training and validating the neural network model, the leave-one-out cross-validation method is used to optimize the selection of hyperparameters on the training set, including the number of hidden layers, the number of neurons in the hidden layer, the number of iterations, the batch size, and the learning rate.

6. A method for optimizing a flood control scheduling scheme for a river basin based on a neural network according to claim 5, characterized in that: In step S4: the optimization algorithm uses the NSGA-Ⅱ algorithm, the objective function is to minimize the downstream water level deviation, and the constraint condition is that the highest water level of the downstream section shall not exceed the downstream section control water level; Among them, the objective function calculation formula is: ； ； ； In the formula, X represents the gate opening combination, is the objective function, is the trained neural network model, Indicates the controlled water level of the downstream section; Constraints: ； ； Optimal solution: ； In the formula, P represents the set of all possible solutions.

Images