An interpretable conceptual rainfall-runoff hybrid prediction method and system

Abstract

The present invention discloses an interpretable conceptual rainfall-runoff hybrid prediction method and system. A GR4J hydrological model is constructed through meteorological data of a target watershed, and an optimization algorithm is used to obtain intermediate runoff variables and runoff data of the target watershed. An initial state factor set is constructed based on the meteorological data and the corresponding runoff variable data and runoff data, and the state factor set is reconstructed using an XGBoost‑SHAP interpretable framework. The reconstructed state factor set is input into a deep learning time series prediction model for training to obtain a predicted runoff process of the target watershed. Compared with the prior art, the present invention can greatly improve the accuracy of watershed runoff prediction.

Description

Technical Field

[0001] The invention relates to the field of hydrological forecasting in hydrology, and in particular to an explainable conceptual rainfall-runoff hybrid prediction method and system. Background Art

[0002] In recent years, rainstorms and floods have occurred frequently, and many parts of my country have suffered from rainstorms, floods, typhoons and other disasters, which have caused huge economic losses to the national economy. With the continuous acceleration of the pace of the "dual carbon" strategic goal, the status of hydropower energy in my country's energy structure will become more prominent, and the role it will play in the field of new energy in the future will become more important. However, the dual effects of climate change and human activities have led to increasingly complex hydrological dynamic processes and internal elements of the water cycle system in the basin, and the uncertainty characteristics and chaotic characteristics of hydrological and meteorological elements have become more significant, which has brought great challenges to the high-precision simulation and prediction of runoff systems. Therefore, it is urgent to conduct in-depth research on the comprehensive forecasting method of nonlinear runoff in the basin under a changing environment, and to explore the mechanism of the influence of basin runoff production and confluence and the new paradigm of flood forecasting under a changing environment. This has important scientific value and practical significance for the optimal allocation of water resources and the assessment of flood control and disaster reduction risks.

[0003] The basin hydrological model driven by physical mechanisms is the core foundation of flood forecasting, mainly including lumped, semi-distributed and distributed hydrological model architectures. Common lumped hydrological models include Xin'anjiang model, TANK model, GR4J model, HYMOD model, etc. However, any existing hydrological model is essentially a simplified simulation of the natural phenomenon of rainfall-runoff, and its construction and parameterization cannot fully capture all influencing factors and their interactions. Especially under the strong influence of global warming and human activities, the coupling effect between meteorological elements, topography, basin characteristics and human intervention involved in the rainfall-runoff process has become increasingly prominent. The rapid development of computer science has led to the widespread application of deep learning in the field of hydrology, but this typical data-driven method is considered to be a black box model, and the model parameter structure lacks physical interpretation. For this reason, hydrological models that rely entirely on physical laws and mathematical equations have problems such as difficulty in accurately understanding and predicting the complex rainfall-runoff formation process affected by multiple factors, and difficulty in effectively utilizing multi-source observation information. Summary of the invention

[0004] The present invention provides an explainable conceptual rainfall-runoff hybrid prediction method and system, which are used to solve the technical problem of low accuracy of rainfall-runoff prediction methods.

[0005] In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0006] An interpretable conceptual rainfall-runoff hybrid forecasting method including:

[0007] Constructing a GR4J hydrological model of the target watershed, and using the GR4J hydrological model to obtain intermediate flow variables and runoff data of the target watershed under different meteorological data;

[0008] A state factor set is constructed based on the meteorological data and the corresponding intermediate runoff variables and runoff data, and the state factor set is reconstructed based on the XGBoost-SHAP interpretable framework;

[0009] The reconstructed state factor set is input into the deep learning time series prediction model for training to obtain the runoff process of the target watershed.

[0010] Preferably, the meteorological data include the surface precipitation and surface evaporation of the target watershed; the intermediate variables of runoff generation include effective precipitation, residual evaporation, precipitation to supplement the runoff generation reservoir, evaporation of the runoff generation reservoir, runoff generation of the runoff generation reservoir, total runoff generation of the runoff generation reservoir, and water storage of the runoff generation reservoir; the runoff data include the total flow of the outlet section of the watershed; and constructing the GR4J hydrological model includes:

[0011] Obtain the surface precipitation P and surface evaporation E of the target basin, and determine the effective precipitation P according to the surface precipitation P and surface evaporation E of the target basin. n and the remaining evaporation capacity E n ;

[0012] Through effective precipitation P n and the remaining evaporation capacity E n Calculate the amount of precipitation P that replenishes the runoff reservoir s and evapotranspiration E of runoff reservoir s ;

[0013] By replenishing the runoff reservoir with precipitation P s and evapotranspiration E of runoff reservoir s Calculate the water storage capacity S of the runoff reservoir t and the flow yield Perc of the flow-producing reservoir;

[0014] The amount of water stored in the runoff reservoir S t and the flow rate Perc of the flow-producing reservoir to calculate the total flow rate P of the flow-producing reservoir r ;

[0015] The total flow of the flow-producing reservoir P r The water volume is divided into the part that flows into the confluence reservoir and the part that flows directly into the basin outlet section according to the preset ratio, and the confluence calculation is performed using the time unit line to obtain the final water volume Q9 flowing into the confluence reservoir and the water volume Q1 flowing into the basin outlet section; based on the final water volume flowing into the confluence reservoir and the water volume flowing into the basin outlet section, the time period water volume exchange volume F is introduced to calculate the total flow of the basin outlet section Q sim .

[0016] Preferably, based on the amount of water that finally flows into the confluence reservoir and the amount of water that flows into the outlet section of the basin, the water exchange volume F during the period is introduced to calculate the total flow rate Q of the outlet section of the basin sim include:

[0017] The water exchange volume F during the introduction period is used to calculate the water volume R of the confluence reservoir;

[0018] Calculate the outflow Q of the confluence reservoir based on the water volume R of the confluence reservoir r ;

[0019] Calculate the outflow Q of the confluence reservoir based on the amount of water Q1 flowing into the outlet section of the basin and the water exchange volume F during the period d ;

[0020] Based on the outflow Q of the confluence reservoir d And the outflow of the confluence reservoir Q r Calculate the total flow Q at the outlet section of the basin sim .

[0021] Preferably, the GR4J hydrological model uses a shuffled composite evolutionary algorithm to calibrate model parameters, wherein surface rainfall and surface evaporation are calculated using the Taylor polygon method.

[0022] Preferably, the state factor set includes:

[0023]

[0024] Among them, t represents the time, X t represents the state factor set at time t; P t is the basin precipitation at time t, E t is the evaporation at time t, is the effective precipitation at time t, is the remaining evaporation at time t, is the precipitation that replenishes the runoff-producing reservoir at time t; is the evapotranspiration E of the runoff reservoir at time t s , Perc t is the flow yield of the runoff-producing reservoir at time t; is the total flow of the flow-producing reservoir at time t; S t is the water storage capacity of the runoff reservoir at time t, It is the total flow of the basin outlet section calculated by GR4J at time t.

[0025] Preferably, the state factor set is reconstructed using an XGBoost-SHAP interpretable framework, including:

[0026] First, set the sliding window n to obtain the expansion state factor set

[0027]

[0028] Secondly, based on the expanded state factor set Build the XGBoost-SHAP interpretable framework:

[0029] Will After dividing the training set and the test set, they are used as the input of the XGBoost model for training. The SHAP Value generated by the XGBoost-SHAP interpretable framework is obtained. The global and local explanations of the input features are calculated, and the global contribution of each feature factor is evaluated. The top N factors ranked by the SHAP Value of the global explanation of the input features are selected as the input of the time series prediction model.

[0030] Preferably, the deep learning time series prediction model is an LSTM neural network, the training set of the LSTM neural network is composed of a reconstructed and normalized state factor set, and the Bayesian algorithm is used to optimize the LSTM model hyperparameters.

[0031] Preferably, the method further comprises the step of evaluating:

[0032] The trained LSTM neural network is evaluated using three indicators: Nash coefficient, root mean square error and Kling-Gupta efficiency coefficient, and whether to optimize the LSTM neural network is determined based on the evaluation results.

[0033] A computer system comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the computer program.

[0034] The present invention has the following beneficial effects:

[0035] 1. The present invention innovatively combines advanced machine learning technology with traditional hydrological principles, and proposes an interpretable conceptual rainfall-runoff hybrid model. The LSTM neural network is used to replace the confluence process of the GR4J hydrological model, which enhances the model's ability to capture nonlinear spatiotemporal dynamics, improves the generalization performance of the model under complex environmental conditions, and expands the applicable scenarios and prediction range of the model. Compared with the traditional conceptual hydrological model GR4J, the accuracy and reliability of runoff forecasts are improved, and the prediction error is effectively reduced through the optimized application of deep learning algorithms. In view of the "black box" characteristics of deep learning models, the XGBoost-SHAP factor evaluation and interpretation framework is introduced to clearly reveal the driving factors behind the model's decision, improve the interpretability of the prediction results, and solve the problem of lack of interpretability in the application of deep learning in the field of hydrology. The present invention not only achieves significant improvements on existing hydrological models at the technical level, but also opens up new paths in theoretical and practical applications, providing a powerful support tool for hydrological research and water resources management.

[0036] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0038] Figure 1 A flowchart of an interpretable conceptual rainfall-runoff mixing model of the present invention;

[0039] Figure 2 This is the GR4J model architecture diagram in the present invention;

[0040] Figure 3 It is a flow chart of the conceptual rainfall-runoff hybrid model GR4J of the present invention;

[0041] Figure 4 This is the XGBoost-SHAP factor importance evaluation diagram of the present invention;

[0042] Figure 5 This is the global distribution of the global SHAP value of the XGBoost-SHAP factor of the present invention. DETAILED DESCRIPTION

[0043] The embodiments of the present invention are described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0044] In this embodiment, an interpretable conceptual rainfall-runoff hybrid prediction method is provided to solve the problems that the traditional conceptual hydrological model has limited application boundaries and cannot accurately characterize the rainfall-runoff process in complex environments, and the data-driven method lacks physical interpretability.

[0045] This invention improves the applicability and generalization performance of the traditional conceptual hydrological model while giving the hybrid model input factors interpretability, which plays an important role in promoting the development of physical mechanism coupled deep learning forecasting. Figure 1-3 As shown, the specific steps include:

[0046] Step S1: Collect measured meteorological data of the target watershed and the corresponding flow generation data and runoff data, wherein the meteorological data include the surface precipitation and surface evaporation of the target watershed; the flow generation data include effective precipitation, residual evaporation, precipitation to supplement the flow generation reservoir, evaporation of the flow generation reservoir, flow generation of the flow generation reservoir, total flow generation of the flow generation reservoir, and water storage capacity of the flow generation reservoir; the runoff data include the total flow at the outlet section of the watershed;.

[0047] Step S2: Based on the selected meteorological data and the corresponding runoff data and runoff data, the GR4J hydrological model is constructed, divided into a training period and a validation period, and the model parameters are calibrated using the shuffled composite evolution (SCE-UA) algorithm. Among them, the surface rainfall data is calculated using the Taylor polygon method. The calculation process of the GR4J hydrological model is as follows:

[0048] GR4J is a conceptual rainfall-runoff model, which is currently widely used in flood forecasting, water resources planning, etc. The GR4J model contains only 4 parameters, and has certain characteristics compared with similar models in terms of model principle and structure. The GR4J model uses two nonlinear reservoirs for runoff calculation, where the first reservoir is called the runoff reservoir and the second reservoir is called the runoff reservoir. The GR4J hydrological model architecture is shown in Figure 1 .

[0049] (1) Abortion stage

[0050] In the runoff generation stage, effective precipitation P is first determined based on the basin surface precipitation P and surface evaporation E. n and the remaining evaporation capacity E n .

[0051]

[0052] Then, through P n and E n Calculate the amount of precipitation P that replenishes the runoff reservoir s and evapotranspiration E of runoff reservoir s :

[0053]

[0054] Where: P s is the increase in water volume of the runoff-producing reservoir; Es is the water consumption of the runoff-producing reservoir; S is the water storage capacity of the runoff-producing reservoir; x1 is the water storage capacity of the runoff-producing reservoir (mm).

[0055] After calculating Es and Ps, the water storage capacity S of the runoff-producing reservoir and the runoff Perc of the runoff-producing reservoir are:

[0056]

[0057] Perc t =S t {1-[1+(4S t / 9x1) 4 ] -1 / 4}(6)

[0058] After deducting Perc, the water storage capacity S of the runoff reservoir and the total runoff P r for:

[0059] S t+1 =S t -Perc t (7)

[0060]

[0061] (2) Confluence stage

[0062] GR4J uses time unit lines for flow routing. r It is divided into two parts, 90% is calculated by unit line UH1, and 10% is calculated by unit line UH2. The former needs to be adjusted again by the confluence reservoir, and the latter is directly collected at the outlet section of the basin. Both unit lines are calculated by S curves (SH1, SH2), and the calculation method is shown in formulas (9) and (10):

[0063]

[0064] Where: j is an integer, representing the jth day. x4 is the unit line confluence time (days).

[0065] The water volumes calculated from the two unit lines (UH1, UH2) are:

[0066] Q9=UH1×0.9×P r (11)

[0067] Q1=UH2×0.1×P r (12)

[0068] In the formula: Q9 refers to the amount of water entering the confluence reservoir; Q1 refers to the amount of water directly collected at the outlet section of the basin.

[0069] At the same time, GR4J introduces the water exchange volume F in each period, and the calculation method is shown in formula (13):

[0070] F t =x2(R t / x3) 7 / 2 (13)

[0071]

[0072] Where: F is the water exchange volume in the period; R is the water volume of the confluence reservoir.

[0073] Outflow of confluence reservoir Q r :

[0074] Q r t =R t {1-[1+(R t / x3) 4 ] -1 / 4} (15)

[0075]

[0076] Where: x2 is the groundwater exchange coefficient (mm); x3 is the confluence reservoir capacity (mm).

[0077] Outflow of confluence reservoir Q d and the total flow rate at the outlet section of the basin Q sim for:

[0078]

[0079] The GR4J model contains four parameters: x1, x2, x3, and x4. At the same time, when calculating runoff, the influence of the initial values ​​of soil moisture S and the water volume R of the confluence reservoir on the runoff simulation accuracy must also be considered.

[0080] Step S3: After the model calibration is completed, the intermediate variables of the GR4J runoff module are calculated based on the optimal parameters x1, x2, x3, and x4 obtained by the model. The state factor set X is constructed based on the intermediate variables of the runoff module (i.e., runoff data) and precipitation and evaporation data (i.e., meteorological data). t .

[0081]

[0082] Step S4: Based on the obtained factor X t , based on the Extreme Gradient Boosting Tree (XGBoost) and Python's interpretable machine learning library SHAP, we build a state factor evaluation and interpretable model based on XGBoost-SHAP.

[0083] First, set the sliding window n and expand the state factor set

[0084]

[0085] XGBoost is an efficient gradient boosting decision tree algorithm. It improves on the original gradient boosting decision tree (GBDT) to greatly improve the model effect. Its core is to adopt the Boosting idea to form a powerful integrated model by integrating multiple weak learners (decision trees). The objective function of XGBoost consists of two parts: the loss function and the regularization term. Given a training data set T = {(x1, y1), (x2, y2), ..., (x n ,y n )}, loss function Regularization term Ω(f k ), then the overall objective function of XGBoost can be recorded as:

[0086]

[0087] Where: is the expression in linear space; i is the i-th sample; k is the k-th tree; ∑ k Ω(f k ) represents the complexity of the number of k trees; is the i-th sample x i The predicted value of

[0088] Although XGBoost has excellent regression and classification performance, its internal mechanism and decision-making process are difficult to understand and are generally considered to be black box models. SHAP plays a vital role in the field of XGBoost regression, which enhances the transparency and interpretability of this widely used machine learning method. SHAP is a post-explanation method that draws on the idea of ​​game theory. SHAP measures the influence of each feature by calculating the marginal contribution of each feature in the model, and then explains the black box model. The marginal contribution is called Shapley Value in SHAP. This is very suitable for explaining black box models such as XGBoost. In addition, SHAP also provides extremely powerful data visualization functions, such as interactive interpreters and decision tree visualization. To this end, the present invention generates SHAP values ​​through Python's SHAP package and TreeExplainer function, and then obtains a dependency graph.

[0089] Step S5: According to the reconstructed state factor X c tThe training set and validation set are divided and, after normalization, input into the LSTM neural network for model training, and the LSTM model hyperparameters are optimized using the Bayesian algorithm.

[0090] Step S6: Model evaluation.

[0091] like Figure 4-5 As shown, the present invention uses three commonly used hydrological indicators for model performance evaluation: Nash coefficient (NSE), root mean square error (RMSE) and Kling-Gupta efficiency coefficient (KGE).

[0092] The NSE calculation formula is as follows:

[0093]

[0094] In the formula, Q obs (i) is the measured flow rate at the i-th moment, m 3 / s;Q pre (i) is the forecast flow at the i-th moment, m 3 / s; N is the number of time periods; is the average value of the measured flow rate, m 3 / s.

[0095] The RMSE calculation formula is as follows:

[0096]

[0097] Where: The meaning of the variable is the same as that of formula (1), and its value range is [0, +∞]. It represents the deviation between the predicted value and the true value.

[0098] KGE measures the Euclidean distance from a point to the optimal point. It mainly consists of three components: correlation coefficient (PCC), bias ratio (BR) and relative rate of change (RV). By calculating KGE and its evaluation factors, a more detailed accuracy evaluation of the forecast flow can be provided. The closer KGE is to 1, the better the fitting effect of the predicted flow process. KGE is calculated as follows:

[0099]

[0100] G1=(σ pre / σ obs -1) 2 (25)

[0101] G2=(μ pre / μ obs -1) 2 (26)

[0102] G3=(r-1) 2 (27)

[0103] Where: pre is the standard deviation of the predicted flow; obs is the standard deviation of the measured flow rate; μ pre is the mean value of the forecast flow; μ obs is the mean of the measured flow; r is the linear correlation coefficient between the predicted flow and the measured flow; G1 is the variance factor; G2 is the mean factor; G3 is the linear correlation factor; KGE has a monotonically decreasing relationship with the three factors.

[0104] In summary, the present invention proposes an interpretable conceptual rainfall-runoff hybrid model, which constructs a dual-mode driving scheme by coupling the classic lumped hydrological model with advanced deep learning technology, and uses neural networks to replace the core confluence process of the GR4J hydrological model, thereby enhancing the generalization ability of traditional hydrological models in complex environments and expanding their application boundaries. At the same time, the present invention introduces the XGBoost-SHAP interpretability framework to conduct an in-depth analysis of the input parameters and intermediate processes of the model. On the basis of improving the accuracy of basin runoff forecasting and the reliability of the model, the hybrid model is given a certain interpretable path, which solves the "black box" problem of deep learning applied to hydrological forecasting. The design idea of ​​the present invention is logically rigorous and easy to operate and implement, and is expected to play an important role in the fields of water resources management, flood prevention and disaster reduction, etc.

[0105] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An interpretable conceptual rainfall-runoff hybrid prediction method, characterized in that: include: Constructing a GR4J hydrological model of the target watershed, and using the GR4J hydrological model to obtain intermediate flow variables and runoff data of the target watershed under different meteorological data; A state factor set is constructed based on the meteorological data and the corresponding intermediate runoff variables and runoff data, and the state factor set is reconstructed based on the XGBoost-SHAP interpretable framework; The reconstructed state factor set is input into the deep learning time series prediction model for training to obtain the runoff process of the target watershed.

2. The interpretable conceptual rainfall-runoff hybrid prediction method according to claim 1 is characterized in that: The meteorological data include the surface precipitation and surface evaporation of the target basin; the intermediate variables of runoff generation include effective precipitation, residual evaporation, precipitation to supplement the runoff generation reservoir, evaporation of the runoff generation reservoir, runoff generation of the runoff generation reservoir, total runoff generation of the runoff generation reservoir, and water storage of the runoff generation reservoir; the runoff data include the total flow of the outlet section of the basin; Building the GR4J hydrological model includes: Obtain the surface precipitation P and surface evaporation E of the target basin, and determine the effective precipitation P according to the surface precipitation P and surface evaporation E of the target basin. n and the remaining evaporation capacity E n ; Through effective precipitation P n and the remaining evaporation capacity E n Calculate the amount of precipitation P required to replenish the runoff-producing reservoir s and evapotranspiration E of runoff reservoir s ; By replenishing the runoff reservoir with precipitation P s and evapotranspiration E of runoff reservoir s Calculate the water storage capacity S of the runoff reservoir t and the flow yield Perc of the flow-producing reservoir; The amount of water stored in the runoff reservoir S t and the flow rate Perc of the flow-producing reservoir to calculate the total flow rate P of the flow-producing reservoir r ; The total flow of the flow-producing reservoir P r The water volume is divided into the part that flows into the confluence reservoir and the part that flows directly into the basin outlet section according to the preset ratio, and the confluence calculation is performed using the time unit line to obtain the final water volume Q9 flowing into the confluence reservoir and the water volume Q1 flowing into the basin outlet section; based on the final water volume flowing into the confluence reservoir and the water volume flowing into the basin outlet section, the time period water volume exchange volume F is introduced to calculate the total flow of the basin outlet section Q sim .

3. The interpretable conceptual rainfall-runoff hybrid prediction method according to claim 2 is characterized in that: Based on the final amount of water flowing into the confluence reservoir and the amount of water flowing into the outlet section of the basin, the water exchange volume F during the period is introduced to calculate the total flow rate Q of the outlet section of the basin sim include: The water exchange volume F during the introduction period is used to calculate the water volume R of the confluence reservoir; Calculate the outflow Q of the confluence reservoir based on the water volume R of the confluence reservoir r ; Calculate the outflow Q of the confluence reservoir based on the amount of water Q1 flowing into the outlet section of the basin and the water exchange volume F during the period d ; Based on the outflow Q of the confluence reservoir d And the outflow of the confluence reservoir Q r Calculate the total flow Q at the outlet section of the basin sim .

4. The interpretable conceptual rainfall-runoff hybrid prediction method according to claim 3 is characterized in that: The GR4J hydrological model uses a shuffled composite evolutionary algorithm to calibrate model parameters, wherein surface rainfall and surface evaporation are calculated using the Taylor polygon method.

5. The interpretable conceptual rainfall-runoff hybrid prediction method according to any one of claims 1 to 4, characterized in that: The state factor set includes: Among them, t represents the time, X t represents the state factor set at time t; P t is the basin precipitation at time t, E t is the evaporation at time t, is the effective precipitation at time t, is the remaining evaporation at time t, is the precipitation that replenishes the runoff-producing reservoir at time t; is the evapotranspiration E of the runoff reservoir at time t s ;Perc t is the flow yield of the runoff-producing reservoir at time t; is the total flow of the flow-producing reservoir at time t; S t is the water storage capacity of the runoff reservoir at time t; It is the total flow of the basin outlet section calculated by GR4J at time t.

6. The interpretable conceptual rainfall-runoff hybrid prediction method according to claim 5, characterized in that: The state factor set is reconstructed using the XGBoost-SHAP interpretable framework, including: First, set the sliding window n to obtain the expansion state factor set Secondly, based on the expanded state factor set Build the XGBoost-SHAP interpretable framework: Will After dividing the training set and the test set, they are used as the input of the XGBoost model for training. The SHAP Value generated by the XGBoost-SHAP interpretable framework is obtained. The global and local explanations of the input features are calculated, and the global contribution of each feature factor is evaluated. The top N factors ranked by the SHAP Value of the global explanation of the input features are selected as the input of the time series prediction model.

7. The interpretable conceptual rainfall-runoff hybrid prediction method according to claim 6 is characterized in that: The deep learning time series prediction model is an LSTM neural network, the training set of the LSTM neural network consists of a reconstructed and normalized state factor set, and the Bayesian algorithm is used to optimize the LSTM model hyperparameters.

8. The interpretable conceptual rainfall-runoff hybrid prediction method according to claim 7, characterized in that: It also includes the evaluation step: The trained LSTM neural network is evaluated using three indicators: Nash coefficient, root mean square error and Kling-Gupta efficiency coefficient, and whether to optimize the LSTM neural network is determined based on the evaluation results.

9. A computer system 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 computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images