A method for predicting outflow from reservoirs considering climate change and artificial regulation

Abstract

The present invention discloses a method for predicting the outflow of a reservoir group considering climate change and artificial regulation, and the steps include: collecting hydrometeorological data of the study basin; clarifying the upstream inlet reservoir, interval reservoir and outlet reservoir according to the location distribution of reservoirs in the study basin; determining the inflow calculation method of reservoirs at different locations and the outflow model of the reservoir; for the outlet reservoir, simulating the outflow of the reservoir according to the outflow of the upstream reservoir and the reservoir inflow obtained by the vector autoregression model or the multivariate linear regression model through the hydrometeorological data, and establishing an outflow residual prediction model based on deep learning to realize the outflow prediction of the outlet reservoir. The present invention considers the impact of the artificial regulation of the reservoir on the downstream reservoir based on the hydraulic connection between the inflow and outflow of the reservoir group, and predicts the outflow process of the downstream reservoir by combining the conceptual rainfall-runoff model, the reservoir scheduling model and the deep learning model.

Description

Technical Field

[0001] The invention relates to the technical field of reservoir flow prediction, and in particular to a method for predicting outflow of a reservoir group taking into account climate change and artificial regulation. Background Art

[0002] In recent years, with the intensification of global climate change, changes in precipitation patterns and the increase in extreme weather events, higher requirements have been placed on the dispatch and management of reservoirs. The outflow of a reservoir is the most direct reflection of the operation of a reservoir. It is necessary to comprehensively consider factors such as the upstream natural water supply and artificial storage. Its prediction accuracy is crucial to the rational use and dispatch of water resources.

[0003] In a reservoir group, the mutual influence and hydraulic coupling relationship between each reservoir makes flow prediction more complicated. Traditional hydrological models usually focus on simulating the hydrological process of a single reservoir or a local basin, without fully considering the changes in upstream natural water conditions and the impact of artificial regulation of upstream inlet reservoirs on downstream reservoirs. They also do not consider the mutual dispatch and flow impact between reservoirs in the reservoir group. Therefore, when faced with complex dispatch operations of multi-reservoir systems, their prediction accuracy often has problems such as insufficient prediction accuracy and difficulty in coping with climate change.

[0004] Deep learning models are gradually being applied to reservoir flow prediction because they can effectively capture complex nonlinear relationships. However, most existing studies apply deep learning models to flow prediction of a single reservoir or data analysis of a certain aspect, ignoring the interaction between reservoir groups. The existing methods of combining traditional reservoir operation models with machine learning are very limited. The hydraulic connection between multiple reservoirs is often ignored, and the impact of upstream natural water on the outflow of a reservoir group is difficult to quantify. Summary of the invention

[0005] The purpose of the present invention is to provide a method for predicting the outflow flow of a reservoir group taking into account climate change and artificial regulation, and to predict the outflow flow of a reservoir group by combining the complex relationship between the upstream inflow of the reservoir group and the outflow flow of the downstream outlet reservoir.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] A method for predicting outflow of a reservoir group considering climate change and artificial storage, the steps of which include:

[0008] Collect hydrometeorological data and reservoir data of the study basin; hydrometeorological data include precipitation, potential evapotranspiration, etc. Reservoir data include water level-storage capacity curve, outflow, inflow, Palmer drought index, etc.;

[0009] Based on the location of the reservoirs in the study area, a generalized diagram of the hydraulic topology of the reservoir group is determined to identify the upstream inlet reservoir, interval reservoir and outlet reservoir;

[0010] For the upstream inlet reservoir, the conceptual rainfall-runoff model is used to calculate the corresponding inflow. For the interval reservoir and the outlet reservoir, the vector autoregression model (VAR) or multivariate linear regression model is used to calculate the corresponding inflow. The reservoir operation model is used to calculate the outflow of the upstream inlet reservoir, interval reservoir and outlet reservoir.

[0011] Based on the outflow of the outlet reservoir calculated by the reservoir dispatching model, the outflow residual of the outlet reservoir is predicted using the outflow residual prediction model;

[0012] The outflow of the export reservoir in the study area calculated by the reservoir scheduling model and the predicted outflow residual are superimposed to obtain the final outflow prediction value of the export reservoir, and the final outflow prediction value of the export reservoir is expressed as the flow of the total basin outlet section.

[0013] Among them, the input data of the conceptual rainfall-runoff model GR4J are the precipitation and potential evapotranspiration data of the catchment area above the upstream inlet reservoir. The meteorological data required by the conceptual rainfall-runoff model GR4J are usually area averages. In areas with limited or uneven data, estimates based on remote sensing data may be used, or interpolation methods based on meteorological stations may be used to obtain the area average of the basin.

[0014] According to the above technical solution, the interval reservoirs are divided into series reservoirs and parallel reservoirs;

[0015] Each parallel reservoir uses a vector autoregression model or a multiple linear regression model to predict its inflow, and takes the outflow of each parallel reservoir as an independent flow source and calculates it to the first reservoir downstream of the two parallel reservoirs;

[0016] The inflow of the downstream series reservoir is predicted using a vector autoregression model or a multivariate linear regression model based on the hydrometeorological data of the upstream inlet reservoir, and the inflow of the series reservoir is used as the inflow of the first reservoir downstream of the series reservoir.

[0017] According to the above technical solution, the vector autoregression model is expressed as follows:

[0018] ;

[0019] ;

[0020] ;

[0021] In the formula, for The endogenous variable at time t+1 is the inflow of the next reservoir at time t+1. and the outflow of the upstream inlet reservoir at time t Together they form a two-dimensional vector, The exogenous variable at time t is the precipitation in the catchment area controlled by the next reservoir. and potential evapotranspiration in the catchment controlled by the next reservoir Together they form a two-dimensional vector, express The endogenous variables at time, express The endogenous variables at time, Represents endogenous variables The corresponding coefficient matrix is, Represents endogenous variables The corresponding coefficient matrix is, represents the lag order, Exogenous variable The corresponding coefficient matrix is, is expressed as a time-dependent error term, is the number of samples.

[0022] Among them, the input data of the vector autoregression model (VAR) are the outflow of the upstream reservoir, the precipitation in the catchment area controlled by the next reservoir, and the potential evapotranspiration data. The stationarity of the input data of the vector autoregression model is tested by using the ADF test (Augmented Dickey-Fuller test). The null hypothesis of the ADF test is that the time series has a unit root, that is, the data is non-stationary; and the alternative hypothesis is that the time series has no unit root, that is, the data is stationary. If the time series is non-stationary, it is usually necessary to convert it into a stationary series through methods such as difference. The Bayesian Information Criterion (BIC) is used to select the optimal lag order of the vector autoregression model.

[0023] According to the above technical solution, the multivariate linear regression model is expressed as follows:

[0024] ;

[0025] In the formula, is the intercept term of the regression model, , and represents the regression coefficient, represents the error term of the regression model.

[0026] According to the above technical solution, there are several interval reservoirs between the upstream inlet reservoir and the outlet reservoir, so the steps of calculating the outflow flow of the outlet reservoir using the reservoir scheduling model include:

[0027] S1. Based on the hydrological and meteorological data of the catchment area above the upstream inlet reservoir, the inflow of the upstream inlet reservoir is predicted by the conceptual rainfall-runoff model GR4J; the upstream reservoir is the total outlet of the basin controlled by the upstream inlet reservoir, and the total outflow of the basin controlled by the upstream inlet reservoir obtained by simulating the conceptual rainfall-runoff model is used as the inflow of the upstream reservoir;

[0028] S2, based on the inflow of the upstream inlet reservoir, the historical data of the water level in front of the reservoir dam, the Palmer drought index and the water level-storage capacity curve of the reservoir, the outflow process of the upstream inlet reservoir is predicted by using a reservoir scheduling model to obtain the outflow of the upstream inlet reservoir;

[0029] S3, based on the outflow of the upstream inlet reservoir and the next interval reservoir The hydrometeorological data in the controlled catchment area are used to predict the interval reservoir using the vector autoregression model (VAR) or the multivariate linear regression model. Inbound flow of

[0030] S4, based on interval reservoir The reservoir dispatching model is used to predict the interval reservoir flow based on the historical data of the reservoir dam water level, the Palmer drought index and the reservoir water level-capacity curve. The outflow process of the interval reservoir Outbound flow;

[0031] S5, based on the interval reservoir Outflow and Interval Reservoirs The hydrometeorological data in the controlled catchment area are used to predict the interval reservoir using the vector autoregression model (VAR) or the multivariate linear regression model. Inbound flow of

[0032] S6, jump to step S4 and continue to execute until the outflow of the outlet reservoir is obtained, and use the mean square error (MSE) and Nash efficiency coefficient (NSE) to evaluate the prediction effect of the conceptual rainfall-runoff model, the vector autoregression model (VAR) or the multivariate linear regression model and the reservoir operation model.

[0033] When step S5 is completed and the execution is continued to step S4, the interval reservoir in the currently executed step S4 The interval reservoir in step S5 of the previous cycle .

[0034] Among them, the mean square error (MSE):

[0035] ;

[0036] In the formula, is the number of samples, For the The actual observed value at the moment, For the The model simulation value at the moment. This means that the model's predictions are completely correct. hour The larger the value, the greater the error of the model prediction.

[0037] Nash efficiency coefficient (NSE):

[0038] ;

[0039] In the formula, represents the mean of the observations, This means that the model prediction is completely accurate. means that the model prediction is no better than using the observed mean, This means that the model's prediction performance is worse than simply using the observed mean.

[0040] According to the above technical solution, the reservoir scheduling model includes a data-driven reservoir scheduling model, a water storage target and release function derivation model and a dynamic partition target outflow model.

[0041] These three reservoir operation models belong to the existing models. Data-driven reservoir operation model: Based on the reservoir inflow, reservoir capacity (water storage), Palmer drought index and time, the model provides a set of regression trees and a classification tree to predict the reservoir outflow. The regression tree is also called the operation module. The operation module of the reservoir outflow prediction is determined by the provided classification tree. Among them, the data-driven reservoir operation model divides the training set, validation set and test set according to the ratio of 64%:16%:20%;

[0042] Water storage target and release function derivation model: Based on the collected time, reservoir capacity (water storage), reservoir outflow, reservoir inflow, and water level data, a normal water storage range that changes weekly is defined for each reservoir, and the reservoir outflow is predicted based on this water storage range.

[0043] Dynamically partitioned target outflow model: Based on the collected reservoir capacity (storage), reservoir inflow, and reservoir outflow data, the reservoir storage is divided into different operating zones. Then, for each defined zone, the reservoir outflow is predicted based on the reservoir storage within the zone.

[0044] According to the above technical solution, the steps of constructing the outbound flow residual prediction model include:

[0045] Obtaining historical data of each reservoir; the historical data of the reservoir includes the reservoir's inflow, precipitation, potential evapotranspiration and the reservoir's outflow;

[0046] Based on the historical data of each reservoir, the reservoir dispatching model is used to predict the outflow of the corresponding reservoir, and the difference between each actual outflow and the flow predicted by the reservoir dispatching model is calculated to obtain the residual of the outflow of the corresponding reservoir;

[0047] Based on the historical data of each reservoir and the residual of its corresponding outflow, the long short-term memory network (LSTM) model, the gated recurrent unit (GRU) model and the convolutional neural network (CNN) model were selected for training, and three corresponding preliminary outflow residual prediction models were obtained respectively;

[0048] Bayesian optimization is used to select the hyperparameters of the three preliminary outbound flow residual prediction models, and minimizing the mean square error (MSE) is used as the objective function. The Nash efficiency coefficient (NSE) and root mean square error (RMSE) are used to evaluate the performance of the three different preliminary outbound flow residual prediction models. The preliminary outbound flow residual prediction model with the best performance is selected as the final outbound flow residual prediction model for outbound flow residual prediction.

[0049] Compared with the prior art, the beneficial effects achieved by the present invention are: by comprehensively considering the impact of climate change and artificial regulation, combined with the conceptual rainfall-runoff model and deep learning method, a more accurate reservoir group outflow prediction scheme is provided. Through accurate modeling of the hydraulic topological relationship of the reservoir group, combined with the reservoir scheduling model and deep learning to correct the outflow residual, it can not only capture the complex impact of upstream water on the outflow of the downstream outlet reservoir, but also can perform flexible prediction under different climate modes, significantly improving the prediction accuracy and model adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0051] Figure 1 It is a flow chart of the steps of a method for predicting outflow of a reservoir group considering climate change and artificial storage in the present invention;

[0052] Figure 2 It is a schematic diagram of upstream-downstream reservoir flow transfer;

[0053] Figure 3 It is a schematic diagram of the hydraulic topology of a reservoir group. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] This plan uses the global climate model data of the Sixth Coupled Model Intercomparison Project (CMIP6), with four shared socioeconomic pathways: SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5. The global climate model (GCM) CanESM5 is selected to collect future precipitation and potential evapotranspiration data, and the impact of climate change on the outflow of the reservoir group is considered.

[0056] First, spatial resampling is performed to adjust the spatial resolution of GCM output to adapt it to the needs of the study area, and the bicubic B-spline interpolation method is used to improve the spatial resolution of future precipitation and potential evapotranspiration data. Then, statistical downscaling is performed using the quantile mapping method to calculate the cumulative probability distribution function (CDF) of the observed and simulated values, respectively. According to the quantiles of the observed data and the climate model data, a mapping function is constructed to convert the output values ​​of the climate model into results that are closer to the statistical characteristics of the historical observation data. The global climate model output adjusted by the quantile mapping method can generate high-resolution climate data for local areas, using the following method: Figure 1 The method used in this paper is to study the impact of upstream water flow changes on the outflow flow of the downstream outlet reservoir. The schematic diagram of upstream-downstream reservoir flow transfer is shown in the figure. Figure 2 shown.

[0057] The specific steps to study the impact of upstream water flow changes on the outflow flow of downstream outlet reservoirs include:

[0058] Step 1: Collect hydrometeorological data such as precipitation and potential evapotranspiration in the study basin, and collect reservoir data such as water level-storage capacity curves, outflow, inflow, and Palmer Drought Index of relevant reservoirs.

[0059] Step 2: Based on the location of the reservoirs in the study area, determine the generalized diagram of the hydraulic topology of the reservoir group. Figure 3 , and then clarify the upstream inlet reservoir, interval reservoir and outlet reservoir.

[0060] Among them, for each parallel reservoir, a vector autoregression model or a multivariate linear regression model can be used to predict its inflow flow, and the outflow flow of each parallel reservoir is taken as an independent flow source and calculated to the first reservoir downstream of the two parallel reservoirs.

[0061] The inflow of the series reservoir is provided by the upstream reservoir. When using the vector autoregression model or the multivariate linear regression model, the inflow of the downstream series reservoir can be predicted based on the outflow, precipitation, potential evapotranspiration and other factors of the upstream reservoir, and the inflow can be used as the inflow of the first reservoir downstream of the series reservoir.

[0062] Step 3: For the upstream inlet reservoir, the conceptual rainfall-runoff model is used to calculate the corresponding inflow flow. For the interval reservoir and the outlet reservoir, the vector autoregression model (VAR) or multivariate linear regression model is used to calculate the corresponding inflow flow, and the reservoir operation model is used to calculate the outflow flow of each reservoir. The specific steps include:

[0063] S1. Based on the hydrological and meteorological data of the catchment area above the upstream inlet reservoir, the inflow of the upstream inlet reservoir is predicted by the conceptual rainfall-runoff model GR4J; the input data of the conceptual rainfall-runoff model GR4J are the precipitation and potential evapotranspiration data of the catchment area above the upstream inlet reservoir.

[0064] The meteorological data required by the conceptual rainfall-runoff model GR4J are usually area averages. In areas with limited or uneven data, estimates based on remote sensing data or interpolation methods of meteorological stations may be used to obtain the area average of the basin.

[0065] The upstream reservoir serves as the total outlet of the basin it controls, and the total outflow of the basin controlled by the upstream inlet reservoir simulated by the conceptual rainfall-runoff model GR4J is used as the inflow of the upstream reservoir.

[0066] S2. Based on the inflow of the upstream inlet reservoir, the historical data of the water level in front of the reservoir dam, the Palmer drought index and the water level-storage capacity curve of the reservoir, the outflow process of the upstream inlet reservoir is predicted by using the reservoir scheduling model to obtain the outflow of the upstream inlet reservoir, specifically:

[0067] If the general data-driven reservoir operation model is used, the required input data should be time, reservoir capacity, reservoir inflow, and Palmer drought index. The running model divides the training set, validation set, and test set into a ratio of 64%:16%:20%. The outflow of the reservoir is predicted by a set of regression trees and a classification tree provided by the model. The regression tree is also called the operation module. The operation module of the reservoir outflow prediction is determined by the provided classification tree.

[0068] If the water storage target and release function derivation model is selected, the required input data are time, reservoir capacity, reservoir inflow, and water level. This model defines a normal water storage range for each reservoir that changes weekly, and predicts the reservoir outflow based on this water storage range.

[0069] If the dynamic partitioned target outflow model is used, the required input data are time, reservoir storage capacity, and reservoir inflow. This model divides the reservoir storage into different operating zones. Then, for each defined zone, the outflow of the reservoir is predicted based on the reservoir storage in the zone.

[0070] S3, based on the outflow of the upstream inlet reservoir and the next interval reservoir The hydrometeorological data in the controlled catchment area are used to predict the interval reservoir using the vector autoregression model (VAR) or the multivariate linear regression model. Inbound flow of

[0071] Among them, the vector autoregression model expression is as follows:

[0072] ;

[0073] ;

[0074] ;

[0075] In the formula, for The endogenous variable at time t+1 is the inflow of the next reservoir at time t+1. and the outflow of the upstream inlet reservoir at time t Together they form a two-dimensional vector, The exogenous variable at time t is the precipitation in the catchment area controlled by the next reservoir. and potential evapotranspiration in the catchment controlled by the next reservoir Together they form a two-dimensional vector, express The endogenous variables at time, express The endogenous variables at time, Represents endogenous variables The corresponding coefficient matrix is, Represents endogenous variables The corresponding coefficient matrix is, represents the lag order, Exogenous variable The corresponding coefficient matrix is, is expressed as a time-dependent error term, is the number of samples.

[0076] The multiple linear regression model expression is as follows:

[0077] ;

[0078] In the formula, is the intercept term of the regression model, , and The regression coefficients are shown. represents the error term of the regression model.

[0079] S4, based on interval reservoir The reservoir dispatching model is used to predict the interval reservoir flow based on the historical data of the reservoir dam water level, the Palmer drought index and the reservoir water level-capacity curve. The outflow process of the interval reservoir outbound flow.

[0080] S5, based on the interval reservoir Outflow and Interval Reservoirs The hydrometeorological data in the controlled catchment area are used to predict the interval reservoir using the vector autoregression model (VAR) or multivariate linear regression model. Inbound flow.

[0081] S6, jump to step S4 and continue to execute until the outflow of the outlet reservoir is obtained. Become the interval reservoir in step S5 of the previous cycle .

[0082] Step 4 is to predict the outflow residual of the outlet reservoir using the outflow residual prediction model based on the outflow of the outlet reservoir calculated by the reservoir scheduling model.

[0083] Among them, the steps of building the outbound flow residual prediction model include:

[0084] Obtaining historical data of each reservoir; the historical data of the reservoir includes the reservoir's inflow, precipitation, potential evapotranspiration and the reservoir's outflow;

[0085] Based on the historical data of each reservoir, the reservoir dispatching model is used to predict the outflow of the corresponding reservoir, and the difference between each actual outflow and the flow predicted by the reservoir dispatching model is calculated to obtain the residual of the outflow of the corresponding reservoir;

[0086] Based on the historical data of each reservoir and its corresponding outflow residual, the long short-term memory network (LSTM) model, the gated recurrent unit (GRU) model and the convolutional neural network (CNN) model were selected for training, and three corresponding preliminary outflow residual prediction models were obtained respectively; the hyperparameters of the three preliminary outflow residual prediction models were selected by Bayesian optimization, and the minimization of the mean square error (MSE) was used as the objective function. The Nash efficiency coefficient (NSE) and the root mean square error (RMSE) were used to evaluate the performance of the three different preliminary outflow residual prediction models, and the preliminary outflow residual prediction model with the best performance was selected as the final outflow residual prediction model for outflow residual prediction.

[0087] Root Mean Square Error (RMSE):

[0088] ;

[0089] In the formula, is the number of samples, For the The actual observed value at the moment, For the The model simulation value at the moment.

[0090] Step 5: Superimpose the outflow of the export reservoir in the study area calculated by the reservoir operation model and the predicted outflow residual to obtain the final outflow forecast value of the export reservoir, that is, the flow of the total basin outlet section.

[0091] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0092] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. 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. A method for predicting outflow from a reservoir group considering climate change and artificial regulation, characterized in that: The prediction steps include: Collect hydrometeorological data and reservoir data of the study basin; Based on the location of the reservoirs in the study area, a generalized diagram of the hydraulic topology of the reservoir group is determined to identify the upstream inlet reservoir, interval reservoir and outlet reservoir; For the upstream inlet reservoir, the conceptual rainfall-runoff model is used to calculate the corresponding inflow. For the interval reservoir and the outlet reservoir, the vector autoregression model or the multivariate linear regression model is used to calculate the corresponding inflow. The reservoir operation model is used to calculate the outflow of the upstream inlet reservoir, interval reservoir and outlet reservoir. The outflow flow of the export reservoir is calculated based on the reservoir dispatching model, and the outflow flow residual prediction model is used to predict the outflow flow residual of the export reservoir; wherein the step of calculating the outflow flow of the export reservoir by using the reservoir dispatching model includes: S1. Based on the hydrological and meteorological data of the catchment area above the upstream inlet reservoir, predict the inflow of the upstream inlet reservoir through a conceptual rainfall-runoff model; S2, based on the inflow of the upstream inlet reservoir, the historical data of the water level in front of the reservoir dam, the Palmer drought index and the water level-storage capacity curve of the reservoir, the outflow process of the upstream inlet reservoir is predicted by using a reservoir scheduling model to obtain the outflow of the upstream inlet reservoir; S3, based on the outflow of the upstream inlet reservoir and the next interval reservoir The hydrological and meteorological data in the controlled catchment area are used to predict the interval reservoir using vector autoregression model or multivariate linear regression model Inbound flow; S4, based on interval reservoir The reservoir dispatching model is used to predict the interval reservoir flow based on the historical data of the reservoir dam water level, the Palmer drought index and the reservoir water level-capacity curve. The outflow process of the interval reservoir Outbound flow; S5, based on the interval reservoir Outflow and Interval Reservoirs The hydrological and meteorological data in the controlled catchment area are used to predict the interval reservoir using vector autoregression model or multivariate linear regression model. Inbound flow; S6, jump to step S4 and continue to execute until the outflow of the outlet reservoir is obtained; wherein, when step S5 is executed and the step S4 is executed, the interval reservoir in the currently executed step S4 Become the interval reservoir in step S5 of the previous cycle ; The outflow of the export reservoir in the study area calculated by the reservoir scheduling model and the predicted outflow residual are superimposed to obtain the final outflow prediction value of the export reservoir, and the final outflow prediction value of the export reservoir is expressed as the flow of the total basin outlet section.

2. A method for predicting outflow from a reservoir group considering climate change and artificial regulation according to claim 1, characterized in that: The interval reservoirs are divided into series reservoirs and parallel reservoirs; Each of the parallel reservoirs uses a vector autoregression model or a multiple linear regression model to predict its inflow, and uses the outflow of each parallel reservoir as an independent flow source to calculate the first reservoir downstream of the two parallel reservoirs; The inflow of the downstream series reservoir is predicted using a vector autoregression model or a multivariate linear regression model based on the hydrometeorological data of the upstream inlet reservoir, and the inflow of the series reservoir is used as the inflow of the first reservoir downstream of the series reservoir.

3. The method for predicting outflow of a reservoir group considering climate change and artificial regulation according to claim 1 is characterized in that: The vector autoregression model expression is as follows: ； ； ； In the formula, for The endogenous variable at time t+1 is the inflow of the next reservoir at time t+1. and the outflow of the upstream inlet reservoir at time t Together they form a two-dimensional vector, The exogenous variable at time t is the precipitation in the catchment area controlled by the next reservoir. and potential evapotranspiration in the catchment controlled by the next reservoir Together they form a two-dimensional vector, express The endogenous variables at time, express The endogenous variables at time, Represents endogenous variables The corresponding coefficient matrix is, Represents endogenous variables The corresponding coefficient matrix is, represents the lag order, Exogenous variable The corresponding coefficient matrix is, is expressed as a time-dependent error term, is the number of samples.

4. The method for predicting outflow of a reservoir group considering climate change and artificial regulation according to claim 1 is characterized in that: The multivariate linear regression model expression is as follows: ； In the formula, is the intercept term of the regression model, , and represents the regression coefficient, represents the error term of the regression model.

5. The method for predicting outflow of a reservoir group considering climate change and artificial regulation according to claim 1 is characterized in that: The reservoir operation model includes a data-driven reservoir operation model, a water storage target and release function derivation model and a dynamic partition target outflow model.

6. The method for predicting outflow of a reservoir group considering climate change and artificial regulation according to claim 1 is characterized in that: The steps of constructing the outbound flow residual prediction model include: Obtaining historical data of each reservoir; the historical data of the reservoir includes the reservoir's inflow, precipitation, potential evapotranspiration and the reservoir's outflow; Based on the historical data of each reservoir, the reservoir dispatching model is used to predict the outflow of the corresponding reservoir, and the difference between each actual outflow and the flow predicted by the reservoir dispatching model is calculated to obtain the residual of the outflow of the corresponding reservoir; Based on the historical data of each reservoir and its corresponding residual outflow, the long short-term memory network model, gated recurrent unit model and convolutional neural network model were selected for training, and three corresponding preliminary outflow residual prediction models were obtained respectively. Bayesian optimization is used to select the hyperparameters of three preliminary outbound flow residual prediction models, and minimizing the mean square error is used as the objective function. The Nash efficiency coefficient and root mean square error are used to evaluate the performance of the three different preliminary outbound flow residual prediction models. The preliminary outbound flow residual prediction model with the best performance is selected as the final outbound flow residual prediction model for outbound flow residual prediction.

Images