Reservoir outlet water temperature rapid prediction method integrated with hydraulic connection

Abstract

The invention belongs to the technical field of reservoir water environment prediction, and discloses a reservoir outlet water temperature rapid prediction method integrated with hydraulic connection. The problems that in an existing reservoir outlet water temperature prediction scheme, numerical simulation efficiency is low, and a machine learning model lacks physical process support of hydraulic connection and a time lag effect, so that robustness and applicability are poor are solved. The method comprises the following steps: collecting reservoir related data and scheduling data; establishing a numerical model to calculate the dam front water body water ages of typical years such as a wet water year, a flat water year and a low water year, screening water age abnormal values by adopting a Raida criterion, and determining the maximum reasonable water age; taking the maximum reasonable water age as a time sequence length of an input feature, and establishing a time sequence prediction model of historical boundary data and ex-warehouse water temperature based on an LSTM algorithm; in practical application, boundary data of the maximum reasonable water age length before a to-be-predicted time period is input into the trained LSTM time sequence prediction model, and a reservoir outlet water temperature prediction result is output.

Description

Technical Field

[0001] This invention belongs to the field of reservoir water environment prediction technology, specifically relating to a method for rapid prediction of reservoir outflow water temperature that incorporates hydraulic connections. Background Technology

[0002] Reservoir outflow water temperature is a key environmental factor affecting the stability of downstream river ecosystems, directly impacting fish reproduction, aquatic habitat quality, crop irrigation adaptability, and the safety of drinking water for residents along the river. Therefore, accurate and rapid prediction of reservoir outflow water temperature is of significant practical importance for optimizing reservoir ecological management, downstream ecological environmental protection, and the rational utilization of water resources.

[0003] Currently, reservoir outflow water temperature prediction technologies are mainly divided into two categories: numerical simulation methods and machine learning-based rapid prediction methods, as detailed below:

[0004] (1) Numerical simulation method based on hydrodynamic-water temperature coupling

[0005] This type of method predicts the outlet water temperature by constructing a refined model to reconstruct the heat exchange and flow processes of the reservoir water. The physical mechanism of this method is clear, and it can achieve high prediction accuracy when supported by ideal data. However, its application has certain limitations: on the one hand, it relies on a large amount of basic data such as high-precision topographic data and refined meteorological data, resulting in high data acquisition costs; on the other hand, the model calculation process is complex, involving multi-physics coupling solutions, which is time-consuming and difficult to meet the actual needs of "rapid response" in reservoir scheduling decisions.

[0006] (2) Fast prediction methods based on machine learning

[0007] This type of method constructs a predictive model by mining the statistical correlation between boundary data such as inflow water temperature, inflow rate, and meteorological conditions and outflow water temperature. Existing machine learning prediction methods typically use common boundary data from numerical simulations as input features, employing a trial-and-error approach to construct input time series data of varying lengths and selecting the best-performing model. However, this trial-and-error method fails to consider the time lag effect of boundary factors on outflow water temperature—reservoir water undergoes a retention process in front of the dam, and changes in boundary conditions require a certain amount of time to be transmitted to the outflow section through hydraulic connections. This results in the selected model potentially only reflecting a strong correlation at the data level, rather than a true correlation at the physical process level. Consequently, the model exhibits poor robustness and applicability, and its prediction accuracy is prone to significant drops when hydrological annual patterns change or reservoir scheduling methods are adjusted.

[0008] In summary, existing methods for predicting reservoir outflow water temperature cannot simultaneously meet the core requirements of physical rationality, prediction accuracy, and rapid response. There is an urgent need to develop a rapid prediction technology that can quantify hydraulic connections and time lag effects, and provide physical process support for machine learning models, so as to provide scientific and reliable technical support for reservoir ecological scheduling and downstream ecological protection. Summary of the Invention

[0009] The technical problem to be solved by this invention is to provide a method for rapid prediction of reservoir outflow water temperature that incorporates hydraulic connections, thereby solving the problems of low efficiency in numerical simulation and poor robustness and applicability of existing reservoir outflow water temperature prediction schemes due to the lack of physical process support from hydraulic connections and time lag effects in machine learning models.

[0010] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0011] A method for rapid prediction of reservoir outflow water temperature incorporating hydraulic connections includes the following steps:

[0012] S1. Collect basic monitoring data and reservoir operation data for typical years. The basic monitoring data includes continuous monitoring data of inflow temperature, inflow rate, outflow rate, reservoir water level, and outflow temperature, as well as synchronous meteorological conditions. The typical year operation data includes high-water years, normal-water years, and low-water years.

[0013] S2. Based on reservoir topographic data, typical year scheduling data and input-output boundary conditions of the year, a hydrodynamic-temperature coupled numerical model is constructed to simulate the water transport and diffusion process in front of the reservoir dam in a typical year and calculate the water age time series. After outlier processing, statistical analysis is performed to obtain the maximum reasonable water age of the reservoir under the typical year scenario.

[0014] S3. Using the maximum reasonable water age as the time series length, select the inflow water temperature, inflow rate, meteorological conditions, outflow rate, and water level data of the corresponding time step before the prediction time as input features, and use the outflow water temperature at the prediction time as the output feature to construct an LSTM time series prediction model and train it.

[0015] S4. Input the boundary data of the maximum reasonable water age length before the period to be predicted into the trained LSTM time series prediction model, and output the reservoir outflow water temperature prediction result; the boundary data includes inflow water temperature, inflow flow, meteorological conditions, outflow flow and water level data.

[0016] Furthermore, in step S1, the time resolution of the basic monitoring data is on a daily or hourly scale, and the continuous monitoring duration is not less than 3 years; the meteorological conditions include air temperature, sunshine duration, wind speed, relative humidity, cloud cover, and dew point temperature; the reservoir scheduling data includes: reservoir operation mode, gate scheduling rules, and beneficial scheduling plan.

[0017] Furthermore, in step S2, the boundary conditions of the hydrodynamic-temperature coupled numerical model are set as the inflow, inflow temperature, meteorological conditions and outflow process of a typical year, and the initial conditions are the reservoir water level and water temperature distribution at the beginning of a typical year.

[0018] Furthermore, in step S2, the outlier handling includes:

[0019] The Laida criterion is used to handle outliers in the calculated water age data. Specifically, by calculating the mean μ and standard deviation σ of the water age data, water age data that exceed the range of [μ-3σ, μ+3σ] are identified as outliers and removed.

[0020] Furthermore, in step S2, statistical analysis yields the maximum reasonable water age for reservoir operation under typical annual scenarios, including:

[0021] The water age data for three typical years—high-water year, normal-water year, and low-water year—were statistically analyzed according to the three vertical positions of the reservoir: surface, middle, and bottom. The longest reasonable water age for each vertical position was selected among all typical years, ultimately yielding three maximum reasonable water ages that correspond one-to-one with the surface, middle, and bottom layers of the reservoir.

[0022] Furthermore, in step S3, the constructed LSTM time series prediction model includes an input layer, 2-4 hidden layers, and an output layer. The input dimension of the input layer is the number of features × the time series length, and the output dimension of the output layer is 1. Each hidden layer has 32-12 neurons, and the activation function is the ReLU function.

[0023] Furthermore, in step S3, the training includes: dividing the collected basic monitoring data into a training set, a validation set, and a test set in a ratio of 6:3:1; training the model using the training set; optimizing the model parameters using the validation set; and finally verifying the model performance using the test set, and judging the model performance using the mean absolute error (MAE) and root mean square error (RMSE).

[0024] Furthermore, in step S2, a hydrodynamic-temperature coupled numerical model is constructed based on reservoir topographic data, typical annual scheduling data, and the input-output boundary conditions of that year, including:

[0025] Construct a two-dimensional basic model of the reservoir facade with an average width:

[0026] ;

[0027] ;

[0028] ;

[0029] ;

[0030] ;

[0031] ;

[0032] Where B is the width of the river; u and w are the velocity components in the longitudinal (x) and vertical (z) directions, respectively; ν e For the molecular viscosity coefficient ν and the turbulent eddy viscosity coefficient ν t The sum of, ν t =C μ k 2 / ε;ρ w ρ is the density of water; p is the pressure; β is the coefficient of thermal expansion of water; ΔT is the change in water temperature; k is the turbulent kinetic energy; ε is the turbulent kinetic energy dissipation rate; σ k and σ ε Prandtl numbers for turbulent kinetic energy and dissipation rate, respectively; C μ C ε1 and C ε2 For each of the empirical constants, we take values ​​of 0.09, 1.44, and 1.92, respectively; G k For the generation of turbulent kinetic energy, G b This is a buoyancy-generating term. In stable water stratification, this term can suppress the generation of turbulent kinetic energy and weaken the downward transfer of heat, making it an important factor in maintaining stable water stratification in reservoirs. ;T w Water temperature; C p φ is the specific heat capacity of water. z σ represents the shortwave solar radiation flux that passes through the surface of the water and across the z-plane. T The Prandtl number represents the temperature.

[0033] Constructing the water age coupling equation:

[0034] ;

[0035] ;

[0036] in, Let V be the density of the water, and V be the volume of the control volume. where t is the age concentration, u is the longitudinal flow velocity, and A is the longitudinal flow velocity. z D represents the vertical water flow area. x D is the longitudinal dispersion coefficient. z Q is the vertical dispersion coefficient; in For a single inflow, Q out For a single outflow flow rate S a For source and sink terms, m is the inflow contribution, and the water age of the newly flowing water body is 0, i.e., m=0;

[0037] By integrating the two-dimensional basic model of the reservoir facade with the water age coupling equation, a hydrodynamic-water temperature coupled numerical model is obtained.

[0038] Furthermore, in step S4, the maximum reasonable water age length is the maximum reasonable water age corresponding to the middle layer of the reservoir.

[0039] The beneficial effects of this invention are:

[0040] (1) Balancing physical plausibility with prediction accuracy:

[0041] This invention quantifies the hydraulic connections and time-lag effects of the water body in front of the reservoir dam by measuring water age. It uses the maximum reasonable water age of the surface, middle, and bottom layers in typical years (wet, normal, and dry) as the basis for the length of the model's input time series. This ensures that the correlation between input features and outflow water temperature is based on the physical process of water retention and hydraulic transfer, making the model's predictions more closely aligned with actual water temperature evolution. Simultaneously, the input series selected based on physical mechanisms can more accurately capture the long-term cumulative impact of boundary factors on outflow water temperature, improving the model's prediction accuracy and stability under different hydrological scenarios and scheduling methods.

[0042] (2) Improve prediction efficiency and adapt to real-time scheduling requirements:

[0043] This invention employs an LSTM time series forecasting model to construct a rapid forecasting framework, eliminating the need for complex hydrodynamic-temperature coupled numerical calculations and avoiding the reliance on massive amounts of basic data and lengthy computational processes inherent in traditional numerical simulation methods. While maintaining sufficient forecasting accuracy, the forecasting speed is improved by 1-2 orders of magnitude compared to traditional numerical simulation methods, enabling rapid response to real-time forecasting needs in reservoir scheduling decisions and providing efficient data support for timely adjustments to scheduling plans.

[0044] (3) Robustness and wide applicability:

[0045] This invention, through typical annual water age analysis covering high-water, normal-water, and low-water years, determines a maximum reasonable water age that can adapt to reservoir operation states under different hydrological year types, ensuring stable predictive performance of the model even when hydrological scenarios change or scheduling methods are adjusted. Furthermore, the method has moderate data resolution requirements (supporting daily or hourly scale data), does not require refined topographic and meteorological data, and is easy to promote and apply in engineering projects. Attached Figure Description

[0046] Figure 1 This is a flowchart of a method for rapid prediction of reservoir outflow water temperature incorporating hydraulic connections, as described in an embodiment of the present invention.

[0047] Figure 2This is a comparison chart of the LSTM prediction model's performance on the validation set for predicting outflow water temperature under three different input sequence lengths in this embodiment of the invention.

[0048] Figure 3 This is a comparison chart of the LSTM prediction model's performance on the test set for predicting outflow water temperature under three different input sequence lengths in this embodiment of the invention. Detailed Implementation

[0049] This invention aims to provide a rapid prediction method for reservoir outflow water temperature that incorporates hydraulic connections, addressing the problems of low efficiency in numerical simulation and poor robustness and applicability in existing reservoir outflow water temperature prediction schemes due to the lack of physical process support from hydraulic connections and time lag effects in machine learning models. Its core idea is to construct a reservoir outflow water temperature prediction method that balances physical rationality and rapid response capability by using water age quantification of hydraulic connections and time lag effects as physical support and combining the time-series prediction advantages of LSTM models. Specifically, the technical means employed by this invention to achieve the above core idea include:

[0050] (1) In view of the problem that existing machine learning models rely on trial and error to determine the length of the input sequence and lack physical process support, this paper quantifies the time lag effect of the influence of boundary factors on the temperature of the outflowing water due to the retention of water in the reservoir by using water age (the retention time of water from the inflow to the dam) to provide a scientific basis for the length of the input sequence of the model at the physical level and avoid false correlations at the data level.

[0051] (2) First, the hydrodynamic-water temperature coupled numerical model is used to simulate the water transport and diffusion process in front of the dam in three typical years, calculate the water age time series and select the maximum reasonable water age; then, the maximum reasonable water age is used as the input time series length to construct an LSTM prediction model. This not only ensures the physical correlation between the input features and the output by means of numerical simulation, but also breaks through the limitation of low efficiency of traditional numerical simulation by means of machine learning.

[0052] (3) By analyzing the water age data of high-water years, normal-water years, low-water years and surface, middle and bottom layers, multiple sets of maximum reasonable water ages are determined so that the model input can be adapted to the hydraulic characteristics of different hydrological year types and vertical positions of the reservoir area, and ensure that the prediction accuracy can still be maintained when the hydrological scenario changes or the scheduling mode is adjusted, thus solving the problem of insufficient robustness of traditional models.

[0053] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0054] This example uses a medium-sized reservoir (total capacity 11.2 × 10⁻⁶ m³ / h) as an example. 8 m 3 Using a dam (68m high) as the research object, the method of this invention was used to predict the outflow water temperature. See also... Figure 1The rapid prediction method for reservoir outflow water temperature incorporating hydraulic connections provided in this embodiment includes the following implementation steps:

[0055] 1. Data Collection

[0056] The core objective of this step is to obtain sufficient duration and multi-dimensional monitoring and scheduling data to provide comprehensive support for subsequent water age calculations and model training, ensuring that the data covers different hydrological year types and reservoir operating states.

[0057] Basic monitoring data collection: In this embodiment, daily-scale monitoring data were collected for three typical years in the study area: dry year, normal year, and wet year. The data included: inflow water temperature range (4.7-17.7℃), inflow flow range (98-2869 m³ / s), outflow flow range (136-2459 m³ / s), outflow water temperature range (7.8-27.3℃), air temperature range (-1.1-30.5℃), wind speed range (0.2-4.0 m / s), relative humidity range (32%-98%), and solar radiation range (14-359 W / m²). 2 ).

[0058] Typical year dispatch data collection: This example uses data from the watershed hydrological stations, selecting 2018 (a high-water year with a total annual runoff of 16.3 × 10⁻⁶). 9 m³), 2015 (normal water year, annual runoff 14.2×10 8 m³), 2011 (dry year, annual runoff 6.8×10 8 (m³) is a typical year, and reservoir gate scheduling logs, beneficial scheduling plans and operation reports for the above years are collected.

[0059] 2. Water age calculation and determination of the maximum reasonable water age

[0060] This step involves reconstructing the water transport process through numerical simulation, quantifying the hydraulic time delay between the reservoir inlet and the dam outlet using water age, and selecting the maximum reasonable water age as the physical basis for the length of the model input sequence.

[0061] In this embodiment, a hydrodynamic-temperature coupling model based on the CE-QUAL-W2 framework is adopted, and the construction process includes:

[0062] (1) Construct a two-dimensional basic model of the reservoir facade with average width, which includes the continuity equation shown in formula (1.1), the momentum equation shown in (1.2a) and (1.2b), the turbulence model k-ε equation shown in (1.3a) and (1.3b), and the energy equation shown in (1.4).

[0063] ; (1.1)

[0064] (1.2a)

[0065] (1.2b)

[0066] (1.3a)

[0067] (1.3b)

[0068] (1.4)

[0069] Where B is the width of the river; u and w are the velocity components in the longitudinal (x) and vertical (z) directions, respectively; ν e For the molecular viscosity coefficient ν and the turbulent eddy viscosity coefficient ν t The sum of, ν t =C μ k 2 / ε;ρ w ρ is the density of water; p is the pressure; β is the coefficient of thermal expansion of water; ΔT is the change in water temperature; k is the turbulent kinetic energy; ε is the turbulent kinetic energy dissipation rate; σ k and σ ε Prandtl numbers for turbulent kinetic energy and dissipation rate, respectively; C μ C ε1 and C ε2 For each of the empirical constants, we take values ​​of 0.09, 1.44, and 1.92, respectively; G k For the generation of turbulent kinetic energy, G b This is a buoyancy-generating term. In stable water stratification, this term can suppress the generation of turbulent kinetic energy and weaken the downward transfer of heat, making it an important factor in maintaining stable water stratification in reservoirs. ;T w Water temperature; C p φ is the specific heat capacity of water. z σ represents the shortwave solar radiation flux that passes through the surface of the water and across the z-plane. T The Prandtl number represents the temperature.

[0070] (2) Construct the water age coupling equation, which includes the age transport control equation shown in formula (2.1) and the source-sink term refinement formula shown in (2.1a).

[0071] ; (2.1)

[0072] ; (2.1a)

[0073] in, Let V be the density of the water, and V be the volume of the control volume. where t is the age concentration, u is the longitudinal flow velocity, and A is the longitudinal flow velocity. z D represents the vertical water flow area. x D is the longitudinal dispersion coefficient. z Q is the vertical dispersion coefficient; in For a single inflow, Q out For a single outflow flow rate S a For source and sink terms, m is the inflow contribution, and the water age of the newly flowing water body is 0, i.e., m=0;

[0074] (3) By integrating the above-mentioned two-dimensional basic model of the reservoir facade with the water age coupling equation, a hydrodynamic-water temperature coupling numerical model is obtained.

[0075] Input settings: Based on the above hydrodynamic-temperature coupling model, input the actual topographic data of the reservoir, set the model calculation domain to the area from the reservoir inlet section to the dam axis, and divide it into 42 sections from the inlet to the dam front. The vertical grid accuracy is 1m, with a total of 121 layers. The boundary conditions are set to the daily scale inflow, inflow temperature, meteorological data and outflow process of a typical year. The initial water level is the measured water level on January 1 of a typical year.

[0076] Water age calculation: A hydrodynamic-temperature coupled model was run to simulate the transport and diffusion process of water in front of the reservoir dam in a typical year, outputting daily-scale water age data in front of the dam. The vertical output accuracy was 10m, and the output range was from 5m to 85m above the surface. The Laida criterion was used to process the water age data, and the mean and standard deviation of water age data for typical years of abundant, normal, and dry seasons were calculated respectively. 27 abnormal water age values ​​outside the range of [μ-3σ, μ+3σ] were removed. The specific processing results are shown in Tables 1, 2, and 3 below. After outlier removal, the average water age in abundant years was 44.32 days, with a maximum water age of 188.01 days; the average water age in normal years was 47.83 days, with a maximum water age of 200.73 days; and the average water age in dry years was 42.1 days, with a maximum water age of 230 days.

[0077] Table 1. Results of Outlier Handling in Dry Years

[0078]

[0079] Table 2 Results of handling outliers during normal water years

[0080]

[0081] Table 3 Results of handling outliers during high-water years

[0082]

[0083] Determination of the maximum reasonable water age: Based on the statistical data of water age in the surface (5m), middle (35m), and bottom (75m) layers during wet, normal, and dry years, the longest effective retention time in the surface water age data of the three typical years is 84 days. Therefore, the maximum reasonable water age T1 for the surface layer (5m) is determined to be 84 days. The longest effective retention time in the middle layer water age data of the three typical years is 106 days. Therefore, the maximum reasonable water age T2 for the middle layer (35m) is determined to be 106 days. The longest effective retention time in the bottom layer water age data of the three typical years is 136 days. Therefore, the maximum reasonable water age T3 for the bottom layer (75m) is determined to be 136 days.

[0084] Three input length sequences were obtained, which will be used as the boundary input condition sequence lengths of the downstream water temperature prediction model.

[0085] 3. Predictive Model Establishment and Validation

[0086] This step utilizes the strong learning capability of the LSTM model on time series data, uses the maximum reasonable water age as the input sequence length, constructs a prediction model, and ensures that the model accuracy meets engineering requirements through training and validation, while also taking prediction efficiency into account.

[0087] (1) Input-output characteristics:

[0088] Eight features were selected as inputs: inflow water temperature, inflow rate, air temperature, sunshine duration, wind speed, relative humidity, outflow rate, and water level for the previous 84 days, 106 days, and 136 days, respectively, and the outflow water temperature on the current day was used as the output.

[0089] (2) LSTM model construction:

[0090] A unified LSTM time series prediction model framework is constructed, adapting to three schemes simply by adjusting the input sequence length. The model structure design is as follows:

[0091] Input layer: The input dimension is the number of features × the length of the time series, where the number of features is 8 and the lengths of the time series are T1, T2, and T3, respectively.

[0092] Hidden layers: Two hidden layers are set, with 64 neurons in the first layer and 32 neurons in the second layer. The ReLU activation function is used (to effectively avoid the gradient vanishing problem and improve the stability of model training).

[0093] Output layer: The output dimension is 1, corresponding to the predicted value of the outflow water temperature.

[0094] Optimizer and loss function: The Adam optimizer (learning rate 0.001, balancing training speed and convergence accuracy) is used, and the loss function is mean squared error (sensitive to the error of regression tasks, suitable for the accuracy requirements of water temperature prediction).

[0095] (3) Model training and validation:

[0096] The data from 2011 to 2017 (2321 days in total) was used as the training set, the data from 2017 to 2020 (1107 days in total) was used as the validation set, and the data from 2020 to 2021 (365 days in total) was used as the test set.

[0097] The training iterations were set to 100. The three sets of input sequence schemes were fed into the model for training. During the training process, the model performance was monitored in real time using the validation set to avoid overfitting.

[0098] The training iterations are 100. The model is trained using the training set, the model parameters are optimized using the validation set, and finally the model performance is evaluated using the test set.

[0099] Accuracy verification: The mean absolute error (MAE) and root mean square error (RMSE) are used as evaluation indicators. The verification standard is MAE ≤ 0.5℃ and RMSE ≤ 0.8℃.

[0100] For a comparison of the LSTM prediction model's performance on the validation set for outflow water temperature under three different input sequence lengths (T1, T2, and T3), please refer to the graph. Figure 2 See the comparison chart of the prediction results for outflow water temperature on the prediction set. Figure 3 It can be seen that, in both the validation and test sets, the simulation effect of reservoir discharge water temperature is best when T2=106 days, with MAE and RMSE being the lowest values ​​among the three operating conditions. This indicates that determining the LSTM boundary input time sequence based on the maximum reasonable water age of the middle layer results in the best model prediction performance.

[0101] 4. Practical Application of the Model

[0102] In this step, the trained LSTM time series prediction model is applied to a real-world scenario. After inputting boundary data, it enables fast and accurate prediction of reservoir outflow water temperature. As can be seen from step 3 above, the optimal time series for the boundary data is based on the maximum reasonable water age of the middle layer.

[0103] Although embodiments of the present invention have been described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present invention, and all such changes and alterations shall not depart from the protection scope of the present invention.

Claims

1. A method for rapid prediction of reservoir outflow water temperature incorporating hydraulic connections, characterized in that, Includes the following steps: S1. Collect basic monitoring data and reservoir operation data for typical years. The basic monitoring data includes continuous monitoring data of inflow temperature, inflow rate, outflow rate, reservoir water level, and outflow temperature, as well as synchronous meteorological conditions. The typical year operation data includes high-water years, normal-water years, and low-water years. S2. Based on reservoir topographic data, typical year scheduling data and input-output boundary conditions of the year, a hydrodynamic-temperature coupled numerical model is constructed to simulate the water transport and diffusion process in front of the reservoir dam in a typical year and calculate the water age time series. After outlier processing, statistical analysis is performed to obtain the maximum reasonable water age of the reservoir under the typical year scenario. S3. Using the maximum reasonable water age as the time series length, select the inflow water temperature, inflow rate, meteorological conditions, outflow rate, and water level data of the corresponding time step before the prediction time as input features, and use the outflow water temperature at the prediction time as the output feature to construct an LSTM time series prediction model and train it. S4. Input the boundary data of the maximum reasonable water age length before the period to be predicted into the trained LSTM time series prediction model, and output the reservoir outflow water temperature prediction result; the boundary data includes inflow water temperature, inflow flow, meteorological conditions, outflow flow and water level data.

2. The method for rapid prediction of reservoir outflow water temperature incorporating hydraulic connections as described in claim 1, characterized in that, In step S1, the time resolution of the basic monitoring data is on a daily or hourly scale, and the continuous monitoring duration is not less than 3 years; the meteorological conditions include air temperature, sunshine duration, wind speed, relative humidity, cloud cover, and dew point temperature; the reservoir scheduling data includes: reservoir operation mode, gate scheduling rules, and beneficial scheduling plan.

3. The method for rapid prediction of reservoir outflow water temperature incorporating hydraulic connections as described in claim 1, characterized in that, In step S2, the boundary conditions of the hydrodynamic-temperature coupled numerical model are set as the inflow, inflow temperature, meteorological conditions and outflow process of a typical year, and the initial conditions are the reservoir water level and water temperature distribution at the beginning of a typical year.

4. The method for rapid prediction of reservoir outflow water temperature incorporating hydraulic connections as described in claim 1, characterized in that, In step S2, the outlier handling includes: The Laida criterion is used to handle outliers in the calculated water age data. Specifically, by calculating the mean μ and standard deviation σ of the water age data, water age data that exceed the range of [μ-3σ, μ+3σ] are identified as outliers and removed.

5. The method for rapid prediction of reservoir outflow water temperature incorporating hydraulic connections as described in claim 1, characterized in that, In step S2, statistical analysis is used to obtain the maximum reasonable water age for reservoir operation under typical annual scenarios, including: The water age data for three typical years—high-water year, normal-water year, and low-water year—were statistically analyzed according to the three vertical positions of the reservoir: surface, middle, and bottom. The longest reasonable water age for each vertical position was selected among all typical years, ultimately yielding three maximum reasonable water ages that correspond one-to-one with the surface, middle, and bottom layers of the reservoir.

6. The method for rapid prediction of reservoir outflow water temperature incorporating hydraulic connections as described in claim 1, characterized in that, In step S3, the constructed LSTM time series prediction model includes an input layer, 2-4 hidden layers, and an output layer. The input dimension of the input layer is the number of features × the length of the time series, and the output dimension of the output layer is 1. Each hidden layer has 32-12 neurons, and the ReLU function is used as the activation function.

7. The method for rapid prediction of reservoir outflow water temperature incorporating hydraulic connections as described in claim 1, characterized in that, In step S3, the training includes: dividing the collected basic monitoring data into a training set, a validation set, and a test set in a ratio of 6:3:1; training the model using the training set; optimizing the model parameters using the validation set; and finally verifying the model performance using the test set. The mean absolute error (MAE) and root mean square error (RMSE) are used to judge the model performance.

8. The method for rapid prediction of reservoir outflow water temperature incorporating hydraulic connections as described in claim 1, characterized in that, In step S2, a hydrodynamic-temperature coupled numerical model is constructed based on reservoir topographic data, typical annual scheduling data, and the input-output boundary conditions of that year, including: Construct a two-dimensional basic model of the reservoir facade with an average width: ； ； ； ； ； ； Where B is the width of the river; u and w are the velocity components in the longitudinal (x) and vertical (z) directions, respectively; ν e For the molecular viscosity coefficient ν and the turbulent eddy viscosity coefficient ν t The sum of, ν t =C μ k 2 / ε;ρ w ρ is the density of water; p is the pressure; β is the coefficient of thermal expansion of water; ΔT is the change in water temperature; k is the turbulent kinetic energy; ε is the turbulent kinetic energy dissipation rate; σ k and σ ε Prandtl numbers for turbulent kinetic energy and dissipation rate, respectively; C μ C ε1 and C ε2 For each of the empirical constants, we take values ​​of 0.09, 1.44, and 1.92, respectively; G k For the generation of turbulent kinetic energy, G b This is a buoyancy-generating term. In stable water stratification, this term can suppress the generation of turbulent kinetic energy and weaken the downward transfer of heat, making it an important factor in maintaining stable water stratification in reservoirs. ;T w Water temperature; C p φ is the specific heat capacity of water. z σ represents the shortwave solar radiation flux that passes through the surface of the water and across the z-plane. T The Prandtl number represents the temperature. Constructing the water age coupling equation: ； ； in, Let V be the density of the water, and V be the volume of the control volume. where t is the age concentration, u is the longitudinal flow velocity, and A is the longitudinal flow velocity. z D represents the vertical water flow area. x D is the longitudinal dispersion coefficient. z Q is the vertical dispersion coefficient; in For a single inflow, Q out For a single outflow flow rate S a For source and sink terms, m is the inflow contribution, and the water age of the newly flowing water body is 0, i.e., m=0; By integrating the two-dimensional basic model of the reservoir facade with the water age coupling equation, a hydrodynamic-water temperature coupled numerical model is obtained.

9. The method for rapid prediction of reservoir outflow water temperature incorporating hydraulic connections as described in claim 5, characterized in that, In step S4, the maximum reasonable water age length is the maximum reasonable water age corresponding to the middle layer of the reservoir.

Images