Method and system for predicting upstream discharged water temperature and regulating and controlling flexible water retaining curtain wall

Abstract

The invention belongs to the technical field of reservoir water temperature prediction and regulation and control, and particularly discloses an upstream discharged water temperature prediction and flexible water retaining curtain wall regulation and control method and system. According to the method, a physical mechanism of an LSTM neuron structure is reconstructed, a water body stratification stability mechanism and a vertical thermal diffusion mechanism are deeply fused in a deep learning network, an improved LSTM model can well predict a water temperature stratification phenomenon of upstream incoming water, and high-precision prediction of vertical water temperature distribution of a reservoir under a complex meteorological condition is realized. A water temperature mixing model calculation formula is created for the first time, and the problem that the discharged water temperature requirement is difficult to calculate is solved. The vertical water temperature predicted by the improved LSTM is combined with the discharged water temperature inverted by the downstream ecological target, and the optimal water taking elevation is compared and determined, so that accurate lifting regulation and control of the flexible water retaining curtain wall are guided, the ecological water temperature requirement of a downstream river channel is strictly guaranteed, meanwhile, water resource waste is effectively avoided, and the power generation benefit and the flux utilization rate of a reservoir are maximized.

Description

Technical Field

[0001] This application belongs to the field of reservoir water temperature prediction and control technology, and more specifically, relates to a method and system for predicting upstream discharge water temperature and controlling flexible water-retaining curtain walls. Background Technology

[0002] The temperature of water released from upstream reservoirs has a significant impact on the temperature of downstream waters. Specifically, the release of cold water from upstream causes a drop in downstream water temperature, and this temperature change poses a serious threat to fish and irrigated crops in the downstream ecosystem. The decrease in water temperature not only affects the survival and reproduction of fish but may also adversely affect the growth and yield of crops. Therefore, precise control of the temperature of upstream released water is crucial for protecting the downstream ecological environment and agricultural activities.

[0003] Currently, a flexible water barrier structure is used to block the lower layer of low-temperature water, allowing the upper layer of normal-temperature water to flow through. However, on the one hand, existing technologies use sensors to collect temperature information and then regulate the water barrier to control the temperature of the discharged water. This method suffers from slow decision-making speed and the possibility of inaccurate data acquisition due to external factors. On the other hand, existing technologies use classical LSTM for vertical water temperature prediction. Due to insufficient modeling of the spatiotemporal coupling characteristics and layered thermal structure of the water body in the vertical direction, the prediction results have limited accuracy in spatial distribution, especially at locations with drastic changes in water temperature gradients. Furthermore, existing water barrier control systems may not be able to precisely control the discharged water flow, resulting in water that does not meet temperature requirements being discharged as well. This inaccurate regulation may lead to the waste of large amounts of water that does not meet temperature requirements, resulting in water resource waste and reduced power generation efficiency. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this application is to provide a method and system for predicting upstream discharge water temperature and regulating flexible water-retaining curtain walls, aiming to solve the problems of traditional water temperature prediction models lacking physical mechanism constraints and having lagging regulation.

[0005] To achieve the above objectives, in a first aspect, this application provides a method for predicting upstream outflow water temperature, comprising: acquiring multi-source time-series meteorological and hydrological data of the reservoir prior to the current time step and measured water temperature sequences at different depths, inputting them into an upstream outflow water temperature prediction model to obtain the predicted vertical water temperature distribution data of the reservoir at the current time step; the upstream outflow water temperature prediction model is obtained by training an improved LSTM based on multi-source time-series meteorological and hydrological data of the reservoir and measured water temperature data at multiple depths, wherein the improved LSTM is used to incorporate the changes in upstream water body stratification stability over time into the genetic phylogenetic tree, and at the same time incorporate the changes in upstream vertical thermal diffusivity over space into the updated hidden state.

[0006] Preferably, the improved LSTM forget gate is as follows:

[0007] in, Forgot Gate Current Time Step Network layer The output, Here is the weight matrix for the forget gate. For the previous time step Network layer The hidden state, For the current time step Network layer Input, For the bias term of the forget gate, This is the time-physical weight matrix.

[0008] Preferably, the upstream stratification stability coefficient The calculation formula is as follows:

[0009] in, It is the hyperbolic tangent function. For the current time step The temperature For the previous time step Upstream surface water temperature, For the current time step wind speed, This is the scaling factor. It is a local constant.

[0010] Preferably, the updated hidden state of the improved LSTM is as follows:

[0011] in,

[0012]

[0013] in, For the current time step Network layer The hidden state, Output the current time step of the gate Network layer The output, It is the hyperbolic tangent function. For the current time step Network layer The final state of the unit, The thermal conduction gating factor is... It is the Sigmoid activation function. This is the space physics weight matrix. For the current time step Upstream vertical vortex expansion coefficient, This is the thermal conduction gate bias vector.

[0014] Preferably, the upstream vertical vortex expansion coefficient The calculation formula is as follows:

[0015] in, It is the Sigmoid activation function. For the current time step The upstream water is deep. For the current time step wind speed, This is the scaling factor. It is a local constant.

[0016] To achieve the above objectives, in a second aspect, this application provides a method for regulating a flexible water-retaining curtain wall, comprising: calculating the downstream water temperature at the upstream intake in the current time step; using the prediction method described in the first aspect to obtain the predicted distribution data of the vertical water temperature of the reservoir in the current time step; comparing the predicted vertical water temperature distribution data with the calculated downstream water temperature at the upstream intake in real time to determine the water layer depth that best matches it, and regulating the flexible water-retaining curtain wall accordingly.

[0017] Preferably, the calculation of the downstream water temperature at the upstream intake at the current time step is as follows: Determine the expected value of downstream cross-section water temperature, the average water depth of downstream channel, the average air temperature during the forecast period, and the average solar radiation during the forecast period corresponding to the current time step; The above parameters are substituted into the longitudinal water temperature inversion model based on flow velocity time delay correction to obtain the water temperature discharged from the upstream intake at the current time step. The longitudinal water temperature inversion model based on flow velocity time delay correction is as follows:

[0018] in, For the current time step Upstream water flow The actual water temperature at the downstream section after time transmission; For the current time step Water temperature discharged from the upstream intake; This is the time it takes for water to flow from upstream to downstream. Each represents the current time step. Average temperature and average solar radiation; The thermal retention coefficient is related to distance and flow velocity. An empirical coefficient for converting solar radiation energy into water temperature changes. This represents the average water depth in the downstream river channel.

[0019] To achieve the above objectives, in a third aspect, this application provides an upstream discharge water temperature prediction system, including a memory and one or more processors; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions; the one or more processors invoke the computer instructions to cause the system to execute the prediction method as described in the first aspect.

[0020] To achieve the above objectives, in a fourth aspect, this application provides a flexible water-retaining curtain wall control system, including a memory and one or more processors; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions; the one or more processors invoke the computer instructions to cause the system to execute the control method as described in the second aspect.

[0021] To achieve the above objectives, in a fifth aspect, this application provides a computer-readable storage medium including instructions that, when executed on an electronic device, cause the electronic device to perform the method as described in the first or second aspect.

[0022] It is understood that the beneficial effects of the third to fifth aspects mentioned above can be found in the relevant descriptions in the first and second aspects mentioned above, and will not be repeated here.

[0023] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art: Firstly, this application reconstructs the physical mechanism of the traditional LSTM neuron structure, deeply integrating the water stratification stability mechanism and the vertical heat diffusion mechanism into the deep learning network. The improved LSTM model can effectively predict the water temperature stratification phenomenon of upstream water, achieving high-precision prediction of the vertical water temperature distribution of reservoirs under complex meteorological conditions, so as to protect the downstream ecological environment and improve the utilization efficiency of water resources.

[0024] Secondly, this application pioneered a water temperature hybrid model calculation formula, which combines time delay, along-path heating, and target inversion to form a physically interpretable, computationally efficient, and real-time control hybrid model. This effectively solves the problems of "lag, inaccuracy, and difficulty in reverse calculation" in traditional methods for dynamic water temperature control, and also solves the problem of difficulty in calculating downstream water temperature requirements.

[0025] Thirdly, this application uses improved LSTM to predict vertical water temperature, combined with downstream ecological target-derived discharge water temperature, to compare and determine the optimal water intake elevation, thereby guiding the precise raising and lowering control of the flexible water-retaining wall. This effectively avoids water waste and maximizes the power generation efficiency and flux utilization rate of the reservoir while strictly ensuring the downstream river's ecological water temperature requirements (such as fish breeding temperature). Attached Figure Description

[0026] Figure 1 This is a flowchart of a method and system for predicting upstream discharge water temperature and regulating flexible water-retaining curtain walls, provided in an embodiment of this application.

[0027] Figure 2 This is a schematic diagram of multi-source time-series meteorological and hydrological data of a reservoir provided in an embodiment of this application.

[0028] Figure 3 This is a schematic diagram of the water temperature time series at different depths provided in the embodiments of this application.

[0029] Figure 4 This is a diagram illustrating the training process of the ST-PC-LSTM model provided in the embodiments of this application.

[0030] Figure 5 This is a comparison chart of predicted and actual water temperatures at different depths provided in the embodiments of this application.

[0031] Figure 6 This is a fitting graph of predicted and actual values ​​provided in the embodiments of this application.

[0032] Figure 7 This is a schematic diagram of water temperature prediction error provided in the embodiments of this application.

[0033] Figure 8 This is a schematic diagram of the generated physical coefficient curve provided in the embodiments of this application. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0035] In this application, the term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A existing alone, A and B existing simultaneously, and B existing alone. In this application, the symbol " / " indicates that the related objects are in an "or" relationship, for example, A / B means A or B.

[0036] In this application, the terms "first" and "second," etc., are used to distinguish different objects, not to describe a specific order of objects. For example, "first response message" and "second response message," etc., are used to distinguish different response messages, not to describe a specific order of response messages.

[0037] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0038] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more, for example, multiple processing units means two or more processing units, multiple elements means two or more elements, etc.

[0039] The embodiments of this application are described below with reference to the accompanying drawings.

[0040] like Figure 1 As shown, this application proposes a method for predicting upstream discharge water temperature and regulating flexible water-retaining curtain walls.

[0041] I. Analysis of downstream water temperature demand and determination of upstream target temperature through inversion.

[0042] Considering the hysteresis of traditional static mixing water temperature models, this application adopts a novel "longitudinal water temperature inversion model based on flow velocity time delay correction" to construct a model that includes time variables. Functions:

[0043] in, Indicates the current time step Upstream water flow The actual water temperature at the downstream section after time transmission; Indicates the current time step The water temperature discharged from the upstream intake; This refers to the time it takes for water to travel from upstream to downstream. The upstream and downstream water temperatures do not change synchronously; there is a dynamic time lag. That is, the lag time is related to the upstream flow. The function; Each represents the current time step. The average air temperature and average solar radiation are used to represent the heating or cooling effect of air temperature and solar radiation on water temperature along the way. The thermal retention coefficient is related to distance and flow velocity. An empirical coefficient for converting solar radiation energy into water temperature changes. The average water depth of the river channel.

[0044] II. Using an improved LSTM, predict the vertical water temperature distribution data of the reservoir at the current time step.

[0045] The information processing order in a traditional LSTM is as follows: First, the forget gate determines the cell state. Which information is discarded; then, the input gate determines which new information needs to be written to the cell state. Simultaneously generate candidate states The new information is then used as its specific content; subsequently, the cell state is updated by combining the results of the forget gate and the input gate. Next, the output gate determines which information to output from the updated cell state; finally, based on the result of the output gate, the hidden state is updated. Complete the processing of the current time step.

[0046] This application reconstructs the physical mechanism of the traditional LSTM neuron structure and deeply integrates the water stratification stability mechanism and the vertical heat diffusion mechanism into the deep learning network to achieve high-precision prediction of the vertical water temperature distribution of reservoirs under complex meteorological conditions, so as to protect the downstream ecological environment and improve the efficiency of water resource utilization.

[0047] The improved LSTM, hereinafter referred to as the Spatiotemporal Physical Coupled Long Short-Term Memory Network (ST-PC-LSTM), comprises the following six parts.

[0048] Part 1: The Improved Forget Gate

[0049]

[0050] in, Forgot Gate Current Time Step Network layer The output, For activation functions, the Sigmoid function is typically used. Here is the weight matrix for the forget gate. For the previous time step Network layer The hidden state, For the current time step Network layer Input, For the bias term of the forget gate, This is the time-physical weight matrix, used to convert the scalar-form layered stability coefficients. Mapped to the forget gate Same vector dimension.

[0051] This application has been approved. This introduces temporal physical constraints into the genetics framework. During model training, the network automatically adjusts based on prediction errors using a backpropagation algorithm. Internal numerical values ​​allow the model to adaptively learn the quantitative impact of physical stability on memory retention. That is, when... When the water body is unstable, This will drive the forget gate to output a smaller value, thus accelerating the forgetting of the old state; when A drastic change occurs; this term directly affects the Sigmoid function, causing... The rapid reduction forces the model to forget its previous hierarchical states and adapt to the new mixed states.

[0052] Part Two: Input Gate.

[0053]

[0054] in, The current time step of the input gate Network layer The output, For the Sigmoid activation function, Here is the weight matrix of the input gate. For the previous time step Network layer The hidden state, For the current time step Network layer Input data, This is the bias term for the input gate.

[0055] Part Three: Candidate States.

[0056]

[0057] in, The weight matrix for the candidate values, For the previous time step Network layer The hidden state, For the current time step Network layer Input data, The bias term for candidate values, For the current time step Network layer Candidate states, It is the hyperbolic tangent function.

[0058] Part Four: Update Unit Status.

[0059]

[0060] in, For the current time step Network layer The cell state, For the current time step of the genetic phylum Network layer The output, The current time step of the input gate Network layer The output, For the current time step Network layer The candidate state.

[0061] Part 5, Output Gate.

[0062]

[0063] in, Output the current time step of the gate Network layer The output, For the Sigmoid activation function, Here is the weight matrix of the output gate. For the previous time step Network layer The hidden state, For the current time step Network layer Input, This is the bias term for the output gate.

[0064] The gating unit is based on the current input (Meteorological and hydrological characteristics) and the state at the previous moment A filtering coefficient between 0 and 1 is calculated, the physical meaning of which is information filtering: not all thermodynamic history information stored in the memory unit directly contributes to the current water temperature prediction.

[0065] The output gate uses adaptive learning to identify the most critical features for the current moment's prediction (such as a sudden cooling or a continuous radiative heating) and to filter out irrelevant background noise.

[0066] Part 6: Improved Update Hidden Status.

[0067]

[0068] in,

[0069] in, For the current time step Network layer The hidden state, Output the current time step of the gate Network layer The output, It is the hyperbolic tangent function. For the current time step Network layer The final state of the unit, The heat conduction gating coefficient is the output of the Sigmoid function, with a value range of (0,1). It acts as a weighting coefficient, directly affecting the hidden state of the previous spatial layer (layer d-1). This determines how much information from the upper layer can flow into the current layer; For the Sigmoid activation function, The space physics weight matrix. It serves as a bridge connecting the physical world and deep learning models, enabling neural networks to "sense" and "respond" to real vertical heat diffusion processes, thereby achieving adaptive water temperature prediction guided by physical mechanisms and significantly improving the robustness and interpretability of the model in complex scenarios. For the current time step Upstream vertical vortex expansion coefficient, is the thermal conduction gate bias vector, which represents the inherent thermal conduction characteristics or background diffusion capacity of the water medium in the vertical direction.

[0070] It should be noted that Equation 8 simulates the upstream heat flux. Specifically, if the upstream vertical vortex expansion coefficient The extent is significant (e.g., strong winds or temperature differences can cause strong expansion). The heat information of the upper layer tends towards 1. A large amount of data will "flow" into the current layer; conversely, it will be blocked, thus achieving physical interaction between layers. Through training, Able to capture those things that cannot be explicitly quantified by physical formulas ( The microscopic heat exchange processes (such as molecular diffusion) are fully covered, serving as a modification and supplement to the physical formulas. Obtained through supervised learning based on gradient descent. During the training phase, the model automatically seeks the optimal weight combination to represent the upstream vertical vortex spread coefficient. The nonlinear relationship between information flow between layers of a neural network. Through adjustment of this weight matrix, the physical phenomenon of strong diffusion (…) The amplification signal is transformed into an enhanced signal for information interaction between adjacent layers in the neural network, thereby reproducing the physical process of heat conduction in the deep learning architecture.

[0071] Preferably, the upstream stratification stability coefficient The calculation formula is as follows:

[0072] in, It is the hyperbolic tangent function. for Current time step The temperature For the previous time step Upstream surface water temperature, For the current time step wind speed, This is a scaling factor used to adjust the range of numerical values. It is a local constant used to ensure computational stability. This indicates high upstream stratification stability. This indicates low stability of the upstream stratification.

[0073] Preferably, the upstream vertical vortex expansion coefficient The calculation formula is as follows:

[0074] in, For activation functions, the Sigmoid function is typically used. For the current time step The upstream water is deep. This is a scaling factor used to adjust the range of numerical values. The larger the value, the more intense the heat exchange between the upper and lower layers.

[0075] III. Practical Applications First, the boundary conditions are determined and then substituted into the "longitudinal water temperature inversion model based on flow velocity time delay correction" to obtain the water temperature discharged from the upstream intake when the target water temperature at the downstream section is met. , This is the current time step.

[0076] Secondly, multi-source time-series data of the reservoir are acquired, including multi-dimensional meteorological and hydrological characteristics such as air temperature, solar radiation, wind speed, water level, and flow rate, as well as measured water temperature sequences at different depths, with the sequences corresponding to the current time step. The previous time steps are input into the trained ST-PC-LSTM to obtain the vertical water temperature of the reservoir at the current time step. The predicted distribution data (i.e., the time-series predicted values ​​of water temperature at each depth).

[0077] Then, the water temperature obtained from the upstream intake is used for inversion. The data is compared in real time with the predicted vertical water temperature distribution data to determine its consistency with the actual water temperature distribution. The optimal water depth is determined, and the flexible water-retaining curtain wall is adjusted accordingly.

[0078] Example I. Determine the boundary conditions.

[0079] Target water temperature at downstream section: Current average water depth in the downstream channel: Average temperature during the forecast period: Average solar radiation during the forecast period: Substituting this into the "longitudinal water temperature inversion model based on velocity time delay correction" described above, we can obtain: That is, in order to meet the downstream ecological water temperature requirement of 8.5℃, and considering the heating effect of air temperature along the way, the water temperature discharged from the upstream intake needs to be controlled at around 7.26℃.

[0080] II. Construction and Training of Predictive Models like Figure 2 As shown, meteorological and hydrological characteristic data of a reservoir from January to April 2024 were collected as model input. Specifically, this includes six dimensions of time-series characteristics: ① Temperature ( ① Temperature fluctuates within the range of -5℃ to 25℃, exhibiting a significant seasonal trend; ② Solar radiation ( ): It exhibits a regular diurnal peak variation, directly affecting the heat absorption of surface water; ③ and ④ relative humidity (attributed to the input vector) ) and wind speed ( ⑤ The reservoir's operating water level exhibits high-frequency random fluctuations, affecting heat exchange at the water-air interface; (Current computation depth) and input vector ): The overall trend is fluctuating upward (approximately 100m to 112m); ⑥ Inflow rate (Q): Affected by rainfall and upstream water diversion, the fluctuation range is large (40-100m³ / s).

[0081] like Figure 3 As shown, measured water temperature data at different depths during the same period were collected as the prediction targets for the model. The curves in the figure show the stratified response of water temperature in the vertical direction: surface water temperature (0m, 10m, red / orange curves) to... Figure 2 The air temperature and radiation response in the water are rapid and fluctuate violently; as the depth increases, the water temperature changes gradually become more gradual, and the deep water temperature (40m, 50m, purple / brown curves) shows obvious thermal inertia and hysteresis.

[0082] This embodiment constructs a spatiotemporally coupled long short-term memory network model and trains it using the aforementioned data. The input features include 6-dimensional meteorological and hydrological time-series data (temperature, solar radiation, relative humidity, wind speed, reservoir operating water level, and inflow) and two physical mechanism parameters calculated based on physical formulas (layered stability coefficient). and vertical vortex diffusion coefficient The labels are multi-depth measured water temperature data collected at the same time (covering real water temperature values ​​at key depths such as 0m, 10m, 20m, 25m, 40m and 50m), which are used to guide the model to master the stratified evolution of water temperature in the vertical direction through supervised learning.

[0083] like Figure 4 The diagram illustrates the training and convergence process of the model. The solid blue line represents the training loss, and the dashed red line represents the validation loss. In the first 10 training epochs, the model quickly learns the data features, and the loss value decreases significantly. Between epochs 10 and 20, the curve shows slight fluctuations. After approximately 25 epochs, both the training and validation losses stabilize and converge to a low level (MSE < 0.05), with a small difference between them, indicating that the model is sufficiently trained and has good generalization ability.

[0084] III. Validation and Analysis of Model Prediction Performance The trained model was applied to the test set (September to November 2024) and evaluated from two dimensions: fitting accuracy and error distribution.

[0085] (1) Multi-depth time series fitting analysis, such as Figure 5 As shown, the comparison results of predicted values ​​(red dashed lines) and measured values ​​(blue solid lines) at five typical depths of 0m, 10m, 25m, 40m, and 50m are presented. Specifically, in the surface region (0m, 10m): despite the greatest interference from the external environment and frequent water temperature fluctuations, the model's prediction curve still closely follows the trend of the measured values. The mean absolute error (MAE) at 0m is controlled at 0.37℃, and the root mean square error (RMSE) is 1.34℃. In the middle layer region (25m, 40m): the model's prediction accuracy is significantly improved, with the MAE at 25m decreasing to 0.14℃ and the RMSE decreasing to 0.53℃, accurately capturing subtle changes in water temperature. In the deep layer region (50m): the model exhibits the best prediction performance, with an MAE of only 0.09℃ and an RMSE as low as 0.33℃. Figure 6 As shown, the predicted curve almost perfectly matches the measured curve. This proves that the model can effectively adapt to the water temperature variation characteristics at different depths.

[0086] (2) Error statistical distribution analysis, such as Figure 7As shown, statistical tests were performed on the prediction errors: The error distribution histogram (left) shows that the prediction errors follow a standard normal distribution, and the mean line (green solid line) highly coincides with the zero error line (red dashed line), with an average error of only 0.17℃, indicating that the model has no systematic bias; The error box plot (right) clearly reveals the rule that "the deeper the depth, the more stable the prediction." As the water depth increases from 0m to 50m, the box length (error dispersion) gradually shortens, and the number of outliers decreases significantly. This result further verifies the robustness of this application in deep water temperature prediction and can provide highly reliable data support for the regulation of deep water temperature in reservoirs.

[0087] (3) Based on the physical coefficient curves generated by the simulation, as shown in the figure Figure 8 As shown, the following significant conclusions can be drawn: ① Layering stability coefficient ( The response characteristics of ) (see above figure): The values ​​exhibit a highly negative correlation with wind speed (dynamic disturbance), showing fluctuations. During periods of lower wind speed and rising temperatures, A value consistently above 0.8 indicates that the water body is in a state of strong stratification; however, during sudden strong wind events (such as when the wind speed exceeds 6 m / s in the simulation sequence). It rapidly drops to 0.1 or even negative values. This change accurately captures the physical process of "dynamic disturbance disrupting thermal stratification," providing a crucial "condition switching signal" for the neural network. ② Vertical eddy diffusion coefficient ( The attenuation characteristics of ) (middle figure): The values ​​exhibit a clear vertical gradient decay pattern. The average diffusion coefficient at the surface (0m) remains above 0.9, indicating that surface heat exchange is strongly driven by external meteorological conditions; however, as the depth increases to 50m, The value significantly decreased to around 0.6 and the fluctuation range narrowed. This verifies the influence of water depth damping effect during downward heat transfer, providing a quantitative basis for the model of "interlayer thermal conduction intensity".

[0088] IV. The Improvement Effect of Physical Indicators on the ST-PC-LSTM Model The dynamic forgetting mechanism takes effect: when it is detected During a sudden drop, the physical gating units in the ST-PC-LSTM are activated, forcing the forget gate to increase the forgetting rate. This allows the model to quickly discard outdated "hierarchical memories" and instead rely on real-time input data, thus solving the problem of prediction lag in traditional models during mixed periods.

[0089] Physical constraints of deep prediction: utilizing Due to the deep attenuation characteristics, the model automatically reduces its sensitivity to high-frequency meteorological noise in the deep network, instead receiving smoothed upper-layer heat information through a "heat conduction gate." This explains why, in the results of the example, the predicted curve for deep water temperature (MAE=0.09℃) can eliminate non-physical oscillations and achieve extremely high smoothness and accuracy.

[0090] Figure 5 The predicted results and the previously calculated temperature of water flowing through the flexible water-retaining curtain wall The comparison revealed that adjusting the water-retaining curtain wall to the relevant temperature range ensured that the discharged water temperature met the requirements, and the changes in the aforementioned physical coefficients directly improved the robustness of the prediction model.

[0091] It should be understood that the above-described device is used to execute the methods in the above embodiments. The implementation principle and technical effect of the corresponding program modules in the device are similar to those described in the above methods. The working process of the device can be referred to the corresponding process in the above methods, and will not be repeated here.

[0092] Based on the methods in the above embodiments, this application provides an electronic device that may include a processor, a communications interface, a memory, and a communication bus, wherein the processor, communications interface, and memory communicate with each other via the communication bus. The processor may invoke logical instructions stored in the memory to execute the methods in the above embodiments.

[0093] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0094] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0095] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0096] It is understood that the processor in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0097] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0098] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0099] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.

[0100] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for predicting upstream outflow water temperature, characterized in that, include: Obtain multi-source time-series meteorological and hydrological data of the reservoir before the current time step and measured water temperature sequences at different depths, input them into the upstream discharge water temperature prediction model, and obtain the vertical water temperature prediction distribution data of the reservoir at the current time step; The upstream discharge water temperature prediction model is obtained by training an improved LSTM based on multi-source time-series data of reservoir meteorological and hydrological data and multi-depth measured water temperature data. The improved LSTM is used to incorporate the changes in upstream water stratification stability over time into the genetic phylogenetic tree, and at the same time incorporate the changes in upstream vertical thermal diffusivity over space into the updated hidden state.

2. The prediction method as described in claim 1, characterized in that, The improved LSTM forget gate is as follows: in, Forgot Gate Current Time Step Network layer The output, Here is the weight matrix for the forget gate. For the previous time step Network layer The hidden state, For the current time step Network layer Input, For the bias term of the forget gate, This is the time-physical weight matrix.

3. The prediction method as described in claim 2, characterized in that, Upstream stratification stability coefficient The calculation formula is as follows: in, It is the hyperbolic tangent function. For the current time step The temperature For the previous time step Upstream surface water temperature, For the current time step wind speed, This is the scaling factor. It is a local constant.

4. The prediction method as described in claim 1, characterized in that, The updated hidden state of the improved LSTM is as follows: in, in, For the current time step Network layer The hidden state, Output the current time step of the gate Network layer The output, It is the hyperbolic tangent function. For the current time step Network layer The final state of the unit, The thermal conduction gating factor is... It is the Sigmoid activation function. This is the space physics weight matrix. For the current time step Upstream vertical vortex expansion coefficient, This is the thermal conduction gate bias vector.

5. The prediction method as described in claim 4, characterized in that, Upstream vertical vortex expansion coefficient The calculation formula is as follows: in, It is the Sigmoid activation function. For the current time step The upstream water is deep. For the current time step wind speed, This is the scaling factor. It is a local constant.

6. A method for controlling a flexible water-retaining curtain wall, characterized in that, include: Calculate the water temperature discharged from the upstream intake at the current time step; Using the prediction method described in any one of claims 1 to 5, the predicted distribution data of the vertical water temperature of the reservoir at the current time step are obtained; The predicted vertical water temperature distribution data and the calculated downstream water temperature from the upstream intake are compared in real time to determine the water layer depth that best matches them, and the flexible water barrier wall is adjusted accordingly.

7. The control method as described in claim 6, characterized in that, The calculation of the downstream water temperature at the current time step is as follows: Determine the expected value of downstream cross-section water temperature, the average water depth of downstream channel, the average air temperature during the forecast period, and the average solar radiation during the forecast period corresponding to the current time step; The above parameters are substituted into the longitudinal water temperature inversion model based on flow velocity time delay correction to obtain the water temperature discharged from the upstream intake at the current time step. The longitudinal water temperature inversion model based on flow velocity time delay correction is as follows: in, For the current time step Upstream water flow The actual water temperature at the downstream section after time transmission; For the current time step Water temperature discharged from the upstream intake; This is the time it takes for water to flow from upstream to downstream. Each represents the current time step. Average temperature and average solar radiation; The thermal retention coefficient is related to distance and flow velocity. An empirical coefficient for converting solar radiation energy into water temperature changes. This represents the average water depth in the downstream river channel.

8. An upstream discharge water temperature prediction system, characterized in that, Includes memory and one or more processors; The memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions; The one or more processors invoke the computer instructions to cause the system to perform the prediction method as described in any one of claims 1 to 5.

9. A flexible water-retaining curtain wall control system, characterized in that, Includes memory and one or more processors; The memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions; The one or more processors invoke the computer instructions to cause the system to perform the control method as described in claim 6 or 7.

10. A computer-readable storage medium, characterized in that, Includes instructions that, when executed on an electronic device, cause the electronic device to perform the method as described in any one of claims 1 to 7.

Images