Phase change heat storage tank control method and system based on step-by-step prediction

Abstract

The invention belongs to the technical field of energy storage, and provides a phase change heat storage tank control method and system based on step-by-step prediction.The phase change heat storage tank control method comprises the steps that firstly, a guide plate is preliminarily adjusted according to the inlet temperature, the center temperature, the outlet temperature and the edge temperature; then, according to a preset MRE-UNet model, prediction and further adjustment of a baffle angle are carried out; wherein the MRE-UNet model comprises a convolutional neural network and a multi-scale feature convolution structure; in the encoding process of the MRE-UNet model, through convolution and pooling operation, extracting input characteristic temperature space-time distribution data layer by layer; in the decoding process, through up-sampling and convolution operation, the output feature map is reconstructed into a high-resolution temperature field prediction map; and embedding a cavity convolution layer in the encoding process, and expanding a receptive field to capture global structures and local details of different scales in the temperature field. The spatial structure information of the temperature field can be captured by means of an MRE-UNet model, and guide control over adjustment of the guide plate is achieved.

Description

Technical Field

[0001] This invention belongs to the field of energy storage technology, and in particular relates to a control method and system for phase change thermal storage tanks based on step-by-step prediction. Background Technology

[0002] With the continuous increase in the penetration rate of renewable energy, phase change thermal energy storage technology has shown great application potential in suppressing new energy fluctuations and realizing peak shaving and valley filling of the power grid due to its advantages such as high energy storage density and near isothermal heat release process.

[0003] In existing phase change thermal energy storage tank control technologies, computational fluid dynamics (CFD) models are used to simulate the internal state, but their computational cost is extremely high and cannot meet the requirements of real-time control. Simple machine learning models (such as fully connected neural networks) are applied for overall energy efficiency prediction, but such models cannot capture the spatial structure information of the temperature field, have limited prediction accuracy, and are difficult to guide specific spatial optimization control (such as baffle adjustment). Summary of the Invention

[0004] To address the aforementioned problems, this invention proposes a phase change thermal storage tank control method and system based on stepwise prediction. This invention utilizes the MRE-UNet model to capture the spatial structure information of the temperature field, thereby achieving guided control of the baffle adjustment.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution: In a first aspect, the present invention provides a phase change thermal storage tank control method based on stepwise prediction, comprising: Acquire temperature and operating data at a preset location of the phase change thermal storage tank; The temperature is divided into inlet temperature, center temperature, outlet temperature, and edge temperature; the guide vane is initially adjusted based on the inlet temperature, center temperature, outlet temperature, and edge temperature. Based on the inlet temperature, center temperature, outlet temperature, and edge temperature, as well as the preset MRE-UNet model, predictions are made and the baffle angle is further adjusted. The MRE-UNet model includes a convolutional neural network and a multi-scale feature convolutional structure. During the encoding process, the MRE-UNet model extracts the spatiotemporal distribution data of the input feature temperature layer by layer through convolution and pooling operations. During the decoding process, the output feature map is reconstructed into a high-resolution temperature field prediction map through upsampling and convolution operations. Furthermore, during the encoding process, dilated convolutional layers are embedded to expand the receptive field in order to capture the global structure and local details at different scales in the temperature field.

[0006] Furthermore, the temperature at the preset location includes the inlet and outlet working fluid temperatures and the temperature inside the tower; the operating data is the inlet and outlet working fluid flow rates.

[0007] Furthermore, during the initial adjustment of the guide vanes, in the heat storage condition: if the center temperature equals the inlet temperature and the center temperature equals the edge temperature, then maintain the current guide vane angle; if the outlet temperature rises and some center temperatures are lower than the outlet temperature, then decrease the guide vane angle to guide the fluid to flow towards the sidewall; if the edge temperature is higher than the center temperature, increase the guide vane angle to guide the fluid to flow towards the center region. In the heat release condition: if the center temperature equals the outlet temperature and the center temperature equals the edge temperature, then maintain the current guide vane angle; if the outlet temperature drops and some center temperatures are higher than the outlet temperature, then increase the guide vane angle to guide the fluid to flow towards the center region; if some edge temperatures change slowly, then decrease the guide vane angle to guide the fluid to flow towards the sidewall.

[0008] Furthermore, during the training of the MRE-UNet model, CFD dynamic simulation is performed, and the CFD temperature field data is normalized. The CFD three-dimensional data is horizontally sliced ​​along the height direction, and the height information and the deflector angle information are used as additional channel inputs. For each height slice, a two-dimensional matrix with the same resolution as the CFD mesh is constructed. The temperature map, height information, and deflector angle information of each height slice are used as inputs, and the complete temperature map is used as the output to train the MRE-UNet model.

[0009] Furthermore, the prediction and baffle angle adjustment include: obtaining the spatiotemporal distribution prediction map of the temperature field inside the thermal storage tank at the current moment according to the MRE-Unet model; evaluating the flow field uniformity based on the spatiotemporal distribution prediction map; and determining the objective function for baffle angle optimization based on the flow field uniformity evaluation results.

[0010] Furthermore, the evaluation indicators include the radial non-uniformity coefficient and the axial temperature non-uniformity: Axial non-uniformity coefficient : ; Radial temperature distribution skewness : ; Based on the above indicators, construct an optimization objective function. J (θ): J (θ)= oh 1· RUI + oh 2· Show ; in, oh 1. oh 2 represents the weight; i The angle of the deflector; The first thermocouple is arranged along the height.N layer; for z i Thermocouple temperature is located at the center of the height. The height at which the thermocouples are arranged. For the first i layer; This represents the average temperature of the thermocouples positioned at the edge; Temperature of the imported working fluid; The outlet working fluid temperature; This refers to the number of edge thermocouples arranged at the same floor height. for z i The temperature of the thermocouples arranged at the edge in terms of height; for z i The average temperature of the thermocouples arranged at the edge of the height.

[0011] Furthermore, the current three-dimensional temperature distribution is obtained based on the MRE-UNet model; the uniformity of the predicted temperature field is evaluated, and the direction of the deflector adjustment is determined based on the previous results; the deflector angle is changed, input into the MRE-UNet model, a new three-dimensional temperature distribution is obtained, and the improvement of the new three-dimensional temperature field is evaluated; the optimal deflector angle is obtained by iterative optimization based on the optimization algorithm.

[0012] Secondly, the present invention also provides a phase change thermal storage tank control system based on stepwise prediction, comprising: The data acquisition module is configured to acquire temperature and operating data at a preset location of the phase change thermal storage tank. The deflector initial adjustment module is configured to: divide the temperature into inlet temperature, center temperature, outlet temperature, and edge temperature; and make initial adjustments to the deflector based on the inlet temperature, center temperature, outlet temperature, and edge temperature. The deflector adjustment module is configured to further adjust the baffle angle based on inlet temperature, center temperature, outlet temperature, and edge temperature, as well as a preset MRE-UNet model. The MRE-UNet model includes a convolutional neural network and a multi-scale feature convolutional structure. During the encoding process, the MRE-UNet model extracts the spatiotemporal distribution data of the input feature temperature layer by layer through convolution and pooling operations. During the decoding process, the output feature map is reconstructed into a high-resolution temperature field prediction map through upsampling and convolution operations. Furthermore, during the encoding process, dilated convolutional layers are embedded to expand the receptive field in order to capture the global structure and local details at different scales in the temperature field.

[0013] Thirdly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the step-by-step prediction-based phase change thermal storage tank control method described in the first aspect.

[0014] Fourthly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the program to implement the steps of the step-by-step prediction-based phase change thermal storage tank control method described in the first aspect.

[0015] Fifthly, the present invention also provides a computer program product, the computer program product comprising a computer program, which, when executed by a processor, implements the steps of the phase change thermal storage tank control method based on step-by-step prediction described in the first aspect.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention first adjusts the baffle plate based on the inlet, center, outlet, and edge temperatures. Then, based on the inlet, center, outlet, and edge temperatures, and a pre-defined MRE-UNet model, it further adjusts the baffle angle for prediction. The MRE-UNet model includes a convolutional neural network and a multi-scale feature convolutional structure. During encoding, the MRE-UNet model extracts the spatiotemporal distribution data of the input feature temperature layer by layer through convolution and pooling operations. During decoding, it reconstructs the output feature map into a high-resolution temperature field prediction map through upsampling and convolution operations. Additionally, it embeds dilated convolutional layers during encoding to expand the receptive field and capture the global structure and local details at different scales within the temperature field. The MRE-UNet model enables the capture of spatial structural information of the temperature field, thus guiding and controlling the adjustment of the baffle plate.

[0017] 2. This invention utilizes the MRE-UNet model, based on high-resolution CFD dynamic simulation and standard experimental data, to predict the operating status of thermal storage tanks with high accuracy and speed, laying the foundation for prediction and optimization, and enhancing the perception of the operating status of thermal storage tanks.

[0018] 3. This invention combines data-driven prediction with physical models. Control decisions are based on predictions of the future state of the system and can coordinate the adjustment of multiple variables such as flow rate, temperature and guide vane angle, transforming from passive response to active optimization, significantly improving control quality and system energy efficiency.

[0019] 4. Compared with other deep learning-based energy storage state prediction methods, this invention establishes the spatiotemporal correlation and quantitative evaluation index of step-by-step prediction results, which can further improve the accuracy and generalization ability of prediction. Attached Figure Description

[0020] The accompanying drawings, which form part of this embodiment, are used to provide a further understanding of this embodiment. The illustrative embodiments and their descriptions are used to explain this embodiment and do not constitute an improper limitation of this embodiment.

[0021] Figure 1 This is a schematic diagram of the thermocouple arrangement in the unit height thermal storage tank of Embodiment 1 of the present invention; Figure 2 This is a flowchart of the method in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the MRE-Unet neural network model structure in Embodiment 1 of the present invention. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0024] Example 1: Phase change thermal storage tanks face two core challenges in practical operation: ① The thermal storage state is difficult to characterize. The "charging" and "releasing" of thermal storage tanks are complex multi-physics coupling processes involving fluid flow, heat transfer, and phase change. Currently, relying on a limited number of sparsely arranged temperature sensors (such as thermocouples) cannot accurately and comprehensively reflect the true spatiotemporal distribution of the three-dimensional temperature field inside the tank, leading to inaccurate operational status assessments and becoming one of the bottlenecks in control optimization. ② Lagging and inefficient control strategies: Traditional control strategies are mostly based on simple thresholds (such as inlet and outlet temperatures) or preset empirical rules, which cannot cope with dynamically changing operating conditions (such as varying heat source temperatures and flow demands). This "sensor-response" control is lagging, making it difficult to optimize system energy efficiency in real time, often leading to uneven melting / solidification fronts, thermal short circuits, and other phenomena, resulting in wasted usable thermal energy and reduced system efficiency.

[0025] In existing technologies, some studies have used computational fluid dynamics (CFD) models to simulate internal states, but their computational cost is extremely high and cannot meet the requirements of real-time control. Other studies have attempted to apply simple machine learning models (such as fully connected neural networks) for overall energy efficiency prediction, but such models cannot capture the spatial structure information of the temperature field, have limited prediction accuracy, and are difficult to guide specific spatial optimization control (such as deflector adjustment).

[0026] Based on this, this embodiment provides a phase change thermal storage tank control method based on step-by-step prediction. This method exhibits good robustness and can accurately predict the energy storage state of the thermal storage tank, fully tapping the energy efficiency potential of the phase change thermal storage system. The control method combines CFD dynamic inversion with machine learning for step-by-step prediction and composite control of the phase change thermal storage tank's energy efficiency. First, a shallow neural network is used to quickly determine the energy storage state and the initial flow regime within the thermal storage tank. Then, an MRE-UNet model is used to achieve high-resolution spatiotemporal prediction of the internal temperature field of the thermal storage tank. Based on the prediction results, the angle of the adjustable guide vanes within the thermal storage tank is continuously optimized, thereby maximizing the energy efficiency of the thermal storage / release process. The method specifically includes: S1. Sparse temperature sensors are installed at key locations in the phase change thermal storage tank to collect system operation data, including inlet and outlet working fluid temperatures. T in , T out ), inlet and outlet working fluid flow rate ( G in , G out ) and the temperature of the tower temperature sensor ( T sensor The layout is as follows Figure 1 (As shown). The thermocouple located at the center is denoted as... T center The thermocouples placed at 0.75R and 0.5R are denoted as follows: T edge - 0.75R , T edge - 0.5R The angle of the deflector is ( ). i 1… i N ).

[0027] To achieve rapid forecasting of thermal storage conditions and preliminary control of the flow diversion device, the temperature dataset is divided into two categories. One category includes the inlet temperature, center temperature, and inlet temperature, denoted as […]. T in , T center1… T centerN、 T out Because the central temperature change is located in the flow core region, it can quickly reflect the position and movement of the phase change front, and is used to quickly determine the current energy storage state (heat storage / heat release level). Another type is […]. T 0.75R1… T 0.75RN ]、[ T0.5R1… T 0.5RN [Edge temperature sets, by comparing with the center temperature dataset] T center1… T centerN The comparison is used to initially determine the uniformity of the flow pattern inside the thermal storage tank and to make preliminary adjustments to the guide vanes. The judgment rules are shown in Table 1. Table 1 Rules for Judging the Uniformity of Flow Pattern in Thermal Storage Tanks

[0028] Design a series of typical operating conditions (different) covering the operation of thermal storage tanks. T in , G in and the angle of the deflector i For pre-designed operating conditions (such as...) T in =90°C G in =2kg / s and θ= (30°), conduct high-fidelity CFD dynamic simulation. After the simulation, output the three-dimensional temperature field spatiotemporal evolution data inside the thermal storage tank under this condition, as the ground truth label in supervised learning.

[0029] Extract the time-varying sequences of parameters such as the center temperature, outlet temperature, and State of Charge (SOC) of the thermal storage tank from the CFD results under various operating conditions. Perform dimensionless processing on the temperature as follows:

[0030] Since the data is a one-dimensional time series, a shallow one-dimensional convolutional neural network can quickly obtain good results. To maintain consistent sequence length, a sliding window is used to truncate a fixed-length temperature time series as the input feature. X k .

[0031]

[0032] in, Fo k For Fourier dimensionless time; ▽ T k The temperature gradient is centered. Fo k Calculate using the following formula:

[0033] in, α Where is the thermal diffusivity, L This refers to the height of the thermal storage tank.

[0034] The energy storage state (the ratio of currently stored energy to maximum storable energy) corresponding to the time series segment and the phase transition front are used as output labels. Y k Using the above ( X k , Y k This involves training and validating the shallow neural network. In practical applications, the energy storage state can be quickly determined using only the collected center temperature, providing support for subsequent optimized control decisions. Furthermore, by comparing the center temperatures... T center With edge temperature T edge A preliminary assessment indicates that the deflector needs to be adjusted in the correct direction.

[0035] S2. Further, full-field prediction and refined baffle angle adjustment are achieved through a center-edge temperature set and the MRE-UNet model. An MRE-UNet module is constructed using convolutional neural networks and multi-scale feature convolution, integrating multi-scale enhanced receptive field and residual attention mechanisms for model training. During encoding, convolution and pooling operations are used to extract the spatiotemporal distribution data of the input feature temperature layer by layer. During decoding, upsampling and convolution operations are used to reconstruct the output high-level feature map into a high-resolution temperature field prediction map. To further improve prediction accuracy and generalization ability, dilated convolutional layers are embedded during encoding to expand the receptive field, effectively capturing global structures (such as hot-cold stratification) and local details (such as phase transition fronts) at different scales in the temperature field. An attention gating unit is connected between encoding and decoding, enabling the model to focus more on key regions related to the phase transition heat transfer process and suppress irrelevant background interference.

[0036] Optionally, the CFD temperature field data is normalized based on the MRE-Unet model. The CFD 3D data is horizontally sliced ​​along the height direction, with height information and deflector angle information used as additional input channels. For each height slice, a 2D matrix with the same resolution as the CFD mesh is constructed, but temperature values ​​are assigned only at sensor locations, with other locations set to 0. The sparse temperature map, height information, and deflector angle information of each height slice are used as input, and the complete temperature map is used as output to train the MRE-UNet model.

[0037] During the implementation phase, the system operating parameters collected in real time are used to form tensors required by the model input, which are then input into the trained MRE-Unet model to output a high-resolution spatiotemporal distribution prediction map of the temperature field inside the thermal storage tank at the current moment through real-time inference.

[0038] Based on the predicted temperature field, flow field uniformity evaluation indices are established. These indices include the radial non-uniformity coefficient and the axial temperature non-uniformity. Axial non-uniformity coefficient RUI:

[0039] Radial temperature distribution skewness (Shew):

[0040] Based on the above indicators, construct an optimization objective function. J (θ): J (θ)= oh 1· RUI + oh 2· Show in, oh 1. oh 2 represents the weight; i The angle of the deflector; The first thermocouple is arranged along the height. N layer; for z i Thermocouple temperature is located at the center of the height. The height at which the thermocouples are arranged. For the first i layer; This represents the average temperature of the thermocouples positioned at the edge; Temperature of the imported working fluid; The outlet working fluid temperature; This refers to the number of edge thermocouples arranged at the same floor height. for z i The temperature of the thermocouples arranged at the edge in terms of height; for z i The average temperature of the thermocouples arranged at the edge of the height.

[0041] Subsequently, other input variables are fixed to generate a set of candidate guide vane angles, which are then input into the trained MRE-Unet model to obtain the corresponding predicted temperature field. For each temperature field, performance metrics are calculated, and the guide vane angle with the optimal metrics is selected and sent to the system's actuators to adjust the system to its optimal operating state. This process is then repeated in a new round of data acquisition, prediction, and optimization to form a closed-loop real-time intelligent optimization control system.

[0042] Optionally, temperature information from measurement points is input into the MRE-UNet model to obtain the current 3D temperature distribution. The uniformity of the predicted temperature field is evaluated, and the direction of the deflector adjustment is determined based on the previous results. With other information fixed, only the deflector angle is changed, and the result is input into the MRE-UNet model to obtain a new 3D temperature distribution. The improvement of the new 3D temperature field is then evaluated. Based on iterative optimization algorithms, the optimal deflector angle is obtained and issued to the execution structure to adjust the deflector angle. The deflector angle is adjusted via an electric mechanism.

[0043] In some embodiments, the method of this embodiment is specifically illustrated through an exemplary application scenario: Application scenario: A vertical cylindrical phase change thermal storage tank for district heating, filled with inorganic hydrated salt phase change material, and equipped with three independently rotatable baffles (angle range 0-180°).

[0044] In the training data preparation phase: ① Design various typical working conditions and use Ansys Fluent to conduct high-precision CFD numerical simulations to generate 500 sets of different parameters ( T in , T out , G in , G out ① Combined 3D temperature field spatiotemporal data. ② Extract virtual temperature measurements at the locations of thermocouples in the thermal storage tank from the CFD simulation results to simulate sparse sensor data. ③ Use two-dimensional symmetrical slices (256×256 pixels) representing the main heat transfer processes in the reduced-order CFD 3D temperature field as ground truth labels. ④ Construct a dataset by fusing and assembling the CFD results at the locations corresponding to the experimental measurements to form training temperature field images, and construct test and training sets.

[0045] An MRE-UNet network was constructed with a 4-level encoder. Dilated convolutional modules with dilation rates of 2 and 4 were embedded in levels 2 and 3 to obtain multi-scale features. All skip connections were integrated with channel attention modules. The model was trained on a dataset using the Adam optimizer. After multiple iterations, the model's prediction error on the test set was below 2%.

[0046] During real-time system operation, a control loop is executed every 15 minutes. At the beginning of each loop, real-time data is collected and input into the trained MRE-UNet module to obtain the predicted temperature field. The PSO algorithm is then invoked to optimize the solution with optimal energy efficiency as the objective, obtaining the optimal operating parameters, which are then sent to the actuators.

[0047] Example 2: This embodiment provides a phase change thermal storage tank control system based on stepwise prediction, including: The data acquisition module is configured to acquire temperature and operating data at a preset location of the phase change thermal storage tank. The deflector initial adjustment module is configured to: divide the temperature into inlet temperature, center temperature, outlet temperature, and edge temperature; and make initial adjustments to the deflector based on the inlet temperature, center temperature, outlet temperature, and edge temperature. The deflector adjustment module is configured to further adjust the baffle angle based on inlet temperature, center temperature, outlet temperature, and edge temperature, as well as a preset MRE-UNet model. The MRE-UNet model includes a convolutional neural network and a multi-scale feature convolutional structure. During the encoding process, the MRE-UNet model extracts the spatiotemporal distribution data of the input feature temperature layer by layer through convolution and pooling operations. During the decoding process, the output feature map is reconstructed into a high-resolution temperature field prediction map through upsampling and convolution operations. Furthermore, during the encoding process, dilated convolutional layers are embedded to expand the receptive field in order to capture the global structure and local details at different scales in the temperature field.

[0048] The working method of the system is the same as the step-by-step prediction-based phase change thermal storage tank control method in Embodiment 1, and will not be repeated here.

[0049] Example 3: This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the phase change thermal storage tank control method based on step-by-step prediction described in Embodiment 1.

[0050] Example 4: This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the program, it implements the steps of the phase change thermal storage tank control method based on step-by-step prediction described in Embodiment 1.

[0051] Example 5: This embodiment provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the steps of the phase change thermal storage tank control method based on step-by-step prediction described in Embodiment 1.

[0052] The above description is merely a preferred embodiment of this practice and is not intended to limit the scope of this practice. Various modifications and variations can be made to this practice by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this practice should be included within the protection scope of this practice.

Claims

1. A phase change thermal storage tank control method based on stepwise prediction, characterized in that, include: Acquire temperature and operating data at a preset location of the phase change thermal storage tank; The temperature is divided into inlet temperature, center temperature, outlet temperature, and edge temperature; The guide vane is initially adjusted based on the inlet temperature, center temperature, outlet temperature, and edge temperature. Based on the inlet temperature, center temperature, outlet temperature, and edge temperature, as well as the preset MRE-UNet model, predictions are made and the baffle angle is further adjusted. The MRE-UNet model includes a convolutional neural network and a multi-scale feature convolutional structure. During the encoding process, the MRE-UNet model extracts the spatiotemporal distribution data of the input feature temperature layer by layer through convolution and pooling operations. During the decoding process, the output feature map is reconstructed into a high-resolution temperature field prediction map through upsampling and convolution operations. Furthermore, during the encoding process, dilated convolutional layers are embedded to expand the receptive field in order to capture the global structure and local details at different scales in the temperature field.

2. The phase change thermal storage tank control method based on stepwise prediction as described in claim 1, characterized in that, The temperature at the preset location includes the inlet and outlet working fluid temperature and the temperature inside the tower; the operating data is the inlet and outlet working fluid flow rate.

3. The phase change thermal storage tank control method based on stepwise prediction as described in claim 1, characterized in that, When making initial adjustments to the guide vanes, in the heat storage condition: if the center temperature equals the inlet temperature and the center temperature equals the edge temperature, then maintain the current guide vane angle; if the outlet temperature rises and some center temperatures are lower than the outlet temperature, then decrease the guide vane angle to guide the fluid towards the sidewalls; if the edge temperature is higher than the center temperature, increase the guide vane angle to guide the fluid towards the central region; in the heat release condition: if the center temperature equals the outlet temperature and the center temperature equals the edge temperature, then maintain the current guide vane angle. If the outlet temperature decreases and the temperature in some parts of the center is higher than the outlet temperature, the angle of the guide vane is increased to guide the fluid to flow towards the center area; if the temperature change is slow in some parts of the edge, the angle of the guide vane is decreased to guide the fluid to flow towards the side wall.

4. The phase change thermal storage tank control method based on stepwise prediction as described in claim 1, characterized in that, During the training of the MRE-UNet model, CFD dynamic simulation is performed, and the CFD temperature field data is normalized. The CFD three-dimensional data is horizontally sliced ​​along the height direction, and the height information and the deflector angle information are used as additional channel inputs. For each height slice, a two-dimensional matrix with the same resolution as the CFD mesh is constructed. The temperature map, height information, and deflector angle information of each height slice are used as inputs, and the complete temperature map is used as the output to train the MRE-UNet model.

5. The phase change thermal storage tank control method based on stepwise prediction as described in claim 1, characterized in that, The prediction and further adjustment of the baffle angle include: obtaining the spatiotemporal distribution prediction map of the temperature field inside the thermal storage tank at the current moment according to the MRE-Unet model; evaluating the flow field uniformity based on the spatiotemporal distribution prediction map; and determining the objective function for optimizing the baffle angle based on the flow field uniformity evaluation results.

6. The phase change thermal storage tank control method based on stepwise prediction as described in claim 5, characterized in that, Evaluation metrics include radial non-uniformity coefficient and axial temperature non-uniformity: Axial non-uniformity coefficient : ； Radial temperature distribution skewness : ； Based on the above indicators, construct an optimization objective function. J (θ): J (θ)= ω 1; RUI + ω 2; Shew ; in, ω 1. ω 2 represents the weight; θ The angle of the deflector; The first thermocouple is arranged along the height. N layer; for z i Thermocouple temperature is located at the center of the height. The height at which the thermocouples are arranged. For the first i layer; This represents the average temperature of the thermocouples positioned at the edge; Temperature of the imported working fluid; The outlet working fluid temperature; This refers to the number of edge thermocouples arranged at the same floor height. for z i The temperature of the thermocouples arranged at the edge in terms of height; for z i The average temperature of the thermocouples arranged at the edge of the height.

7. The phase change thermal storage tank control method based on stepwise prediction as described in claim 6, characterized in that, The current three-dimensional temperature distribution is obtained based on the MRE-UNet model; the uniformity of the predicted temperature field is evaluated, and the direction of the deflector adjustment is determined based on the previous results; the deflector angle is changed, and the MRE-UNet model is input to obtain a new three-dimensional temperature distribution, and the improvement of the new three-dimensional temperature field is evaluated; the optimal deflector angle is obtained by iterative optimization based on the optimization algorithm.

8. A phase change thermal storage tank control system based on stepwise prediction, characterized in that, include: The data acquisition module is configured to acquire temperature and operating data at a preset location of the phase change thermal storage tank. The deflector initial adjustment module is configured to: divide the temperature into inlet temperature, center temperature, outlet temperature, and edge temperature; and make initial adjustments to the deflector based on the inlet temperature, center temperature, outlet temperature, and edge temperature. The deflector adjustment module is configured to further adjust the baffle angle based on inlet temperature, center temperature, outlet temperature, and edge temperature, as well as a preset MRE-UNet model. The MRE-UNet model includes a convolutional neural network and a multi-scale feature convolutional structure. During the encoding process, the MRE-UNet model extracts the spatiotemporal distribution data of the input feature temperature layer by layer through convolution and pooling operations. During the decoding process, the output feature map is reconstructed into a high-resolution temperature field prediction map through upsampling and convolution operations. Furthermore, during the encoding process, dilated convolutional layers are embedded to expand the receptive field in order to capture the global structure and local details at different scales in the temperature field.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor executes the program, it implements the steps of the phase change thermal storage tank control method based on step prediction as described in any one of claims 1-7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the phase change thermal storage tank control method based on step-by-step prediction as described in any one of claims 1-7.

Images