Solid-liquid mixed electric heating heat storage agent optimization method and system

Abstract

The application provides a solid-liquid mixed electric heating and heat storage agent optimization method and system, and relates to the fields of energy thermal engineering and artificial intelligence. The method pre-trains a deep neural network by using simulation data of a source field, and freezes hidden layer parameters to extract general width features. Elastic network regularization and an improved alternating direction multiplier method are used to calculate output weights, so that high-precision modeling and prediction of operation characteristics of a solid-liquid mixed electric heating and heat storage device are realized under small sample conditions. In order to reduce abandoned wind and light and optimize system performance, a data-driven improved transfer learning agent model method is used. The system comprises a solid-liquid mixed electric heating and heat storage device, which is used for quickly responding to wind and light fluctuations and long-time peak regulation, involves multi-physical field coupling, and the operation characteristics of the device are influenced by multiple factors such as fluid flow rate and electric power distribution. The application can realize real-time optimization of system operation parameters and development of control strategies under complex working conditions, and has high engineering application value.

Description

Technical Field

[0001] This invention relates to the fields of energy thermal engineering and artificial intelligence, specifically to a method and system for optimizing solid-liquid hybrid electric heating and thermal storage. Background Technology

[0002] With the rapid development of renewable energy, especially the increasing penetration of wind and solar power in the power system, how to effectively absorb and utilize these highly volatile and intermittent power sources has become a pressing technical problem to be solved for the safe and stable operation of the power grid and the integration of a high proportion of renewable energy. Wind and solar power output fluctuates rapidly on millisecond to minute scales, and exhibits significant power and energy mismatches on hourly to daytime and even seasonal scales. Relying solely on conventional peak-shaving units and simple electro-thermal conversion processes is insufficient to achieve smooth regulation.

[0003] Therefore, energy storage and thermal energy storage technologies are becoming increasingly important in integrated energy systems. Electric heating / thermal energy storage devices can convert surplus wind and solar power into thermal energy for storage, and release it on demand during peak grid load periods or when renewable energy output is insufficient, thus achieving time-shifting and flexible scheduling of electrical and thermal energy. However, most existing electric heating / thermal energy storage systems are designed for a single time scale, focusing on using molten salt for long-term peak shaving from hours to days, or using electric boilers for minute-level power regulation. They lack synergistic consideration of rapid fluctuations from milliseconds to minutes and energy transport from hours to days / seasons, making it difficult to balance rapid response capabilities with large-capacity, long-term energy storage.

[0004] Traditional thermal energy storage technologies, such as molten salt thermal storage systems, rely on the high melting point and large specific heat capacity of molten salt to achieve large-scale heat storage. The heat is then released to a working fluid (such as water or thermal oil) via a heat exchanger to generate steam or a high-temperature heat transfer medium, making them suitable for energy regulation on hourly to diurnal scales. Solid thermal storage media such as cast iron have high density, good thermal conductivity, and low cost, allowing for long-term stable operation under high-temperature conditions, making them suitable for constructing simple, long-life solid thermal storage units. However, these single-medium or single-function thermal storage systems are insufficient in responding to drastic fluctuations in wind and solar power output on a millisecond to minute scale and providing rapid short-term buffering capabilities. Furthermore, simply configuring fast electro-thermal conversion devices such as electric boilers makes it difficult to achieve large-capacity energy transfer and peak shaving on hourly to seasonal scales.

[0005] Therefore, how to construct a hybrid electric heating / thermal storage system that organically integrates a fast-response wind and solar power fluctuation device and a long-term peak-shaving device in the same device or system, where the former is used to absorb wind and solar power output fluctuations at the millisecond to minute level, and the latter is used to realize energy transport at the hour to day / season level, so as to have both fast power response capability and large-capacity long-term energy storage and peak-shaving capability, has become an urgent problem to be solved in this field.

[0006] Furthermore, when a hybrid electric heating / heat storage system employs a cast iron-liquid metal solid-liquid hybrid electric heating / heat storage device to fully absorb wind and solar energy fluctuations, the device involves multi-physics coupling during operation, including complex mechanisms such as electric field, heat conduction, and liquid metal flow. Its operating characteristics are influenced by multiple factors, including liquid metal temperature and flow rate, electric power distribution, and device structural parameters. Currently, three-dimensional multiphysics numerical simulation is commonly used to model and optimize the parameters of such devices. However, these models are computationally intensive and have slow convergence speeds, making it difficult to meet the real-time operation and control requirements of engineering scenarios. On-site experimental measurements are constrained by wiring conditions, high-temperature and high-pressure environments, and safety concerns, making it impossible to collect data on a large scale across the entire operating range. Therefore, solid-liquid hybrid electric heating / heat storage devices often only yield a small number of simulation or experimental samples under discrete operating conditions, and these samples exhibit significant sparse and non-uniform distribution in multi-dimensional spaces such as temperature, flow rate, and electric power.

[0007] Existing methods employ data-driven surrogate modeling, which utilize limited simulation or experimental data to establish machine learning models capable of rapidly predicting the input-output relationships of electrothermal / thermal storage devices involving solid-liquid mixtures of cast iron and liquid metal. These models approximate three-dimensional numerical models, enabling real-time optimization of operating parameters and formulation of control strategies. However, electrothermal / thermal storage devices involving solid-liquid mixtures of cast iron and liquid metal exhibit high nonlinearity, sparse data sample distribution, and unmeasurable state variables. Traditional machine learning models, such as support vector machines and neural networks, suffer from limitations in generalization, accuracy, and stability.

[0008] Therefore, transfer learning, as a technique that leverages existing domain knowledge to improve the learning effect of new tasks, has attracted widespread attention, especially in industrial applications where samples are insufficient or experimental costs are high. However, for electric heating / heat storage devices with solid-liquid mixtures of cast iron and liquid metal, existing data-driven surrogate models and traditional transfer learning methods have significant shortcomings in expressing complex multi-physical mechanisms, utilizing sparse target domain samples, suppressing overfitting and negative transfer, improving robustness, and meeting real-time online updates. There is an urgent need to propose an improved transfer learning surrogate modeling method for the operating characteristics of electric heating / heat storage devices with solid-liquid mixtures of cast iron and liquid metal. This method should fully inherit the multi-physical mechanism knowledge of the source domain, efficiently complete the transfer and adaptation to the target operating conditions with only a small number of parameter updates, and possess good sparsity, generalization, and engineering deployability, providing reliable support for real-time operation and control under conditions of full absorption of wind and solar power fluctuations. Summary of the Invention

[0009] To address the aforementioned technical challenges, this invention proposes a solid-liquid hybrid electric heating / storage proxy optimization method and system, combining wind and solar power generation, molten salt energy storage, liquid metal heat exchange, cast iron thermal storage, and electric heating technologies. The method employs a data-driven improved transfer learning proxy optimization approach, significantly reducing computational costs. It achieves high-precision modeling and prediction of the operating characteristics of the solid-liquid hybrid electric heating / storage device under small sample conditions, enabling real-time optimization of system operating parameters and formulation of control strategies, thus possessing high application value. The system includes a solid-liquid hybrid electric heating / storage device for rapid response to wind and solar power fluctuations and long-term peak shaving. Through multi-stage heat exchange and a hybrid thermal storage method combining liquid metal and cast iron, it achieves efficient thermal storage and peak shaving capabilities. This system can effectively solve the problems of high volatility in renewable energy and difficulties in power system dispatch, providing a novel solution for the efficient utilization of green energy in the future.

[0010] This invention combines the advantages of energy thermal engineering and artificial intelligence to address the problems of high cost, poor real-time performance, and insufficient generalization ability of traditional three-dimensional numerical calculations and data-driven models. It constructs a novel surrogate model optimization method for hybrid electric heating / thermal storage systems with engineering applicability. On one hand, by rationally designing the flow and thermal field coupling structure of cast iron and liquid metal, efficient heat exchange and storage of the solid-liquid mixture are achieved, improving the electrothermal conversion efficiency of the device. On the other hand, an improved transfer learning-based surrogate model construction method is proposed. First, a deep neural network is pre-trained using source domain data, and the hidden layer parameters are frozen to extract general width features. Then, combined with a small number of samples from the target system, the output weights are optimized by introducing an elastic network regularization structure and the alternating direction multiplier method (ADMM). This achieves high-precision modeling and prediction of the operating characteristics of the cast iron-liquid metal solid-liquid hybrid electric heating / thermal storage device under small sample conditions, significantly improving the model's generalization ability, stability, and sparse interpretability. This method is suitable for rapid modeling and intelligent optimization control of complex energy equipment.

[0011] To achieve the above objectives, the present invention adopts the following technical solution:

[0012] A method for optimizing a solid-liquid hybrid electric heating thermal storage agent includes the following steps:

[0013] Based on the operational requirements of the electric heating / heat storage device in the solid-liquid hybrid electric heating and heat storage system, optimization variables are determined, with electrothermal conversion efficiency as the objective function, and temperature, pressure, and flow rate constraints are set.

[0014] A three-dimensional multiphysics numerical model is constructed and simulated to obtain a source domain dataset.

[0015] Based on the source domain dataset, samples are generated by Latin hypercube sampling, backpropagated to a pre-trained deep neural network and its feature extraction layer is frozen. An elastic network regularization structure is introduced into the output layer, and the output weight matrix is ​​solved by an improved alternating direction multiplier method to establish an improved transfer learning surrogate model.

[0016] A training set is constructed by collecting a small number of samples from the target domain. Based on the frozen feature extraction layer, the output layer parameters are fine-tuned to achieve rapid adaptation to the working conditions of the target domain.

[0017] The performance metrics of candidate runtime parameters are quickly predicted using a trained improved transfer learning agent model.

[0018] In the optimization search phase, the trained improved transfer learning agent model is replaced with the three-dimensional multiphysics numerical model, and the particle swarm algorithm is used for global optimization. The optimal running strategy is obtained through iterative search. When the prediction deviation of the working point exceeds the threshold, the working point is added to the target domain training set and the training optimization process is repeated to achieve adaptive iterative update.

[0019] The system dynamically manages the solid-liquid hybrid electric heating and heat storage system by adjusting the electric heating power, liquid metal inlet temperature, and flow rate in real time according to the optimal operating strategy, and by combining sensor feedback for closed-loop regulation.

[0020] This invention also provides a solid-liquid hybrid electric heating and heat storage system, employing the aforementioned optimized method for solid-liquid hybrid electric heating and heat storage, comprising: a low-temperature molten salt tank, a first valve, a first high-temperature molten salt pump, a molten salt electric heater, and a high-temperature molten salt tank connected in sequence; the outlet of the high-temperature molten salt tank is connected in sequence to a second valve, a second high-temperature molten salt pump, and a molten salt-water vapor generator, with the molten salt side outlet of the molten salt-water vapor generator flowing back to the low-temperature molten salt tank to form a closed loop; the outlet of the liquid metal storage tank is connected in sequence to a third valve, a high-temperature liquid metal pump, and an electric heating / heat storage device, with the liquid metal outlet of the electric heating / heat storage device connected to a liquid metal-steam generator, and the liquid metal side outlet of the liquid metal-steam generator flowing back to the liquid metal storage tank to form a closed loop; the steam side of the molten salt-water vapor generator is connected to the steam pipeline of the liquid metal-steam generator; a new energy power generation device supplies power to the molten salt electric heater and the electric heating / heat storage device respectively; a control unit is connected to the sensors on the electric heating / heat storage device via signal lines, and is connected to the power supplies, valves, and pumps of each electric heater via control lines.

[0021] Beneficial effects:

[0022] 1. This invention employs a multi-stage heat exchange structure involving molten salt, liquid metal, and cast iron. Compared to traditional single-stage heat exchange methods, this effectively improves the system's heat transfer efficiency. Through heat exchange between molten salt and water, secondary heat exchange between steam and liquid metal, and precise control of the cast iron-liquid metal exchange, heat loss is reduced, and high-temperature heat energy can be stably provided in a short time, thereby improving the overall system efficiency.

[0023] 2. Liquid metal has higher thermal conductivity and can transfer heat faster at the same temperature compared to molten salt, avoiding the thermal lag problem of molten salt energy storage systems at high temperatures, thereby improving the speed and accuracy of heat exchange.

[0024] 3. Liquid metals can store more heat at high temperatures, while cast iron, as a stable heat storage material, provides reliable temperature control and continuous high-temperature output, which significantly enhances the heat storage capacity of the entire system.

[0025] 4. This invention utilizes wind and solar power to drive an electric heater, converting green energy into efficient thermal storage. This not only reduces the consumption of traditional fossil fuels but also optimizes the thermal storage process through power dispatch and load forecasting, improving the system's operational flexibility and response speed. When wind and solar power fluctuate, the system can achieve rapid peak shaving by precisely controlling heat output, reducing grid instability.

[0026] 5. To address the issues of long computation time and difficulty in meeting real-time optimization requirements in traditional 3D numerical simulations, this invention introduces a data-driven surrogate model, significantly reducing modeling and simulation costs. Furthermore, in the face of constraints in industrial applications such as scarce samples and strong system nonlinearity, this invention proposes an improved transfer learning strategy. This involves extracting general-width features by freezing the hidden layers and introducing an elastic network regularization term in the output layer to enhance the sparsity and stability of the model, avoiding overfitting and improving the model's generalization ability and interpretability in complex scenarios. Finally, an improved Alternating Direction Multiplier Method (ADMM) is proposed for efficient analytical solution of the output weights, further improving training efficiency and model controllability. This enables the established surrogate model to have rapid prediction and parameter optimization capabilities, making it suitable for operational modeling, online prediction, and intelligent control of complex thermal systems such as electric heating / heat storage devices with solid-liquid hybrid cast iron and liquid metal. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of a solid-liquid hybrid electric heating and heat storage system according to an embodiment of the present invention.

[0028] Figure 2 This is a schematic diagram of an electric heating / heat storage device for solid-liquid mixing of cast iron and liquid metal, according to an embodiment of the present invention.

[0029] The attached figures are labeled as follows: 1-low temperature molten salt tank, 2-first valve, 3-first high temperature molten salt pump, 4-molten salt electric heater, 5-high temperature molten salt tank, 6-second valve, 7-second high temperature molten salt pump, 8-molten salt-steam generator, 9-power supply for first electric heater, 10-liquid metal-steam generator, 11-liquid metal storage tank, 12-third valve, 13-high temperature liquid metal pump, 14-electric heating / heat storage device, 15-power supply for second electric heater, 16-new energy power generation device, 17-control unit, A-electric heater, B-cast iron, C-liquid metal flow channel, D-temperature sensor, E-pressure sensor, F-liquid metal inlet, G-liquid metal outlet. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other. The invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0031] like Figure 1 and Figure 2 As shown, an embodiment of the present invention provides a solid-liquid hybrid electric heating and heat storage system with a multi-stage heat exchange structure, including a low-temperature molten salt tank 1, a first valve 2, a first high-temperature molten salt pump 3, a molten salt electric heater 4, a high-temperature molten salt tank 5, a second valve 6, a second high-temperature molten salt pump 7, a molten salt-water vapor generator 8, a first electric heater power supply 9, a liquid metal-steam generator 10, a liquid metal storage tank 11, a third valve 12, a high-temperature liquid metal pump 13, an electric heating / heat storage device 14, a second electric heater power supply 15, a new energy power generation device 16, and a control unit 17.

[0032] The outlet of the cryogenic molten salt tank 1 is connected in sequence to the first valve 2 and the first high-temperature molten salt pump 3 via pipelines. The first valve 2 is installed on the outlet pipeline of the cryogenic molten salt tank 1 and is used to control the flow rate of molten salt flowing downstream from the cryogenic molten salt tank 1. The downstream of the first valve 2 is connected to the first high-temperature molten salt pump 3 via a pipeline. The outlet of the first high-temperature molten salt pump 3 is connected to the inlet of the molten salt electric heater 4 via a pipeline. The inlet of the molten salt electric heater 4 is connected to the outlet of the first high-temperature molten salt pump 3, and its outlet is connected to the high-temperature molten salt tank 5 via a pipeline. The power supply terminal of the molten salt electric heater 4 is connected to the power supply 9 of the first electric heater. The inlet of the high-temperature molten salt tank 5 is connected to the outlet pipeline of the molten salt electric heater 4. The outlet of the high-temperature molten salt tank 5 is connected in sequence to the second valve 6 and the second high-temperature molten salt pump 7 via pipelines. The second valve 6 is installed at the outlet of the high-temperature molten salt tank 5 and is used to control the flow rate of molten salt supplied to the molten salt-steam generator 8. The downstream of the second valve 6 is connected to the second high-temperature molten salt pump 7. The second high-temperature molten salt pump 7 is located downstream of the second valve 6, and its outlet is connected to the molten salt side inlet of the molten salt-steam generator 8 via a pipeline. The molten salt side inlet of the molten salt-steam generator 8 is connected to the outlet of the second high-temperature molten salt pump 7, and the molten salt side outlet of the molten salt-steam generator 8 is connected back to the low-temperature molten salt tank 1 via a return pipeline, forming a closed-loop molten salt circulation loop. The steam side of the molten salt-steam generator 8 is connected to the steam pipeline of the liquid metal-steam generator 10, jointly supplying steam. The first electric heater power supply 9 supplies power to the molten salt electric heater 4 via a cable and is connected to the new energy power generation device 16 via a power line, utilizing wind and solar energy to drive the molten salt heating.

[0033] The liquid metal-side inlet of the liquid metal-steam generator 10 is connected to the outlet of the electric heating / heat storage device 14 via a pipeline, and the liquid metal-side outlet is connected back to the liquid metal storage tank 11 via a return pipeline, forming part of the liquid metal circulation. The steam side of the liquid metal-steam generator 10 is connected to the steam pipeline of the molten salt-water steam generator 8. The outlet of the liquid metal storage tank 11 is connected in sequence to the third valve 12 and the high-temperature liquid metal pump 13 via a pipeline, and the outlet of the high-temperature liquid metal pump 13 is then connected to the liquid metal inlet of the electric heating / heat storage device 14. The liquid metal return pipeline of the liquid metal-steam generator 10 is connected to the liquid metal storage tank 11, so that the liquid metal circulates between the liquid metal storage tank 11, the high-temperature liquid metal pump 13, the electric heating / heat storage device 14, and the liquid metal-steam generator 10.

[0034] like Figure 2 As shown, the electric heating / heat storage device 14 includes an electric heater A, a cast iron B, a liquid metal flow channel C, a temperature sensor D, a pressure sensor E, a liquid metal inlet F, and a liquid metal outlet G. Several electric heaters A are uniformly embedded inside the cast iron B.

[0035] Electric heater A is connected to the power supply 15 of the second electric heater and the control unit 17 to heat the cast iron B inside the device, forming a solid heat storage body. A liquid metal flow channel C is arranged in the electric heating / heat storage device 14, surrounding the cast iron B. One end of the liquid metal flow channel C is connected to the liquid metal inlet F, and the other end is connected to the liquid metal outlet G. The liquid metal enters from end F, flows through the liquid metal flow channel C through the cast iron B, exchanges heat with the heat storage body, and then flows out from the liquid metal outlet G. The liquid metal enters the liquid metal-steam generator 10 and releases heat to the steam side.

[0036] Temperature sensor D extends into the electric heating / heat storage device 14 to measure the temperature of the cast iron and liquid metal inside. Temperature sensor D is connected to control unit 17 via a signal line, which adjusts the electric heating power and liquid metal flow rate based on the temperature signal. Pressure sensor extends from the top of the electric heating / heat storage device 14 to detect the pressure inside. It is connected to control unit 17 via a signal line to monitor and protect the system's operational safety.

[0037] therefore, Figure 2 The electric heater (A), cast iron (B), liquid metal flow channel (C), temperature sensor (D), pressure sensor (E), liquid metal inlet (F), and liquid metal outlet (G) form an integrated channel structure within the electric heating / heat storage device 14, comprising electric heating, cast iron heat storage, and liquid metal heat exchange. This structure connects to the liquid metal inlet (F) and liquid metal outlet (G). Figure 1 The liquid metal circulation loop is connected to the temperature sensor D, pressure sensor E and control unit 17 to form a closed loop control.

[0038] like Figure 2 As shown, the electric heating / heat storage device 14 is internally equipped with an electric heater A and a cast iron for heat storage B, and is connected to the second electric heater power supply 15 and the control unit 17. The new energy power generation device 16 includes a wind turbine and photovoltaic modules, which supply power to the first electric heater power supply 9 and the second electric heater power supply 15 through power lines, providing new energy power to the molten salt electric heater 4 and the electric heating / heat storage device 14. The control unit 17 is connected to the temperature sensor D and pressure sensor E on the electric heating / heat storage device 14 through signal lines to collect the temperature and pressure signals of the electric heating / heat storage device 14 in real time. The control unit 17 is also connected to the first electric heater power supply 9, the second electric heater power supply 15, the first valve 2, the first high-temperature molten salt pump 3, the second valve 6, the second high-temperature molten salt pump 7, the third valve 12, and the high-temperature liquid metal pump 13 through control lines to realize the start-up and shutdown and power regulation of the molten salt circuit and the liquid metal circuit.

[0039] During normal operation, the new energy power generation unit 16, including wind and solar power, supplies power to the electric heating equipment in the system via the second electric heater power supply 15. The electric heating equipment includes a molten salt electric heater 4 and an electric heater A, which heat molten salt and cast iron B, respectively. The cryogenic molten salt is first fed into the molten salt electric heater 4 through a cryogenic molten salt tank 1 via a first valve 2 and a first high-temperature molten salt pump 3. In the molten salt electric heater 4, the cryogenic molten salt is heated to 540°C and then flows into the high-temperature molten salt tank 5 for storage. Using a molten salt-water steam generator 8, the high-temperature molten salt transfers heat to the water through heat transfer, generating steam at 495°C.

[0040] The electric heater A in the electric heating / storage device 14 converts the electricity provided by wind and solar power generation into heat energy, which heats the cast iron B in the device. The cast iron B heats up rapidly and stores the heat energy, transferring it to a liquid metal (such as sodium or a sodium-potassium alloy). The liquid metal absorbs the heat released by the cast iron through heat exchange, ensuring that the heat energy is stored stably and used in the subsequent power generation process.

[0041] The steam generated after the molten salt exchanges heat with water vapor is then subjected to a secondary heat exchange with liquid metal in the liquid metal-steam generator 10, producing 560°C steam to drive a steam turbine for power generation. After the heat exchange is complete, the liquid metal flows back to the liquid metal storage tank 11, forming a closed-loop cycle to ensure the continuous utilization of the liquid metal's thermal energy. This closed-loop cycle maintains efficient electro-thermal conversion and a stable steam supply.

[0042] Electric heater A absorbs or mitigates the second-to-minute output fluctuations of wind and solar power, while molten salt electric heater 4 achieves peak-to-valley shifts on an hourly to day-to-day scale, enabling full utilization of wind and solar renewable energy. Through a multi-stage thermal storage module consisting of molten salt, liquid metal, and cast iron B, renewable energy fluctuations are stored, and during peak electricity demand, heat release is regulated to balance the grid load. If wind and solar power is insufficient, the system adjusts the flow rate and inlet temperature of the molten salt and liquid metal, and adjusts the distribution of electric heating power according to real-time electricity demand, ensuring the system can respond quickly to demand within a short timeframe.

[0043] Preferably, the temperature of the molten salt in the low-temperature molten salt tank 1 is 290°C.

[0044] Preferably, the molten salt electric heater 4 consists of four shell-and-tube heat exchangers, with the tube side being the electric heater and the shell side being the molten salt. The molten salt is heated and stored through convection heat transfer and heat conduction.

[0045] Preferably, the molten salt in the high-temperature molten salt tank 5 is heated to 540°C after passing through the molten salt electric heater.

[0046] Preferably, the molten salt-water steam generator 8 is used to preheat the feedwater, evaporate it, and perform a first superheat, heating the steam to 495°C.

[0047] Preferably, the liquid metal-steam generator 10 increases the steam temperature from 495°C to 560°C.

[0048] Preferably, the liquid metal storage tank 11 is used to store liquid metal.

[0049] Preferably, the electric heating / heat storage device 14 adopts a cast iron heating-liquid metal heat storage module.

[0050] This invention provides a proxy optimization method for the aforementioned solid-liquid hybrid electric heating / heat storage system, wherein the solid and liquid are preferably cast iron-liquid metal. This method is used to proxy optimize the electric heating / heat storage device while considering both computational efficiency and prediction accuracy, thereby obtaining the optimal operating strategy for the cast iron-liquid metal solid-liquid hybrid electric heating / heat storage device. The method includes the following steps:

[0051] Step 1: Based on the operating requirements and application scenarios of the electric heating / heat storage device with solid-liquid mixture of cast iron and liquid metal, determine the optimization variables, objective function, and constraints of the electric heating / heat storage device with solid-liquid mixture of cast iron and liquid metal.

[0052] Step 2: After determining the optimization variables, objective function, and constraints, a three-dimensional multiphysics numerical model is constructed for the electric heating / thermal storage device with a solid-liquid mixture of cast iron and liquid metal. This model includes the coupled physical processes of heat conduction and liquid metal flow, and details the geometry and material properties of the cast iron, electric heater, liquid metal flow channels, and insulation structure within the device. Based on this, and given boundary conditions such as inlet temperature, inlet velocity, outlet pressure, electric heating power in each section, and outer wall heat transfer coefficient under actual operating conditions, high-fidelity numerical simulation software is used to solve for the temperature field, flow field, and power characteristics under different operating conditions, providing source and target domain data for the subsequent construction of the surrogate model.

[0053] Step 3: Based on the concept of transfer learning, a large-scale dataset is first constructed in the source domain, including multiphysics numerical simulation data or historical operating data based on the electric heating / heat storage device of cast iron-liquid metal solid-liquid hybrid system. This dataset is then used to pre-train a deep neural network. Through backpropagation, multi-layer nonlinear mapping relationships are learned to obtain a general "width feature" that characterizes the heat transfer-flow coupling mechanism. After pre-training, the feature extraction layer parameters of the deep neural network are frozen, and retraining is performed only on the output layer. An elastic network regularization structure that balances sparsity and smoothness is introduced into the output layer. By simultaneously including L1 and L2 regularization terms, the model's anti-overfitting ability is improved. An improved alternating direction multiplier method (ADMM) is used to optimize the output layer with elastic network constraints, obtaining sparse and stable output weights. Finally, through forward propagation, the frozen features and the obtained sparse and stable output weights are linearly combined to form an improved transfer learning surrogate model, enabling efficient prediction of key performance indicators such as the electrothermal conversion efficiency of the electric heating / heat storage device of cast iron-liquid metal solid-liquid hybrid system.

[0054] Step 4: On a three-dimensional multiphysics numerical model or experimental platform of the cast iron-liquid metal solid-liquid hybrid electric heating / heat storage device, following a sampling strategy of Latin hypercube sampling or Sobol sequence, select several representative operating condition combinations to obtain input parameters such as liquid metal inlet temperature, inlet flow rate, and electric power of each heating circuit, along with corresponding output responses such as electrothermal conversion efficiency and temperature distribution. Combine these target domain samples with the feature structure of the pre-trained deep neural network from Step 3 to form a training set for fine-tuning the improved transfer learning surrogate model. Use this training set to retrain the output layer, updating the weight parameters of the elastic network, so that the improved transfer learning surrogate model can still accurately reflect the real operating characteristics of the cast iron-liquid metal solid-liquid hybrid electric heating / heat storage device even under small sample conditions (target domain samples are small sample data).

[0055] Step 5: During the optimization process, a trained improved transfer learning surrogate model is used to replace the computationally intensive three-dimensional multiphysics numerical model to quickly predict the electrothermal conversion efficiency and other constraint-related outputs for each set of candidate operating parameters. By calculating the objective function and verifying the constraints based on the prediction results of the improved transfer learning surrogate model, rapid energy assessment is achieved, which can significantly reduce the computation time of a single assessment and greatly improve the overall optimization efficiency.

[0056] Step 6: Based on the rapid evaluation capability in Step 5, the particle swarm optimization algorithm is used to globally optimize decision variables such as liquid metal inlet temperature, inlet flow rate, and input electrical power of each heating circuit. First, the positions and velocities of the particle swarm are initialized, and the position vector of each particle is considered as a set of candidate operating strategies. Then, in each iteration, an improved transfer learning surrogate model is called to calculate the objective function value of each particle, and feasible solutions are selected according to constraints, updating individual extrema and global extrema. The velocity and position update formulas of the particle swarm optimization algorithm are then used for iterative searching until the maximum number of iterations or convergence criteria are met, thereby obtaining the optimal operating strategy for the cast iron-liquid metal solid-liquid hybrid electric heating / heat storage device, including the optimal liquid metal inlet temperature, flow rate, and input electrical power allocation of each heating circuit.

[0057] Step 7: During the optimization process described above, if the deviation between the prediction results of the improved transfer learning surrogate model and the calculated values ​​of the 3D multiphysics numerical model or the actual measured data exceeds a preset threshold, it is determined that the current training data is insufficient in describing certain regions. In this case, the operating points with larger errors and their neighboring regions are selected as new sample points. The actual output is obtained through high-fidelity simulation or experimentation and added to the training dataset. The surrogate model training in Step 4 and the optimization search in Steps 5 and 6 are then re-executed. Through this adaptive iterative mechanism of "prediction-correction-retraining-reoptimization," the target domain samples are gradually enriched, the prediction accuracy and robustness of the improved transfer learning surrogate model are improved, and the quality of the global optimal solution searched by the particle swarm optimization algorithm is continuously improved, ultimately approximating the optimal operating strategy of the hybrid electric heating / thermal storage system.

[0058] Step 8: The obtained optimal operating strategy includes key parameters such as the liquid metal inlet temperature, liquid metal flow rate, and power distribution of each heating circuit. The control unit 17 of the electric heating / heat storage device adjusts the electric heater power output, liquid metal inlet temperature, and flow rate in real time according to this optimal operating strategy, achieving dynamic control of the system's operating status. This ensures that the electrothermal conversion efficiency of the cast iron-liquid metal solid-liquid mixing electric heating / heat storage device is maximized while meeting safety constraints.

[0059] Specifically, in step 1, the optimization variables are the inlet temperature and flow rate of the liquid metal, and the input electrical power of each loop. In step 1, the cast iron absorbs heat from the electric heater and transfers it to the liquid metal, while the liquid metal transfers heat and exchanges heat with the steam to generate high-temperature steam. The most suitable optimization objective should simultaneously reflect the synergistic effect of the liquid metal and the cast iron; therefore, the objective function is the electrothermal conversion efficiency, expressed by the formula:

[0060] (1)

[0061] in, For electrothermal conversion efficiency; For the quality of cast iron; The specific heat capacity at constant pressure of cast iron; This represents the temperature difference before and after heating the cast iron. This refers to the mass flow rate of the liquid metal. The specific heat capacity at constant pressure of liquid metal, The temperature difference between the inlet and outlet of the liquid metal; This is the input power for electric heating.

[0062] Specifically, in step 1, corresponding constraints are given based on the equipment structural parameters and operating boundaries, including the maximum pipe wall temperature of the electric heater, the safe pressure, the maximum temperature of the cast iron, and the temperature and flow range of the liquid metal.

[0063] Specifically, in step 3, an improved transfer learning proxy model is constructed. Deep feature training, including convolutional layers and fully connected layers, is performed on the collected source domain data, and its parameters are frozen. Based on a small number of samples from the target domain, a regression problem with elastic network regularization is solved in the linear output layer to obtain the sparse output weight vector. The prediction method expression of the constructed improved transfer learning proxy model is as follows:

[0064] (2)

[0065] in, To improve the thermoelectric conversion efficiency of the electric heating / thermal storage device for cast iron-liquid metal solid-liquid mixture predicted by the transfer learning surrogate model; The feature mapping matrix; This is the output weight matrix.

[0066] To solve for the output weight matrix in formula (2), we can explicitly obtain it using the elastic network objective function:

[0067] (3)

[0068] in, The final output is the weight matrix; Denotes the variable that minimizes the objective function. The value of ; This represents the number of training samples; The square of the L2 norm; It is a norm; and This represents the regularization coefficient that controls smoothness and sparsity.

[0069] Introducing auxiliary variables, namely the sparse surrogate vectors generated by the splitting. ,constraint Solving formula (3) transforms it into an augmented Lagrange expression:

[0070] (4)

[0071] in, Represents the augmented Lagrange function; The sparse proxy vectors generated by the splitting; To constrain The corresponding scaling dual variable; For adaptively updated augmented Lagrange multipliers.

[0072] During the iterative solution process, the following settings are configured:

[0073] (5)

[0074] (6)

[0075] (7)

[0076] in, and Indicates the solution to the first and the step; These are intermediate variables after extrapolation. This is an element-level soft threshold function.

[0077] The expression for over-relaxation extrapolation is:

[0078] (8)

[0079] in, The relaxation coefficient takes a value of [value missing]. .

[0080] After the above iterations converge, a sparse and smooth final output weight matrix is ​​obtained. During prediction, only forward computation is required. This allows for the efficient output of the electrothermal conversion efficiency of the cast iron-liquid metal solid-liquid mixture electric heating / heat storage device, where... This represents the predicted electrothermal conversion efficiency.

[0081] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for optimizing solid-liquid hybrid electric heating and heat storage agents, characterized in that, Includes the following steps: Based on the operational requirements of the electric heating / heat storage device in the solid-liquid hybrid electric heating and heat storage system, optimization variables are determined, with electrothermal conversion efficiency as the objective function, and temperature, pressure, and flow rate constraints are set. A three-dimensional multiphysics numerical model is constructed and simulated to obtain a source domain dataset. Based on the source domain dataset, samples are generated by Latin hypercube sampling, backpropagated to a pre-trained deep neural network and its feature extraction layer is frozen. An elastic network regularization structure is introduced into the output layer, and the output weight matrix is ​​solved by an improved alternating direction multiplier method to establish an improved transfer learning surrogate model. The process of constructing the improved transfer learning agent model is as follows: Based on the concept of transfer learning, a large-scale dataset is first constructed in the source domain, including multiphysics numerical simulation data or historical operation data of an electric heating / heat storage device based on solid-liquid mixing of cast iron and liquid metal. This dataset is then used to pre-train a deep neural network. Through backpropagation, multi-layer nonlinear mapping relationships are learned to obtain a general "width feature" that characterizes the heat transfer-flow coupling mechanism. After pre-training, the feature extraction layer parameters of the deep neural network are frozen, and retraining is performed only on the output layer. An elastic network regularization structure that balances sparsity and smoothness is introduced into the output layer. By simultaneously including L1 and L2 regularization terms, the model's anti-overfitting ability is improved. An improved alternating direction multiplier method is used to optimize the output layer with elastic network constraints, obtaining sparse and stable output weights. Finally, the frozen features and the obtained sparse and stable output weights are linearly combined through forward propagation to form an improved transfer learning surrogate model. A training set is constructed by collecting a small number of samples from the target domain. Based on the frozen feature extraction layer, the output layer parameters are fine-tuned to achieve rapid adaptation to the working conditions of the target domain. The performance metrics of candidate runtime parameters are quickly predicted using a trained improved transfer learning agent model. In the optimization search phase, the trained improved transfer learning agent model is replaced with the three-dimensional multiphysics numerical model, and the particle swarm algorithm is used for global optimization. The optimal running strategy is obtained through iterative search. When the prediction deviation of the working point exceeds the threshold, the working point is added to the target domain training set and the training optimization process is repeated to achieve adaptive iterative update. The system dynamically manages the solid-liquid hybrid electric heating and heat storage system by adjusting the electric heating power, liquid metal inlet temperature, and flow rate in real time according to the optimal operating strategy, and by combining sensor feedback for closed-loop regulation.

2. The method for optimizing solid-liquid hybrid electric heating and heat storage agent according to claim 1, characterized in that, The optimization variables include the liquid metal inlet temperature, flow rate, and electric heating power of each heating circuit.

3. The method for optimizing solid-liquid hybrid electric heating and heat storage agent according to claim 1, characterized in that, The experimental data was collected as the target domain. A small number of samples of the target domain were obtained by using Latin hypercube sampling or Sobol sequence sampling strategies. Representative working condition combinations were selected to obtain the input parameter and output response pairs as the fine-tuning training set. The output layer is retrained using the fine-tuned training set to update the weight parameters of the elastic network.

4. The method for optimizing solid-liquid hybrid electric heating and heat storage agent according to claim 1, characterized in that, The global optimization process of the particle swarm optimization algorithm includes: initializing the positions and velocities of the particle swarm, using each particle's position vector as a candidate running strategy, calling an improved transfer learning surrogate model to calculate the objective function value, selecting feasible solutions based on constraints, updating individual extrema and global extrema, and iteratively searching according to the velocity update and position update formulas until the convergence criterion is met.

5. The method for optimizing solid-liquid hybrid electric heating and heat storage agent according to claim 1, characterized in that, The adaptive iterative update includes: when the deviation between the prediction results of the improved transfer learning surrogate model and the working point of the three-dimensional multiphysics numerical model or experimental data exceeds a preset threshold, the working point and its neighboring area are used as new sample points. The real output is obtained through simulation or experiment and added to the fine-tuning training set. The training and optimization search of the improved transfer learning surrogate model are then re-executed.

6. The method for optimizing solid-liquid hybrid electric heating and heat storage agent according to claim 1, characterized in that, Temperature, pressure, and flow rate constraints include the maximum wall temperature limit for the electric heater, the system safety pressure limit, the maximum temperature limit for cast iron, the temperature range for liquid metal, and the flow rate range for liquid metal.

7. A solid-liquid hybrid electric heating and heat storage system, employing the solid-liquid hybrid electric heating and heat storage proxy optimization method according to any one of claims 1-6, characterized in that, include: The system is connected in sequence to a cryogenic molten salt tank, a first valve, a first high-temperature molten salt pump, a molten salt electric heater, and a high-temperature molten salt tank. The outlet of the high-temperature molten salt tank is connected in sequence to a second valve, a second high-temperature molten salt pump, and a molten salt-steam generator. The molten salt side outlet of the molten salt-steam generator flows back to the cryogenic molten salt tank, forming a closed loop. The outlet of the liquid metal storage tank is connected in sequence to a third valve, a high-temperature liquid metal pump, and an electric heating / heat storage device. The liquid metal outlet of the electric heating / heat storage device is connected to a liquid metal-steam generator. The liquid metal side outlet of the liquid metal-steam generator flows back to the liquid metal storage tank, forming a closed loop. The steam side of the molten salt-steam generator is connected to the steam pipeline of the liquid metal-steam generator. The new energy power generation device supplies power to the molten salt electric heater and the electric heating / heat storage device, respectively. The control unit is connected to the sensors on the electric heating / heat storage device via signal lines and to the power supplies, valves, and pumps of each electric heater via control lines.

8. The solid-liquid hybrid electric heating and heat storage system according to claim 7, characterized in that, The electric heating / heat storage device is equipped with an electric heater, cast iron, a liquid metal flow channel, a temperature sensor, and a pressure sensor. The liquid metal flow channel surrounds the cast iron and has a liquid metal inlet and a liquid metal outlet. The temperature sensor and pressure sensor extend into the electric heating / heat storage device and are connected to the control unit. Several electric heaters are evenly embedded inside the cast iron.

9. A solid-liquid hybrid electric heating and heat storage system according to claim 7, characterized in that, The control unit collects temperature and pressure signals from the electric heating / heat storage device in real time via signal lines, adjusts the power of the molten salt electric heater and the electric heating / heat storage device via control lines, controls the opening of the first valve, the second valve, and the third valve, as well as the flow rates of the first high-temperature molten salt pump, the second high-temperature molten salt pump, and the high-temperature liquid metal pump, thereby achieving dynamic control of the operating status of the solid-liquid hybrid electric heating and heat storage system.

Images