Alkaline electrolytic water hydrogen production energy management system based on superposition and coupling of multiple algorithms, hydrogen production dynamic optimization method and electronic equipment

Abstract

The invention discloses an energy management system and a hydrogen production dynamic optimization method for alkaline electrolytic water hydrogen production based on superposition and coupling of multiple algorithms, and electronic equipment. The system comprises a hybrid modeling module, a layered optimization controller, a cloud edge collaborative incremental learning module and an uncertainty management module. The hybrid modeling module constructs an electrolytic cell digital twinborn body by fusing a physical model and a data driving model; the hierarchical optimization controller is divided into a top layer reinforcement learning scheduling unit, a middle layer model prediction control unit and a bottom layer self-adaptive PID control unit according to a time scale; the cloud edge collaborative incremental learning module deploys a lightweight edge model and is connected with a cloud high-fidelity simulator; and the uncertainty management module integrates Bayesian deep learning and a robust optimization algorithm. The influence of different working conditions on the electrolytic cell, the power module and the auxiliary equipment can be evaluated in advance in the virtual environment, and the actual debugging risk and energy waste are reduced.

Description

Technical Field

[0001] This invention relates to the field of alkaline water electrolysis for hydrogen production. More specifically, this invention relates to an energy management system, a dynamic optimization method for hydrogen production dynamics, and electronic equipment for alkaline water electrolysis using a combination of multiple algorithms. Background Technology

[0002] Alkaline electrolysis has a long history of application in the industrial sector, but when faced with dynamic operating conditions coupled with renewable energy sources (such as wind and solar power), there is still room for optimization in terms of start-up time, partial load efficiency, and dynamic response characteristics. With the increasing demand for "green electricity to hydrogen," energy dispatch and efficiency improvement of alkaline electrolysis systems are required under fluctuating energy input conditions, which places higher demands on energy management. Hydrogen production plants or integrated energy scenarios often involve the coordinated operation of the power grid, renewable energy sources, energy storage devices, and electrolyzers; traditional "extensive" operating methods are difficult to adapt to rapid fluctuations and optimization needs.

[0003] Industrial-grade alkaline water hydrogen production systems typically employ automation methods such as PLC / DCS to monitor and adjust basic parameters such as temperature, pressure, voltage, and current of the electrolyzer. Most of these systems use static or single-rule control logic, which lacks the dynamic optimization capability to handle fluctuations in renewable energy and complex load changes, and cannot provide in-depth analysis of the energy utilization efficiency of the entire process. Summary of the Invention

[0004] The purpose of this invention is to provide an energy management system, a dynamic optimization method for hydrogen production via alkaline water electrolysis, and electronic equipment that combines multiple algorithms. This system uses real-time data input to correct the model and feeds optimization instructions back to the physical system. In a virtual environment, it can assess the impact of different operating conditions on the electrolyzer, power module, and auxiliary equipment in advance, reducing actual debugging risks and energy waste. Furthermore, by combining sensor data and multiphysics simulation, it can monitor the health status of key equipment (electrolyzer, power electronic devices, etc.) and predict failure trends.

[0005] The technical solution adopted by this invention to solve this technical problem is: an energy management system for alkaline water electrolysis hydrogen production by superimposing and coupling multiple algorithms, including: a hybrid modeling module, a hierarchical optimization controller, a cloud-edge collaborative incremental learning module, and an uncertainty management module; The hybrid modeling module constructs a digital twin of the electrolyzer by integrating the physical model and the data-driven model; The hierarchical optimization controller is divided into a top-level reinforcement learning scheduling unit, a middle-level model prediction control unit, and a bottom-level adaptive PID control unit according to the time scale. The cloud-edge collaborative incremental learning module deploys a lightweight edge model and connects to a cloud-based high-fidelity simulator. The uncertainty management module integrates Bayesian deep learning and robust optimization algorithms.

[0006] As a further aspect of the present invention, the hybrid modeling module achieves the fusion of the physical model and the data-driven model in the following manner: Physically Constrained Data-Driven Unit: Construct a neural network based on physical equation constraints, using the Butler-Volmer equation and thermodynamic equation of the electrolyzer as constraints of the loss function, and train the network using labeled data of current density, temperature, pressure, and hydrogen concentration to make the network output satisfy the electrochemical laws. Reduced-order data units for physics guidance: High-dimensional physical data of the flow / temperature field of the electrolyzer are generated through computational fluid dynamics (CFD) simulation; Key physical features are extracted using intrinsic orthogonal decomposition (POD) to construct a low-dimensional dynamic model; The dimensionality-reduced physical features are input into an LSTM network to predict the dynamic response of hydrogen production efficiency under fluctuating power.

[0007] As a further aspect of the present invention, the working logic of the hierarchical optimization controller is as follows: Top-level reinforcement learning scheduling unit: with wind and solar power generation forecast, electricity price, and hydrogen storage capacity as the state space, electrolyzer start-up and shutdown and power allocation as the action space, and minimizing hydrogen production cost as the reward function, a power allocation reference trajectory is generated through the DQN algorithm; Mid-level model prediction control unit: Based on the wind and solar power prediction for a certain period of time and the real-time status of the electrolyzer, the power command is continuously optimized under the constraints of efficiency-load curve and safe temperature threshold. The underlying adaptive PID control unit dynamically adjusts the PID parameters to control the DC / DC converter based on feedback from the electrolytic cell temperature / pressure sensor, thereby stabilizing the input power.

[0008] As a further aspect of the present invention, the mid-level model prediction control unit is a robust MPC (RMPC), whose objective function considers the worst-case scenarios of wind and solar fluctuations and equipment failures, and forcibly satisfies the safety threshold for all scenarios.

[0009] As a further aspect of the present invention, the cloud-edge collaborative incremental learning module includes: A lightweight LSTM model deployed at the edge executes millisecond-level control commands; The ANSYS Twin Builder multiphysics simulator, running in the cloud, is used for model parameter calibration; Incremental learning unit: Based on Gaussian process regression (GPR), it predicts the efficiency decay of electrolyzers and dynamically absorbs new data to update model parameters through exponential decay weights.

[0010] As a further aspect of the present invention, when the incremental learning unit assigns weights to new and old data, it satisfies the following constraints: , where α is the attenuation coefficient, 0 < α < 1; t is the time interval between the data acquisition time and the current time (unit: days).

[0011] As a further aspect of the present invention, the uncertainty management module includes: Bayesian deep learning unit: Monte Carlo Dropout technique is used in the neural network to output the confidence interval of the prediction results; when the width of the confidence interval of wind and solar power prediction exceeds the threshold, a conservative control strategy of reducing the load of the electrolyzer is triggered.

[0012] This invention also provides a method for dynamic optimization of hydrogen production systems, comprising the following steps: S1. Dimensionality reduction of electrolyzer fluid dynamics (CFD) simulation data is performed by intrinsic orthogonal decomposition (POD) to construct a low-dimensional dynamic model; S2. Embed physical equation constraints into the PINN training process to fine-tune network parameters online to adapt to changes in operating conditions; S3. Generate a long-term scheduling strategy based on the DQN algorithm, and decompose it into minute-level control instructions through MPC rolling optimization. S4. Use adaptive PID to adjust power electronic devices in real time to stabilize the operating parameters of the electrolytic cell.

[0013] As a further aspect of the present invention, physical constraints are achieved in step S2 by adding a physical consistency term to the loss function of the neural network, which forces the network output to satisfy the electrochemical equation of the electrolytic cell and the law of conservation of mass.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the program to implement the steps of the method.

[0015] The present invention has at least the following beneficial effects: the system achieves dynamic adaptation of the electrolyzer to the second-level fluctuations of renewable energy through a hierarchical optimized control architecture, compresses the power fluctuation response time, and reduces the wind / solar curtailment rate; at the same time, it ensures that the prediction results conform to the electrochemical laws based on physical constraint hybrid modeling.

[0016] This system innovatively integrates cloud-edge collaborative incremental learning with Bayesian uncertainty management. Real-time optimization and control are achieved using a lightweight LSTM model at the edge, while high-fidelity simulation in the cloud enables online parameter calibration, allowing the system to adapt to electrolyzer performance degradation (catalyst aging, etc.) with minimal model prediction errors. A robust MPC strategy triggered by confidence intervals ensures equipment safety and extends electrolyzer lifespan even in extreme scenarios with large wind and solar forecast deviations.

[0017] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0018] Figure 1 This is a flowchart of the dynamic optimization method for hydrogen production according to the present invention. Detailed Implementation

[0019] The present invention will now be described in detail and completely with reference to the accompanying drawings. Those skilled in the art will be able to implement the present invention based on these descriptions. Before describing the present invention with reference to the accompanying drawings, it should be particularly noted that the technical solutions and features provided in various parts of the present invention, including the following description, can be combined with each other without conflict.

[0020] Furthermore, the embodiments of the present invention described below are generally only some, not all, of the embodiments of the present invention. Therefore, all other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention.

[0021] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. The specific implementation process is as follows: like Figure 1 As shown, this invention provides an energy management system for alkaline water electrolysis hydrogen production using a combination of multiple algorithms, including: a hybrid modeling module, a hierarchical optimization controller, a cloud-edge collaborative incremental learning module, and an uncertainty management module; The hybrid modeling module constructs a digital twin of the electrolyzer by integrating the physical model and the data-driven model; The hierarchical optimization controller is divided into a top-level reinforcement learning scheduling unit, a middle-level model prediction control unit, and a bottom-level adaptive PID control unit according to the time scale. The cloud-edge collaborative incremental learning module deploys a lightweight edge model and connects to a cloud-based high-fidelity simulator. The uncertainty management module integrates Bayesian deep learning and robust optimization algorithms.

[0022] This technical solution may also include the following technical details to better achieve the technical effect: The hybrid modeling module achieves the fusion of the physical model and the data-driven model in the following ways: Physically Constrained Data-Driven Unit: A neural network based on physical equation constraints is constructed, using the Butler-Volmer equation and thermodynamic equation of the electrolyzer as constraints in the loss function. The network is trained using labeled data of current density, temperature, pressure, and hydrogen concentration to ensure that the network output conforms to electrochemical laws. The training data comes from real-time sensors of the electrolyzer's current density, temperature, pressure, and hydrogen concentration, with a sampling frequency of 1Hz. During online fine-tuning, the network weights are updated every 5 minutes using real-time sensor data.

[0023] Reduced-order data units for physics guidance: High-dimensional physical data of the flow field / temperature field of the electrolyzer are generated by computational fluid dynamics (CFD) simulation. The CFD simulation adopts the standard k-ε turbulence model with a meshing accuracy of 1 mm. The first 15 principal mode features (cumulative energy ratio > 90%) are extracted by POD. The LSTM network has an input dimension of 20 and outputs the predicted hydrogen production efficiency for the next 30 seconds.

[0024] Key physical features are extracted using intrinsic orthogonal decomposition (POD) to construct a low-dimensional dynamic model; The dimensionality-reduced physical features are input into an LSTM network to predict the dynamic response of hydrogen production efficiency under fluctuating power.

[0025] This technical solution may also include the following technical details to better achieve the technical effect: the working logic of the hierarchical optimization controller operates according to the following parameters in this embodiment: Top-level reinforcement learning scheduling unit: With wind and solar power generation forecasts, electricity prices, and hydrogen storage capacity as the state space, electrolyzer start-up / shutdown and power allocation as the action space, and minimizing hydrogen production costs as the reward function, a 24-hour power allocation reference trajectory is generated through the DQN algorithm; The state space of the DQN algorithm includes: 0-24 hour wind and solar power generation forecast curves, time-of-use electricity price data, and remaining capacity of hydrogen storage tanks (range 0-100%); The action space is defined as electrolyzer start-up / shutdown commands and power allocation ratios (10%-100%).

[0026] Mid-level model predictive control unit: Based on wind and solar power forecasts for a certain future time period and the real-time status of the electrolyzer, it continuously optimizes power commands under the constraints of efficiency-load curves and safe temperature thresholds. The continuous optimization cycle is 15 minutes, and the power commands are updated every 5 minutes; the safe temperature threshold is set to 80℃, and the efficiency-load curves are stored in a lookup table format.

[0027] The underlying adaptive PID control unit dynamically adjusts PID parameters to control the DC / DC converter based on feedback from the electrolytic cell temperature / pressure sensors, stabilizing the input power. The dynamic adjustment rules are as follows: when the electrolytic cell temperature exceeds 75℃, the proportional gain Kp is automatically reduced by 20%; when the pressure fluctuation exceeds ±5kPa, the integral time constant Ti is increased by 15%.

[0028] This technical solution may also include the following technical details to better achieve the technical effect: the mid-level model prediction control unit is a robust MPC (RMPC), whose objective function considers the worst-case scenarios of wind and solar power fluctuations and equipment failures, and forcibly satisfies the safety thresholds for all scenarios. In this embodiment, the extreme scenario library includes: a sudden drop of 50% in wind and solar power, cooling water pump failure, and DC / DC converter overcurrent; the safety thresholds are forcibly satisfied with an electrolytic cell temperature ≤85℃ and a pressure ≤2.5MPa.

[0029] This technical solution may also include the following technical details to better achieve the technical effect: The cloud-edge collaborative incremental learning module includes: The lightweight LSTM model deployed at the edge is deployed in an embedded industrial controller to execute millisecond-level control instructions with an execution cycle of ≤10ms. The ANSYS Twin Builder multiphysics simulator running in the cloud performs model calibration every 8 hours to calibrate model parameters and generate flow field distribution maps inside the electrolyzer.

[0030] Incremental learning unit: Based on Gaussian process regression (GPR), it predicts the efficiency decay of electrolyzers and dynamically updates the model parameters by incorporating new data through exponential decay weights. GPR uses daily hydrogen production data to update the efficiency decay model; the decay coefficient α = 0.95.

[0031] This technical solution may also include the following technical details to better achieve the technical effect: When the incremental learning unit assigns weights to the new and old data, it satisfies the following constraints: , where α is the attenuation coefficient, 0 < α < 1; t is the time interval between the data acquisition time and the current time (unit: days).

[0032] This technical solution may also include the following technical details to better achieve the technical effect: The uncertainty management module includes: Bayesian deep learning unit: Monte Carlo Dropout technique is used in the neural network to sample 50 times to generate confidence intervals; when the width of the wind and solar power prediction interval exceeds ±20%, the power of the electrolyzer is triggered to drop to 80% of the rated value.

[0033] In another embodiment, the method for dynamic optimization of hydrogen production in the system includes the following steps: S1. The dimensionality of the electrolyzer fluid dynamics (CFD) simulation data is reduced by intrinsic orthogonal decomposition (POD) to construct a low-dimensional dynamic model; the POD basis vectors are calculated from the CFD simulation dataset (containing 100 transient conditions).

[0034] S2. Embed physical equation constraints into the PINN training process and fine-tune network parameters online to adapt to changes in operating conditions; online fine-tuning uses real-time current and voltage data, with the learning rate fixed at 0.001.

[0035] S3. Generate a long-term scheduling strategy based on the DQN algorithm, and decompose it into minute-level control instructions through MPC rolling optimization. S4. The IGBT switching frequency of the DC / DC converter is controlled by an adaptive PID output signal to stabilize the operating parameters of the electrolyzer.

[0036] This technical solution may also include the following technical details to better achieve the technical effect: In step S2, physical constraints are achieved in the following way: a physical consistency term is added to the loss function of the neural network, which forces the network output to satisfy the electrochemical equation of the electrolytic cell and the law of conservation of mass. In this embodiment, the bv equation and the energy conservation equation are used.

[0037] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and embodiments shown and described herein.

Claims

1. An energy management system for alkaline water electrolysis hydrogen production using a combination of multiple superimposed and coupled algorithms, characterized in that, include: Hybrid modeling module, hierarchical optimization controller, cloud-edge collaborative incremental learning module, and uncertainty management module; The hybrid modeling module constructs a digital twin of the electrolyzer by integrating the physical model and the data-driven model; The hierarchical optimization controller is divided into a top-level reinforcement learning scheduling unit, a middle-level model prediction control unit, and a bottom-level adaptive PID control unit according to the time scale. The cloud-edge collaborative incremental learning module deploys a lightweight edge model and connects to a cloud-based high-fidelity simulator. The uncertainty management module integrates Bayesian deep learning and robust optimization algorithms.

2. The energy management system for alkaline water electrolysis hydrogen production using multiple superimposed and coupled algorithms as described in claim 1, characterized in that, The hybrid modeling module achieves the fusion of the physical model and the data-driven model in the following ways: Physically Constrained Data-Driven Unit: Construct a neural network based on physical equation constraints, using the Butler-Volmer equation and thermodynamic equation of the electrolyzer as constraints of the loss function, and train the network using labeled data of current density, temperature, pressure, and hydrogen concentration to make the network output satisfy the electrochemical laws. Reduced-order data units for physics guidance: High-dimensional physical data of the flow field / temperature field of the electrolyzer are generated through computational fluid dynamics simulation. Key physical features are extracted using intrinsic orthogonal decomposition to construct a low-dimensional dynamic model; The dimensionality-reduced physical features are input into an LSTM network to predict the dynamic response of hydrogen production efficiency under fluctuating power.

3. The energy management system for alkaline water electrolysis hydrogen production using multiple superimposed and coupled algorithms as described in claim 1, characterized in that, The working logic of the hierarchical optimization controller is as follows: Top-level reinforcement learning scheduling unit: with wind and solar power generation forecast, electricity price, and hydrogen storage capacity as the state space, electrolyzer start-up and shutdown and power allocation as the action space, and minimizing hydrogen production cost as the reward function, a power allocation reference trajectory is generated through the DQN algorithm; Mid-level model prediction control unit: Based on the wind and solar power prediction for a certain period of time and the real-time status of the electrolyzer, the power command is continuously optimized under the constraints of efficiency-load curve and safe temperature threshold. The underlying adaptive PID control unit dynamically adjusts the PID parameters to control the DC / DC converter based on feedback from the electrolytic cell temperature / pressure sensor, thereby stabilizing the input power.

4. The energy management system for alkaline water electrolysis hydrogen production using multiple superimposed and coupled algorithms as described in claim 3, characterized in that, The mid-level model prediction control unit is a robust MPC, whose objective function considers the worst-case scenarios of wind and solar fluctuations and equipment failures, and forces the satisfaction of the safety threshold for all scenarios.

5. The energy management system for alkaline water electrolysis hydrogen production using multiple superimposed and coupled algorithms as described in claim 1, characterized in that, The cloud-edge collaborative incremental learning module includes: A lightweight LSTM model deployed at the edge executes millisecond-level control commands; The ANSYS Twin Builder multiphysics simulator, running in the cloud, is used for model parameter calibration; Incremental learning unit: Based on Gaussian process regression, it predicts the efficiency decay of electrolyzers and dynamically absorbs new data to update model parameters through exponential decay weights.

6. The energy management system for alkaline water electrolysis hydrogen production using multiple superimposed and coupled algorithms as described in claim 5, characterized in that, When the incremental learning unit assigns weights to new and old data, it satisfies the following constraints: , where α is the attenuation coefficient, 0 < α < 1; t is the time interval between the data acquisition time and the current time.

7. The energy management system for alkaline water electrolysis hydrogen production using multiple superimposed and coupled algorithms as described in claim 1, characterized in that, The uncertainty management module includes: Bayesian deep learning unit: Monte Carlo Dropout technique is used in the neural network to output the confidence interval of the prediction results; when the width of the confidence interval of wind and solar power prediction exceeds the threshold, a conservative control strategy of reducing the load of the electrolyzer is triggered.

8. The method for dynamic optimization of hydrogen production according to any one of claims 1 to 7, characterized in that, Including the following steps: S1. Dimensionality reduction of electrolytic cell fluid dynamics simulation data is achieved by intrinsic orthogonal decomposition to construct a low-dimensional dynamic model; S2. Embed physical equation constraints into the PINN training process to fine-tune network parameters online to adapt to changes in operating conditions; S3. Generate a long-term scheduling strategy based on the DQN algorithm, and decompose it into minute-level control instructions through MPC rolling optimization. S4. Use adaptive PID to adjust power electronic devices in real time to stabilize the operating parameters of the electrolytic cell.

9. The method for dynamic optimization of hydrogen production in the system as described in claim 8, characterized in that, In step S2, physical constraints are achieved by adding a physical consistency term to the loss function of the neural network, which forces the network output to satisfy the electrochemical equation of the electrolyzer and the law of conservation of mass.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 8-9.

Images