Electrolysis reactor self-adaptive power supply regulation and control method based on digital twinborn model

Abstract

The invention discloses an electrolytic reactor self-adaptive power supply regulation and control method based on a digital twinborn model, which belongs to the technical field of electrolytic reactor power supply regulation and control, and establishes a dynamic mapping relation between a physical system state and measurable sensor data through combination of computational fluid dynamics (CFD) and machine learning. The state of a key reaction area between the cathode and the anode is sensed and predicted in real time; based on the digital twin, the limiting current density and the relaxation time are calculated, the power supply strategy is optimized, self-adaptive intelligent control over system power supply is achieved, and it is guaranteed that the electrochemical system operates under the working conditions of the maximum current efficiency and the minimum energy consumption all the time in the working process. Specifically, the method comprises the following steps: constructing a multi-working-condition CFD simulation data set; constructing a neural network model and reconstructing spatial information; calculating the limiting current density and the relaxation time in real time; the invention discloses a real-time intelligent accurate regulation and control strategy for equipment current.

Description

Technical Field

[0001] This invention belongs to the field of electrolyzer power supply regulation technology, specifically relating to an adaptive power supply regulation method for electrolyzers based on a digital twin model. Background Technology

[0002] An electrolytic reactor is a device that uses electrochemical technology to drive specific chemical reactions, and it has wide applications in wastewater treatment, industrial synthesis, and biosensing. In practical applications, the core issues of this technology and device are electrolytic reaction efficiency and energy consumption, with the applied current density being a key parameter affecting both. High current density increases reaction efficiency but also increases electricity consumption and the probability of side reactions; while low current density maintains high current efficiency but significantly reduces reaction efficiency. Therefore, accurately determining the supply current of the electrochemical reaction system to maximize current utilization efficiency while minimizing energy consumption is of significant research and application value.

[0003] During electrolysis, when the system is in the mass transfer control phase, the reactant concentration on the electrode surface will decrease to a steady-state value after a certain relaxation time. This steady-state concentration decreases as the supply current increases. When the supply current increases to a certain threshold, the steady-state concentration of reactants on the electrode surface decreases to 0. At this point, the reaction rate of the reactants is at its maximum, and the current efficiency is highest. This current threshold is the limiting current of the reaction system. Continuing to increase the current not only does not benefit the target reaction but also exacerbates side reactions and wastes energy. Therefore, controlling the supply current of the system to reach the limiting current is key to improving electrolysis efficiency and reducing electrolysis energy consumption.

[0004] However, the limiting current of the reaction system changes in real time, simultaneously controlled by the reactant concentration and flow rate in the bulk solution. In actual electrolysis, commonly used power supply methods (constant current and constant voltage) cannot continuously maintain a supply current equal to the limiting current. The limiting current must be calculated in real time based on the reactant concentration in the bulk solution and the flow and mass transfer within the reactor, thereby adjusting the supply current. Existing methods attempt to determine the average mass transfer coefficient based on the Sherwood relation, but these assume constant coefficients and primarily consider concentration effects, neglecting the decisive role of forced convection and differences in flow distribution on mass transfer at the electrode surface. Furthermore, since the electrolysis reaction occurs between the anode and cathode, there is currently a lack of experimental means to directly measure the bulk concentration and flow rate in the electrode gap, making it difficult to accurately calculate the electrode surface mass transfer coefficient and thus determine the real-time limiting current. Computational fluid dynamics (CFD) technology can accurately calculate mass transfer and flow field information in electrochemical reactors. CFD technology can directly extract flux from the wall concentration gradient, calculate the mass transfer coefficient according to the definition of the mass transfer coefficient, and thus determine the limiting current density of the system. However, CFD solutions to the control equations are time-consuming and cannot meet the requirements of real-time control.

[0005] With the development of industrial intelligence, digital twin technology has become an important tool for process industry optimization. By fusing simulation models with sensor data, real-time mapping and prediction of physical systems can be achieved. For electrolysis reactors, a typical multi-physics coupled system, joint modeling based on CFD and machine learning is an effective way to realize digital twins. In particular, by combining intrinsic orthogonal decomposition (POD) to reduce the order of the complex flow and concentration fields obtained by CFD, and then using machine learning models to establish the mapping relationship between sensor observations and POD coefficients, key physical features can be preserved while quickly achieving full-field prediction and order reconstruction. This not only allows for the dynamic acquisition of limiting current and relaxation time under different operating conditions, but also enables real-time optimization of power supply strategies, providing a feasible path and theoretical support for the intelligent control of electrolysis reactors. Summary of the Invention

[0006] To overcome the shortcomings of existing technologies, this invention provides an adaptive power supply control method for electrolytic reactors based on a digital twin model. By combining computational fluid dynamics (CFD) and machine learning, a dynamic mapping relationship between the physical system state and measurable sensor data is established, enabling real-time perception and prediction of the state of key reaction regions between the anode and cathode. Based on the digital twin, the limiting current density and relaxation time are calculated to optimize the power supply strategy, achieving adaptive intelligent control of the system power supply and ensuring that the electrochemical system operates at maximum current efficiency and minimum energy consumption during operation.

[0007] The adaptive power supply control method for an electrolysis reactor based on a digital twin model includes the following steps: Step 1: Construct a multi-condition CFD simulation dataset; Step 2: Neural network model construction and spatial information reconstruction; Step 3: Real-time calculation of limiting current density and relaxation time; Step 4: Real-time intelligent and precise control strategy for equipment current.

[0008] Furthermore, in step 1, the steady-state flow field and concentration field of the reactor under different initial boundary conditions are simulated by solving the Navier–Stokes equations and the reactant diffusion equations; the obtained simulation data includes the three-dimensional velocity vector distribution U(…) inside the reactor. x , y , z , t ) and concentration field C ( x , y , z , t Focusing on the boundary layer conditions within the anode-cathode gap region, a large-scale dataset containing the mapping relationship between input parameters and output features is established through batch simulations under different working conditions for training machine learning models.

[0009] Specifically, the specific operations of step 1 are as follows: Step 1-1: Determine the reactor's geometric parameters and inlet boundary conditions: electrode spacing g Average inlet velocity and inlet reactant concentration ; Steps 1-2: Select the turbulence model and the dilute mass transport equation, and set the electrode surface to concentration constraint. ; Steps 1-3: For each operating condition, calculate the steady-state or quasi-steady-state transient time slices, and derive the three-dimensional velocity vector distribution U( x , y , z , t ) and concentration field C ( x , y , z , t This allows us to obtain a CFD computation dataset, which will be used for subsequent neural network model construction and spatial information reconstruction.

[0010] Furthermore, in step 2, machine learning technology is used to establish a mapping relationship between sensor measurement point data and POD coefficient, so that in actual operation, the global velocity and concentration distribution in the anode-cathode gap can be quickly predicted and reconstructed in order of magnitude based on the real-time monitoring values ​​of the flow rate and concentration sensors.

[0011] Specifically, step 2 is performed as follows: Step 2-1: Read the velocity and concentration information at the sensor locations in the CFD dataset for each operating condition, and combine it with the corresponding reactor geometric parameters and inlet boundary conditions as the dataset for input features; Step 2-2: Unify the velocity field and concentration field derived from CFD onto the same mesh on a selected section, and perform POD on the velocity field and concentration field respectively to obtain the previous... The dataset consists of 1 principal modalities and their coefficients, serving as the output features. Steps 2-3: Standardize all inputs / outputs and divide them into training / validation / test datasets, with the training set comprising 70%-80%, the validation set 10%-20%, and the test set 10%-20%. Steps 2-4: Select the multilayer perceptron regression model, with the input being "sensor measurement points + boundary conditions + geometric parameters" and the output being "POD coefficient vector". Use the mean square error of the POD coefficients as the loss function and employ the adaptive gradient descent method to train and optimize the prediction model.

[0012] Furthermore, in step 3, the velocity and concentration information at the sensor location, along with the reactor's geometric parameters and inlet boundary conditions, are used as real-time inputs to obtain the POD coefficient output by the machine learning model. The POD coefficient is then upgraded to three-dimensional spatial information using an online reconstruction formula. Additionally, the flow rate and concentration information at the electrode surface are read to obtain the corresponding mass transfer coefficient, and the limiting current density at the electrode is calculated in real-time. j lim Meanwhile, based on Fick's law, the thickness of the Nernst diffusion layer is... Calculate the relaxation time of the reactants on the electrode surface after they are depleted, i.e., the time required to reach a steady-state concentration.

[0013] Furthermore, in step 4, the target supply current is set as the limit current density based on the real-time operating condition prediction, and the current setting is kept unchanged during the relaxation time to ensure sufficient steady-state establishment of interface mass transfer; the current controller adopts a stepped update or PID control strategy with hysteresis to make the actual supply current as close as possible to the limit current without exceeding the limit current.

[0014] Compared with the prior art, the present invention has the following advantages: (1) Accurate modeling of unmeasurable regions: The method of this invention uses measurable regions to model and predict the reactant concentration and flow rate in the electrode gap region, thereby achieving a “digital twin” of the core reaction region; (2) Real-time identification of extreme working conditions: The method of the present invention no longer relies on experience or average mass transfer coefficient, but dynamically calculates the limiting current to reflect the changes in real working conditions; (3) Reduce energy consumption and improve efficiency: The method of the present invention can control the power supply current to always be in the high efficiency range of the reactor, avoiding energy waste and side reactions; (4) Intelligent closed-loop control mechanism: This invention integrates sensing, modeling, calculation and control, and has adaptive and intelligent adjustment capabilities. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a two-dimensional cross-sectional view of the reactor in an embodiment of the present invention; Figure 3 This is a linear regression diagram of the average velocity and concentration on the electrode surface calculated by CFD for all data in the embodiments of the present invention, and the average velocity and concentration reconstructed by the upgraded prediction model. Figure 4 This is a diagram illustrating the flow rate variation in an embodiment of the present invention; Figure 5 This diagram illustrates the intelligent control strategy of the reactor in an embodiment of the present invention. Detailed Implementation

[0016] 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 specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0017] This invention integrates "CFD + POD order reduction / upgrade + machine learning + electrochemical theory" to achieve real-time perception and prediction of the state of key reaction zones in an electrolysis reactor. It provides an adaptive power supply control method for electrolysis reactors based on a digital twin model, such as... Figure 1 As shown, the specific steps include: (a) Step 1: Construct a multi-condition CFD simulation dataset.

[0018] By solving the Navier–Stokes (N–S) equations and the reactant diffusion equations, the steady-state flow field and concentration field of the reactor under different initial boundary conditions (including influent velocity, electrode spacing, reactant concentration, etc.) were simulated. The obtained simulation data includes the three-dimensional velocity vector distribution U(x, y, z, t) inside the reactor and the concentration field C(x, y, z, t), with particular attention paid to the boundary layer situation in the anode-cathode gap region (i.e., the main electrochemical reaction region). Through batch simulations under different operating conditions, a large-scale dataset was established, containing the mapping relationship between input parameters (such as influent velocity and reactant concentration, electrode spacing, velocity and concentration information at sensor locations, etc.) and output features (global velocity distribution and concentration distribution), for training machine learning models. The specific operation is as follows: (1) Determine the reactor geometric parameters and inlet boundary conditions: electrode spacing g, average inlet velocity and inlet reactant concentration .

[0019] (2) Select the turbulence model (preferably the Ke model) and the dilute mass transport equation, and set the electrode surface to concentration constraint ( ).

[0020] (3) For each working condition, calculate the steady state (or the transient time slice that reaches the quasi-steady state), derive the three-dimensional velocity vector distribution U(x, y, z, t) and concentration field C(x, y, z, t), and obtain the CFD calculation dataset for subsequent neural network model construction and spatial information reconstruction.

[0021] (ii) Step 2: Neural network model construction and spatial information reconstruction.

[0022] The reactor geometric parameters and inlet boundary conditions, as well as the velocity and concentration information at fixed sensor locations, from the aforementioned CFD dataset are extracted as the input dataset. Due to the large amount of global spatial information, the spatial field is first subjected to dimensionality reduction using Orthogonal Eigenfactor Decomposition (POD). The POD coefficient vector after dimensionality reduction is then used as the output dataset to establish the total dataset for machine learning training. The dataset is divided into training and testing sets, and a prediction model is constructed using a Multilayer Perceptron (MLP) regression model. The specific operation is as follows: (1) Read the velocity and concentration information at the sensor location in the CFD dataset for each working condition, and combine it with the corresponding reactor geometric parameters and inlet boundary conditions as the dataset of input features.

[0023] (2) Unify the velocity field and concentration field derived from CFD onto the same grid on a selected section, and perform POD on the velocity field and concentration field respectively to obtain the previous... dominant mode , and its coefficient , This dataset is used as the output feature.

[0024] (3) Standardize all inputs / outputs and divide them into training / validation / test datasets, with the training set accounting for 70%-80% (preferably 70%), the validation set accounting for 10%-20% (preferably 15%), and the test set accounting for 10%-20% (preferably 15%).

[0025] (4) Select a multilayer perceptron (MLP) regression model. The input is “sensor measurement points + boundary conditions + geometric parameters”, and the output is “POD coefficient vector”. The mean square error (MSE) of the POD coefficients is used as the loss function, and the adaptive gradient descent method is used to train and optimize the prediction model.

[0026] (III) Step 3: Real-time calculation of limiting current density and relaxation time.

[0027] Using velocity and concentration information at the sensor location, along with reactor geometry and inlet boundary conditions, as real-time inputs, the POD coefficient is obtained from the machine learning model. This POD coefficient is then upgraded to three-dimensional spatial information via an online reconstruction formula. Furthermore, flow velocity and concentration information at the electrode surface are read to determine the corresponding mass transfer coefficient, and the limiting current density at the electrode is calculated in real-time. Meanwhile, based on Fick's law, the thickness of the Nernst diffusion layer is... Calculate the relaxation time of reactant depletion at the electrode surface The time required to reach a steady-state concentration (approaching 0) is calculated as follows: , in, The mass transfer coefficient from the bulk solution to the electrode surface; The reactant diffusion rate is used. The relaxation time is used to determine the system's response period to fluctuations in flow rate and concentration, providing a timescale reference for current regulation.

[0028] Specifically, the formula for calculating the limiting current density is: , In the formula, This represents the concentration of the reactants in the bulk solution. The number of electrons transferred; This is Faraday's constant, with a value of 96487 C·mol⁻¹. -1 .

[0029] In practice, mass transfer on the electrode surface is affected by a combination of diffusion and convection. The solution is affected by convection-diffusion effects. Unlike traditional semi-empirical formulas based on Sherwood, accurate calculation requires... The value of requires solving for the flow field and mass transfer field near the electrode. The defining formula is: , , In the formula, is the steady-state concentration of the reactants on the electrode surface at a certain current density, which is 0 when it is the limiting current density; J is the mass transfer flux of the reactants on the electrode surface, which is obtained from the wall normal gradient; D is the diffusion coefficient of the reactants.

[0030] Electrocatalytic oxidation reactions take place between the anode and cathode. Physical quantities such as the flow rate and mass transfer flux of the bulk solution between the anode and cathode are usually impossible to measure. Traditional methods often approximate the limiting current density by assuming the boundary layer thickness or the average mass transfer coefficient. However, such methods cannot reflect the true flow characteristics of the system under different operating conditions (such as different flow rates and electrode spacing).

[0031] (iv) Step 4: Real-time intelligent and precise control strategy for equipment current.

[0032] The aforementioned prediction model and the limiting current density calculation method are integrated with the relaxation time calculation method to form the equipment's intelligent computing core. This core sets the target supply current as the limiting current density based on the real-time operating condition prediction and maintains this current setting without updating it during the relaxation time to ensure sufficient steady-state establishment of interface mass transfer. The current controller employs a stepped update or PID control strategy with a hysteresis loop to ensure that the actual supply current approaches the limiting current as closely as possible without exceeding it, thereby maximizing the reaction rate, optimizing current efficiency, minimizing side reactions, and maximizing energy utilization.

[0033] This invention acquires the velocity and concentration fields of the anode-cathode gap region in a reactor under different operating conditions through CFD simulation. Intrinsic orthogonal decomposition (POD) is used to reduce the order of the high-dimensional physical field, generating a small number of modal coefficients for characterization. Furthermore, machine learning techniques are combined to establish a mapping relationship between sensor measurement data and POD coefficients. This allows for rapid prediction and reconstructing of the global velocity and concentration distribution in the anode-cathode gap based on real-time monitoring values ​​from flow rate and concentration sensors during actual operation. Based on the reconstructed physical field, Fick's law and electrochemical theory are used to calculate the reactant flux on the electrode surface, thereby obtaining the limiting current density and the corresponding relaxation time. The controller updates the supply current within the relaxation time scale based on this information, employing a stepped or hysteresis control strategy to ensure the supply current approximates the limiting current as closely as possible without exceeding it. This maximizes the reaction rate, optimizes current efficiency, minimizes side reactions, and maximizes energy utilization.

[0034] Example Taking a hypochlorous acid electrosynthesis reactor as an example, an adaptive power supply control method for the electrolytic reactor based on a digital twin model is constructed to adjust the optimal current of the reactor in real time according to the flow conditions. The reactor consists of a shell and two electrode plates, positive and negative. The reaction zone is the gap between the electrode plates. A hydrochloric acid solution of a certain concentration enters the reaction zone at a certain flow rate. The hydrochloric acid undergoes a chlorine evolution reaction at the anode, thereby producing hypochlorous acid disinfectant. Velocity and concentration sensors are installed at the inlet and outlet of the reaction zone, respectively. The two-dimensional cross-sectional diagram of the specific structure is shown below. Figure 2 As shown.

[0035] The solution density is approximately 1000 kg / m³ 3 The dynamic viscosity is 10. 3 Pa·s, Cl − The species diffusion coefficient D is 2.0 × 10⁻⁶. −9 m 2 ⋅s −1 The Ke turbulence model and the dilute mass transport equation are selected, and the electrode surface is set to concentration constraint. A constant concentration (0.1 g / L) was set for inflow, and four different flow rates (0.1 m / s, 0.5 m / s, 1 m / s, 2 m / s) were calculated. Additionally, hydrochloric acid solution was inflowed at a rate of 1 m / s, and four different concentrations (0.05 g / L, 0.1 g / L, 0.5 g / L, 1 g / L) were calculated, resulting in a total of eight operating conditions. The global velocity field and concentration field were solved using CFD, and the average bulk velocity and concentration at the electrode surface, as well as the chloride ion flux J at the electrode surface, were also calculated, as shown in the table below.

[0036] Data points <![CDATA[U 入口 (m / s)]]> <![CDATA[C 入口 (g / L)]]> <![CDATA[C 入口 (mol / m 3 )]]> <![CDATA[C 出口 (mol / m 3 )]]> <![CDATA[C 平均 (mol / m 3 )]]> <![CDATA[J 平均 (mol / m 2 / s)]]> 1 0.1 0.1 2.742732 2.719261 2.703613 2.75E-05 2 0.5 0.1 2.742732 2.732235 2.725237 1.05E-04 3 1 0.1 2.742732 2.73531 2.730361 1.86E-04 4 2 0.1 2.742732 2.737483 2.733985 3.31E-04 5 1 0.05 1.371366 1.36951 1.368273 9.3E-05 6 1 0.1 2.742732 2.73531 2.730361 1.86E-04 7 1 0.5 13.71366 13.5281 13.4044 9.3E-04 8 1 1 27.42732 26.68509 26.19028 1.86E-03 The reactor geometry parameters and inlet boundary conditions from the above CFD dataset, as well as the velocity and concentration information at fixed sensor locations, are extracted as the input dataset, as shown in the table below.

[0037]

[0038] To reduce the dimensionality of the spatial field, perform intrinsic orthogonal decomposition (POD). Unify the velocity and concentration fields derived from CFD onto the same grid on a selected section. Perform POD on the velocity and concentration fields separately to obtain the previous... dominant mode , and its coefficient , The dataset used as the output features is shown in the table below.

[0039]

[0040] Then, all inputs / outputs are standardized and divided into training / validation / test datasets, with the training set comprising 70%-80% (preferably 70%), the validation set 10%-20% (preferably 15%), and the test set 10%-20% (preferably 15%). A multilayer perceptron (MLP) regression model is selected, with inputs consisting of "sensor measurement points + boundary conditions + geometric parameters" and outputs as "POD coefficient vectors". The mean squared error (RMSE) of the POD coefficients is used as the loss function, and the adaptive gradient descent method is employed to train and optimize the prediction model. Using the velocity and concentration information at the sensor location, as well as the reactor geometric parameters and inlet boundary conditions, as real-time inputs, the machine learning model outputs the POD coefficients. The POD coefficients are then upgraded to three-dimensional spatial information using an online reconstruction formula, and the flow velocity and concentration information on the electrode surface are read, as shown in the table below.

[0041]

[0042] The linear regression plot of the average velocity and concentration on the electrode surface calculated by CFD for all data with the average velocity and concentration reconstructed from the upgraded prediction model is shown below. Figure 3 As shown.

[0043] By establishing a good prediction model and an upgraded reconstruction method, the flow field information on the electrode surface can be obtained quickly in real time through the data information obtained by the sensor, and then the reactant diffusion flux, mass transfer coefficient, limiting current density and relaxation time can be calculated, as shown in the following formula.

[0044] The defining formula is: , , In the formula, This represents the steady-state concentration of the reactants on the electrode surface at a certain current density; it is 0 when the limiting current density is reached. J The mass transfer flux of the reactants at the electrode surface is derived from the wall normal gradient; D is the diffusion coefficient of the reactants, which is 2.0 × 10⁻⁶. −9 m 2 ⋅s −1 .

[0045] The formula for calculating the limiting current density is: , In the formula, This represents the concentration of the reactants in the bulk solution. The number of electrons transferred; This is Faraday's constant, with a value of 96487 C·mol⁻¹. -1 ; is the mass transfer coefficient from the bulk solution to the electrode surface.

[0046] Meanwhile, based on Fick's law, the thickness of the Nernst diffusion layer... Calculate the relaxation time of reactant depletion at the electrode surface The time required to reach a steady-state concentration (approaching 0) is calculated as follows: , For example, if the chloride ion concentration of the solution entering the reactor is 0.1 g / L, the flow rate change is as follows: Figure 4 As shown.

[0047] The calculated reactant flux, limiting current density, and relaxation time are shown in the table below.

[0048]

[0049] At this point, it is only necessary to set the target supply current to this limiting current density and maintain this current setting without updating it during the relaxation time to ensure sufficient steady-state establishment of interfacial mass transfer. The intelligent control strategy for this reactor is then as follows: Figure 5 As shown.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An adaptive power supply control method for an electrolytic reactor based on a digital twin model, characterized in that, Includes the following steps: Step 1: Construct a multi-condition CFD simulation dataset; Step 2: Neural network model construction and spatial information reconstruction; Step 3: Real-time calculation of limiting current density and relaxation time; Step 4: Real-time intelligent and precise control strategy for equipment current.

2. The adaptive power supply control method for an electrolytic reactor based on a digital twin model according to claim 1, characterized in that, In step 1, the steady-state flow field and concentration field of the reactor under different initial boundary conditions are simulated by solving the Navier–Stokes equations and the reactant diffusion equations; the obtained simulation data includes the three-dimensional velocity vector distribution U( x , y , z , t ) and concentration field C ( x , y , z , t Focusing on the boundary layer conditions within the anode-cathode gap region, a large-scale dataset containing the mapping relationship between input parameters and output features is established through batch simulations under different working conditions for training machine learning models.

3. The adaptive power supply control method for an electrolysis reactor based on a digital twin model according to claim 2, characterized in that, The specific steps of step 1 are as follows: Step 1-1: Determine the reactor's geometric parameters and inlet boundary conditions: electrode spacing g Average inlet velocity and inlet reactant concentration ; Steps 1-2: Select the turbulence model and the dilute mass transport equation, and set the electrode surface to concentration constraint. ; Steps 1-3: For each operating condition, calculate the steady-state or quasi-steady-state transient time slices, and derive the three-dimensional velocity vector distribution U( x , y , z , t ) and concentration field C ( x , y , z , t This allows us to obtain a CFD computation dataset, which will be used for subsequent neural network model construction and spatial information reconstruction.

4. The adaptive power supply control method for an electrolytic reactor based on a digital twin model according to claim 1, characterized in that, In step 2, machine learning technology is used to establish a mapping relationship between sensor measurement point data and POD coefficient, so that in actual operation, the global velocity and concentration distribution in the anode-cathode gap can be quickly predicted and reconstructed in order of magnitude based on the real-time monitoring values ​​of the flow rate and concentration sensors.

5. The adaptive power supply control method for an electrolysis reactor based on a digital twin model according to claim 4, characterized in that, The specific steps for step 2 are as follows: Step 2-1: Read the velocity and concentration information at the sensor locations in the CFD dataset for each operating condition, and combine it with the corresponding reactor geometric parameters and inlet boundary conditions as the dataset for input features; Step 2-2: Unify the velocity field and concentration field derived from CFD onto the same mesh on a selected section, and perform POD on the velocity field and concentration field respectively to obtain the previous... The dataset consists of 1 principal modalities and their coefficients, serving as the output features. Steps 2-3: Standardize all inputs / outputs and divide them into training / validation / test datasets, with the training set comprising 70%-80%, the validation set 10%-20%, and the test set 10%-20%. Steps 2-4: Select the multilayer perceptron regression model, with the input being "sensor measurement points + boundary conditions + geometric parameters" and the output being "POD coefficient vector". Use the mean square error of the POD coefficients as the loss function and employ the adaptive gradient descent method to train and optimize the prediction model.

6. The adaptive power supply control method for an electrolytic reactor based on a digital twin model according to claim 1, characterized in that, In step 3, the velocity and concentration information at the sensor location, along with the reactor's geometric parameters and inlet boundary conditions, are used as real-time inputs to obtain the POD coefficient output by the machine learning model. The POD coefficient is then upgraded to three-dimensional spatial information using an online reconstruction formula. Furthermore, the flow rate and concentration information at the electrode surface are read to determine the corresponding mass transfer coefficient, and the limiting current density at the electrode is calculated in real-time. j lim Meanwhile, based on Fick's law, the thickness of the Nernst diffusion layer is... Calculate the relaxation time of the reactants on the electrode surface after they are depleted, i.e., the time required to reach a steady-state concentration.

7. The adaptive power supply control method for an electrolysis reactor based on a digital twin model according to claim 1, characterized in that, In step 4, the target supply current is set to the limit current density based on the real-time operating condition prediction, and the current setting is kept unchanged during the relaxation time to ensure sufficient steady-state establishment of interface mass transfer. The current controller adopts a stepped update or PID control strategy with hysteresis to make the actual supply current as close as possible to the limit current without exceeding the limit current.

Images