Alkaline electrolytic cell local gas content enrichment risk assessment method, system, medium and equipment

Abstract

The invention provides an alkaline electrolytic cell local gas content enrichment risk assessment method and system, a medium and equipment. The method comprises the steps that S1, a CFD model suitable for alkaline electrolytic cell simulation is constructed; s2, on the basis of the constructed CFD model, constructing an electrolytic cell agent model suitable for predicting the voltage state and the gas content distribution of the alkaline electrolytic cell; s3, correcting undetermined parameters of the electrolytic cell agent model to obtain a corrected electrolytic cell agent model; s4, predicting the spatial distribution characteristics of the gas holdup by using the corrected agent model of the electrolytic cell; and S5, evaluating the local gas content enrichment risk based on the predicted gas content spatial distribution characteristics.

Description

Technical Field

[0001] The present invention relates to the technical field of local gas holdup enrichment risk assessment. Specifically, it relates to a method, system, medium and device for assessing the local gas holdup enrichment risk of an alkaline electrolyzer. More specifically, it relates to a method, system, medium and device for assessing the local gas holdup enrichment risk of an alkaline electrolyzer based on an artificial neural network. Background Art

[0002] Alkaline electrolyzers have broad application prospects. In the context of renewable energy, it is necessary to further improve the performance and lifespan of electrolyzers. Modeling and simulation analysis is an important means to improve the design and operation plan of electrolyzers. By establishing an appropriate model and evaluating the operating conditions of electrolyzers under different conditions, it not only helps to improve the system energy efficiency but also provides a theoretical basis for fault diagnosis, lifespan prediction, and operation and maintenance management.

[0003] Patent document CN117252032B (application number: 202311490394.9) discloses a method for constructing a digital twin of an alkaline electrolytic water hydrogen production system, including the following steps: Step 1, obtain the system parameters of the actual alkaline electrolytic water hydrogen production system and the target performance of the digital twin; Step 2, establish multiple simulation models according to the system parameters; Step 3, obtain an initial digital twin based on the simulation models; Step 4, perform data docking between the initial digital twin and the actual alkaline electrolytic water hydrogen production system, test and adjust the initial digital twin until the initial digital twin meets the target performance; use the obtained initial digital twin that meets the target performance as the digital twin of the actual alkaline electrolytic water hydrogen production system. Although this model in this patent integrates the comprehensive effects of multiple factors such as electricity, chemistry, and environment, it cannot characterize the spatial distribution characteristics of multiple physical fields inside the electrolyzer and is difficult to analyze and evaluate problems such as local excessive enrichment of gas holdup.

[0004] CFD models are suitable for analyzing the interaction between multiple physical fields and have high spatial resolution of calculation results, and have been widely cited in the field of alkaline electrolyzer design and operation. However, the CFD solution speed is slow and often requires a large amount of computing resources and time, which limits its fast response ability in practical engineering applications. Therefore, how to improve the computational efficiency of the model while maintaining high accuracy has become a key issue in the design, operation, and management of electrolyzers. Summary of the Invention

[0005] Aiming at the deficiencies in the prior art, the purpose of the present invention is to provide a method, system, medium and device for assessing the local gas holdup enrichment risk of an alkaline electrolyzer.

[0006] According to a method for assessing the local gas holdup enrichment risk of an alkaline electrolyzer provided by the present invention, it includes:

[0007] Step S1: Construct a CFD model applicable to the simulation of alkaline electrolyzers;

[0008] Step S2: Based on the constructed CFD model, construct an electrolyzer surrogate model applicable to predicting the voltage state and gas holdup distribution of alkaline electrolyzers;

[0009] Step S3: Modify the undetermined parameters of the electrolyzer surrogate model to obtain a modified electrolyzer surrogate model;

[0010] Step S4: Use the modified electrolyzer surrogate model to predict the spatial distribution characteristics of gas holdup;

[0011] Step S5: Evaluate the risk of local gas holdup enrichment based on the predicted spatial distribution characteristics of gas holdup.

[0012] Preferably, the said Step S2 includes:

[0013] Step S2.1: Considering comprehensively the changes in the undetermined parameters and operating parameters of the CFD model for the simulation of alkaline electrolyzers, randomly generate a series of working parameter combinations within the feasible region to form a set W; for each generated working parameter w i , apply the CFD model to calculate the corresponding voltage state y volt,i and multi-physical field distribution y field,i , to form a voltage state set Y volt and a multi-physical field set Y field ;

[0014] Step S2.2: Construct a voltage prediction network f w→volt , use the working parameter set W as the input data set and the voltage state set Y volt as the output data set, train the voltage prediction network f w→volt to obtain a trained voltage prediction network f w ′ →volt ; the trained voltage prediction network f w ′ →volt is used to predict the voltage state under specific working parameters w i ;

[0015] y volt,net,i = f′ w→volt (w i )

[0016] Step S2.3: Construct a multi-physical field prediction network f w→field , use the working parameter set W as the input data set and the multi-physical field Y field as the output data set, train the multi-physical field prediction network f w→field to obtain a trained multi-physical field prediction network f′w→field ; The trained multi - physical - field prediction network f′ w→field is used to predict the distribution of physical quantities at a position in the space under specific working parameters w i ; ;

[0017]

[0018] During the prediction process, the multi - physical - field y field,net,i,j is characterized as a linear combination of z basis functions. The basis - function network h z takes the spatial coordinates as the input and characterizes the spatial distribution characteristics of the basis functions. The weight network g z takes w i as the input and characterizes the influence of the working parameters on the weights of each basis function.

[0019] Preferably, the step S3 includes:

[0020] Step S3.1: Change the current density, design M sets of working conditions to form a set G. For each working condition g i , measure the voltage state y volt,exp,i corresponding to the working condition through experiments, and use it as the target for CFD model parameter identification;

[0021] Step S3.2: Take the electrode reaction kinetic parameters and the diaphragm conductivity as undetermined parameters, and randomly generate N sets of undetermined parameters to form a set P; for each set of generated undetermined parameters p j , apply the CFD model to calculate the voltage states y volt,CFD,i,j under a series of working conditions to form a set Y p→volt ;

[0022] Step S3.3: Construct a parameter - estimation network f p→volt , use the set P of undetermined parameters as the input data set, and the set Y of voltage states p→volt as the output data set, and train the parameter - estimation network f p→volt to obtain the trained parameter - estimation network f′ p→volt ; The trained parameter - estimation network f′ p→volt is used to predict the voltage state of the alkaline electrolyzer under the combination of a specific working condition s i and parameters p j ;

[0023] y volt,net,i,j = f p→volt (s i , p j )

[0024] Step S3.4: For each set of undetermined parameters p j, compare the corresponding CFD calculation results with the experimental data to obtain the current optimal combination p of undetermined parameters cur ,

[0025]

[0026] Step S3.5: Apply the improved Levenberg-Marquardt method to iteratively update the optimal combination of undetermined parameters;

[0027] p next = p cur + Δp

[0028] where Δp is calculated by the following formula

[0029] Δp = -(J T J + μI) -1 J T r(p cur,best )

[0030] where J is the Jacobi matrix;

[0031]

[0032] Based on the constructed parameter estimation network f p→volt , using the backpropagation property of the gradient in the neural network, obtain the derivatives of the predicted voltage with respect to each undetermined parameter under each working condition, and approximately replace the direct calculation results of the CFD model, that is, let

[0033]

[0034] Step S3.6: For the updated optimal combination p of undetermined parameters next , apply the CFD model to calculate the working voltage y i under each working condition g volt,CFD,i,next ; Add p next and y volt,CFD,i,next to the data sets P and Y p→volt ; Repeat steps S3.3 to S3.6 until the error is less than the specified value.

[0035] Preferably, the said step S4 includes:

[0036] Step S4.1: Take the obtained undetermined parameters and the operating parameters measured by the sensors to form the working parameter w, and predict the gas holdup spatial distribution characteristics through the trained multi-physics prediction network f' w→field ; Wherein, the operating parameters measured by the sensors include temperature, pressure and flow rate;

[0037] Step S4.2: Perform risk zoning on the preset cross-section according to the gas holdup distribution. Among them, the area where the gas holdup is in the range of 0% - 40% is classified as a low-risk area, the area where the gas holdup is in the range of 40% - 70% is classified as a medium-risk area, and the area where the gas holdup is in the range of 70% - 100% is classified as a high-risk area; count the proportion of different areas for the preset cross-section.

[0038] Step S4.3: Make a warning judgment based on the proportion of the area. If the area with medium risk and above exceeds 10%, or the area with high risk and above exceeds 3%, a warning message is issued.

[0039] Among them, the preset interface includes the diaphragm surface, the electrode surface, and the center position of the flow channel.

[0040] According to an alkaline electrolyzer local gas holdup enrichment risk assessment system provided by the present invention, it includes:

[0041] Module M1: Construct a CFD model suitable for alkaline electrolyzer simulation;

[0042] Module M2: Based on the constructed CFD model, construct an electrolyzer surrogate model suitable for predicting the voltage state and gas holdup distribution of the alkaline electrolyzer;

[0043] Module M3: Modify the undetermined parameters of the electrolyzer surrogate model to obtain a modified electrolyzer surrogate model;

[0044] Module M4: Use the modified electrolyzer surrogate model to predict the gas holdup spatial distribution characteristics;

[0045] Module M5: Evaluate the local gas holdup enrichment risk based on the predicted gas holdup spatial distribution characteristics.

[0046] Preferably, the Module M2 includes:

[0047] Module M2.1: Considering the changes in the undetermined parameters and operating parameters of the CFD model for alkaline electrolyzer simulation comprehensively, randomly generate a series of working parameter combinations within the feasible domain to form a set W; for each generated working parameter w i , apply the CFD model to calculate the corresponding voltage state y volt,i and the multi-physical field distribution y field,i , to form a voltage state set Y volt and a multi-physical field set Y field ;

[0048] Module M2.2: Construct a voltage prediction network f w→volt , use the working parameter set W as the input data set, and the voltage state set Y volt as the output data set, for the voltage prediction network f w→voltTrain to obtain the trained voltage prediction network f w ′ →volt ; The trained voltage prediction network f w ′ →volt is used to predict the voltage state at specific working parameters w i ;

[0049] y volt,net,i = f′ w→volt (w i )

[0050] Module M2.3: Construct a multi-physical field prediction network f w→field , using the set of working parameters W as the input data set and the multi-physical field Y field as the output data set, train the multi-physical field prediction network f w→field to obtain the trained multi-physical field prediction network f′ w→field ; The trained multi-physical field prediction network f′ w→field is used to predict the distribution of physical quantities at a position in the space at specific working parameters w i ; During the prediction process, the multi-physical field y

[0051]

[0052] is characterized as a linear combination of z basis functions. The basis function network h field,net,i,j takes the spatial coordinates z as the input to characterize the spatial distribution characteristics of the basis functions. The weight network g takes w z as the input to characterize the influence of the working parameters on the weights of each basis function. i ;

[0053] Preferably, the module M3 includes:

[0054] Module M3.1: Change the current density, design M sets of working conditions to form a set G, for each working condition g i , measure the corresponding voltage state y volt,exp,i under the working condition through experiments, and use it as the target for CFD model parameter identification;

[0055] Module M3.2: Use the electrode reaction kinetic parameters and the diaphragm conductivity as undetermined parameters, randomly generate N sets of undetermined parameters to form a set P; for each set of generated undetermined parameters p j , apply the CFD model to calculate the voltage states y volt,CFD,i,j under a series of working conditions to form a set Y p→volt ;

[0056] Module M3.3: Construct a parameter estimation network fp→volt , taking the set of undetermined parameters P as the input data set and the set of voltage states Y p→volt as the output data set, training the parameter estimation network f p→volt to obtain the trained parameter estimation network f' p→volt ; The trained parameter estimation network f' p→volt is used to predict the voltage state of the alkaline electrolyzer under the combination of a specific working condition s i and the parameter p j ;

[0057] y volt,net,i,j = f p→volt (s i , p j )

[0058] Module M3.4: For each set of undetermined parameters p j , compare its corresponding CFD calculation result with the experimental data to obtain the current optimal combination of undetermined parameters p cur ,

[0059]

[0060] Module M3.5: Apply the improved Levenberg-Marquardt method to iteratively update the optimal combination of undetermined parameters;

[0061] p next = p cur + Δp

[0062] where Δp is calculated by the following formula

[0063] Δp = -(J T J + μI) -1 J T r(p cur,best )

[0064] where J is the Jacobi matrix;

[0065]

[0066] Based on the constructed parameter estimation network f p→volt , using the backpropagation property of the gradient in the neural network, obtain the derivative of the predicted voltage with respect to each undetermined parameter under each working condition, approximately substituting the direct calculation result of the CFD model, that is, let

[0067]

[0068] Module M3.6: For the updated optimal combination of undetermined parameters p next , apply the CFD model to calculate each working condition g iThe working voltage y below volt,CFD,i,next ; Incorporate p next and y volt,CFD,i,next into the data sets P and Y p→volt ; Repeat from module M3.3 to module M3.6 until the error is less than the specified value.

[0069] Preferably, the module M4 includes:

[0070] Module M4.1: Combine the obtained undetermined parameters and the operating parameters measured by the sensor to form the working parameter w, and predict the gas holdup spatial distribution characteristics through the trained multi-physical field prediction network f′ w→field ; Among them, the operating parameters measured by the sensor include temperature, pressure, and flow rate;

[0071] Module M4.2: Perform risk zoning on the preset cross-section according to the gas holdup distribution. Among them, the area with a gas holdup of 0% - 40% is divided into a low-risk area, the area with a gas holdup of 40% - 70% is divided into a medium-risk area, and the area with a gas holdup of 70% - 100% is divided into a high-risk area; Count the proportion of different areas for the preset cross-section.

[0072] Module M4.3: Make a warning judgment based on the proportion of the area. If the area with medium risk or above exceeds 10%, or the area with high risk or above exceeds 3%, then send a warning message.

[0073] Among them, the preset interface includes the diaphragm surface, the electrode surface, and the center position of the flow channel.

[0074] According to an electronic device provided by the present invention, the electronic device includes a memory and at least one processor, and instructions are stored in the memory;

[0075] The at least one processor calls the instructions in the memory so that the electronic device executes each step of the above-mentioned method for risk assessment of local gas holdup enrichment in an alkaline electrolyzer.

[0076] According to a computer-readable storage medium provided by the present invention, instructions are stored on the computer-readable storage medium, and when the instructions are executed by a processor, each step of the above-mentioned method for risk assessment of local gas holdup enrichment in an alkaline electrolyzer is implemented.

[0077] Compared with the prior art, the present invention has the following beneficial effects:

[0078] 1. By combining advanced artificial neural network technology with computational fluid dynamics (CFD) methods, the present invention successfully solves the technical problem of the difficult real-time and accurate assessment of the spatial distribution of gas holdup in alkaline electrolyzers. Due to the limitations of complex hydrodynamic behaviors and multiphase flow characteristics inside the electrolyzers, traditional methods often struggle to accurately capture the dynamic changes in gas phase distribution, resulting in large errors in assessment results. However, by introducing an artificial neural network, the present invention can efficiently process and learn a large amount of data generated by CFD simulations, thereby achieving real-time and accurate prediction of the gas holdup distribution in the electrolyzer.

[0079] 2. The present invention uses CFD simulations to obtain detailed information on the internal flow field of the electrolyzer, including the flow, mixing, and distribution of gas and liquid. Then, through deep learning and optimization of these data by an artificial neural network, the real-time calculation of the degree of local gas phase enrichment is finally achieved. It can not only accurately reflect the dynamic changes inside the electrolyzer but also promptly identify potential safety risks, such as the attenuation of electrolyzer performance, the excessive hydrogen content in oxygen, and the accelerated aging of equipment. Through early warning of these risks, operators can take corresponding measures to avoid accidents, thereby significantly improving the safety and operational stability of the electrolyzer.

[0080] 3. The application of the present invention also effectively extends the service life of the equipment. By real-time monitoring and optimizing the operating state of the electrolyzer, problems such as local overheating and corrosion caused by gas phase enrichment are reduced, and the loss rate of the equipment is lowered. This not only reduces maintenance costs but also improves the overall operating efficiency of the electrolyzer, bringing significant economic benefits to industrial production. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Other features, objects, and advantages of the present invention will become more apparent by reading the detailed description of non-limiting embodiments with reference to the following drawings:

[0082] Figure 1 It is a hardware architecture diagram of the local gas holdup enrichment warning system for alkaline electrolyzers.

[0083] Figure 2 It is a software flow chart of the local gas holdup enrichment warning for alkaline electrolyzers.

[0084] Figures 3a to 3b It is a cloud map of the spatial distribution of gas holdup at the key cross-section of the alkaline electrolyzer.

[0085] Figures 4a to 4b It is a schematic diagram of the risk assessment results of the local gas holdup enrichment degree of the alkaline electrolyzer. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0086] The present invention will be described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several changes and improvements can still be made. These all belong to the protection scope of the present invention.

[0087] Embodiment 1

[0088] A method for risk assessment of local gas holdup enrichment in an alkaline electrolyzer based on an artificial neural network provided by the present invention includes:

[0089] Step S1: Construct a CFD model suitable for alkaline electrolyzer simulation;

[0090] Step S2: Construct a surrogate model suitable for predicting the voltage state and gas holdup distribution of the alkaline electrolyzer;

[0091] Step S3: Construct a parameter correction module suitable for online correction of undetermined parameters of the surrogate model;

[0092] Step S4: Construct a safety assessment module suitable for evaluating the degree of local gas phase enrichment and potential safety risks.

[0093] Specifically, the step S2 adopts:

[0094] Step S2.1: Considering comprehensively the changes of undetermined parameters and operating parameters of the alkaline electrolyzer CFD model, randomly generate a series of working parameter combinations within the feasible region to form a set W; for each generated working parameter w i , apply the CFD model to calculate the corresponding voltage state y volt,i and multi-physical field distribution y field,i , to form a voltage state set Y volt and a multi-physical field set Y field ;

[0095] Step S2.2: Construct a voltage prediction network f w→volt , take the working parameter set W as the input data set, and the voltage state set Y volt as the output data set, and train the network. The trained f w→volt network is suitable for predicting the voltage state under specific working parameters w i ,

[0096] y volt,net,i = f w→volt (w i )

[0097] Step S2.3: Construct a multi-physical field prediction network f w→field , take the working parameter set W as the input data set, and the multi-physical field Yfield Using the output data set, the network is trained to predict w under specific working parameters i in the space of the distribution of physical quantities at the position of

[0098]

[0099] In the formula, the sub-networks g z and h z are used to preprocess w i and respectively, where z is the output dimension of the sub-network, representing the spatial coordinates corresponding to the physical field;

[0100] Specifically, step S3 adopts:

[0101] Step S3.1: Change the current density, design M sets of working conditions to form a set G, for each working condition g i , measure the voltage state y volt,exp,i corresponding to the working condition through experiments, as the target for CFD model parameter identification;

[0102] For the full range of working conditions, the objective function is:

[0103]

[0104] Step S3.2: Using the electrode reaction kinetic parameters and the membrane conductivity as undetermined parameters, randomly generate N sets of undetermined parameters to form a set P; for each set of generated undetermined parameters p j , apply the CFD model to calculate the voltage state y volt,CFD,i,j in the series of working conditions to form a set Y p→volt ;

[0105] Step S3.3: Construct a parameter estimation network f p→volt , using the set P of undetermined parameters as the input data set and the set Y of voltage states p→volt as the output data set, train the network to predict the voltage state of the alkaline electrolyzer under the combination of specific working conditions s i and parameters p j ,

[0106] y volt,net,i,j = f p→volt (s i , p j )

[0107] Step S3.4: For each set of undetermined parameters p j , compare its corresponding CFD calculation results with the experimental data to obtain the current optimal combination of undetermined parameters p cur ,

[0108]

[0109] Step S3.5: Apply the improved Levenberg - Marquardt method to iteratively update the optimal combination of undetermined parameters.

[0110] p next = p cur + Δp

[0111] where Δp is calculated by the following formula

[0112] Δp = -(H T H + μI) -1 J T r(p cur,best )

[0113] where J is the Jacobi matrix.

[0114]

[0115] In the traditional Levenberg - Marquardt algorithm, the derivatives of each term in the matrix are generally obtained by direct calculation based on the model. However, the CFD model of the alkaline electrolyzer is complex, and it is difficult to directly calculate the derivatives of the calculated voltage under each working condition with respect to each undetermined parameter based on the model. Therefore, based on the constructed parameter estimation network f p→volt , using the backpropagation property of the gradient in the neural network, the derivatives of the predicted voltage under each working condition with respect to each undetermined parameter are obtained, approximately substituting the direct calculation results of the CFD model, that is, let

[0116]

[0117] Step S3.6: For the updated optimal combination of undetermined parameters p next , apply the CFD model to calculate the working voltage y i under each working condition g volt,CFD,i,next ; Add p next and y volt,cFD,i,next to the data sets P and Y p→volt ; Repeat steps S3.3 to S3.6 until the error is less than the specified value.

[0118] Specifically, step S4 adopts:

[0119] Step S4.1: Take the undetermined parameters obtained in step S3 and the operating parameters (temperature, pressure, flow rate) measured by the sensors to form the working parameter w, and predict the spatial distribution characteristics of the gas holdup through the multi - physical - field prediction network f w→field obtained in step S2.

[0120] Step S4.2: Considering that the electrodes and the diaphragm are key components of the alkaline electrolyzer, the flow of the alkaline solution and its wetting of the electrodes and the diaphragm are crucial for the long-term stable operation of the alkaline electrolyzer. Preferably, the surface of the diaphragm, the surface of the electrodes, and the center position of the flow channel are taken as the key cross-sections;

[0121] Step S4.3: Perform risk zoning on the key cross-sections according to the gas holdup distribution. Among them, the area where the gas holdup is in the range of 0% - 40% is classified as a low-risk area, the area where the gas holdup is in the range of 40% - 70% is classified as a medium-risk area, and the area where the gas holdup is in the range of 70% - 100% is classified as a high-risk area; For each key cross-section, count the proportion of different areas.

[0122] Step S4.4: Make a warning judgment based on the proportion of the area. If the area with medium risk or above exceeds 10%, or the area with high risk or above exceeds 3%, a warning message is issued.

[0123] The present invention also provides a risk assessment system for local gas holdup enrichment in an alkaline electrolyzer based on an artificial neural network. The risk assessment system for local gas holdup enrichment in an alkaline electrolyzer based on an artificial neural network can be implemented by executing the process steps of the risk assessment method for local gas holdup enrichment in an alkaline electrolyzer based on an artificial neural network. That is, those skilled in the art can understand the risk assessment method for local gas holdup enrichment in an alkaline electrolyzer based on an artificial neural network as a preferred implementation manner of the risk assessment system for local gas holdup enrichment in an alkaline electrolyzer based on an artificial neural network.

[0124] Example 2

[0125] Example 2 is a preferred example of Example 1

[0126] The following provides a specific implementation process using the above method, as follows:

[0127] In this embodiment, facing the renewable energy scenario, the safety of the local gas holdup enrichment degree of the alkaline electrolyzer under typical working conditions is evaluated. According to the method and device provided by the present invention patent, the implementation process includes three aspects: system deployment, system operation, and emergency management.

[0128] In terms of system deployment, the deployment of system software and hardware is completed. In terms of software, first, draw a geometric model according to the structure of the alkaline electrolyzer, as shown in Table x; Secondly, establish a CFD simulation model according to the physical and chemical processes in the alkaline electrolyzer, as shown in Table 1; Then, generate batch data based on the constructed CFD model for the training of the surrogate model of the alkaline electrolyzer to form a surrogate model module; Finally, couple the surrogate model module, the parameter correction module, and the evaluation and warning module to form software and arrange it in the industrial control computer. In terms of hardware, connect the industrial control computer with the alkaline electrolyzer, the alarm, and the display to form a system architecture with data and instruction transmission functions, such asFigure 1 as shown

[0129] Table 1 Main Structural Parameters of Alkaline Electrolyzer

[0130]

[0131] Table 2 Conservation Equations of CFD Model for Alkaline Electrolyzer

[0132]

[0133] In terms of system operation, the software and hardware of the system cooperate. The system process is as Figure 2 shown. First, the sensors of the alkaline electrolyzer transmit the necessary operating parameters to the industrial control computer. Two typical operating conditions during the operation of the alkaline electrolyzer are shown in Table 2. Second, the parameter correction module of the industrial control computer calls the proxy model module to predict the voltage based on the received data and compares it with the measured voltage, and online corrects the proxy model according to the comparison result; then, it calls the proxy model module to predict the gas holdup distribution characteristics at key cross-sections, such as Figures 3a to 3b ; finally, it calls the evaluation and early warning module to evaluate each key cross-section, and the evaluation results are as Figures 4a to 4b shown. For operating condition 1, the proportions of low-risk areas at the three key cross-sections of the diaphragm surface, electrode surface, and flow channel center are 98.97%, 99.12%, and 99.19% respectively, and the proportions of medium-risk and high-risk areas are both less than 1%. For operating condition 2, the proportions of medium-risk areas at the three cross-sections are 32.62%, 35.2%, and 35.29%, and the proportions of high-risk areas are 48%, 41.66%, and 41.55% respectively; finally, the industrial control computer transmits the evaluation results to the display and judges whether to issue an alarm. In operating condition 1, most areas of each key cross-section are in the low-risk area, and the proportions of the medium-risk area and high-risk area do not exceed the set threshold, so no alarm is issued. In operating condition 2, the proportions of areas above medium-risk and high-risk areas at each key cross-section both exceed the set threshold, so an alarm is issued.

[0134] Table 3 Typical Chamber Operating Conditions

[0135]

[0136] In terms of emergency management, the operation and maintenance personnel take corresponding measures according to the system evaluation results and alarm status. For operating condition 1, since the early warning system does not issue an alarm, no operation is required; for operating condition 2, the early warning system issues an alarm, and measures such as reducing the load or increasing the flow rate need to be taken to prevent problems such as attenuation of electrolyzer performance, excessive hydrogen content in oxygen, and accelerated aging caused by excessive local gas holdup.

[0137] Those skilled in the art know that, in addition to implementing the systems, devices and their respective modules provided by the present invention in the form of pure computer-readable program codes, it is entirely possible to make the systems, devices and their respective modules provided by the present invention be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. by logically programming the method steps. Therefore, the systems, devices and their respective modules provided by the present invention can be regarded as a kind of hardware components, and the modules included therein for implementing various programs can also be regarded as the structures within the hardware components; the modules for implementing various functions can also be regarded as either software programs for implementing the methods or the structures within the hardware components.

[0138] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. Without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A method for risk assessment of local gas holdup enrichment in an alkaline electrolyzer, characterized in that, Including: Step S1: Construct a CFD model suitable for alkaline electrolyzer simulation; Step S2: Based on the constructed CFD model, construct an electrolyzer surrogate model suitable for predicting the voltage state and gas holdup distribution of the alkaline electrolyzer; Step S3: Correct the undetermined parameters of the electrolyzer surrogate model to obtain a corrected electrolyzer surrogate model; Step S4: Use the corrected electrolyzer surrogate model to predict the spatial distribution characteristics of gas holdup; Step S5: Evaluate the risk of local gas holdup enrichment based on the predicted spatial distribution characteristics of gas holdup.

2. The method for evaluating the risk of local gas holdup enrichment in an alkaline electrolyzer according to claim 1, wherein The said Step S2 includes: Step S2.1: Considering comprehensively the undetermined parameters of the CFD model for alkaline electrolyzer simulation and the variations of operating parameters, a series of combinations of working parameters are randomly generated within the feasible region to form a set W; for each generated working parameter w i , the corresponding voltage state y volt,i and multi-physical field distribution y field,i are calculated using the CFD model, forming a voltage state set Y volt and a multi-physical field set Y field ; Step S2.2: Construct the voltage prediction network f w→volt , using the set of working parameters W as the input data set and the set of voltage states Y volt as the output data set, train the voltage prediction network f w→volt to obtain the trained voltage prediction network f w ′ →volt ; the trained voltage prediction network f w ′ →volt is used to predict the voltage state under specific working parameters w i ; y volt,net,i = f' w→volt (w i ) Step S2.3: Construct the multi-physical field prediction network f w→field , using the set of working parameters W as the input data set and the multi-physical field Y field as the output data set, train the multi-physical field prediction network f w→field to obtain the trained multi-physical field prediction network f′ w→field ; the trained multi-physical field prediction network f′ w→field is used to predict the distribution of physical quantities at positions in the space under specific working parameters w i ; ​ During the prediction process, the multi-physical field y field,net,i,j is represented as a linear combination of z basis functions, and the basis function network h z takes the spatial coordinates as the input to characterize the spatial distribution characteristics of the basis functions, and the weight network g z takes w i as the input to characterize the influence of the working parameters on the weights of the basis functions.

3. The method for evaluating the risk of local gas holdup enrichment in an alkaline electrolyzer according to claim 1, wherein The said Step S3 includes: Step S3.1: Change the current density, design M sets of working conditions to form a set G, and for each working condition g i , measure the corresponding voltage state y volt,exp,i through experiments, and use it as the target for the parameter identification of the CFD model; Step S3.2: Using the electrode reaction kinetic parameters and the membrane conductivity as undetermined parameters, randomly generate N sets of undetermined parameters to form a set P; for each generated set of undetermined parameters p j , apply the CFD model to calculate the voltage states y volt,CFD,i,j to form a set Y p→volt ; Step S3.3: Construct the parameter estimation network f p→volt , using the set of undetermined parameters P as the input data set and the set of voltage states Y p→volt as the output data set, train the parameter estimation network f p→volt to obtain the trained parameter estimation network f' p→volt ; The trained parameter estimation network f' p→volt is used to predict the voltage state of the alkaline electrolyzer under the combination of a specific working condition s i and the parameter p j . y volt,net,i,j = f p→volt (s i , p j ) Step S3.4: For each group of undetermined parameters p j , compare its corresponding CFD calculation result with the experimental data to obtain the current optimal combination of undetermined parameters p cur , Step S3.5: Apply the improved Levenberg-Marquardt method to iteratively update the optimal combination of undetermined parameters; p next = p cur + Δp wherein, Δp is calculated by the following formula, Δp = -(J T J + μI) -1 J T r(p cur,best ) wherein, J is the Jacobi matrix; Based on the constructed parameter estimation network f p→volt , using the backpropagation characteristic of the gradient in the neural network, the derivatives of the predicted voltage under various working conditions with respect to each undetermined parameter are obtained, approximately replacing the direct calculation results of the CFD model, that is, let Step S3.6: For the updated optimal pending parameter combination p next , use the CFD model to calculate the operating voltage y i under each working condition g volt,CFD,i,next ; Add p next and y volt,CFD,i,next to the data sets P and Y p→volt ; Repeat steps S3.3 to S3.6 until the error is less than the specified value.

4. The method for evaluating the risk of local gas holdup enrichment in an alkaline electrolyzer according to claim 1, wherein The said Step S4 includes: Step S4.1: Take the obtained undetermined parameters and the operating parameters measured by the sensor to form the working parameter w, and predict the spatial distribution characteristics of the gas holdup through the trained multi-physical field prediction network f'; where the operating parameters measured by the sensor include temperature, pressure, and flow rate; w→field ​ Step S4.2: Conduct risk zoning for the preset cross-section according to the gas holdup distribution. The area where the gas holdup is in the range of 0% - 40% is classified as a low-risk area, the area where the gas holdup is in the range of 40% - 70% is classified as a medium-risk area, and the area where the gas holdup is in the range of 70% - 100% is classified as a high-risk area; count the proportion of different areas for the preset cross-section. Step S4.3: Conduct early warning judgment based on the proportion of areas. If the medium-risk and above areas exceed 10%, or the high-risk and above areas exceed 3%, an early warning message is issued. wherein, the said preset interface includes the diaphragm surface, the electrode surface, and the flow channel center position.

5. A risk assessment system for local gas holdup enrichment in an alkaline electrolyzer, characterized in that, Including: Module M1: Construct a CFD model suitable for alkaline electrolyzer simulation; Module M2: Based on the constructed CFD model, construct an electrolyzer surrogate model suitable for predicting the voltage state and gas holdup distribution of the alkaline electrolyzer; Module M3: Correct the undetermined parameters of the electrolyzer surrogate model to obtain a corrected electrolyzer surrogate model; Module M4: Use the corrected electrolyzer surrogate model to predict the spatial distribution characteristics of gas holdup; Module M5: Evaluate the risk of local gas holdup enrichment based on the predicted spatial distribution characteristics of gas holdup.

6. The risk assessment system for local gas holdup enrichment in an alkaline electrolyzer according to claim 5, wherein The said Module M2 includes: Module M2.1: Considering the changes in the undetermined parameters and operating parameters of the CFD model for alkaline electrolyzer simulation, a series of working parameter combinations are randomly generated within the feasible region to form a set W; for each generated working parameter w i , the corresponding voltage state y volt,i and multi-physical field distribution y field,i are calculated using the CFD model, forming a voltage state set Y volt and a multi-physical field set Y field ; Module M2.2: Construct the voltage prediction network f w→volt , using the set of working parameters W as the input data set and the set of voltage states Y volt as the output data set, train the voltage prediction network f w→volt to obtain the trained voltage prediction network f w ′ →volt ; the trained voltage prediction network f w ′ →volt is used to predict the voltage state under specific working parameters w i ; y volt,net,i = f' w→volt (w i ) Module M2.3: Constructing the multi-physics prediction network f w→field , taking the set of working parameters W as the input data set, and the multi-physics Y field as the output data set, training the multi-physics prediction network f w→field to obtain the trained multi-physics prediction network f' w→field ; the trained multi-physics prediction network f' w→field is used to predict the distribution of physical quantities at the position in the space of a specific working parameter w i ; ​ During the prediction process, the multi-physical field y field,net,i,j is represented as a linear combination of z basis functions, and the basis function network h z takes the spatial coordinates as the input to represent the spatial distribution characteristics of the basis functions, and the weight network g z takes w i as the input to represent the influence of the working parameters on the weights of the basis functions.

7. The local gas holdup enrichment risk assessment system for an alkaline electrolyzer according to claim 5, wherein The said Module M3 includes: Module M3.1: Change the current density, design M sets of working conditions to form a set G, for each working condition g i , measure the corresponding voltage state y through experiments volt,exp,i , which is used as the target for parameter identification of the CFD model; Module M3.2: Using the electrode reaction kinetic parameters and the diaphragm conductivity as undetermined parameters, randomly generate N sets of undetermined parameters to form a set P; for each generated set of undetermined parameters p j , apply the CFD model to calculate the voltage state y under a series of working conditions volt,CFD,i,j to form a set Y p→volt ; Module M3.3: Construct the parameter estimation network f p→volt , using the set of undetermined parameters P as the input data set and the set of voltage states Y p→volt as the output data set, train the parameter estimation network f p→volt to obtain the trained parameter estimation network f' p→volt ; the trained parameter estimation network f' p→volt is used to predict the voltage state of the alkaline electrolyzer under the combination of a specific working condition s i and the parameter p j ; y volt,net,i,j = f p→vo1t (s i , p j ) Module M3.4: For each set of undetermined parameters p j , compare the corresponding CFD calculation results with the experimental data to obtain the current optimal combination of undetermined parameters p cur , Module M3.5: Apply the improved Levenberg-Marquardt method to iteratively update the optimal combination of undetermined parameters; p next = p cur + Δp wherein, Δp is calculated by the following formula, Δp = -(J T J + μI) -1 J T r(p cur,best ) wherein, J is the Jacobi matrix; Based on the constructed parameter estimation network f p→volt , using the backpropagation property of gradients in the neural network, the derivatives of the predicted voltage with respect to each undetermined parameter under various operating conditions are obtained, approximately substituting the direct calculation results of the CFD model, that is, let Module M3.6: Update the optimal undetermined parameter combination p next , using CFD model to calculate each working condition g i Working voltage y volt,CFD,i,next ; next and y volt,CFD,i,next Join datasets P and Y p→volt ; Repeat module M3.3 to module M3.6 until the error is less than the specified value.

8. The risk assessment system for local gas holdup enrichment of an alkaline electrolyzer according to claim 5, characterized in that, The said Module M4 includes: Module M4.1: The obtained undetermined parameters and the operating parameters measured by the sensors are combined to form the working parameter w, and the multi-physical field prediction network f′ after training is used w→field to predict the spatial distribution characteristics of the gas holdup; wherein, the operating parameters measured by the sensors include temperature, pressure, and flow rate; Module M4.2: Conduct risk zoning for the preset cross-section according to the gas holdup distribution. The area where the gas holdup is in the range of 0% - 40% is classified as a low-risk area, the area where the gas holdup is in the range of 40% - 70% is classified as a medium-risk area, and the area where the gas holdup is in the range of 70% - 100% is classified as a high-risk area; count the proportion of different areas for the preset cross-section. Module M4.3: Conduct early warning judgment based on the proportion of areas. If the medium-risk and above areas exceed 10%, or the high-risk and above areas exceed 3%, an early warning message is issued. wherein, the said preset interface includes the diaphragm surface, the electrode surface, and the flow channel center position.

9. An electronic device, the electronic device includes a memory and at least one processor, and instructions are stored in the memory; The at least one processor calls the instructions in the memory to cause the electronic device to execute each step of the method for evaluating the risk of local gas holdup enrichment in an alkaline electrolyzer as described in any one of claims 1-4.

10. A computer-readable storage medium, on which instructions are stored, characterized in that, When the instructions are executed by the processor, each step of the method for evaluating the risk of local gas holdup enrichment in an alkaline electrolyzer as described in any one of claims 1-4 is implemented.

Images