Multi-physics field coupling analysis method and related device for co-electrolysis battery stack

Abstract

The present application belongs to the field of electrochemical technology and discloses a multi-physics field coupling analysis method for a co-electrolysis battery stack and its related device. The method includes: fusing the velocity field, potential field, and temperature field to obtain a physical field tensor; mapping the physical field tensor to the initial state of the Hamiltonian system, and obtaining a physical field state prediction value through a physical field prediction model; using the statistical characteristics of the physical field state prediction value as macroscale parameters, and inputting the velocity field and macroscale parameters in the physical field state prediction value into a graph attention network to obtain mesoscale parameters; sampling and calculating the expected value of mesoscopic bubble transport data to obtain microscale parameters; inputting the macro, meso, and microscale parameters, the physical field state prediction value, and the corresponding physical field measured values ​​into a correction model to obtain a correction value for the multi-physics field, and correcting the physical field state prediction value using the correction value for the multi-physics field. The present application improves the accuracy of multi-physics field state prediction.

Description

Technical Field

[0001] The present application relates to the field of electrochemical technology, and in particular to a multi-physics field coupling analysis method for a co-electrolysis cell stack and related devices. Background Art

[0002] Co-electrolysis cell stacks (such as alkaline electrolyzers and PEM electrolyzers) are the core equipment for green hydrogen production. Their performance is affected by the coupling of multiple physical fields such as electrolyte flow, electrochemical reaction, and heat conduction. Current technology has the following bottlenecks:

[0003] 1) Insufficient multi-physics coupling accuracy: Traditional methods use a one-way sequential solution (e.g., fluid flow followed by electrochemistry), which leads to error accumulation and fails to capture the nonlinear interactions between turbulence, reaction, and heat transfer, resulting in high prediction errors.

[0004] 2) Energy conservation and real-time performance defects: Discrete methods such as the explicit Euler method introduce energy dissipation, and fixed time steps are difficult to adapt to rapid changes in electrolysis operating conditions (such as start-stop and load fluctuations);

[0005] 3) Cross-scale modeling fragmentation: The macro-meso-micro scale parameter transfer relies on empirical interpolation and lacks coupling based on physical mechanisms (such as the effect of bubble transport on catalyst activity). Summary of the Invention

[0006] In order to solve the above problems, the present application provides a multi-physics field coupling analysis method for a co-electrolysis cell stack and related devices.

[0007] In view of this, the first aspect of the present application provides a multi-physics field coupling analysis method for a co-electrolysis cell stack, comprising:

[0008] Perform data fusion on the collected velocity field of the electrolyte, the potential field of the electrochemical reaction, and the temperature field in the electrolytic cell to obtain a fused physical field tensor;

[0009] Mapping the physical field tensor to the initial state of the Hamiltonian system, inputting the initial state of the Hamiltonian system into the physical field prediction model for state prediction, and obtaining a physical field state prediction value;

[0010] The statistical characteristics of the physical field state prediction value are obtained as macroscale parameters. The velocity field and macroscale parameters in the physical field state prediction value are input into the graph attention network for feature extraction to obtain mesoscale parameters; mesoscale parameters include relaxation time and bubble force;

[0011] Monte Carlo sampling and expectation value calculation are performed on the mesoscopic bubble transport data to obtain the microscopic scale parameters; the mesoscopic bubble transport data includes the mesoscopic scale parameters and the mesoscopic microfluidic chip observation data;

[0012] The macroscopic scale parameters, mesoscopic scale parameters, microscopic scale parameters, physical field state prediction values ​​and corresponding physical field measured values ​​are input into the correction model to obtain the correction amount of multiple physical fields, and the physical field state prediction values ​​are corrected by the correction amount of multiple physical fields.

[0013] Optionally, data fusion is performed on the collected velocity field of the electrolyte, the potential field of the electrochemical reaction, and the temperature field in the electrolytic cell to obtain a fused physical field tensor, including:

[0014] Perform convolution operation on the velocity field weight parameter and the velocity field of the electrolyte to obtain the velocity field characteristics;

[0015] Perform element-wise product operation on the potential field weight parameter and the potential field of the electrochemical reaction to obtain the potential field characteristics;

[0016] Perform a tensor product operation on the temperature field weight parameter and the temperature field in the electrolytic cell to obtain the temperature field characteristics;

[0017] The velocity field features, electric potential field features, temperature field features and bias terms are summed and mapped through the activation function to obtain the fused physical field tensor.

[0018] Optionally, the method further includes: optimizing the velocity field weight parameter, the potential field weight parameter, the temperature field weight parameter, and the bias term:

[0019] Calculate the physical field prediction error based on the physical field state prediction value and the corresponding physical field measured value;

[0020] The velocity field weight parameters, potential field weight parameters, temperature field weight parameters and bias terms are updated using the physical field prediction error.

[0021] Optionally, the physical field prediction model is:

[0022]

[0023]

[0024] Where, is the physical field state at the nth time step, is the physical field state at the n-1th time step, is the time step of the n-1th time step, J is the symplectic structure matrix, is the multi-field coupled Hamiltonian, Ω represents the integration area, the three-dimensional physical space of the battery stack, represents the integral, dV represents the volume element; is the gradient operator, is the electrolyte density, U is the velocity field, is the dielectric constant, is the electric potential field, T is the temperature field, k is the thermal conductivity coefficient, The physical field state Hamiltonian gradient under .

[0025] Optionally, the time step determination process includes:

[0026] Calculate the deviation between the Hamiltonian gradient of the current time step and the Hamiltonian gradient of the previous time step to obtain the degree of change of the Hamiltonian gradient of the current time step;

[0027] The long short-term memory network is used to predict the length of the next time step based on the degree of change of the Hamiltonian gradient of the current time step and the time step of the current time step.

[0028] Optionally, the calibration model is:

[0029]

[0030] Where, Discrete operators for multi-physics governing equations, including the Navier-Stokes equations, the Butler-Volmer equations, and the Fourier heat conduction equations; is the physical field state; is the predicted value of the physical field state, is the measured value of the physical field state; is a multi-scale parameter, is the macro-scale parameter, is the mesoscopic scale parameter, is the microscopic scale parameter, is the residual weight; It is a residual model consisting of a 3D convolutional layer, a graph attention network, and a fully connected layer.

[0031] Optionally, the process of obtaining the residual term weight is:

[0032] Calculate the deviation between the predicted value of the physical field state and the measured value of the physical field state;

[0033] The bias value and multi-scale parameters are input into the multi-layer perceptron for feature mapping, and the mapping result is normalized by the activation function to obtain the residual term weight.

[0034] A second aspect of the present application provides a multi-physics field coupling analysis device for a co-electrolysis cell stack, comprising:

[0035] A data fusion unit is used to fuse the collected velocity field of the electrolyte, the potential field of the electrochemical reaction, and the temperature field in the electrolytic cell to obtain a fused physical field tensor;

[0036] A state prediction unit is used to map the physical field tensor to the initial state of the Hamiltonian system, input the initial state of the Hamiltonian system into the physical field prediction model to perform state prediction, and obtain a physical field state prediction value;

[0037] A first feature extraction unit is used to obtain statistical features of the physical field state prediction value as macroscale parameters, input the velocity field and macroscale parameters in the physical field state prediction value into the graph attention network for feature extraction, and obtain mesoscale parameters; the mesoscale parameters include relaxation time and bubble force;

[0038] The second feature extraction unit is used to perform Monte Carlo sampling and expectation value calculation on the mesoscopic bubble transport data to obtain microscopic scale parameters; the mesoscopic bubble transport data includes mesoscopic scale parameters and mesoscopic microfluidic chip observation data;

[0039] The correction unit is used to input macroscopic scale parameters, mesoscopic scale parameters, microscopic scale parameters, physical field state prediction values ​​and corresponding physical field measured values ​​into the correction model, obtain the correction amount of multiple physical fields, and correct the physical field state prediction value through the correction amount of multiple physical fields.

[0040] A third aspect of the present application provides an electronic device, the device comprising a processor and a memory;

[0041] The memory is used to store program code and transmit the program code to the processor;

[0042] The processor is used to execute the multi-physics field coupling analysis method for a co-electrolysis cell stack described in any one of the first aspects according to the instructions in the program code.

[0043] In a fourth aspect, the present application provides a computer-readable storage medium for storing program code, which, when executed by a processor, implements the multi-physics field coupling analysis method for a co-electrolysis cell stack as described in any one of the first aspects.

[0044] It can be seen from the above technical solutions that this application has the following advantages:

[0045] The multi-physics field coupling analysis method for co-electrolysis cell stacks provided in the present application fuses data of the physical field, potential field and temperature field and then predicts the physical field state to realize multi-physics field coupling prediction, which improves the problem that the traditional method adopts a one-way sequential solution, resulting in error accumulation, failure to capture the nonlinear interaction of turbulence-reaction-heat transfer, and high prediction error; the acquired macro-scale parameters, meso-scale parameters, and micro-scale parameters are transferred to the correction model, the macroscopic field drives the mesoscopic LBM model, and the mesoscopic bubble dynamics are fed back to the microscopic catalyst surface reaction rate to form a closed-loop coupling; and, the real-time sensor data and the model prediction value are fused to eliminate the cumulative error introduced by the simplification of boundary conditions or numerical discretization, and the sensor noise and model deviation are learned through the residual network to improve the prediction robustness.

[0046] Furthermore, this application ensures the energy conservation characteristics of the multi-physics field coupling system through an adaptive symplectic algorithm, avoiding the energy divergence problem of the traditional data-driven model; and adopts hybrid operators such as three-dimensional convolution and tensor product to realize nonlinear interaction modeling of velocity-potential-temperature fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0048] Figure 1 A schematic flow chart of a multi-physics field coupling analysis method for a co-electrolysis cell stack provided in an embodiment of the present application;

[0049] Figure 2 A schematic structural diagram of a multi-physics field coupling analysis device for a co-electrolysis cell stack provided in an embodiment of the present application;

[0050] Figure 3 A schematic structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to help those skilled in the art better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0052] For easier understanding, please refer to Figure 1 , the embodiment of the present application provides a multi-physics field coupling analysis method for a co-electrolysis cell stack, comprising:

[0053] Step 110: fusing the collected velocity field of the electrolyte, the potential field of the electrochemical reaction, and the temperature field in the electrolytic cell to obtain a fused physical field tensor;

[0054] The velocity field of the electrolyte in the co-electrolysis cell stack can be measured by experimental PIV (Particle Image Velocimetry) , the potential field during electrochemical reaction can be collected by electrochemical workstation , the temperature field in the electrolytic cell was measured by infrared thermal imager , To measure the size of the space.

[0055] Perform data fusion on velocity field, electric potential field and temperature field to obtain the fused physical field tensor :

[0056]

[0057] in, are the weight parameters (weight matrices) of velocity field, electric potential field and temperature field respectively, b is the bias term, is the activation function, is the convolution operation, is the element-wise product operation, is a tensor product operation. Through three-dimensional convolution Capture the turbulent eddy characteristics of the velocity field; by element-wise product Strengthen the effect of potential gradient on ion migration; through tensor product Modeling anisotropic heat conduction (directionality of Fourier thermal conductivity due to crystal structure).

[0058] Physical field tensor The basic form of comes from tensor field theory (mathematical physics) and is used to describe the set of field quantities in a multi-physics coupled system. The traditional form is: (Simple splicing of velocity field, electric potential field, and temperature field). This application introduces the credibility weight of sensor data on this basis. and the bias term b to solve the noise interference problem.

[0059] The weight parameters and bias items can be set according to the actual situation, or the following optimization method can be used to obtain the weight parameters and bias items. The process of obtaining the weight parameters can be:

[0060] First, initialize the weight parameters. The initialization of each weight parameter uses the same normal distribution 、 、 ;n in is the number of input channels. The number of input channels for the velocity field U is 3 (three-dimensional velocity components), and the number of input channels for the electric potential field and the temperature field is 1. The three weight parameters are initialized with the same normal distribution to ensure that the input features of different physical fields (velocity field, electric potential field, temperature field) have similar numerical ranges before fusion, avoiding a certain mode dominating the training process. However, the input channel n in The difference in actual initialization parameters leads to differences in the operator types (convolution / product / tensor product) of multimodal fusion, and ultimately achieves differentiated modeling of physical field characteristics.

[0061] Secondly, optimize the weight parameters. The weight parameters are updated by the back-propagation algorithm combined with the Adam optimizer. The loss function is the weighted root mean square of the multi-physics field prediction error:

[0062]

[0063] in, is the weight coefficient of the prediction error of each physical field, which is determined by grid search; 、 、 They are the velocity field, electric potential field, and temperature field predicted by the physical field prediction model; 、 、 are the true velocity field, true electric potential field, and true temperature field measured experimentally; is the L2 norm (Euclidean norm), which is used to quantify the overall difference between the two physical fields. , calculate the predicted velocity field and the real velocity field at all spatial points ( grid) and three-dimensional components ( The root mean square error (RMS) of the velocity in the x / y / z directions is expressed in m / s.

[0064] The process of obtaining the bias term b can be:

[0065] First, initialization: the bias term b is initialized to a zero vector to avoid deviations introduced by the initial state;

[0066] Secondly, dynamic adjustment: During the training process, the bias term b is updated synchronously with the weight parameters, and the calculation formula is:

[0067]

[0068] Where, is the learning rate.

[0069] Activation Function Determination of:

[0070] In order to avoid gradient vanishing and enhance nonlinear expression capabilities, the embodiment of the present application preferably adopts the LeakyReLU activation function, whose formula is:

[0071]

[0072] This activation function introduces a small gradient (slope 0.01) in the negative interval to alleviate the problem of neuron "death".

[0073] The embodiment of the present application constructs a hybrid modal tensor space, and jointly models the three field data of electrolyte flow (Navier-Stokes), electrochemical reaction (Butler-Volmer), and heat conduction (Fourier) through a multimodal network.

[0074] Step 120: Map the physical field tensor to the initial state of the Hamiltonian system, input the initial state of the Hamiltonian system into the physical field prediction model to perform state prediction, and obtain a physical field state prediction value;

[0075] Through the fully connected layer The physical field tensor Mapping to the initial state of the Hamiltonian system , , are the trainable parameters of the fully connected layer.

[0076] Assume that the fused physical field tensor The dimension is ( is the total number of channels), and the channels store different physical field characteristics in sequence:

[0077] Channel 0-2: Velocity field The three-dimensional components of );

[0078] Channel 3: Reserved field (may be used for future expansion);

[0079] Channel 4: Electric potential field (single-channel scalar);

[0080] Channel 5: Temperature field (single-channel scalar).

[0081] in, , extract the first three channels (index 0, 1, 2), corresponding to the three-dimensional components of the velocity field, and construct the initial velocity field ; , extract the 5th channel (index 4) as the initial electric potential field ; , extract the 6th channel (index 5) as the initial temperature field .

[0082] Traditional Hamiltonian It only contains kinetic energy (fluid) and potential energy (electric field). This application adds a thermodynamic entropy term on this basis. Therefore, the multi-field coupled Hamiltonian of the Hamiltonian system in the embodiment of the present application is for:

[0083]

[0084] Where dV represents the volume element, an infinitesimal volume unit in the integration space, used for the energy density integration of the continuous medium; Ω represents the integration area, the three-dimensional physical space of the battery stack, represents the integral; is the kinetic energy of electrolyte flow, is the electrolyte density, is the dielectric constant, which reflects the storage capacity of electric field energy in the medium and is related to the electric potential field distribution; is the electric field potential energy (driving potential energy of electrochemical reaction), is the electric potential field, electrochemical reaction (charge transfer potential driven by it), is the electric potential gradient; Thermal energy related to thermodynamic entropy (energy representation of temperature field diffusion); It is the heat conduction coefficient, a measure of the diffusion capacity of the temperature field, and obeys Fourier's heat conduction law. It unifies the energy of multiple physical fields into a Hamiltonian system, providing a theoretical basis for solving conservation problems.

[0085] The embodiment of the present application constructs a physical field prediction model based on the multi-field coupled Hamiltonian. The physical field prediction model uses a symplectic difference format to obtain a time-space discrete solution. The physical field prediction model is specifically:

[0086]

[0087]

[0088] Where, is the physical field state at the nth time step, is the physical field state at the n-1th time step, is the time step length of the n-1th time step; J is the symplectic structure matrix (antisymmetric matrix), which is derived from symplectic geometry (mathematics) to ensure the energy conservation of numerical simulation. Here, the traditional explicit Euler method is changed to the symplectic algorithm to avoid energy dissipation; is the Hamiltonian gradient, The physical field state Hamiltonian gradient under. This application is based on symplectic geometry theory, through the inverse matrix J of the symplectic structure matrix -1 Adjust the Hamiltonian gradient direction to ensure that the discretized numerical solution strictly maintains energy conservation and avoids long-term simulation distortion caused by energy dissipation in the traditional explicit Euler method.

[0089] This application uses an adaptive time step strategy, dynamically predicting the time step through a long short-term memory (LSTM) network: when the Hamiltonian gradient changes dramatically (such as in turbulent or intensely reactive regions), the time step is reduced to improve accuracy; conversely, the time step is increased to speed up the calculation, improving the stability issues of fixed-step algorithms in rigid systems (such as numerical oscillations caused by sudden temperature changes). The prediction process of the adaptive time step is as follows:

[0090]

[0091] The above formula originally comes from LSTM time series prediction (computer science) and rigid system stability theory (numerical analysis). Hamiltonian gradient drive: using Hamiltonian gradient changes as input features, LSTM learns the nonlinear relationship between gradient mutations and time steps, replacing traditional heuristic rules.

[0092] The degree of gradient change between the Hamiltonian gradient at the n-1th time step and the Hamiltonian gradient at the n-2th time step and the step size at the n-1th time step The input is fed into the LSTM network for time step prediction. The LSTM network uses a large step size in stable areas (such as uniform flow fields) and automatically switches to a small step size in mutation areas (such as bubble generation interfaces), thereby obtaining the step size of the nth time step. , the initial time step It can be set based on experience. The LSTM network predicts the step size by historical Hamiltonian gradient changes to avoid divergence of rigid systems.

[0093] The initial state of the Hamiltonian system Input into the above physical field prediction model to perform multi-physical field coupling prediction and obtain the physical field state prediction value of the nth time step .

[0094] Step 130: Obtain statistical features of the physical field state prediction value as macroscopic scale parameters, input the velocity field and macroscopic scale parameters in the physical field state prediction value into the graph attention network for feature extraction, and obtain mesoscopic scale parameters;

[0095] Get the predicted value of the physical field state at the nth time step The statistical characteristics of the macro-scale parameters (such as average flow rate , maximum potential gradient , temperature fluctuation , Reynolds number Re, vorticity, etc.).

[0096] Among them, the macro-scale parameters The calculation process can be:

[0097]

[0098]

[0099] Where, is the temperature value at the macro grid point (i, j) (unit: K); N x ×N y is the macro grid resolution (e.g. 100×100); is the temperature field In space N x ×N y The original scale is just a statistical mechanics parameter, and this application adds temperature values ​​to form a multi-physics field comprehensive scale analysis.

[0100] The velocity field in the physical field state prediction value of the nth time step and macroscale parameters Input into the graph attention network (GAT) for feature mapping to obtain mesoscopic scale parameters , the mesoscopic scale parameters include relaxation time and bubble force. Among them, the core operation process of the graph attention network GAT includes:

[0101]

[0102]

[0103] Where, is the attention weight of node i and its neighbor node j; is the attention weight matrix, which is a trainable parameter; 、 is the characteristic vector of nodes i and j, including the velocity field and macroscale parameters ; is the set of neighbor nodes of node i; is the feature transformation matrix, which is a trainable parameter; is an activation function (such as a ReLU function). The graph attention network is used to process graph structure data. Traditional graph attention networks process node attributes (such as molecular structures), while the embodiment of the present application includes velocity fields and macro-scale parameters.

[0104] Among them, the training process of the graph attention network is:

[0105] The velocity field is interpolated from the macroscopic grid (mm level) to the mesoscopic grid (μm level) to retain the high resolution of the vortex core region; statistics (such as turbulent kinetic energy) are injected into the macroscopic scale parameters to improve the generalization of migration; paired data sets generated by numerical simulation or PIV experimental data are Input into the GAT network and calculate the loss value:

[0106]

[0107] Where, 、 Relaxation time and bubble force predicted for graph attention network (characterizing momentum exchange between bubbles and electrolyte flow at mesoscopic scale); 、 Reference values ​​of mesoscopic relaxation time and bubble force generated for numerical simulation or PIV experimental data; is the L2 norm, used to penalize the relaxation time The prediction deviation is emphasizing the optimization of smoothness; is the L1 norm, used to enhance the bubble force Sparsity constraints to avoid overfitting; is a weighting coefficient used to balance the contribution of the two losses: The larger the value, the more significant the impact of bubble force prediction accuracy on the total loss, which can be determined by grid search or cross validation. .

[0108] Can output in real time to the LBM model (lattice Boltzmann method), driving the simulation of mesoscopic bubble transport.

[0109] Step 140: performing Monte Carlo sampling and expectation value calculation on the mesoscopic bubble transport data to obtain microscopic scale parameters; the mesoscopic bubble transport data includes the mesoscopic scale parameters and mesoscopic microfluidic chip observation data;

[0110] Mesoscopic bubble transport data (including mesoscopic scale parameters , mesoscopic microfluidic chip observation data x, such as local concentration field) are sampled by Monte Carlo, and then the expected value is calculated to obtain the microscopic scale parameters (surface reaction characteristics calculated based on mesoscopic data, such as microscopic catalyst reaction energy barriers), , is an expectation value operator that maps mesoscopic bubble transport statistics (such as average collision frequency) to microscopic reaction probabilities. For example:

[0111]

[0112] Where, F( ) represents the sampling result obtained by Monte Carlo sampling of mesoscopic bubble transport data; N is the number of sampling times.

[0113] Step 150: Input the macroscopic scale parameters, mesoscopic scale parameters, microscopic scale parameters, physical field state prediction values ​​and corresponding physical field measured values ​​into the correction model to obtain the correction amount of multiple physical fields, and correct the physical field state prediction values ​​using the correction amount of multiple physical fields.

[0114] Through the above process, the multi-scale parameters of the nth time step can be obtained And the predicted value of the physical field state at the nth time step , the measured value of the physical field state can be obtained through sensors (such as 1kHz sampling of potential / temperature).

[0115] The calibration model in the embodiment of the present application is:

[0116]

[0117] The above formula is derived from data assimilation theory (earth science) and residual learning (deep learning). This application expands the traditional residual term into a multi-scale parameter Weighted. Where z is the physical field state, is the predicted value of the physical field state, that is, the predicted value of the physical field state at the nth time step obtained in step 120 ; 2 is the residual term weight. This embodiment of the application combines the 3D convolution layer and the graph attention network to build a residual model , the nonlinear deviations between the simulated and measured values ​​(such as sensor noise and model simplification errors) are learned through the residual model. Maintain physical conservation (such as mass and energy conservation) and inherit the Hamiltonian system characteristics in step 120; the second term The deviation between the model prediction value and the measured data is dynamically corrected through the deep learning residual network to compensate for the unmodeled effects (such as catalyst aging and electrolyte impurities).

[0118] is the discretized operator of the original multiphysics governing equation, which includes the following components:

[0119] 1) Navier-Stokes equations (velocity field ):

[0120]

[0121] Where, is the electrolyte density, is the dynamic viscosity, is the pressure field, is the gradient operator. This equation is used to ensure the conservation of mass / momentum in the velocity field;

[0122] 2) Butler-Volmer equation (potential field )

[0123]

[0124] Where, is the electrical conductivity, which reflects the charge transfer ability of the electrolyte / electrode material; is the Ohmic loss term, which describes the diffusion and dissipation of charge in the medium; is the exchange current density, which characterizes the charge transfer rate at the electrode-electrolyte interface and is related to the catalyst activity. This equation is used to constrain the potential field to conform to the electrochemical reaction current distribution;

[0125] 3) Fourier heat conduction equation (temperature field ):

[0126]

[0127] Where, is the specific heat capacity; k is the thermal conductivity; The heat diffusion term describes the conduction of heat energy from a high-temperature region to a low-temperature region. This equation is used to ensure energy conservation in the temperature field and avoid unphysical solutions (such as negative temperatures). The Navier-Stokes equations, the Butler-Volmer equations, and the Fourier heat conduction equation are among the existing equations.

[0128] Calibration model It consists of a 3D convolution layer (extracting spatial error features), a graph attention network (fusing multi-scale parameters), and a fully connected layer (outputting residual tensors). The 3D convolution layer has a kernel size of 3×3×3 and 64 channels, which is used to extract local error features (such as temperature deviation in hot spot areas); the graph attention layer fuses multi-scale parameters into the image. Injected as node attributes to dynamically adjust feature weights.

[0129] Will and Align by grid points and calculate absolute error point by point ; Then generate the error feature map through the 3D convolution layer ; In the graph attention layer, Participate in attention weight calculation as node features:

[0130]

[0131] Where, is the multi-scale parameter corresponding to the grid point j, is the attention weight matrix; the final output residual , corresponding to the correction amount of velocity field, electric potential field and temperature field.

[0132] Dynamically generated by a multi-layer perceptron (MLP), the input is the error amplitude and multi-scale parameters :

[0133]

[0134] The sigmoid function is used to constrain , avoid overcorrection; high error areas (such as hot spots) , strong correction; low error area , retaining the predictions of the physical model. MLP is used to quantify the local error intensity of each physical field (for example, the error in the hot spot area of ​​the temperature field is significantly higher than that in the uniform area). The softmax function normalizes the weight distribution to ensure the stability of the correction process (the total weight is constrained to a reasonable range). Weight Real-time update to achieve error-sensitive adaptive correction.

[0135] The embodiment of this application uses the macroscopic size parameter As the convolution kernel in the 3D convolution layer ( is the initial value of the 3D convolution kernel, used to extract multi-scale features from the velocity field U) and optimize the velocity field Correction accuracy; mesoscopic scale parameter (such as bubble distribution) Enhance the potential correction of bubble clusters (bubble shielding effect) through the graph attention mechanism; micro-scale parameters (such as catalyst activity) by controlling Sensitivity to the current density at the electrode-electrolyte interface.

[0136] Multi-scale parameters participate The generation of error-operating condition coordinated control is realized, for example, according to When the bubble coverage is detected to be >30%, Automatically increase the correction weight of the temperature field (bubbles lead to worsening local heat dissipation). Transfer cross-scale correlations to avoid the destruction of mesoscopic-microscopic balance after correction, and the corrected multi-physics state Need to meet , that is, compatible with the microscopic simulation results. This application integrates real-time sensor data with model predictions to eliminate the cumulative errors introduced by simplified boundary conditions or numerical discretization; through the residual network Learn from sensor noise and model bias (such as changes in electrolyte composition) to improve prediction robustness.

[0137] The update formula for the multiphysics state is:

[0138]

[0139] The above formula is derived from the explicit update format of the symplectic algorithm (mathematical geometry). This application embeds data-driven residuals in the traditional symplectic format. is the multi-physics field state after the k+1th update, that is, the corrected multi-physics field state; the integration step is predicted by the long short-term memory network in the previous step to ensure numerical stability. The initial time value of is the multi-physics state prediction value z at the nth time step output in step 120 n ,Right now:

[0140]

[0141] If the corrected multiphysics state violates mass / energy conservation (e.g. Or T<0), roll back to the update result of the previous step and stop updating.

[0142] The corrected multi-physics field state finally output can be directly used for system monitoring and real-time display of the multi-physics field distribution of the electrolyzer.

[0143] Taking over-temperature protection as an example, the temperature field can be extracted from the corrected multi-physical field state to identify local hot spots (temperature exceeding 358K). When the temperature exceeds the first threshold (such as 360K), a first-level warning is issued and the current density is reduced (such as 5%). The rectifier output current can be adjusted through the programmable logic controller (PLC); when the temperature exceeds the second threshold (such as 365K), the power supply is urgently cut off and the cooling system is started.

[0144] The corrected multi-physics state can also be used to optimize electrolysis efficiency. Taking current density adjustment as an example, the potential field is extracted from the corrected multi-physics state. , and then calculate the electrode overpotential; if the electrode overpotential is greater than 0.25V, it is determined that the catalyst activity has decreased and the current distribution needs to be optimized; pulse width modulation (PWM) can be used to make the current density standard deviation less than the current density threshold (such as ).

[0145] By accurately predicting the temperature gradient and potential distribution inside the fuel cell stack, it guides the flow channel structure design and catalyst coating scheme, thereby extending battery life; real-time identification of local overheating or abnormal electrolyte concentration, supports millisecond-level fault warning, and reduces maintenance costs; and provides multi-physics field coupling simulation interfaces for hydrogen vehicles, energy storage power stations, etc., accelerating the R&D cycle of new energy equipment.

[0146] This embodiment of the application achieves cross-scale parameter transfer based on the fusion of physical mechanisms. The macroscopic field (Navier-Stokes equations) drives the mesoscopic LBM model (D3Q19 lattice Boltzmann method), and the mesoscopic bubble dynamics are fed back to the microscopic catalyst surface reaction rate, forming a closed-loop coupling. Macro-mesoscopic parameter transfer reduces the size of the mesoscopic simulation domain, speeding up the calculation by more than 10 times compared to full-scale direct numerical simulation (DNS), thereby improving efficiency. The analysis method provided in this embodiment of the application has the following advantages:

[0147] Physical constraint embedding: An adaptive symplectic algorithm is used to ensure energy conservation in multi-field coupled systems, avoiding the energy divergence problem of traditional data-driven models.

[0148] Cross-modal feature fusion: Using hybrid operators such as 3D convolution and tensor product to achieve nonlinear interaction modeling of velocity, potential, and temperature fields;

[0149] Multi-scale dynamic connection: By transferring macroscopic-mesoscopic-microscopic parameters (e.g., bubble coverage → catalyst activity correction), the scale gap of empirical formulas can be broken;

[0150] Online closed-loop optimization: Integrates 1kHz high-frequency sensor data to achieve real-time model correction and adapt to the changing operating conditions of the electrolyzer.

[0151] The embodiment of the present application significantly improves the accuracy of multi-physical field coupling analysis of co-electrolysis battery stacks through a closed-loop system of multimodal data fusion, conservation space-time solution, cross-scale modeling and dynamic correction.

[0152] Please refer to Figure 2 The present application also provides a multi-physics field coupling analysis device for a co-electrolysis cell stack, which is characterized by comprising:

[0153] The data fusion unit 210 is used to fuse the collected velocity field of the electrolyte, the potential field of the electrochemical reaction, and the temperature field in the electrolytic cell to obtain a fused physical field tensor;

[0154] A state prediction unit 220 is used to map the physical field tensor to the initial state of the Hamiltonian system, input the initial state of the Hamiltonian system into the physical field prediction model to perform state prediction, and obtain a physical field state prediction value;

[0155] A first feature extraction unit 230 is used to obtain statistical features of the physical field state prediction value as macro-scale parameters, input the velocity field and macro-scale parameters in the physical field state prediction value into the graph attention network for feature extraction, and obtain meso-scale parameters; the meso-scale parameters include relaxation time and bubble force;

[0156] A second feature extraction unit 240 is used to perform Monte Carlo sampling and expected value calculation on the mesoscopic bubble transport data to obtain microscopic scale parameters; the mesoscopic bubble transport data includes the mesoscopic scale parameters and mesoscopic microfluidic chip observation data;

[0157] The correction unit 250 is used to input macroscopic scale parameters, mesoscopic scale parameters, microscopic scale parameters, physical field state prediction values ​​and corresponding physical field measured values ​​into the correction model, obtain the correction amount of multiple physical fields, and correct the physical field state prediction value through the correction amount of multiple physical fields.

[0158] As a further improvement, the data fusion unit 210 is specifically configured to perform a convolution operation on the velocity field weight parameter and the velocity field of the electrolyte to obtain velocity field features;

[0159] Perform element-wise product operation on the potential field weight parameter and the potential field of the electrochemical reaction to obtain the potential field characteristics;

[0160] Perform a tensor product operation on the temperature field weight parameter and the temperature field in the electrolytic cell to obtain the temperature field characteristics;

[0161] The velocity field features, electric potential field features, temperature field features and bias terms are summed and mapped through the activation function to obtain the fused physical field tensor.

[0162] As a further improvement, the device further includes: a weight optimization unit for optimizing the velocity field weight parameter, the potential field weight parameter, the temperature field weight parameter and the bias term;

[0163] The weight optimization unit is specifically used to calculate the physical field prediction error based on the physical field state prediction value and the corresponding physical field measured value; and update the velocity field weight parameter, electric potential field weight parameter, temperature field weight parameter and bias term through the physical field prediction error.

[0164] This application fuses data of the physical field, potential field and temperature field and then predicts the physical field state to achieve multi-physical field coupling prediction, which improves the traditional method of using one-way sequential solution, resulting in error accumulation, inability to capture the nonlinear interaction of turbulence-reaction-heat transfer, and high prediction error. The acquired macro-scale parameters, meso-scale parameters, and micro-scale parameters are transferred to the correction model, the macroscopic field drives the mesoscopic LBM model, and the mesoscopic bubble dynamics are fed back to the microscopic catalyst surface reaction rate to form a closed-loop coupling. In addition, the real-time sensor data and the model prediction value are fused to eliminate the cumulative error introduced by the simplification of boundary conditions or numerical discretization, and the sensor noise and model deviation are learned through the residual network to improve the prediction robustness.

[0165] Please refer to Figure 3 , an embodiment of the present application further provides an electronic device, the device including a processor 310 and a memory 320;

[0166] The memory 320 is used to store program codes and transmit the program codes to the processor 310;

[0167] The processor 310 is configured to execute the multi-physics coupling analysis method for the co-electrolysis cell stack in the aforementioned method embodiment according to instructions in the program code.

[0168] An embodiment of the present application also provides a computer-readable storage medium, which is used to store program code. When the program code is executed by a processor, it implements the multi-physics field coupling analysis method for a co-electrolysis cell stack in the aforementioned method embodiment.

[0169] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0170] In the specification of this application and the above-mentioned drawings, the terms "first," "second," "third," "fourth," etc. (if any) are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. In addition, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus that includes a series of steps or elements is not necessarily limited to those steps or elements explicitly listed, but may include other steps or elements not explicitly listed or inherent to such process, method, product, or apparatus.

[0171] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships can exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or plural.

[0172] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0173] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0174] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0175] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for executing all or part of the steps of the method described in each embodiment of the present application through a computer device (which can be a personal computer, server, or network device, etc.). The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (full name: Read-Only Memory, English abbreviation: ROM), random access memory (full name: Random Access Memory, English abbreviation: RAM), disk or optical disk, and other media that can store program code.

[0176] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-physics field coupling analysis method for a co-electrolysis cell stack, characterized in that: include: Perform data fusion on the collected velocity field of the electrolyte, the potential field of the electrochemical reaction, and the temperature field in the electrolytic cell to obtain a fused physical field tensor; Mapping the physical field tensor to the initial state of the Hamiltonian system, inputting the initial state of the Hamiltonian system into the physical field prediction model for state prediction, and obtaining a physical field state prediction value; The statistical characteristics of the physical field state prediction value are obtained as macroscale parameters. The velocity field and macroscale parameters in the physical field state prediction value are input into the graph attention network for feature extraction to obtain mesoscale parameters; mesoscale parameters include relaxation time and bubble force; Monte Carlo sampling and expectation value calculation are performed on the mesoscopic bubble transport data to obtain the microscopic scale parameters; the mesoscopic bubble transport data includes the mesoscopic scale parameters and the mesoscopic microfluidic chip observation data; The macroscopic scale parameters, mesoscopic scale parameters, microscopic scale parameters, physical field state prediction values ​​and corresponding physical field measured values ​​are input into the correction model to obtain the correction amount of multiple physical fields, and the physical field state prediction values ​​are corrected by the correction amount of multiple physical fields.

2. The multi-physics field coupling analysis method for a co-electrolysis cell stack according to claim 1, characterized in that: The collected velocity field of the electrolyte, the potential field of the electrochemical reaction, and the temperature field in the electrolytic cell are fused to obtain the fused physical field tensor, including: Perform convolution operation on the velocity field weight parameter and the velocity field of the electrolyte to obtain the velocity field characteristics; Perform element-wise product operation on the potential field weight parameter and the potential field of the electrochemical reaction to obtain the potential field characteristics; Perform a tensor product operation on the temperature field weight parameter and the temperature field in the electrolytic cell to obtain the temperature field characteristics; The velocity field features, electric potential field features, temperature field features and bias terms are summed and mapped through the activation function to obtain the fused physical field tensor.

3. The multi-physics field coupling analysis method for a co-electrolysis cell stack according to claim 2, characterized in that: The method further includes optimizing the velocity field weight parameter, the potential field weight parameter, the temperature field weight parameter, and the bias term, specifically including: Calculate the physical field prediction error based on the physical field state prediction value and the corresponding physical field measured value; The velocity field weight parameters, potential field weight parameters, temperature field weight parameters and bias terms are updated using the physical field prediction error.

4. The multi-physics field coupling analysis method for a co-electrolysis cell stack according to claim 1, characterized in that: The physical field prediction model is: Where, is the physical field state at the nth time step, is the physical field state at the n-1th time step, is the time step of the n-1th time step, J is the symplectic structure matrix, is the multi-field coupled Hamiltonian, Ω represents the integration area, the three-dimensional physical space of the battery stack, represents the integral, dV represents the volume element; is the gradient operator, is the electrolyte density, U is the velocity field, is the dielectric constant, is the electric potential field, T is the temperature field, k is the thermal conductivity coefficient, The physical field state Hamiltonian gradient under .

5. The multi-physics field coupling analysis method for a co-electrolysis cell stack according to claim 4, characterized in that: The process of determining the time step includes: Calculate the deviation between the Hamiltonian gradient of the current time step and the Hamiltonian gradient of the previous time step to obtain the degree of change of the Hamiltonian gradient of the current time step; The long short-term memory network is used to predict the length of the next time step based on the degree of change of the Hamiltonian gradient of the current time step and the time step of the current time step.

6. The multi-physics field coupling analysis method for a co-electrolysis cell stack according to claim 1, characterized in that: The calibration model is: Where, Discrete operators for multi-physics governing equations, including the Navier-Stokes equations, the Butler-Volmer equations, and the Fourier heat conduction equations; is the physical field state; is the predicted value of the physical field state, is the measured value of the physical field state; is a multi-scale parameter, is the macro-scale parameter, is the mesoscopic scale parameter, is the microscopic scale parameter, is the residual weight; It is a residual model consisting of a 3D convolutional layer, a graph attention network, and a fully connected layer.

7. The multi-physics field coupling analysis method for a co-electrolysis cell stack according to claim 6, characterized in that: The process of obtaining the residual term weight is: Calculate the deviation between the predicted value of the physical field state and the measured value of the physical field state; The bias value and multi-scale parameters are input into the multi-layer perceptron for feature mapping, and the mapping result is normalized by the activation function to obtain the residual term weight.

8. A multi-physics field coupling analysis device for a co-electrolysis cell stack, characterized in that: include: A data fusion unit is used to fuse the collected velocity field of the electrolyte, the potential field of the electrochemical reaction, and the temperature field in the electrolytic cell to obtain a fused physical field tensor; A state prediction unit is used to map the physical field tensor to the initial state of the Hamiltonian system, input the initial state of the Hamiltonian system into the physical field prediction model to perform state prediction, and obtain a physical field state prediction value; A first feature extraction unit is used to obtain statistical features of the physical field state prediction value as macroscale parameters, input the velocity field and macroscale parameters in the physical field state prediction value into the graph attention network for feature extraction, and obtain mesoscale parameters; the mesoscale parameters include relaxation time and bubble force; The second feature extraction unit is used to perform Monte Carlo sampling and expectation value calculation on the mesoscopic bubble transport data to obtain microscopic scale parameters; the mesoscopic bubble transport data includes mesoscopic scale parameters and mesoscopic microfluidic chip observation data; The correction unit is used to input macroscopic scale parameters, mesoscopic scale parameters, microscopic scale parameters, physical field state prediction values ​​and corresponding physical field measured values ​​into the correction model, obtain the correction amount of multiple physical fields, and correct the physical field state prediction value through the correction amount of multiple physical fields.

9. An electronic device, characterized in that: The device includes a processor and a memory; The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the multi-physics field coupling analysis method for a co-electrolysis cell stack according to any one of claims 1 to 7 according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store program code, and when the program code is executed by a processor, it implements the multi-physics field coupling analysis method for a co-electrolysis cell stack according to any one of claims 1 to 7.

Images