Performance analysis method and test system of solid oxide fuel cell

Abstract

The invention discloses a performance analysis method and a test system for a solid oxide fuel cell, and solves the problem that the aging state of the cell and the influence of the charge-discharge cycle of the cell on the service life of the cell cannot be comprehensively reflected due to the fact that the service life of the cell is analyzed by only relying on cell voltage data in the existing method. The method comprises the following steps: constructing and training a battery simulation model based on a particle swarm algorithm, outputting a prediction sequence parameter, extracting a sensitivity coefficient of the prediction sequence parameter by a battery health assessment model, and calculating to obtain an operation state value representing the current operation state of a battery and a health state value representing the current health state; according to the method, the battery simulation model is constructed and trained based on the particle swarm optimization, and the time sequence parameters are accurately predicted and evaluated through the battery simulation model, so that the battery health evaluation model evaluates and calculates the operation state value, the health state value and the predicted life coefficient of the battery in combination with the prediction evaluation result; the prediction result based on the battery simulation model-battery health assessment model is reliable and accurate.

Description

Technical Field

[0001] The invention belongs to the technical field of fuel cells, and in particular relates to a performance analysis method and a testing system for solid oxide fuel cells. Background Art

[0002] At present, fuel cells can be divided into five categories according to the type of electrolyte: alkaline fuel cells (AFC), phosphoric acid fuel cells (PAFC), molten carbonate fuel cells (MCFC), solid oxide fuel cells (SOFC) and proton exchange membrane fuel cells (PEMFC). Currently, the main types shipped in the market are proton exchange membrane fuel cells and solid oxide fuel cells. Solid oxide fuel cells are widely used, mainly in portable power supplies, distributed power generation / combined heat and power systems, high-performance power supplies, large power stations and other fields.

[0003] The performance of solid fuel cells will affect their lifespan. Efficient energy conversion, optimal materials, reasonable operating conditions, high-performance electrocatalysts, stable microstructures, good mechanical properties and precise coupled field analysis are all conducive to extending the service life of solid fuel cells. Therefore, performance analysis of fuel cells is very necessary.

[0004] Among them, Chinese patent CN111310387B discloses a fuel cell life prediction method. First, the adaptive variational mode decomposition technology is used to analyze the original fuel cell voltage data to improve the adaptability to different voltage change characteristics; based on the fuel cell voltage decay characteristics and sample entropy theory, the VMD algorithm decomposition subsequence is reconstructed to better reflect the battery aging characteristic information; a deep belief network prediction model is established, but the existing method only relies on battery voltage data to analyze the battery life and cannot fully reflect the battery aging state and the impact of battery charge and discharge cycles on battery life, resulting in large errors in the analysis results and inaccurate evaluation; in response to the above problems, we propose a performance analysis method and test system for solid oxide fuel cells. Summary of the invention

[0005] The purpose of the present invention is to address the shortcomings of the prior art and provide a performance analysis method and test system for solid oxide fuel cells, which solves the problem that the existing method only relies on battery voltage data to analyze battery life and cannot fully reflect the battery aging state and the impact of battery charge and discharge cycles on battery life, resulting in large errors in the analysis results and inaccurate evaluation.

[0006] The present invention is achieved by providing a method for analyzing the performance of a solid oxide fuel cell, the method comprising:

[0007] Obtaining fuel cell performance test timing parameters, wherein the timing parameters include battery voltage data, current data, current density, battery temperature, gas composition parameters, and battery material characterization data, and preprocessing the timing parameters to obtain a real-time preprocessing set;

[0008] Building and training a battery simulation model based on a particle swarm algorithm, loading a real-time preprocessing set, executing the battery simulation model with the real-time preprocessing set as input, and outputting prediction sequence parameters;

[0009] A battery health assessment model is constructed based on support vector machine combined with integrated empirical mode decomposition. The battery health assessment model is iteratively trained to obtain prediction sequence parameters. The battery health assessment model extracts the sensitivity coefficients of the prediction sequence parameters and determines the sensitivity coefficients associated with the current operating status, current health status and remaining predicted life of the battery.

[0010] The battery health assessment model performs forward uncertainty analysis on the battery simulation model based on the Monte Carlo algorithm, calculates the operating status value that characterizes the current operating status of the battery and the health status value of the current health status, and calculates and evaluates the predicted life coefficient based on the operating status value and the health status value.

[0011] Preferably, the method for preprocessing the timing parameters specifically includes:

[0012] Loading time series parameters, processing missing values ​​of time series parameters, wherein the missing value processing includes feature removal, mean interpolation method and spline interpolation method processing;

[0013] Obtaining the time series parameters after missing value processing, extracting the nonlinear feature parameters and image feature parameters in the time series parameters, removing the nonlinear feature parameters and image feature parameters in the time series parameters, and forming a linear feature parameter set;

[0014] Parameter decomposition of nonlinear characteristic parameters based on modal decomposition algorithm;

[0015] Among them, the modal decomposition algorithm is a wavelet transform algorithm. When the wavelet transform algorithm performs parameter decomposition on nonlinear characteristic parameters:

[0016] Get the nonlinear characteristic parameter x(t f ), set the wavelet basis function and decomposition level of the wavelet transform algorithm;

[0017] The wavelet basis function is expressed as:

[0018]

[0019] Among them, a is the scale parameter of the wavelet basis function, x(t f ) is the input value of the nonlinear characteristic parameter, k is the position index value of the nonlinear characteristic parameter;

[0020] Based on continuous wavelet transform, the nonlinear characteristic parameters are decomposed into high-frequency detail components and low-frequency approximate components at the first decomposition level;

[0021] At the second decomposition level, low-frequency approximate components are obtained, and the low-frequency approximate components are decomposed again based on continuous wavelet transform until a preset decomposition level is reached, and multiple groups of high-frequency approximate components and a group of low-frequency approximate components obtained within the preset decomposition level are integrated to obtain a parameter decomposition set of nonlinear characteristic parameters.

[0022] Preferably, the method for preprocessing the timing parameters further includes:

[0023] Based on the improved convolutional neural network, the image feature parameters are traversed to achieve the image feature parameter transformation and obtain the image feature transformation set;

[0024] Among them, the self-attention mechanism module is introduced into the improved convolutional neural network, and the convolutional neural network is improved through the self-attention mechanism module;

[0025] Obtain parameter decomposition set, image feature transformation set, and linear feature parameter set, perform lumped integration transformation on the parameter decomposition set, image feature transformation set, and linear feature parameter set based on the arithmetic mean method, and obtain a real-time preprocessing set including a real-time sequence, a mean sequence, and an extreme value sequence.

[0026] Preferably, the method for implementing image feature parameter transformation by traversing image feature parameters based on an improved convolutional neural network specifically includes:

[0027] Obtain image feature parameters, and extract clear image features from the image feature parameters;

[0028] The clear image features are refined and decomposed, and the feature decomposition time is marked. The output function is recorded as:

[0029]

[0030] Among them, t f (x) is the characteristic decomposition time, x(t p ) is the refined decomposition result of the clear image features, M is the number of factors affecting the decomposition of image feature information, C M is the influence weight of image feature information decomposition, γ is the fuzzy denoising coefficient when performing fine decomposition of clear image features;

[0031]

[0032] Among them, W x The weight matrix of the refined decomposition result of the clear image features in the image feature parameter set, is to improve the data gradient of the convolutional neural network, and e is the decomposition error coefficient when the improved convolutional neural network performs fine decomposition on the clear image features;

[0033] Improve the convolutional neural network to perform dilated convolution on the refined decomposition results to increase the receptive field;

[0034] The dilated convolution result is expressed as:

[0035]

[0036] Among them, J(x t ) is the result of dilated convolution, f(i) is the size of the convolution matrix in the improved convolutional neural network, n is the size of the convolution kernel, d is the expansion factor during dilated convolution, and m is the number of convolution kernel convolution layers;

[0037] The dilated convolution result after the dilated convolution is input into the self-attention layer. The self-attention layer redistributes the weights of the dilated convolution result based on the self-attention mechanism to obtain a set of image feature transformations.

[0038] Preferably, the method for constructing and training a battery simulation model based on a particle swarm algorithm specifically includes:

[0039] Taking the equivalent circuit model as the initial model of the battery simulation model, traversing the real-time preprocessing set, and extracting the equivalent circuit associated parameters from the real-time preprocessing set;

[0040] Build an equivalent circuit model based on the Simulink model, and configure the fuzzy logic balance controller and leader particle dynamic selection strategy in the equivalent circuit model;

[0041] Setting the charge and discharge current of the battery simulation model, simulating the voltage response of the battery simulation model under the simulated working conditions, and obtaining equivalent circuit simulation parameters;

[0042] Obtain equivalent circuit associated parameters and equivalent circuit simulation parameters, verify the accuracy of equivalent circuit simulation parameters based on the equivalent circuit associated parameters, and determine whether the parameter similarity meets the preset similarity threshold. If so, the group of equivalent circuit simulation parameters will not be adjusted. If not, a fuzzy logic balancing controller will be used to adjust the battery simulation model.

[0043] Preferably, the method of executing the battery simulation model and outputting the prediction sequence parameters specifically includes:

[0044] Obtain a real-time preprocessing set, and the battery simulation model extracts the equivalent circuit associated parameters in the real-time preprocessing set;

[0045] Taking the equivalent circuit associated parameters as input, the battery simulation model performs a global sensitivity analysis on the equivalent circuit simulation parameters based on the Sobol′ model, and selects the equivalent circuit simulation parameters with a total sensitivity of more than 0.005 as sensitive parameters;

[0046] Among them, the sensitivity index of the equivalent circuit simulation parameter is calculated by the following formula:

[0047]

[0048] Among them, S x is the sensitivity of the equivalent circuit simulation parameter, x(t) is the input value of the equivalent circuit simulation parameter, V(Y) is the unconditional variance of the output, Y is the response function of the equivalent circuit simulation parameter, E x(t) is the expected value of the equivalent circuit simulation parameter, V x(t) is the variance operator of the input equivalent circuit simulation parameters;

[0049] Obtain at least one set of sensitive parameters and equivalent circuit simulation parameters corresponding to the sensitive parameters, and calculate output prediction sequence parameters based on the objective constraint function of the battery simulation model;

[0050] Among them, the objective constraint function is:

[0051]

[0052] Where f(x) is the output prediction sequence parameter, n is the number of simulation cycles t, k n is the forward reaction rate coefficient corresponding to the equivalent circuit simulation parameter, c is the number of parameter categories corresponding to the equivalent circuit simulation parameter, E t-1 is the battery activation energy of the previous simulation cycle, R is the gas constant, T is the battery temperature of the previous simulation cycle, and σ is the coupling coefficient of the objective constraint function.

[0053] Preferably, the method for constructing a battery health assessment model based on a support vector machine combined with integrated empirical mode decomposition specifically includes:

[0054] Taking the predicted sequence parameters as model samples, the model samples are divided into training sets and test sets;

[0055] The initial model of the battery health assessment model is constructed in combination with the Arrhenius law. The initial model is a neural network model that combines the support vector machine and the particle swarm PSO algorithm. The particle swarm PSO algorithm is used to extract and determine the sensitivity coefficient of the training set to obtain the sensitivity coefficient set corresponding to the training set.

[0056] Add the sensitivity coefficient set to the training set, calculate the model gradient, and iteratively update the initial model until the loss function of the initial model reaches Nash equilibrium;

[0057] Get the test set, use the test set as input, execute the initial model, and use the test set to determine whether the loss function of the initial model converges. If converged, output the trained battery health assessment model.

[0058] Preferably, the calculation obtains an operating state value characterizing the current operating state of the battery and a health state value characterizing the current health state, and the calculation and evaluation of the predicted life coefficient based on the operating state value and the health state value specifically includes:

[0059] Obtain prediction sequence parameters, and the battery health assessment model extracts sensitivity coefficients of the prediction sequence parameters;

[0060] Among them, the sensitivity coefficient extraction function is:

[0061]

[0062] Among them, G(x) is the minimization and maximization decision function, f τ (x) is the dimension-reduced sequence of the prediction sequence parameters, λ is the extraction coefficient of the sensitivity coefficient extraction function;

[0063]

[0064] Among them, f(x) is the input value of the prediction sequence parameter, R is the data dimension of the input value of the prediction sequence parameter, is the mean of the predicted sequence parameters;

[0065] determining sensitivity coefficients associated with the battery's current operating condition, current health, and remaining predicted life;

[0066] The battery health assessment model performs forward uncertainty analysis on the battery simulation model based on the Monte Carlo algorithm, and calculates the operating state value that represents the current operating condition of the battery and the health state value of the current health condition;

[0067] The running status value and health status value are calculated by the following formula:

[0068]

[0069] Among them, Y Z is the battery operating status value at time t, Q d is the initial capacity of the battery, u x is the number of battery charge and discharge cycles, κ is the battery capacity attenuation rate;

[0070]

[0071] Calculate and evaluate the predicted life coefficient based on the operating status value and health status value;

[0072] Among them, the predicted life coefficient is calculated by the following formula:

[0073]

[0074] Among them, α is the number of prediction sequence parameters, and β is the data dimension value of the prediction sequence parameters.

[0075] The present invention also provides a performance testing system for a solid oxide fuel cell, the performance testing system for a solid oxide fuel cell specifically comprising:

[0076] A preprocessing module is used to obtain the timing parameters of the fuel cell performance test, preprocess the timing parameters, and obtain a real-time preprocessing set;

[0077] A parameter prediction module, which constructs and trains a battery simulation model based on a particle swarm algorithm, loads a real-time preprocessing set, takes the real-time preprocessing set as input, executes the battery simulation model, and outputs prediction sequence parameters;

[0078] Model building module, which builds a battery health assessment model based on support vector machine combined with integrated empirical mode decomposition, iteratively trains the battery health assessment model, obtains prediction sequence parameters, and extracts sensitivity coefficients of the prediction sequence parameters from the battery health assessment model to determine the sensitivity coefficients associated with the current operating status, current health status, and remaining predicted life of the battery;

[0079] The battery evaluation module uses the battery health evaluation model to perform forward uncertainty analysis on the battery simulation model based on the Monte Carlo algorithm, calculates the operating status value that characterizes the current operating status of the battery and the health status value of the current health status, and calculates and evaluates the predicted life coefficient based on the operating status value and the health status value.

[0080] Preferably, the preprocessing module comprises:

[0081] A missing value processing unit is used to load time series parameters and process missing values ​​of time series parameters;

[0082] A parameter classification unit is used to obtain the time series parameters after missing value processing, extract the nonlinear feature parameters and image feature parameters in the time series parameters, remove the nonlinear feature parameters and image feature parameters in the time series parameters, and form a linear feature parameter set;

[0083] Nonlinear decomposition unit, which performs parameter decomposition of nonlinear characteristic parameters based on modal decomposition algorithm;

[0084] An image transformation unit, which traverses image feature parameters based on an improved convolutional neural network, realizes image feature parameter transformation, and obtains an image feature transformation set;

[0085] The parameter integration unit is used to obtain a parameter decomposition set, an image feature transformation set, and a linear feature parameter set, and performs lumped integration transformation on the parameter decomposition set, the image feature transformation set, and the linear feature parameter set based on the arithmetic mean method to obtain a real-time preprocessing set including a real-time sequence, a mean sequence, and an extreme value sequence.

[0086] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0087] In an embodiment of the present invention, a battery simulation model is constructed and trained based on a particle swarm algorithm, and the timing parameters are accurately predicted and evaluated through the battery simulation model, so that the battery health assessment model evaluates and calculates the battery's operating status value, health status value and predicted life coefficient in combination with the predicted evaluation results. The prediction results based on the battery simulation model-battery health assessment model are relatively reliable and accurate, and can be used as a basis for evaluating battery performance. This overcomes the problem that the prior art only relies on battery voltage data to analyze battery life and cannot fully reflect the battery aging status and the impact of battery charge and discharge cycles on battery life, resulting in large errors in the analysis results and inaccurate evaluations.

[0088] In the embodiments of the present invention, a modal decomposition algorithm is used to decompose the nonlinear characteristic parameters, and an improved convolutional neural network is used to transform the image characteristic parameters, thereby improving the efficiency of parameter decomposition and transformation by leveraging the advantages of the modal decomposition algorithm and the improved convolutional neural network, thereby significantly improving the performance, interpretability and generalization ability of the battery simulation model, and reducing the computational complexity.

[0089] In the embodiment of the present invention, the equivalent circuit model is used as the initial model of the battery simulation model. The equivalent circuit model is built based on the Simulink model, and a fuzzy logic balancing controller and a leader particle dynamic selection strategy are configured in the equivalent circuit model to simplify the analysis of the positive and negative poles of the battery. The constructed battery simulation model is relatively simple to calculate and has high simulation accuracy, so that the battery prediction sequence parameters can be accurately evaluated.

[0090] In an embodiment of the present invention, a battery health assessment model and a training method thereof are provided. The battery health assessment model is constructed in combination with the Arrhenius law, and the particle swarm PSO algorithm is used to extract and determine the sensitivity coefficient of the training set, so that the battery health assessment model obtained by combining the support vector machine and the particle swarm PSO algorithm training can improve the prediction accuracy and generalization ability of the model, and can quickly find the optimal solution, thereby shortening the training time and reducing the consumption of computing resources, so as to more accurately evaluate the battery life. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 It is a schematic diagram of the implementation flow of the performance analysis method of the solid oxide fuel cell provided by the present invention.

[0092] Figure 2 A schematic diagram of the implementation flow of the method for preprocessing timing parameters is shown.

[0093] Figure 3 A schematic diagram of the implementation process of the method for traversing image feature parameters based on an improved convolutional neural network to achieve image feature parameter transformation is shown.

[0094] Figure 4 A schematic diagram of the implementation process of the method for constructing and training a battery simulation model based on a particle swarm algorithm is shown.

[0095] Figure 5 A schematic diagram of the implementation flow of the method for executing the battery simulation model and outputting prediction sequence parameters is shown.

[0096] Figure 6 A schematic diagram of the implementation process of the method for constructing a battery health assessment model based on support vector machine combined with integrated empirical mode decomposition is shown.

[0097] Figure 7 The invention shows a flowchart of the method for calculating the operating state value representing the current operating state of the battery and the health state value of the current health state, and calculating and evaluating the predicted life coefficient based on the operating state value and the health state value.

[0098] Figure 8 It is a structural schematic diagram of a performance testing system for a solid oxide fuel cell provided by the present invention. DETAILED DESCRIPTION

[0099] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of this application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0100] The embodiment of the present invention provides a method for analyzing the performance of a solid oxide fuel cell. Figure 1 The schematic diagram of the implementation process of the performance analysis method of the solid oxide fuel cell is shown. The performance analysis method of the solid oxide fuel cell specifically includes:

[0101] Step S10, obtaining timing parameters of fuel cell performance test, wherein the timing parameters include battery voltage data, current data, current density, battery temperature, gas composition parameters, battery material characterization data, and preprocessing the timing parameters to obtain a real-time preprocessing set;

[0102] It should be noted that the timing parameters include but are not limited to battery voltage data, current data, current density, battery temperature, gas composition parameters, and battery material characterization data that characterize the performance of the fuel cell. Battery material characterization can analyze the microstructure of electrodes and electrolytes through X-ray diffraction (XRD), scanning electron microscopy (SEM), etc., so as to understand the impact of battery material characteristics on battery performance.

[0103] Step S20, constructing and training a battery simulation model based on a particle swarm algorithm, loading a real-time preprocessing set, executing the battery simulation model with the real-time preprocessing set as input, and outputting prediction sequence parameters;

[0104] Step S30, constructing a battery health assessment model based on a support vector machine combined with integrated empirical mode decomposition, iteratively training the battery health assessment model, obtaining prediction sequence parameters, and extracting sensitivity coefficients of the prediction sequence parameters by the battery health assessment model to determine the sensitivity coefficients associated with the current operating status, current health status, and remaining predicted life of the battery;

[0105] Step S40, the battery health assessment model performs forward uncertainty analysis on the battery simulation model based on the Monte Carlo algorithm, calculates the operating state value characterizing the current operating state of the battery and the health state value of the current health state, and calculates and evaluates the predicted life coefficient based on the operating state value and the health state value.

[0106] In an embodiment of the present invention, a battery simulation model is constructed and trained based on a particle swarm algorithm, and the timing parameters are accurately predicted and evaluated through the battery simulation model, so that the battery health assessment model evaluates and calculates the battery's operating status value, health status value and predicted life coefficient in combination with the predicted evaluation results. The prediction results based on the battery simulation model-battery health assessment model are relatively reliable and accurate, and can be used as a basis for evaluating battery performance. This overcomes the problem that the prior art only relies on battery voltage data to analyze battery life and cannot fully reflect the battery aging status and the impact of battery charge and discharge cycles on battery life, resulting in large errors in the analysis results and inaccurate evaluations.

[0107] The embodiment of the present invention provides a method for preprocessing timing parameters. Figure 2 The figure shows a schematic diagram of the implementation process of the method for preprocessing the timing parameters, and the method for preprocessing the timing parameters specifically includes:

[0108] Step S101, loading the timing parameters, and processing the missing values ​​of the timing parameters, wherein the missing value processing includes feature removal, mean interpolation method and spline interpolation method processing;

[0109] It should be noted that correlation analysis can also be used to handle missing values ​​of time series parameters. Missing value processing can improve data integrity, ensure the accuracy of analysis and modeling, and thus improve the reliability of results.

[0110] Step S102, obtaining the time series parameters after the missing value processing, extracting the nonlinear feature parameters and image feature parameters in the time series parameters, removing the nonlinear feature parameters and image feature parameters in the time series parameters, and forming a linear feature parameter set;

[0111] In an embodiment of the present invention, by classifying and processing the nonlinear feature parameters, image feature parameters, and linear feature parameters in the time series parameters, it is possible to facilitate the analysis of different types of data, thereby enhancing the robustness of the data and reducing data prediction evaluation errors.

[0112] Step S103, performing parameter decomposition on the nonlinear characteristic parameters based on a modal decomposition algorithm;

[0113] Step S104, traversing the image feature parameters based on the improved convolutional neural network, implementing image feature parameter transformation, and obtaining an image feature transformation set;

[0114] Among them, the self-attention mechanism module is introduced into the improved convolutional neural network, and the convolutional neural network is improved through the self-attention mechanism module;

[0115] Step S105, obtaining a parameter decomposition set, an image feature transformation set, and a linear feature parameter set, and performing a lumped integration transformation on the parameter decomposition set, the image feature transformation set, and the linear feature parameter set based on the arithmetic mean method to obtain a real-time preprocessing set including a real-time sequence, a mean sequence, and an extreme value sequence.

[0116] In the embodiments of the present invention, a modal decomposition algorithm is used to decompose the nonlinear characteristic parameters, and an improved convolutional neural network is used to transform the image characteristic parameters, thereby improving the efficiency of parameter decomposition and transformation by leveraging the advantages of the modal decomposition algorithm and the improved convolutional neural network, thereby significantly improving the performance, interpretability and generalization ability of the battery simulation model, and reducing the computational complexity.

[0117] Among them, the modal decomposition algorithm is a wavelet transform algorithm. When the wavelet transform algorithm performs parameter decomposition on nonlinear characteristic parameters:

[0118] Step S1031, obtaining nonlinear characteristic parameters x(t f ), set the wavelet basis function and decomposition level of the wavelet transform algorithm;

[0119] The wavelet basis function is expressed as:

[0120]

[0121] Wherein, a is the scale parameter of the wavelet basis function. In this embodiment, the scale parameter can be 1 or 2. f ) is the input value of the nonlinear characteristic parameter, k is the position index value of the nonlinear characteristic parameter, and the position index value is greater than 1;

[0122] Step S1032, decomposing the nonlinear characteristic parameters into high-frequency detail components and low-frequency approximation components at a first decomposition level based on continuous wavelet transform;

[0123] Step S1033, at the second decomposition level, obtain low-frequency approximate components, decompose the low-frequency approximate components again based on continuous wavelet transform until a preset decomposition level is reached, integrate multiple groups of high-frequency approximate components and a group of low-frequency approximate components obtained within the preset decomposition level, and obtain a parameter decomposition set of nonlinear characteristic parameters.

[0124] The embodiment of the present invention provides a method for transforming image feature parameters by traversing image feature parameters based on an improved convolutional neural network. Figure 3 The figure shows a schematic diagram of a method for implementing image feature parameter transformation by traversing image feature parameters based on an improved convolutional neural network. The method for implementing image feature parameter transformation by traversing image feature parameters based on an improved convolutional neural network specifically includes:

[0125] Step S1041, obtaining image feature parameters, and extracting clear image features from the image feature parameters;

[0126] It should be noted that when extracting clear image features in image feature parameters, a ResNet network, an ORB algorithm or a BOF algorithm can be used to extract clear image features.

[0127] Step S1042, refine and decompose the clear image features, and mark the feature decomposition time. The output function is recorded as:

[0128]

[0129] Among them, t f (x) is the characteristic decomposition time, x(t p ) is the refined decomposition result of the clear image feature, M is the number of factors affecting the decomposition of the image feature information, in this embodiment, the number of factors affecting the image feature information includes but is not limited to image size, color, contour edge, texture information, color histogram, moment feature, entropy information, C Mis the image feature information decomposition influence weight, which may be 0.1-0.56, and γ is the fuzzy denoising coefficient when performing fine decomposition on clear image features. In this embodiment, the fuzzy denoising constant may be 0.1-0.4;

[0130]

[0131] Among them, W x The weight matrix of the refined decomposition result of the clear image features in the image feature parameter set, To improve the data gradient of the convolutional neural network, e is the decomposition error coefficient when the convolutional neural network is improved to refine the decomposition of the clear image features. In this embodiment, the decomposition error coefficient may be 0.02-0.05;

[0132] Step S1043, improving the convolutional neural network to perform dilated convolution on the refined decomposition result to increase the receptive field;

[0133] The dilated convolution result is expressed as:

[0134]

[0135] Among them, J(x t ) is the result of the dilated convolution, f(i) is the size of the convolution matrix in the improved convolutional neural network, the size of the convolution matrix is ​​calculated by inputting the matrix size, the convolution kernel size, the convolution step size, and the padding, n is the size of the convolution kernel, in this embodiment, the size of the convolution kernel can be 3*3, d is the expansion factor during the dilated convolution, m is the number of convolution kernel convolution layers, and the number of convolution layers can be 3;

[0136] In step S1044, the dilated convolution result after the dilated convolution is input into the self-attention layer. The self-attention layer redistributes the weights of the dilated convolution result based on the self-attention mechanism to obtain a set of image feature transformations.

[0137] It should be noted that when the self-attention layer redistributes the weights of the dilated convolution results based on the self-attention mechanism, the self-attention mechanism can mine the dependencies within the image parameter sequence based on the weight scoring mechanism, effectively improving the model's ability to process complex sequence data.

[0138] The embodiment of the present invention provides a method for constructing and training a battery simulation model based on a particle swarm algorithm. Figure 4 The schematic diagram of the implementation process of the method for constructing and training a battery simulation model based on a particle swarm algorithm is shown. The method for constructing and training a battery simulation model based on a particle swarm algorithm specifically includes:

[0139] Step S201, using an equivalent circuit model as an initial model of a battery simulation model, traversing a real-time preprocessing set, and extracting equivalent circuit associated parameters from the real-time preprocessing set;

[0140] It should be noted that the use of an equivalent circuit model to simulate the nonlinear working characteristics of solid oxide fuel cells is relatively simple in calculation and has high simulation accuracy. It is currently the most practical solid oxide fuel cell modeling method. The parameters of the equivalent circuit model can be relatively easily identified through standard electrical testing methods. These tests can provide the resistance, capacitance and other parameters required by the model, making the model establishment and verification process more convenient.

[0141] Step S202, building an equivalent circuit model based on the Simulink model, and configuring a fuzzy logic balance controller and a leader particle dynamic selection strategy in the equivalent circuit model;

[0142] In this embodiment, MATLAB / Simulink tools can be used to build an equivalent circuit model based on the Simulink model, so that the equivalent circuit model of the battery can be easily built and simulated. For example, a 2RC model is used to realize the relationship between SOC and OCV and the change of RC parameters through a table lookup method.

[0143] Step S203, setting the charge and discharge current of the battery simulation model, simulating the voltage response of the battery simulation model under the simulated working condition, and obtaining equivalent circuit simulation parameters;

[0144] Step S204, obtaining equivalent circuit associated parameters and equivalent circuit simulation parameters, and verifying the accuracy of the equivalent circuit simulation parameters based on the equivalent circuit associated parameters;

[0145] Step S205, determining whether the parameter similarity meets a preset similarity threshold;

[0146] In this embodiment, the similarity threshold may be set to ±0.05.

[0147] Step S206: If the conditions are met, the set of equivalent circuit simulation parameters is not adjusted;

[0148] If not, the battery simulation model is adjusted using a fuzzy logic balancing controller and the process returns to step S203.

[0149] In the embodiment of the present invention, the equivalent circuit model is used as the initial model of the battery simulation model. The equivalent circuit model is built based on the Simulink model, and a fuzzy logic balancing controller and a leader particle dynamic selection strategy are configured in the equivalent circuit model to simplify the analysis of the positive and negative poles of the battery. The constructed battery simulation model is relatively simple to calculate and has high simulation accuracy, so that the battery prediction sequence parameters can be accurately evaluated.

[0150] The embodiment of the present invention provides a method for executing the battery simulation model and outputting prediction sequence parameters. Figure 5 The schematic diagram of the implementation flow of the method for executing the battery simulation model and outputting the prediction sequence parameters is shown. The method for executing the battery simulation model and outputting the prediction sequence parameters specifically includes:

[0151] Step S301, obtaining a real-time preprocessing set, and extracting equivalent circuit associated parameters in the real-time preprocessing set by a battery simulation model;

[0152] Step S302, taking the equivalent circuit associated parameters as input, the battery simulation model performs a global sensitivity analysis on the equivalent circuit simulation parameters based on the Sobol′ model, and selects the equivalent circuit simulation parameters with a total sensitivity of more than 0.005 as sensitive parameters;

[0153] In this embodiment, the battery simulation model performs global sensitivity analysis on equivalent circuit simulation parameters based on the Sobol' model to evaluate the importance of equivalent circuit associated parameters. Sobol' is a global sensitivity calculation method based on variance. At the same time, when screening equivalent circuit simulation parameters that meet the total sensitivity of more than 0.005 as sensitive parameters, the current density sensitivity range is [50-1600], the battery porosity sensitivity range is [0.15-0.68], the anode flow sensitivity range is [0.2-1], the cathode flow sensitivity range is [0.5-1], the battery temperature sensitivity range is [600-1000], and the electrolyte thickness sensitivity range is [20-550].

[0154] Among them, the sensitivity index of the equivalent circuit simulation parameter is calculated by the following formula:

[0155]

[0156] Among them, S x is the sensitivity of the equivalent circuit simulation parameter, x(t) is the input value of the equivalent circuit simulation parameter, V(Y) is the unconditional variance of the output, Y is the response function of the equivalent circuit simulation parameter, in this embodiment, the response function can be a frequency response function or an admittance function, E x(t) is the expected value of the equivalent circuit simulation parameter, V x(t) is the variance operator of the input equivalent circuit simulation parameters;

[0157] Step S303, obtaining at least one set of sensitive parameters and equivalent circuit simulation parameters corresponding to the sensitive parameters, and calculating output prediction sequence parameters based on the objective constraint function of the battery simulation model;

[0158] Among them, the objective constraint function is:

[0159]

[0160] Where f(x) is the output prediction sequence parameter, n is the number of simulation cycles t, k n is the forward reaction rate coefficient corresponding to the equivalent circuit simulation parameter. In this embodiment, the forward reaction rate constant can be determined by linear sweep voltammetry (LSV). c is the number of parameter categories corresponding to the equivalent circuit simulation parameter. In this embodiment, the number of parameter categories is 3-10. E t-1 is the battery activation energy of the previous simulation cycle, R is the gas constant, T is the battery temperature of the previous simulation cycle, σ is the coupling coefficient of the objective constraint function, and the coupling coefficient of the objective constraint function is set to 0.02-0.03.

[0161] The embodiment of the present invention provides a method for constructing a battery health assessment model based on a support vector machine combined with integrated empirical mode decomposition. Figure 6 The schematic diagram of the implementation process of the method for constructing a battery health assessment model based on a support vector machine combined with integrated empirical mode decomposition is shown. The method for constructing a battery health assessment model based on a support vector machine combined with integrated empirical mode decomposition specifically includes:

[0162] Step S401, taking the predicted sequence parameters as model samples, dividing the model samples into a training set and a test set;

[0163] It should be noted that the ratio of the training set to the test set can be 3:1, and the model sample sampling method can be a random cross sampling method.

[0164] Step S402, constructing an initial model of the battery health assessment model in combination with the Arrhenius law, wherein the initial model is a neural network model combining a support vector machine and a particle swarm PSO algorithm, and using the particle swarm PSO algorithm to extract and determine the sensitivity coefficient of the training set to obtain a sensitivity coefficient set corresponding to the training set;

[0165] In this embodiment, the supervision algorithm of the initial model is the radial basis function of the support vector machine, the activation function of the initial model is the Sidmoid activation function, the loss function is the cross entropy loss function, and the learning rate of the initial model can be set to 5e-2.

[0166] Step S403, adding the sensitivity coefficient set to the training set, calculating the model gradient, and iteratively updating the initial model until the loss function of the initial model reaches Nash equilibrium;

[0167] Step S404, obtaining a test set, taking the test set as input, executing the initial model, and using the test set to determine whether the loss function of the initial model converges;

[0168] Step S405: if converged, output the trained battery health assessment model.

[0169] If it has not converged, the learning rate of the initial model is reinitialized to make the loss function converge until the accuracy meets the requirement, and then return to step S403.

[0170] In an embodiment of the present invention, a battery health assessment model and a training method thereof are provided. The battery health assessment model is constructed in combination with the Arrhenius law, and the particle swarm PSO algorithm is used to extract and determine the sensitivity coefficient of the training set, so that the battery health assessment model obtained by combining the support vector machine and the particle swarm PSO algorithm training can improve the prediction accuracy and generalization ability of the model, and can quickly find the optimal solution, thereby shortening the training time and reducing the consumption of computing resources, so as to more accurately evaluate the battery life.

[0171] The embodiment of the present invention provides a method for calculating an operating state value representing the current operating state of a battery and a health state value representing the current health state, and calculating and evaluating a life prediction coefficient based on the operating state value and the health state value. Figure 7 The method for calculating the running state value and the health state value of the current health state of the battery to characterize the current running state of the battery is shown, and the method for calculating the estimated life prediction coefficient based on the running state value and the health state value is shown. The method for calculating the running state value and the health state value to characterize the current running state of the battery to characterize the current health state of the battery is shown, and the estimated life prediction coefficient is calculated based on the running state value and the health state value, specifically including:

[0172] Step S501, obtaining prediction sequence parameters, and extracting sensitivity coefficients of the prediction sequence parameters by the battery health assessment model;

[0173] Among them, the sensitivity coefficient extraction function is:

[0174]

[0175] Wherein, G(x) is a minimax decision function. In this embodiment, the minimax decision function may be an iteratively deepened minimax algorithm, f τ (x) is the dimension reduction sequence of the prediction sequence parameters, λ is the extraction coefficient of the sensitivity coefficient extraction function, and the extraction coefficient of the sensitivity coefficient extraction function can be 0.1-0.6;

[0176]

[0177] Among them, f(x) is the input value of the prediction sequence parameter, R is the data dimension of the input value of the prediction sequence parameter, is the mean of the predicted sequence parameters;

[0178] Step S502, determining a sensitivity coefficient associated with the current operating condition, current health condition, and remaining predicted life of the battery;

[0179] Step S503, the battery health assessment model performs forward uncertainty analysis on the battery simulation model based on the Monte Carlo algorithm, and calculates an operating state value representing the current operating state of the battery and a health state value representing the current health state;

[0180] The running status value and health status value are calculated by the following formula:

[0181]

[0182] Among them, Y Z is the battery operating status value at time t, Q d is the initial capacity of the battery, u x is the number of battery charge and discharge cycles, κ is the battery capacity attenuation rate, which is measured in years and can be 1-10%;

[0183]

[0184] Step S504, calculating and evaluating the predicted life coefficient based on the operating state value and the health state value;

[0185] Among them, the predicted life coefficient is calculated by the following formula:

[0186]

[0187] Among them, α is the number of prediction sequence parameters, and β is the data dimension value of the prediction sequence parameters.

[0188] In an embodiment of the present invention, a forward uncertainty analysis is performed on a battery simulation model through a Monte Carlo algorithm, and an operating status value characterizing the current operating condition of the battery is calculated, so that the operating status value accurately evaluates and characterizes the battery's state of charge, power state, functional state, and information state. At the same time, a comprehensive analysis of the current health state and predicted life of the battery is performed in combination with the health status value, which helps to understand the remaining service life of the battery, facilitates timely replacement or repair of the battery, and ensures the reliability of the system.

[0189] The embodiment of the present invention provides a performance testing system for a solid oxide fuel cell. Figure 8 The structural schematic diagram of the solid oxide fuel cell performance test system is shown, and the solid oxide fuel cell performance test system specifically includes:

[0190] The preprocessing module 100 is used to obtain the timing parameters of the fuel cell performance test, preprocess the timing parameters, and obtain a real-time preprocessing set;

[0191] The parameter prediction module 200 constructs and trains a battery simulation model based on a particle swarm algorithm, loads a real-time preprocessing set, takes the real-time preprocessing set as input, executes the battery simulation model, and outputs a prediction sequence parameter;

[0192] The model building module 300 builds a battery health assessment model based on a support vector machine combined with integrated empirical mode decomposition, iteratively trains the battery health assessment model, obtains prediction sequence parameters, and extracts sensitivity coefficients of the prediction sequence parameters from the battery health assessment model to determine the sensitivity coefficients associated with the current operating status, current health status, and remaining predicted life of the battery;

[0193] The battery evaluation module 400 uses a battery health evaluation model to perform forward uncertainty analysis on a battery simulation model based on a Monte Carlo algorithm, calculates an operating state value representing the current operating state of the battery and a health state value representing the current health state, and calculates an evaluation prediction life coefficient based on the operating state value and the health state value.

[0194] It should be noted that the preprocessing module 100, the parameter prediction module 200, the model building module 300, and the battery evaluation module 400 can use 5G communication or DTU communication to realize data exchange and transmission.

[0195] In this embodiment, the preprocessing module 100 includes:

[0196] A missing value processing unit 110 is used to load timing parameters and process missing values ​​of timing parameters;

[0197] The parameter classification unit 120 is used to obtain the time series parameters after the missing value processing, extract the nonlinear feature parameters and image feature parameters in the time series parameters, remove the nonlinear feature parameters and image feature parameters in the time series parameters, and form a linear feature parameter set;

[0198] A nonlinear decomposition unit 130 performs parameter decomposition on nonlinear characteristic parameters based on a modal decomposition algorithm;

[0199] An image transformation unit 140 traverses image feature parameters based on an improved convolutional neural network to transform image feature parameters and obtain an image feature transformation set;

[0200] The parameter integration unit 150 is used to obtain a parameter decomposition set, an image feature transformation set, and a linear feature parameter set, and performs a lumped integration transformation on the parameter decomposition set, the image feature transformation set, and the linear feature parameter set based on the arithmetic mean method to obtain a real-time preprocessing set including a real-time sequence, a mean sequence, and an extreme value sequence.

[0201] Another aspect of the embodiments of the present invention further provides a computer-readable storage medium, which stores computer program instructions, and the computer program instructions can be executed by a processor. When the computer program instructions are executed, the method of any one of the above embodiments is implemented.

[0202] In another aspect of the embodiments of the present invention, a computer device is provided. The computer device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the method of any one of the above embodiments is implemented.

[0203] As a non-volatile computer-readable storage medium, the memory can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as the program instructions / modules corresponding to the performance analysis method of the solid oxide fuel cell in the embodiment of the present application. The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required for at least one function. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the local module via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0204] Finally, it should be noted that the computer-readable storage medium (e.g., memory) herein may be a volatile memory or a nonvolatile memory, or may include both a volatile memory and a nonvolatile memory. As an example and not by way of limitation, a nonvolatile memory may include a read-only memory (ROM), a programmable ROM (PROM), an electrically programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), or a flash memory. A volatile memory may include a random access memory (RAM), which may act as an external cache memory. As an example and not by way of limitation, RAM may be obtained in a variety of forms, such as synchronous RAM (DRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The storage devices of the disclosed aspects are intended to include, but are not limited to, these and other suitable types of memory.

[0205] In summary, the present invention provides a performance analysis method and a test system for a solid oxide fuel cell. In an embodiment of the present invention, a battery simulation model is constructed and trained based on a particle swarm algorithm, and the timing parameters are accurately predicted and evaluated through the battery simulation model, so that the battery health assessment model evaluates and calculates the battery's operating status value, health status value and predicted life coefficient in combination with the predicted evaluation results. The prediction results based on the battery simulation model-battery health assessment model are relatively reliable and accurate, and can be used as a basis for evaluating battery performance. This overcomes the problem that the prior art only relies on battery voltage data to analyze battery life and cannot fully reflect the battery aging status and the impact of the battery charge and discharge cycle on the battery life, resulting in large errors in the analysis results and inaccurate evaluation.

[0206] It should be noted that, for the above-mentioned embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described order of actions, because according to the present invention, some steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0207] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field can still combine, add, delete or make other adjustments to the features in the various embodiments of the present invention according to the circumstances without conflict, without making creative work, so as to obtain different other technical solutions that do not deviate from the concept of the present invention in essence, and these technical solutions also belong to the scope of protection of the present invention.

Claims

1. A method for analyzing the performance of a solid oxide fuel cell, characterized in that: The performance analysis method of the solid oxide fuel cell comprises: Obtaining fuel cell performance test timing parameters, wherein the timing parameters include battery voltage data, current data, current density, battery temperature, gas composition parameters, and battery material characterization data, and preprocessing the timing parameters to obtain a real-time preprocessing set; Building and training a battery simulation model based on a particle swarm algorithm, loading a real-time preprocessing set, executing the battery simulation model with the real-time preprocessing set as input, and outputting prediction sequence parameters; A battery health assessment model is constructed based on support vector machine combined with integrated empirical mode decomposition. The battery health assessment model is iteratively trained to obtain prediction sequence parameters. The battery health assessment model extracts the sensitivity coefficients of the prediction sequence parameters and determines the sensitivity coefficients associated with the current operating status, current health status and remaining predicted life of the battery. The battery health assessment model performs forward uncertainty analysis on the battery simulation model based on the Monte Carlo algorithm, calculates the operating status value that characterizes the current operating status of the battery and the health status value of the current health status, and calculates and evaluates the predicted life coefficient based on the operating status value and the health status value.

2. The method for analyzing the performance of a solid oxide fuel cell according to claim 1, wherein: The method for preprocessing the timing parameters specifically includes: Loading time series parameters, processing missing values ​​of time series parameters, wherein the missing value processing includes feature removal, mean interpolation method and spline interpolation method processing; Obtaining the time series parameters after missing value processing, extracting the nonlinear feature parameters and image feature parameters in the time series parameters, removing the nonlinear feature parameters and image feature parameters in the time series parameters, and forming a linear feature parameter set; Parameter decomposition of nonlinear characteristic parameters based on modal decomposition algorithm; Among them, the modal decomposition algorithm is a wavelet transform algorithm. When the wavelet transform algorithm performs parameter decomposition on nonlinear characteristic parameters: Get the nonlinear characteristic parameter x(t f ), set the wavelet basis function and decomposition level of the wavelet transform algorithm; The wavelet basis function is expressed as: Among them, a is the scale parameter of the wavelet basis function, x(t f ) is the input value of the nonlinear characteristic parameter, k is the position index value of the nonlinear characteristic parameter; Based on continuous wavelet transform, the nonlinear characteristic parameters are decomposed into high-frequency detail components and low-frequency approximate components at the first decomposition level; At the second decomposition level, low-frequency approximate components are obtained, and the low-frequency approximate components are decomposed again based on continuous wavelet transform until a preset decomposition level is reached, and multiple groups of high-frequency approximate components and a group of low-frequency approximate components obtained within the preset decomposition level are integrated to obtain a parameter decomposition set of nonlinear characteristic parameters.

3. The method for analyzing the performance of a solid oxide fuel cell according to claim 2, wherein: The method for preprocessing the timing parameters specifically includes: Based on the improved convolutional neural network, the image feature parameters are traversed to achieve the image feature parameter transformation and obtain the image feature transformation set; Among them, the self-attention mechanism module is introduced into the improved convolutional neural network, and the convolutional neural network is improved through the self-attention mechanism module; Obtain parameter decomposition set, image feature transformation set, and linear feature parameter set, perform lumped integration transformation on the parameter decomposition set, image feature transformation set, and linear feature parameter set based on the arithmetic mean method, and obtain a real-time preprocessing set including a real-time sequence, a mean sequence, and an extreme value sequence.

4. The method for analyzing the performance of a solid oxide fuel cell according to claim 3, wherein: The method for implementing image feature parameter transformation based on traversing image feature parameters by an improved convolutional neural network specifically includes: Obtain image feature parameters, and extract clear image features from the image feature parameters; The clear image features are refined and decomposed, and the feature decomposition time is marked. The output function is recorded as: Among them, t f (x) is the characteristic decomposition time, x(t p ) is the refined decomposition result of the clear image features, M is the number of factors affecting the decomposition of image feature information, C M is the influence weight of image feature information decomposition, γ is the fuzzy denoising coefficient when performing fine decomposition of clear image features; Among them, W x The weight matrix of the refined decomposition result of the clear image features in the image feature parameter set, is to improve the data gradient of the convolutional neural network, and e is the decomposition error coefficient when the improved convolutional neural network performs fine decomposition on the clear image features; Improve the convolutional neural network to perform dilated convolution on the refined decomposition results to increase the receptive field; The dilated convolution result is expressed as: Among them, J(x t ) is the result of dilated convolution, f(i) is the size of the convolution matrix in the improved convolutional neural network, n is the size of the convolution kernel, d is the expansion factor during dilated convolution, and m is the number of convolution kernel convolution layers; The dilated convolution result after the dilated convolution is input into the self-attention layer. The self-attention layer redistributes the weights of the dilated convolution result based on the self-attention mechanism to obtain a set of image feature transformations.

5. The method for analyzing the performance of a solid oxide fuel cell according to claim 1, wherein: The method for constructing and training a battery simulation model based on a particle swarm algorithm specifically includes: Taking the equivalent circuit model as the initial model of the battery simulation model, traversing the real-time preprocessing set, and extracting the equivalent circuit associated parameters from the real-time preprocessing set; Build an equivalent circuit model based on the Simulink model, and configure the fuzzy logic balance controller and leader particle dynamic selection strategy in the equivalent circuit model; Setting the charge and discharge current of the battery simulation model, simulating the voltage response of the battery simulation model under the simulated working conditions, and obtaining equivalent circuit simulation parameters; Obtain equivalent circuit associated parameters and equivalent circuit simulation parameters, verify the accuracy of equivalent circuit simulation parameters based on the equivalent circuit associated parameters, and determine whether the parameter similarity meets the preset similarity threshold. If so, the group of equivalent circuit simulation parameters will not be adjusted. If not, a fuzzy logic balancing controller will be used to adjust the battery simulation model.

6. The method for analyzing the performance of a solid oxide fuel cell according to claim 5, wherein: The method of executing the battery simulation model and outputting the prediction sequence parameters specifically includes: Obtain a real-time preprocessing set, and the battery simulation model extracts the equivalent circuit associated parameters in the real-time preprocessing set; Taking the equivalent circuit associated parameters as input, the battery simulation model performs a global sensitivity analysis on the equivalent circuit simulation parameters based on the Sobol′ model, and selects the equivalent circuit simulation parameters with a total sensitivity of more than 0.005 as sensitive parameters; Among them, the sensitivity index of the equivalent circuit simulation parameter is calculated by the following formula: Among them, S x is the sensitivity of the equivalent circuit simulation parameter, x(t) is the input value of the equivalent circuit simulation parameter, V(Y) is the unconditional variance of the output, Y is the response function of the equivalent circuit simulation parameter, E x(t) is the expected value of the equivalent circuit simulation parameter, V x(t) is the variance operator of the input equivalent circuit simulation parameters; Obtain at least one set of sensitive parameters and equivalent circuit simulation parameters corresponding to the sensitive parameters, and calculate output prediction sequence parameters based on the objective constraint function of the battery simulation model; Among them, the objective constraint function is: Where f(x) is the output prediction sequence parameter, n is the number of simulation cycles t, k n is the forward reaction rate coefficient corresponding to the equivalent circuit simulation parameter, c is the number of parameter categories corresponding to the equivalent circuit simulation parameter, E t-1 is the battery activation energy of the previous simulation cycle, R is the gas constant, T is the battery temperature of the previous simulation cycle, and σ is the coupling coefficient of the objective constraint function.

7. The method for analyzing the performance of a solid oxide fuel cell according to claim 1, wherein: The method for constructing a battery health assessment model based on a support vector machine combined with integrated empirical mode decomposition specifically includes: Taking the predicted sequence parameters as model samples, the model samples are divided into training sets and test sets; The initial model of the battery health assessment model is constructed in combination with the Arrhenius law. The initial model is a neural network model that combines the support vector machine and the particle swarm PSO algorithm. The particle swarm PSO algorithm is used to extract and determine the sensitivity coefficient of the training set to obtain the sensitivity coefficient set corresponding to the training set. Add the sensitivity coefficient set to the training set, calculate the model gradient, and iteratively update the initial model until the loss function of the initial model reaches Nash equilibrium; Get the test set, use the test set as input, execute the initial model, and use the test set to determine whether the loss function of the initial model converges. If converged, output the trained battery health assessment model.

8. The method for analyzing the performance of a solid oxide fuel cell according to claim 7, wherein: The calculation obtains an operating state value representing the current operating state of the battery and a health state value representing the current health state, and calculates and evaluates a predicted life coefficient based on the operating state value and the health state value, specifically including: Obtain prediction sequence parameters, and the battery health assessment model extracts sensitivity coefficients of the prediction sequence parameters; Among them, the sensitivity coefficient extraction function is: Among them, G(x) is the minimization and maximization decision function, f τ (x) is the dimension-reduced sequence of the prediction sequence parameters, λ is the extraction coefficient of the sensitivity coefficient extraction function; Among them, f(x) is the input value of the prediction sequence parameter, R is the data dimension of the input value of the prediction sequence parameter, is the mean of the predicted sequence parameters; determining sensitivity coefficients associated with the battery's current operating condition, current health, and remaining predicted life; The battery health assessment model performs forward uncertainty analysis on the battery simulation model based on the Monte Carlo algorithm, and calculates the operating state value that represents the current operating condition of the battery and the health state value of the current health condition; The running status value and health status value are calculated by the following formula: Among them, Y Z is the battery operating status value at time t, Q d is the initial capacity of the battery, u x is the number of battery charge and discharge cycles, κ is the battery capacity attenuation rate; Calculate and evaluate the predicted life coefficient based on the operating status value and health status value; Among them, the predicted life coefficient is calculated by the following formula: Among them, α is the number of prediction sequence parameters, and β is the data dimension value of the prediction sequence parameters.

9. A solid oxide fuel cell performance testing system, used to implement the solid oxide fuel cell performance analysis method according to any one of claims 1 to 8, characterized in that: The solid oxide fuel cell performance testing system specifically comprises: A preprocessing module is used to obtain the timing parameters of the fuel cell performance test, preprocess the timing parameters, and obtain a real-time preprocessing set; A parameter prediction module, which constructs and trains a battery simulation model based on a particle swarm algorithm, loads a real-time preprocessing set, takes the real-time preprocessing set as input, executes the battery simulation model, and outputs prediction sequence parameters; Model building module, which builds a battery health assessment model based on support vector machine combined with integrated empirical mode decomposition, iteratively trains the battery health assessment model, obtains prediction sequence parameters, and extracts sensitivity coefficients of the prediction sequence parameters from the battery health assessment model to determine the sensitivity coefficients associated with the current operating status, current health status, and remaining predicted life of the battery; The battery evaluation module uses the battery health evaluation model to perform forward uncertainty analysis on the battery simulation model based on the Monte Carlo algorithm, calculates the operating status value that characterizes the current operating status of the battery and the health status value of the current health status, and calculates and evaluates the predicted life coefficient based on the operating status value and the health status value.

10. The solid oxide fuel cell performance testing system according to claim 9, characterized in that: The preprocessing module comprises: A missing value processing unit is used to load time series parameters and process missing values ​​of time series parameters; A parameter classification unit is used to obtain the time series parameters after missing value processing, extract the nonlinear feature parameters and image feature parameters in the time series parameters, remove the nonlinear feature parameters and image feature parameters in the time series parameters, and form a linear feature parameter set; Nonlinear decomposition unit, which performs parameter decomposition of nonlinear characteristic parameters based on modal decomposition algorithm; An image transformation unit, which traverses image feature parameters based on an improved convolutional neural network, realizes image feature parameter transformation, and obtains an image feature transformation set; The parameter integration unit is used to obtain a parameter decomposition set, an image feature transformation set, and a linear feature parameter set, and performs lumped integration transformation on the parameter decomposition set, the image feature transformation set, and the linear feature parameter set based on the arithmetic mean method to obtain a real-time preprocessing set including a real-time sequence, a mean sequence, and an extreme value sequence.

Images