A Life Prediction Method for a Hybrid-Driven SOFC Cogeneration System

Abstract

The present invention belongs to the technical field of solid fuel cells, and specifically relates to a method for predicting the life of a hybrid-driven SOFC combined heat and power system, including the establishment of a dynamic decay model of the SOFC system and the establishment of a life prediction model of the hybrid-driven SOFC system. The life of the system is predicted by simulating the dynamic decay of the SOFC combined heat and power system and performing machine learning, and then the performance of the entire SOFC combined heat and power system is predicted to analyze the influence of the interaction of each component on the system life under long-term operation of the system. The method of the present invention has a short test time, high accuracy, cost savings, and speeds up the R & D progress; it provides an important reference for the long-life operation of the system, can clearly analyze the influence of BOP components on the entire life of the system, and points out the direction for the long-life operation of the system; by using the method of machine learning, the accidental results that may exist in a single simulation are avoided, making the results universal.

Description

Technical Field

[0001] The invention belongs to the technical field of solid fuel cells, and particularly relates to a method for predicting the life of a hybrid-driven SOFC combined heat and power system. Background Technique

[0002] As a renewable secondary energy carrier, hydrogen energy is regarded as the cleanest energy with the greatest development potential in the 21st century. As an energy carrier with zero carbon emissions, it is receiving more and more attention. Solid Oxide Fuel Cell (SOFC) is a new generation of disruptive power generation technology and a key way to build a new energy security system, and an important starting point for achieving the dual-carbon goal. Its fuel is hydrogen and oxygen that are directly or indirectly generated, and an electrochemical reaction occurs inside the battery to generate electric energy and a large amount of heat. In order to improve the overall fuel utilization rate, it is necessary to use BOP (Balance of Plants, BOP) components such as heat exchangers to realize the recovery and utilization of heat. Therefore, in practice, the operation of SOFC often takes the form of an SOFC combined heat and power system.

[0003] At present, the prediction research on the life of SOFC often focuses on the electrochemical decay of the stack itself, without considering that the performance decay of BOP components will change the operating environment of the stack and then affect the life of the stack. Therefore, on the basis of analyzing the decay of the stack, considering the decay mechanism of coupling typical BOP components, a dynamic decay model of the SOFC combined heat and power system is formed. In addition, since the mechanism of SOFC is still not clear at present, simply relying on a mathematical model cannot accurately predict the remaining life of the system. Therefore, a data-driven + model-driven data processing method is used for machine learning, and based on this, the life of the SOFC system is predicted, which can provide an important reference for the optimal design of the SOFC system and also point the way for the long-life operation of the system. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method for predicting the life of a hybrid-driven SOFC combined heat and power system, which predicts the life of the system by simulating the dynamic decay of the SOFC combined heat and power system and performing machine learning, and then predicts the performance of the entire SOFC combined heat and power system, and analyzes the influence of the interaction of each component on the system life under long-term operation of the system.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for predicting the life of a hybrid-driven SOFC combined heat and power system, including the establishment of a dynamic decay model of the SOFC system and the establishment of a life prediction model of the hybrid-driven SOFC system; wherein the establishment of the dynamic decay model of the SOFC system includes the following steps:

[0007] S0: Determine the specific process of the SOFC thermoelectric system and the typical BOP components used;

[0008] S1: Analyze the performance degradation principle of the stack and establish a degradation model of the stack;

[0009] S2: Analyze the performance degradation principle of typical BOP components and establish a degradation model of typical BOP components;

[0010] S3: Based on the degradation model, use MATLAB software to build a dynamic degradation model of the system and verify it;

[0011] S4: Perform life prediction based on the established neural network.

[0012] The life prediction model of the hybrid-driven SOFC system is mainly based on the data results of a large number of different operating conditions simulated by the SOFC dynamic degradation model. Using the multi-dimensional time series prediction method, combined with the established neural network model for machine learning, after training and verification, the life prediction model of the SOFC system is obtained. The specific steps include:

[0013] 1) Set different boundary condition parameters to obtain the operating data set of the SOFC system under different operating conditions;

[0014] 2) Perform correlation analysis to screen the parameters that have a greater impact on the system life;

[0015] 3) Build a suitable life prediction neural network model based on the characteristics of the SOFC system operating data;

[0016] 4) Use the collected data set to train and verify the neural network. When the error meets within 10%, it is defined that the model training is successful, and the life prediction model of the system is obtained.

[0017] Preferably, in step S0, the modeling of the SOFC thermoelectric system mainly considers components such as the SOFC stack, reformer, mixer, water evaporator, heat exchanger, and tail gas combustor; The specific SOFC working process is that the fuel gas on the anode side passes through a desulfurizer for desulfurization (sulfur-containing is not considered in the modeling, assuming the sulfur content in the fuel is 0), exchanges heat through a heat exchanger and then enters the mixer to mix with water vapor (assuming the gas mixture is uniform), and then enters the reformer for reforming. The reformed gas enters the anode of the stack.

[0018] Preferably, in step S0, the cathode air exchanges heat through a heat exchanger and then enters the cathode of the stack.

[0019] Preferably, in step S0, the anode outlet tail gas first enters the tail gas combustor for combustion. The high-temperature flue gas from the combustion first provides a suitable reaction temperature for the reforming reactor, and then preheats the fuel gas before entering the mixer.

[0020] Preferably, in step S0, the cathode outlet tail gas preheats the inlet air.

[0021] Preferably, in step S1, the attenuation of the SOFC stack is mainly considered for the attenuation under continuous operating conditions. The specific electrochemical attenuation model is:

[0022] ;

[0023] ;

[0024] ;

[0025] ;

[0026] ;

[0027] ;

[0028] ;

[0029] ;

[0030] ;

[0031] ;

[0032] γ i = k i 72 X [ D P − ( D P + D s ) n ] n D s 2 D P 2 ( 1 − 1 − X 2 ) , i = a , c ;

[0033] Wherein, is the actual voltage of the SOFC stack, is the number of single cells in the stack, is the starting voltage of a single cell, is the voltage attenuation rate, is the fitting coefficient, obtained by fitting according to the actual performance of the cell, is the fuel utilization rate, is the damage variable, is the temperature inside the stack, is the ideal operating temperature of the stack, is the natural constant 2.718, i is the current density, E N is the Nernst voltage, η ohm is the ohmic loss, η act is the activation loss, η con is the concentration loss, E0 is the standard electric potential, F is the Faraday constant (96485 C mol -1 ), R is the universal gas constant (8.314 J mol -1 K -1 ), p is the gas pressure, R cell is the resistance, i L is the exchange current density, i oa is the anodic exchange current density, i oc is the cathodic exchange current density, γ is the pre-exponential factor corresponding to the anodic and cathodic exchange current densities (where the subscript with a represents the anodic one, and the subscript with c represents the cathodic one), P ref is the atmospheric pressure, E act,a is the anodic activation energy, E act,c is the cathodic activation energy, k i is the exchange current density coefficient, X is the ratio of the grain contact neck length to the grain size, D p is the pore diameter, D s is the grain size, n is the electrode porosity.

[0034] Preferably, in step S1, the temperature, pressure, and composition changes are calculated according to the following equations:

[0035] ;

[0036] ; ;

[0037] ;

[0038] ;

[0039] ;

[0040] where, N out is the outlet molar flow rate, N in is the inlet molar flow rate, R iis the molar reaction rate, N is the molar content in the component, X is the mole fraction, C V is the specific heat capacity at constant volume, R is the universal gas constant, h is the enthalpy, is the component i enthalpy value of, is the standard enthalpy value of component i at 298.15K.

[0041] Preferably, in step S2, the equations for the temperature, pressure, and components of the typical BOP components are the same as those in S1.

[0042] Preferably, in step S2, the performance degradation of the heat exchanger is mainly the following equation:

[0043] ;

[0044] ;

[0045] Among them, Q in is the heat transfer amount, S area is the heat transfer area, h gs is the convective heat transfer coefficient, T _ban is the heat transfer baffle temperature, T _hotgas is the heat transfer fluid, Φ is the heat transfer degradation coefficient, h ex is the initial convective heat transfer coefficient.

[0046] Preferably, in step S2, a multi-stage modeling is performed for the chemical reactions occurring inside the reformer. The reactions occurring inside the reformer are:

[0047] Reforming reaction:

[0048] ;

[0049] Water-gas shift reaction:

[0050] ;

[0051] Among them, the reforming reaction is assumed to be completely reformed initially, that is:

[0052] ;

[0053] Among them, r SRis the reforming reaction rate, N fuel,in is the molar flow rate at the fuel inlet, is the methane concentration at the fuel inlet.

[0054] Subsequently, considering the influence of carbon deposition, the reforming rate is gradually decreased with time. Among them, MD (Methane Decomposition reaction) is the methane cracking reaction, and BR (Boundouard reaction) is the reverse Boudouard reaction, which are specifically as follows:

[0055] ;

[0056] ;

[0057] ;

[0058] ; ;

[0059] Among them, is the reforming reaction activity, is the molar flow rate at the reformer inlet, is the inlet methane content, is the thermal conductivity, K BR、 K MD is the adsorption coefficient, is the carbon deposition concentration (mol / m 3 ) r is the reaction rate (mol / m 3 / s), are the reaction rates of the reverse Boudouard reaction and the methane cracking reaction respectively, p is the gas pressure, M C is the molar mass (kg / mol).

[0060] In step S2, the attenuation modeling of other components is not considered temporarily. Only the changes in internal components, temperature, and pressure are considered, and the modeling is carried out according to S1.

[0061] Preferably, in step S3, the variables that affect each other are connected by combining the SIMULINK models of the stack and each component to form a dynamic attenuation model of the system.

[0062] Preferably, in step S3, the model is adjusted and verified by combining the actual operation data to ensure that the error between the model result and the actual situation is within 10%.

[0063] The life prediction model of the hybrid-driven SOFC system is mainly based on the simulation results of a large number of data under different working conditions of the SOFC dynamic decay model. Using the multi-dimensional time series prediction method and combining with the established neural network model for machine learning, the life prediction model of the SOFC system is obtained after training and verification. In step S3, the SIMULINK models of the stack and each component are combined to connect the mutually influencing variables to form the dynamic decay model of the system; the model is adjusted and verified by combining the actual operation data to ensure that the error between the model result and the actual situation is within 10%.

[0064] Preferably, during the establishment process of the SOFC system life prediction model, a large number of operation data under different working conditions are calculated using the SOFC dynamic decay model as the data set for machine learning; the correlation analysis of system variables is carried out, and typical variables are selected to study the SOFC system and establish the life prediction model.

[0065] Preferably, for the SOFC stack data, the first 90% of the data is selected as the training set and the validation set, and the last 10% of the data is used as the test set; according to the divided intervals, random initialization is performed before the training of each model for comparison within the interval and between intervals.

[0066] Preferably, the fuel cell life prediction is divided into two stages: the learning-training stage and the prediction-inference stage. The specific process is as follows:

[0067] ① Model construction: Build various neural network models and set relevant structure parameters;

[0068] ② Input the training set for model learning: Use the training set to train the model;

[0069] ③ Evaluate the model on the validation set: The trained model makes inferences on the validation set, and calculates indicators such as mean square error, root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination for comprehensive evaluation;

[0070] ④ Parameter tuning: Optimize the model structure parameters and training parameters according to the results of model evaluation;

[0071] ⑤ The model predicts the final result on the test set: The optimized model makes a final evaluation on the test set.

[0072] The neural network models are trained, verified, and tested respectively on different intervals. The evaluation indicators of each model use the commonly used mean absolute error MAE, mean square error MSE, root mean square error RMSE, mean absolute percentage error MAPE, and coefficient of determination R2 to comprehensively evaluate the prediction performance of the model from multiple angles. The calculation methods of each error are as follows:

[0073] ;

[0074] ;

[0075] ;

[0076] ;

[0077] ;

[0078] Among them, a is the number of samples, j is the specific value serial number in each sample, Y j is the sample serial number j is the true value when is the sample serial number j is the predicted value when is the average value of the true values of the samples.

[0079] After obtaining the life prediction model, it is necessary to compare the output of the life prediction model with the data of the actual SOFC system. If the error requirement is met, the accuracy of the life prediction model can be determined. If the accuracy requirement is not met, the composition of the life prediction model needs to be adjusted, and training, verification, and testing are carried out again to obtain the life prediction model of the system after meeting the error requirement. The model training is considered successful when the error is within 10%.

[0080] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0081] The test time is short and the cost is saved. Compared with the long-term operation experiment of the actual system, the expected design life of the system can be obtained through the simulation method for the system life prediction, saving the experimental cost and accelerating the R & D progress;

[0082] It provides an important reference basis for the long-term operation of the system. Through the dynamic decay simulation and life prediction of the system, the influence of the BOP components on the entire life of the system can be clearly analyzed, indicating the direction for the long-term operation of the system;

[0083] The accuracy is relatively high. By using the machine learning method, the accidental results that may exist in a single simulation are avoided, making the results have universality and relatively high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 is the flowchart for establishing the life prediction model of the SOFC system.

[0085] Figure 2 is the flowchart of the SOFC combined heat and power system of the present invention.

[0086] Figure 3 is the flowchart for debugging and verifying the life prediction model.

[0087] Figure 4 This is the verification result of the SOFC system life prediction model. Specific implementation manners

[0088] The attached drawings are only for illustrative purposes; for better explaining this embodiment, some components in the attached drawings will be omitted, enlarged or reduced, which do not represent the dimensions of the actual product; for those skilled in the art, some well-known structures and their descriptions in the attached drawings may be omitted, therefore, it cannot be understood as a limitation to the present invention.

[0089] In order to make the technical means, creative features, achieved purposes and functions of the present invention easy to understand, the present invention will be further described in detail below with reference to the attached drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. The present invention will be further described in detail below with reference to the attached drawings.

[0090] Embodiment 1

[0091] As Figure 1 shown, a life prediction method for a hybrid-driven SOFC combined heat and power system includes the establishment of a dynamic decay model of the SOFC system and the establishment of a life prediction model of the hybrid-driven SOFC system; among which, the establishment of the dynamic decay model of the SOFC system includes the following steps.

[0092] S0: Determine the specific process of the SOFC thermoelectric system and the typical BOP components used.

[0093] As Figure 2 shown, in step S0, the modeling of the SOFC thermoelectric system mainly considers components such as the SOFC stack, reformer, mixer, water evaporator, heat exchanger, and tail gas combustion chamber; the specific SOFC working process is that the fuel gas on the anode side undergoes desulfurization through a desulfurizer (sulfur-containing is not considered in the modeling, assuming the sulfur content in the fuel is 0), exchanges heat through a heat exchanger and then enters the mixer to be mixed with water vapor (assuming the gas is evenly mixed), and then enters the reformer for reforming. The reformed gas enters the anode of the stack. The cathode air exchanges heat through a heat exchanger and then enters the cathode of the stack.

[0094] The tail gas at the anode outlet first enters the tail gas combustion chamber for combustion. The high-temperature flue gas from the combustion first provides a suitable reaction temperature for the reforming reactor, and then preheats the fuel gas before entering the mixer.

[0095] The tail gas at the cathode outlet preheats the inlet air.

[0096] S1: Analyze the performance decay principle of the stack and establish a decay model of the stack.

[0097] In step S1, the degradation of the SOFC stack is mainly considered under continuous operating conditions. The specific electrochemical degradation model is as follows:

[0098] ,

[0099] ;

[0100] ;

[0101] ;

[0102] ;

[0103] ;

[0104] ;

[0105] ;

[0106] ;

[0107] ;

[0108] γ i = k i 72 X [ D P − ( D P + D s ) n ] n D s 2 D P 2 ( 1 − 1 − X 2 ) , i = a , c ;

[0109] Among them, is the actual voltage of the SOFC stack, is the number of single cells in the stack, is the initial voltage of a single cell, is the voltage degradation rate, is the fitting coefficient, obtained by fitting according to the actual performance of the cell, is the fuel utilization rate, is the damage variable, is the temperature inside the stack, is the ideal operating temperature of the stack, is the natural constant 2.718, i is the current density, E N is the Nernst voltage, η ohm is the ohmic loss, η act is the activation loss, η con is the concentration loss, E 0 is the standard electrode potential, F is the Faraday constant (96485 C mol -1 ), Ris a general other constant (8.314 J mol -1 K -1 ), p is the gas pressure, R cell is the resistance, i L is the exchange current density, i oa is the anodic exchange current density, i oc is the cathodic exchange current density, γ is the pre-exponential factor corresponding to the anodic and cathodic exchange current densities (where the subscript with a represents the anodic one, and the subscript with c represents the cathodic one), P ref is the atmospheric pressure, E act,a is the anodic activation energy, E act,c is the cathodic activation energy, k i is the exchange current density coefficient, X is the ratio of the grain contact neck length to the grain size, D p is the pore diameter, D s is the grain size, n is the electrode porosity.

[0110] In step S1, the temperature, pressure, and composition changes are calculated according to the following equations:

[0111] ;

[0112] ;

[0113] ;

[0114] ;

[0115] ;

[0116] ;

[0117] Among them, N out is the outlet molar flow rate, N in is the inlet molar flow rate, R i is the molar reaction rate, N is the molar content in the component, X is the mole fraction,C V is the specific heat capacity at constant volume, R is the universal gas constant, h is the enthalpy.

[0118] S2: Analyze the performance decay principle of typical BOP components and establish a decay model for typical BOP components.

[0119] The equations for the temperature, pressure, and composition of typical BOP components are as in S1.

[0120] The main performance decay of the heat exchanger is the following equation:

[0121] ;

[0122] ;

[0123] Among them, Q in is the heat transfer amount, S area is the heat transfer area, h gs is the convective heat transfer coefficient, T _ban is the heat transfer baffle temperature, T _hotgas is the heat transfer fluid, Φ is the heat transfer decay coefficient, h ex is the initial heat transfer coefficient.

[0124] In step S2, perform multi-stage modeling of the chemical reactions occurring inside the reformer. The reactions occurring inside the reformer are:

[0125] Reforming reaction:

[0126] .

[0127] Water-gas shift reaction:

[0128] .

[0129] Among them, the reforming reaction is assumed to be completely reformed initially, that is:

[0130] ;

[0131] Among them, r SR is the reforming reaction rate, N fuel,in is the molar flow rate at the fuel inlet, is the methane concentration at the fuel inlet.

[0132] Subsequently, considering the influence of carbon deposition, the reforming rate is set to gradually decrease with time. Among them, MD (Methane Decomposition reaction) is the methane cracking reaction, and BR (Boundouard reaction) is the reverse Boudouard reaction, specifically as follows:

[0133] ;

[0134] ;

[0135] ;

[0136] ; .

[0137] Among them, is the reforming reaction activity, is the molar flow rate at the reformer inlet, is the methane content at the inlet, is the thermal conductivity, K BR、 K MD is the adsorption coefficient, is the carbon deposition concentration (mol / m 3 ), r is the reaction rate (mol / m 3 / s), p is the gas pressure, M C is the molar mass (kg / mol).

[0138] The decay modeling of other components is not considered for the time being. Only the changes in internal components, temperature, and pressure are considered, and the modeling is carried out according to the equation of S1.

[0139] S3: Based on the decay model, use MATLAB software to build a dynamic decay model of the system and verify it.

[0140] Connect the variables that affect each other by combining the SIMULINK models of the stack and each component to form a dynamic decay model of the system.

[0141] In step S3, adjust and verify the model in combination with the actual operation data to ensure that the error between the model result and the actual value is within 5%.

[0142] S4: Perform life prediction based on the built neural network.

[0143] Such as Figure 3As shown in the figure, the life prediction model of the hybrid-driven SOFC system is mainly based on the simulation data results of the SOFC dynamic decay model under a large number of different working conditions. Using the multi-dimensional time series prediction method and combining with the established neural network model for machine learning, the life prediction model of the SOFC system is obtained after training and verification. The specific steps are as follows:

[0144] 1) Set different boundary condition parameters to obtain the SOFC system operation data sets under different working conditions;

[0145] Since a large amount of data is required for the training and debugging of the neural network, but the continuous operation in actual production takes a long time, the SOFC dynamic decay model is used to calculate a large amount of operation data under different working conditions as the data set for machine learning.

[0146] 2) Conduct correlation analysis to screen out the parameters that have a greater impact on the system life;

[0147] The change curves of each parameter with time during the operation of the SOFC system are obtained. However, there are many variables in the whole system, and it is impossible to consider and analyze all of them. It is necessary to conduct correlation analysis on these variables, analyze the correlation coefficients between each parameter, and thus select several typical variables to study the SOFC system and establish a life prediction model.

[0148] The Pearson correlation coefficient is used to measure the degree of association between variables. The calculation of the correlation coefficient is as follows:

[0149] ;

[0150] In the formula: cov(X,Y) is the covariance of variables X and Y; DX is the variance of variable X, a is the number of samples, j is the specific value serial number in each sample, Y j is the sample serial number j is the true value when is the sample serial number j is the predicted value when is the average value of the true values of the samples.

[0151] The value range of the Pearson correlation coefficient is [-1,1]. The closer its absolute value is to 1, the stronger the linear correlation between the two variables; the values with a coefficient above 0.5 are selected as the parameters that have a greater impact on the system life.

[0152] 3) Build a suitable life prediction neural network model based on the characteristics of the SOFC system operation data;

[0153] Based on the multi-dimensional characteristics and long-term characteristics of the SOFC system operation data, LSTM (Long-Short-Term Memory) is selected for machine learning in the main body. To ensure the health of the data used for machine learning, CNN (Convolutional Neural Network) is selected to smooth and filter the data before entering the LSTM. In order to reduce the computational complexity, the original data is normalized by Max-Min normalization. In order to make the factors that have a greater impact on the system life play a greater role, an Attention mechanism is added to strengthen the influence of key parameters.

[0154] 4) Use the collected data set to train and validate the neural network. If the error meets within 10%, the model training is defined as successful, and the system life prediction model is obtained.

[0155] Since the operation of the SOFC stack is mainly divided into three stages: activation stage, stable operation stage, and rapid failure stage, the data set is accordingly divided into the activation + stable operation interval A, the rapid failure interval B, and the total interval C.

[0156] To avoid data leakage and strictly conform to the front-back logic of time series prediction, 90% of the data in each of the three intervals A, B, and C is equally selected as the training set (70%) and the validation set (20%), and the last 10% of the data is used as the test set. Random initialization is performed before the training of each model compared within and between intervals.

[0157] Fuel cell life prediction is mainly divided into two stages: learning-training stage and prediction-inference stage. The specific process is explained as follows:

[0158] ① Model construction: Build various neural network models and set relevant structure parameters.

[0159] ② Input the training set for model learning: Use the training set to train the model.

[0160] ③ Evaluate the model on the validation set: The trained model infers the validation set, and calculates indicators such as mean square error, root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination for comprehensive evaluation.

[0161] ④ Parameter tuning: Optimize the model structure parameters and training parameters according to the results of model evaluation.

[0162] ⑤ The model predicts the final result on the test set: The optimized model performs the final evaluation on the test set.

[0163] The neural network model is trained, validated, and tested on three intervals respectively. The evaluation metrics for each model are the commonly used Mean Absolute Error (MAE), Mean Square Error (MSE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Coefficient of Determination (R-Square, R2). The prediction performance of the model is comprehensively evaluated from multiple perspectives. The calculation methods for each error are as follows:

[0164] ;

[0165] ;

[0166] ;

[0167] ;

[0168] .

[0169] Among them, a is the number of samples, j is the specific value serial number in each sample, Y j is the sample serial number j is the true value when is the sample serial number j is the predicted value when is the average value of the true values of the samples.

[0170] In each group of experiments, by adjusting the above parameters, the model with the best performance on the validation set is selected to infer the test set.

[0171] After obtaining the life prediction model, it is necessary to compare the output of the life prediction model with the data of the actual SOFC system. If the error requirements are met, the accuracy of the life prediction model can be determined. If the accuracy requirements are not met, the composition of the life prediction model needs to be adjusted, and the training process and validation need to be carried out again. After meeting the error requirements, the life prediction model of the system is obtained.

[0172] As Figure 4 shown, it is a comparison chart of the actual operating voltage decay over time of the SOFC system and the results of the life prediction model when the operating temperature is 750 °C, the operating current is 30 A, the anode inlet gas is methane, and the cathode gas is air.

[0173] Certainly, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions, or substitutions made by those skilled in the art within the scope of the essence of the present invention should also fall within the protection scope of the present invention.

Claims

1. A life prediction method for a hybrid-driven SOFC cogeneration system, characterized in that: The method includes establishing a dynamic attenuation model of a SOFC system and establishing a life prediction model of a hybrid-driven SOFC system; wherein the establishment of the dynamic attenuation model of a SOFC system includes the following steps: S0: Determine the specific process of the SOFC thermoelectric system and the typical BOP components used; S1: Analyze the performance attenuation principle of the stack and establish the attenuation model of the stack; the attenuation of the SOFC stack is considered under continuous operation conditions. The specific electrochemical attenuation model is: V=N cell ×V cell ·(1-v); V cell =E N -or ohm -or act -or con ; E0=1.253-2.4516×10 -4 T; or ohm =i·R cell ; Where V is the actual voltage of the SOFC stack, N cell is the number of cells in the battery stack, V cell is the starting voltage of a single cell, ν is the voltage decay rate; C1, C2, C3 are fitting coefficients, which are fitted according to the actual battery performance, Fu is the fuel utilization rate, ω is the damage variable, T is the temperature inside the battery stack, T0 is the ideal operating temperature of the battery stack, e is the natural constant 2.718; i is the current density, E N is the Nernst voltage, η ohm is the ohmic loss, η act is the activation loss, η con is the concentration loss, E0 is the standard potential; F is the Faraday constant, which is 96485C mol -1 ; R is a universal other constant, which is 8.314 Jmol -1 K -1 ; p is the gas pressure, R cell is the resistance, i L is the exchange current density, i oa is the anodic exchange current density, i oc is the cathode exchange current density, γ is the pre-exponential factor corresponding to the exchange current density of the anode and cathode, where a represents the anode, c represents the cathode, and P ref is atmospheric pressure, E act,a is the anode activation energy, E act,c is the cathode activation energy, k i is the exchange current density coefficient, X is the ratio of the grain contact neck length to the grain size, D p is the aperture, D s is the grain size, n is the electrode porosity; S2: Analyze the performance attenuation principle of typical BOP components and establish an attenuation model of typical BOP components; The heat exchanger performance degradation equation is: ∑Q in =S area h gs (T _ban -T _hotgas ); h gs =Φ·h ex ; Among them, Q in is the heat transfer, S area is the heat exchange area, h gs is the convective heat transfer coefficient, T _ban is the heat exchange partition temperature, T _hotgas is the heat transfer fluid, Φ is the heat transfer attenuation coefficient, h ex is the initial convective heat transfer coefficient; S3: Based on the attenuation model, the dynamic attenuation model of the system is built and verified using MATLAB software; the SIMULINK model of the battery stack and each component is combined to connect the variables that affect each other to form a dynamic attenuation model of the system; the model is adjusted and verified based on actual operating data to ensure that the error between the model result and the actual result is within 10%; S4: Lifespan prediction based on the constructed neural network; The establishment of the SOFC system life prediction model includes the following steps: 1) Setting different boundary condition parameters to obtain the SOFC system operation data set under different working conditions; 2) Correlation analysis to select parameters that have a greater impact on system life; 3) Build a suitable life prediction neural network model based on the characteristics of SOFC system operation data; 4) Using the collected data set to train and verify the neural network, the model training is defined as successful if the error is within 10%, and the life prediction model of the system is obtained; In the process of establishing the SOFC system life prediction model, the SOFC dynamic attenuation model is used to calculate a large amount of operating data under different working conditions as a data set for machine learning; correlation analysis is performed on system variables, and typical variables are selected to study the SOFC system and establish a life prediction model; For SOFC stack data, the first 90% of the data were selected as training and validation sets, and the last 10% of the data were selected as test sets. According to the divided intervals, each model for comparison within the interval and between intervals was randomly initialized before training. Fuel cell life prediction is divided into two stages: learning-training stage and prediction-inference stage. The specific process is as follows: ① Model building: build various neural network models and set relevant structural parameters; ② Input the training set for model learning: Use the training set to train the model; ③Evaluate the model on the validation set: The trained model is inferred on the validation set, and the mean square error, root mean square error, mean absolute error, mean absolute percentage error and determination coefficient are calculated for comprehensive evaluation; ④ Parameter adjustment: optimize the model structure parameters and training parameters according to the results of model evaluation; ⑤ The model predicts the final result on the test set: the tuned model performs a final evaluation on the test set; The neural network models were trained, verified and tested in different intervals. The evaluation indicators of each model used the commonly used mean absolute error MAE, mean square error MSE, root mean square error RMSE, mean absolute percentage error MAPE and determination coefficient R2 to comprehensively evaluate the prediction performance of the model from multiple angles. The calculation method of each error is as follows: Among them, a is the number of samples, j is the specific value number in each sample, and Y j is the true value of sample number j, is the predicted value for sample number j, is the average value of the true value of the sample; After obtaining the life prediction model, the output of the life prediction model is compared with the data of the actual SOFC system. If the error requirements are met, the accuracy of the life prediction model can be determined. If the error requirements are not met, the composition of the life prediction model needs to be adjusted, and re-training, verification and testing are performed to obtain the life prediction model of the system after the error requirements are met.

2. The method for predicting the life of a hybrid-driven SOFC cogeneration system according to claim 1, characterized in that: In step S0, the specific SOFC working process is that the anode side fuel gas is desulfurized by the desulfurizer, enters the mixer to mix with water vapor after heat exchange in the heat exchanger, and then enters the reformer for reforming. The reformed gas enters the anode of the stack.

3. The method for predicting the life of a hybrid-driven SOFC cogeneration system according to claim 2, characterized in that: The cathode air enters the cathode of the fuel cell stack after heat exchange in the heat exchanger; the exhaust gas at the anode outlet first enters the exhaust gas combustion chamber for combustion. The high-temperature flue gas first provides a suitable reaction temperature for the reforming reactor, and then preheats the fuel gas before entering the mixer; the exhaust gas at the cathode outlet preheats the inlet air.

4. The method for predicting the life of a hybrid-driven SOFC cogeneration system according to claim 1, characterized in that: In step S2, the chemical reactions occurring inside the reformer are modeled, wherein the reactions occurring inside the reformer are: Reforming reaction: Water gas shift reaction: The reforming reaction is assumed to be completely reformed initially, that is: Among them, r SR is the reforming reaction rate, N fuel,in is the fuel inlet molar flow rate, is the fuel inlet methane concentration; Considering the effect of carbon deposition, the reforming rate is set to decrease gradually over time, where MD is the methane cracking reaction and BR is the reverse Boudouard reaction. The specific equation is as follows: Among them, w SR is the reforming reaction activity, N fuel，in is the reformer inlet molar flow rate, is the inlet methane content, k SR , k MD , k H , k BR , k CO , is the thermal conductivity, K BR , K MD is the adsorption coefficient; C SR is the carbon deposit concentration, in mol / m 3 ; r is the reaction rate, r BR 、r MD are the reaction rates of the reverse Boudouard reaction and methane cracking reaction, respectively, in mol / m 3 / s; p is the gas pressure; M C is the molar mass in kg / mol.

Images