A method and device for predicting the remaining life of a fuel cell based on a hybrid model

Abstract

The present invention discloses a method and device for predicting the remaining life of a fuel cell based on a hybrid model. The method comprises the following steps: S1, constructing a monitoring data set and performing preprocessing; S2, using a filtering algorithm to decompose the stack voltage in the monitoring data set, and constructing a calendar aging voltage data set and a reversible aging voltage data set; S3, constructing a calendar aging voltage model and a reversible aging voltage model; S4, using the calendar aging voltage model and the reversible aging voltage model to perform iterative rolling predictions on the calendar aging voltage and the reversible aging voltage, respectively, and respectively obtain calendar aging voltage estimation values ​​and reversible aging voltage estimation values ​​at future moments, and superimposing the calendar aging voltage estimation values ​​and the reversible aging voltage estimation values ​​to obtain the final fuel cell stack prediction voltage, thereby realizing the prediction of the remaining life of the fuel cell. Compared with the traditional method, the present invention improves the prediction accuracy of the remaining life of the fuel cell.

Description

Technical Field

[0001] The present invention belongs to the technical field of fuel cell state detection and health management, and specifically relates to a method and device for predicting the remaining service life of a fuel cell based on a hybrid model. Background Art

[0002] Proton exchange membrane fuel cells (PEMFCs) using hydrogen as fuel have a series of advantages such as cleanliness and environmental protection, high power density, low temperature operation, and low noise. They are one of the energy devices with broad application prospects. Life prediction of proton exchange membrane fuel cells (PEMFCs) using hydrogen as fuel plays a vital role in ensuring the long-term and healthy operation of PEMFCs.

[0003] At present, there are three main methods for predicting the remaining life of PEMFC:

[0004] 1. Model-based prediction method, which is a mathematical model based on the aging mechanism, such as empirical model, semi-mechanistic and semi-empirical model, etc.

[0005] 2. Data-based prediction methods use historical aging data to train prediction models and learn nonlinear mapping relationships between data, such as using machine learning, neural networks, signal processing methods, etc.

[0006] 3. Hybrid methods integrate the characteristics of mechanism, empirical model and data-driven methods, hoping to further improve the accuracy and robustness of the model.

[0007] However, among the above three PEMFC life prediction methods, the model-based prediction method has less training data, strong versatility, and is easy to understand and debug, but it has a large amount of calculation and is not suitable for online application. In addition, due to the complexity and uncertainty of the failure mechanism, it is difficult to establish an accurate and reliable mechanism aging model; the data-based prediction method does not require the construction of a complex mechanism model, which is conducive to online deployment, but requires extensive and comprehensive training data and has the problem of poor versatility; the hybrid method can alleviate the limitations of training data and the lack of knowledge of the mechanism model, and is usually more accurate than a single method, but the cost is a more complex model structure.

[0008] Therefore, it is necessary to develop accurate and efficient methods to predict the remaining life of fuel cells, so as to improve their durability, reduce maintenance and costs, and promote the further popularization and development of proton exchange membrane fuel cell technology, which is of great significance. Summary of the invention

[0009] In view of the above defects and improvement needs of the prior art, the purpose of the present invention is to provide a method and device for predicting the remaining service life of a fuel cell based on a hybrid model. The method can learn the trend information of calendar aging voltage and the recovery information of reversible aging from aging test data. The established prediction model has strong robustness and generalization ability, and improves the prediction accuracy of the remaining service life of the fuel cell compared with the traditional method.

[0010] The objective of the present invention is achieved through the following technical solutions:

[0011] According to a first aspect of the present specification, a method for predicting the remaining life of a fuel cell based on a hybrid model is provided, comprising the following steps:

[0012] S1, obtaining the operation monitoring data of the fuel cell, constructing the monitoring data set and performing preprocessing;

[0013] S2, using a filtering algorithm to decompose the battery stack voltage in the monitoring data set to obtain a calendar aging voltage and a reversible aging voltage, and constructing a calendar aging voltage data set from the calendar aging voltage and a reversible aging voltage data set from the reversible aging voltage;

[0014] S3, using polarization curve data obtained by polarization curve testing of the fuel cell, establishing a semi-mechanism aging model based on an empirical polarization equation; using a calendar aging voltage data set, combined with the semi-mechanism aging model with parameter identification processing based on a genetic algorithm, and constructing a calendar aging voltage model based on a nonlinear filter algorithm;

[0015] Using the reversible aging voltage dataset and the start time of the polarization curve test of the fuel cell, combined with the automatic machine learning method, a reversible aging voltage model was constructed based on the long short-term memory neural network;

[0016] S4, using the calendar aging voltage model and the reversible aging voltage model to iteratively and rollingly predict the calendar aging voltage in the calendar aging voltage data set in step S2 and the reversible aging voltage in the reversible aging voltage data set, respectively, to obtain the calendar aging voltage estimate value and the reversible aging voltage estimate value at the future moment, and to superimpose the calendar aging voltage estimate value and the reversible aging voltage estimate value to obtain the final fuel cell stack prediction voltage, so as to realize the remaining service life prediction of the fuel cell.

[0017] Furthermore, in the step S1, the preprocessing is specifically to downsample the stack voltage data in the monitoring data set, and the downsampling interval is 1 hour.

[0018] Furthermore, the step S2 specifically includes the following steps:

[0019] S21, using a local weighted regression filtering algorithm to smooth the stack voltage data in the monitoring data set;

[0020] The local weighted regression filtering algorithm includes weighted linear least squares method and second-order polynomial model. The window width of the filter is set to 300. The weight coefficient of the data point in the filter window is calculated by the weight function. The weight coefficient is defined as follows:

[0021]

[0022] Among them, w i is the weight coefficient, Δ(x) is the window width, x is the horizontal coordinate of the point to be fitted, indicating the measurement time of the data, and x i is the horizontal coordinate of the data point in the window, |·| is the absolute value, and the smoothing value of the fitting point (x, y) It can be obtained by weighted least square method; y is the ordinate of the point to be fitted, indicating the stack voltage; Represents the smoothed value of the stack voltage; the objective function J(a, b) minimized by the weighted least squares method in the local weighted regression filtering algorithm is defined as follows:

[0023]

[0024] Where m is the number of data points in the window, w i is the weight coefficient, y i is the function value corresponding to the data point, indicating the stack voltage, a and b are the parameters of the regression curve, x i is the horizontal coordinate of the data point in the window, indicating time; after obtaining the parameters a and B of the non-parametric regression curve of the stack voltage monitoring data, the smoothed value of the current voltage data point can be obtained

[0025]

[0026] S22, stack voltage V st (t) After filtering by S21, the calendar aging voltage V representing the aging trend of the fuel cell is obtained. c (t), t represents the measurement time of voltage, V st (t) minus V c (t) can get the reversible aging voltage V r (t), the final fuel cell aging voltage decomposition expression is as follows:

[0027] V st (t) = V c (t)+V r (t)

[0028] Therefore, a calendar aging voltage data set is constructed from the obtained calendar aging voltage and a reversible aging voltage data set is constructed from the reversible aging voltage.

[0029] Furthermore, the construction of the calendar aging voltage model in step S3 specifically includes the following steps:

[0030] S311, using polarization curve data obtained by polarization curve testing of the fuel cell, a semi-mechanism aging model is established based on a semi-empirical formula of the polarization equation and an aging attenuation factor is introduced. The semi-empirical formula of the polarization equation can be expressed as follows:

[0031]

[0032] Among them, V c (t) is the calendar aging voltage at time t, t represents a certain moment in the operation of the fuel cell, N is the number of cells in the fuel cell stack, E ocv is the open circuit voltage, i(t) is the fuel cell output current at time t, R ohm is the ohmic internal resistance, a is the Tafel constant, b is the concentration polarization constant, i 0 (t) is the exchange current density at time t, i L (t) is the limiting current density at time t, and T is the stack operating temperature;

[0033] According to the meaning of the parameters of the semi-empirical formula of the polarization equation and the changes of the parameters under different aging conditions, the ohmic internal resistance R is selected. ohm and the limiting current density i L As the aging parameter, the aging factor α is introduced to represent the change of the aging parameter, and its expression is as follows:

[0034]

[0035] Among them, α(t) is used as the aging factor to characterize the aging state of the fuel cell at time t, R 0 is the initial value of the ohmic internal resistance, i L0 is the initial value of the limiting current; the functional relationship between the external voltage of the fuel cell and the aging time can be obtained by combining the above formulas, which is used as a semi-mechanism aging model and expressed as follows:

[0036]

[0037] Among them, V c (t) is the calendar aging voltage at time t, t represents a certain moment in the operation of the fuel cell, N is the number of cells in the fuel cell stack, E ocv is the open circuit voltage, i(t) is the output current of the fuel cell at time t, α(t) is the aging factor to characterize the aging state of the fuel cell at time t, R 0is the initial value of the ohmic internal resistance, a is the Tafel constant, B is the concentration polarization constant, i 0 (t) is the exchange current density at time t, i L0 is the initial value of the limiting current, T is the operating temperature of the stack;

[0038] S312, performing parameter identification on the semi-mechanism aging model established in step S311 based on a genetic algorithm, and obtaining the optimal solution of the parameters through the fitness function in the genetic algorithm, as follows:

[0039] First, the parameter identification problem is transformed into an optimization problem, and the fitness function of the genetic algorithm is expressed as:

[0040]

[0041] Where k is the sampling point of the polarization curve, V c，k is the actual value of voltage, To identify the fitting value of the model; there are 6 dimensions of parameters that need to be identified: [E ocv , R 0 , a, b, i 0 ,i L0 ];

[0042] The genetic algorithm first randomly initializes the population, sets the crossover rate and mutation rate, and then calculates the fitness function f of the population. fitmess The size of the variables is used to perform selection, crossover, and mutation operations, and the optimal solution is selected after repeated operations to obtain the parameter identification results;

[0043] S313, based on the semi-mechanism aging model processed by parameter identification in step S312, a third-order aging attenuation factor is introduced, as follows:

[0044] The nonlinear system of the fuel cell aging process is expressed as follows:

[0045]

[0046] In the formula, x k-1 、x k They represent the aging state of the fuel cell at time k-1 and time k respectively, and u k-1 、u k They represent the external input at time k-1 and time k respectively, and y k is the system output at time k, w k-1 、v k represents the system process noise at time k-1 and the system measurement noise at time k, respectively. f(·) and g(·) represent the state transfer matrix and output matrix of the fuel cell aging model. The external input is the current load i k , the system output y kis the fuel cell calendar aging voltage V c，k ;

[0047] Define the third-order aging factor x k =[α k , β k , γ k ] T , the discrete time state space equation of the fuel cell aging process can be obtained:

[0048]

[0049] Among them, Δt is the sampling time interval, α k and α k-1 As the aging factor, β represents the aging state at time k and time k-1 respectively. k and β k-1 They represent the aging rates at time k and time k-1 respectively, γ k and γ k-1 They represent the second-order derivatives of the aging state at time k and time k-1, respectively, and E ocv is the open circuit voltage, i k is the fuel cell load current at time k, R 0 is the initial value of the ohmic internal resistance, a is the Tafel constant, B is the concentration polarization constant, i 0 is the exchange current density, i L0 is the initial value of the limiting current density, T is the operating temperature of the battery stack, N is the number of battery stack cells, and w k-1 、v k They represent the system process noise at time k-1 and the system measurement noise at time k respectively; the external input is the current load i k , the system output y k is the fuel cell calendar aging voltage V c，k ;

[0050] By linearizing the nonlinear system using Taylor polynomials, the Jacobian matrix of the nonlinear system can be obtained and used for iterative calculation of the extended Kalman filter algorithm. The extended Kalman filter algorithm consists of three parts: initialization, prior estimation-prediction, and posterior estimation-update. The algorithm formula is as follows:

[0051] (1) Initialization:

[0052]

[0053] (2) Prior estimation-prediction:

[0054] x k|k-1 =Ax k-1|k-1 , P k|k-1 =AP k-1|k-1 AT +Q;

[0055] (3) Posterior estimation-update:

[0056] x k|k =x k|k-1 +K k (y k -g(x k ,u k )), P k|k =(IK k H k ) k|k-1 ;

[0057] Among them, x 0|0 , P 0|0 They represent the initial system state and the initial system covariance matrix respectively, E represents the expectation, Var represents the covariance matrix, x(t 0 ) indicates t 0 The system state at the moment, H k represents the Jacobian matrix, g(·) represents the output matrix of the fuel cell aging model, x k represents the aging state of the fuel cell at time k, u k represents the external input of the system at time k, x k|k-1 represents the prior estimate of the system state at time k, x k-1|k-1 represents the posterior estimate of the system state at time k-1, A represents the system matrix, A T represents the transpose of the system matrix, P k|k-1 represents the prior estimate of the covariance matrix at time k, P k-1|k-1 represents the posterior estimate of the covariance matrix at time k-1, Q represents the process error covariance matrix, K k represents the Kalman gain at time k, represents the transpose of the Jacobian matrix, R represents the measurement error covariance matrix, x k|k represents the posterior estimate of the system state at time k, y k represents the measured output of the system, P k|k represents the posterior estimate of the covariance matrix at time k, and I represents the identity matrix;

[0058] The calendar aging voltage dataset was used for training, and a calendar aging voltage model of the fuel cell was constructed.

[0059] Furthermore, the construction of the reversible aging voltage model in step S3 specifically includes the following steps:

[0060] S321, combining the reversible aging voltage in the reversible aging voltage data set with the start time of the polarization curve test of the fuel cell to construct a reversible aging data set, and performing data standardization processing on the reversible aging data set to obtain a standardized reversible aging data set;

[0061] S322, using an automatic machine learning method to automatically optimize the hyperparameters of the long short-term memory neural network to obtain an optimized long short-term memory neural network;

[0062] S323, sending the standardized reversible aging data set in step S321 to the optimized long short-term memory neural network in step S322 for training to construct a reversible aging voltage model.

[0063] Furthermore, the specific operations of step S4 are as follows:

[0064] S41, using the calendar aging voltage model and the reversible aging voltage model to perform iterative rolling prediction on the calendar aging voltage in the calendar aging voltage data set and the reversible aging voltage in the reversible aging voltage data set in step S2, respectively, wherein when performing iterative rolling prediction, we use the estimated value of the model in the previous prediction step As part of the input variables, get the estimated probability of the model for the next forecast step By iterating and rolling forecasting in this way, voltage forecasts can be realized for multiple steps in the future, thereby obtaining calendar aging voltage estimates. and the estimated reversible aging voltage

[0065] S42, estimated value of calendar aging voltage and the estimated reversible aging voltage The final fuel cell stack predicted voltage is obtained by superposition. The calculation formula is as follows:

[0066]

[0067] Among them, the failure threshold point of the remaining service life prediction of the fuel cell is selected, and the remaining service life RUL of the fuel cell is the voltage failure time t at which the predicted failure threshold point is located. f Subtract the current prediction time t k :

[0068] RUL=t f -t k

[0069] This enables prediction of the remaining service life of the fuel cell.

[0070] Furthermore, the RUL prediction results were compared and analyzed based on the relative error, RMSE, and MAPE evaluation indicators, where RMSE is the mean square error of the prediction results and MAPE is the mean absolute percentage error.

[0071] According to the second aspect of the present specification, a fuel cell remaining life prediction device based on a hybrid model is provided, comprising a memory and one or more processors, wherein the memory stores executable code, and when the processor executes the executable code, it is used to implement the fuel cell remaining life prediction method based on the hybrid model as described in the first aspect.

[0072] The beneficial effects of the present invention are:

[0073] (1) The present invention establishes a hybrid prediction model for the remaining service life of a fuel cell based on the idea of ​​decomposition prediction. After the original voltage is filtered and decomposed, the calendar aging and reversible aging information of the fuel cell are predicted based on the extended Kalman filter algorithm and the long short-term memory neural network, respectively. The advantages of the two methods are integrated to improve the prediction accuracy and the generalization ability of the model.

[0074] (2) Based on the semi-empirical formula of the polarization equation, the present invention introduces a third-order aging factor to characterize the aging information of the fuel cell, thereby more accurately describing the voltage calendar aging process; and introduces the prior time information of the polarization test to achieve more accurate modeling and analysis of the voltage recovery phenomenon.

[0075] (3) The present invention utilizes a genetic algorithm to identify a semi-empirical formula of the polarization equation, and utilizes an automatic machine learning algorithm (AutoML) to achieve automatic optimization of the hyperparameters of the long short-term memory neural network, thereby improving the prediction accuracy and training effect of the model.

[0076] (4) Among the three existing PEMFC life prediction methods, the model-based prediction method has less training data, strong versatility, and is easy to understand and debug, but it has a large amount of calculation and is not suitable for online application. In addition, due to the complexity and uncertainty of the failure mechanism, it is difficult to establish an accurate and reliable mechanism aging model; the data-based prediction method does not require the construction of a complex mechanism model, which is conducive to online deployment, but requires extensive and comprehensive training data and has the problem of poor versatility; the hybrid method can alleviate the limitations of training data and the lack of knowledge of the mechanism model, and is usually more accurate than a single method, but the cost is a more complex model structure. The development of an accurate and efficient fuel cell remaining life prediction method and device has improved its durability, reduced maintenance and costs, and promoted the further popularization and development of proton exchange membrane fuel cell technology, which is of great significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0078] Figure 1 A flowchart of a method for predicting the remaining useful life of a fuel cell based on a hybrid model provided as an exemplary embodiment;

[0079] Figure 2 A schematic diagram of a method for predicting the remaining useful life of a fuel cell based on a hybrid model provided by an exemplary embodiment;

[0080] Figure 3 A diagram of prediction results of a calendar aging voltage model in aging test data under a fuel cell operation monitoring data set provided by an exemplary embodiment;

[0081] Figure 4 A diagram of prediction results of a hybrid prediction model taking into account voltage reversibility in aging test data under a fuel cell operation monitoring data set provided by an exemplary embodiment;

[0082] Figure 5 A structural diagram of a fuel cell remaining useful life prediction device based on a hybrid model provided as an exemplary embodiment; DETAILED DESCRIPTION

[0083] In order to better understand the technical solution of the present application, the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0084] It should be clear that the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0085] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "an", "the" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0086] like Figure 1 and Figure 2 As shown, in one embodiment, a method for predicting the remaining life of a fuel cell based on a hybrid model is provided, comprising the following steps:

[0087] S1, obtain the operation monitoring data of the fuel cell, construct a monitoring data set and perform preprocessing; in one embodiment, obtain the open source fuel cell monitoring operation data set of the French Fuel Cell Laboratory (FCLAB), and the open source fuel cell monitoring operation data set includes the stack voltage, load current, stack temperature, gas pressure, gas flow, etc.

[0088] In order to improve the data processing speed, the data is downsampled to reduce the data size. For example, in one embodiment, the preprocessing in step S1 is specifically to downsample the stack voltage data in the monitoring data set, and the downsampling interval is 1 hour.

[0089] S2, using a filtering algorithm to decompose the stack voltage in the monitoring data set to obtain a calendar aging voltage and a reversible aging voltage, and constructing a calendar aging voltage data set from the calendar aging voltage and a reversible aging voltage data set from the reversible aging voltage; wherein the calendar aging voltage and the reversible aging voltage are respectively used as a voltage calendar aging component and a voltage reversible aging component.

[0090] In one embodiment, step S2 specifically includes the following steps:

[0091] S21, selecting a local weighted regression filtering algorithm (LOESS) in the filtering algorithm, and using the local weighted regression filtering algorithm to smooth the stack voltage data in the monitoring data set;

[0092] The local weighted regression filtering algorithm includes a weighted linear least squares method and a second-order polynomial model; by setting the window width of the filter, the original signal containing noise and a large number of spikes can be effectively filtered (smoothed). In this embodiment, the window width of the filter is set to 300, and the weight coefficient of the data point in the filter window is calculated by the weight function. The weight coefficient is defined as follows:

[0093]

[0094] Among them, w i is the weight coefficient, Δ(x) is the window width, x is the horizontal coordinate of the point to be fitted, indicating the measurement time of the data, and x i is the horizontal coordinate of the data point in the window, |·| is the absolute value, and the smoothing value of the fitting point (x, y) It can be obtained by weighted least square method; y is the ordinate of the point to be fitted, indicating the stack voltage; Represents the smoothed value of the stack voltage; the objective function J(a, b) minimized by the weighted least squares method in the local weighted regression filtering algorithm is defined as follows:

[0095]

[0096] Where m is the number of data points in the window, wi is the weight coefficient, y i is the function value corresponding to the data point, indicating the stack voltage, a and b are the parameters of the regression curve, x i is the horizontal coordinate of the data point in the window, indicating time; after obtaining the parameters a and B of the non-parametric regression curve of the stack voltage monitoring data, the smoothed value of the current voltage data point can be obtained (i.e. the smoothed value of the voltage):

[0097]

[0098] S22, stack voltage V st (t) After filtering by S21, the calendar aging voltage V representing the aging trend of the fuel cell is obtained. c (t), t represents the measurement time of voltage, V st (t) minus V c (t) can get the reversible aging voltage V r (t), the final fuel cell aging voltage decomposition expression is as follows:

[0099] V st (t) = V c (t)+V r (t)

[0100] Therefore, a calendar aging voltage data set is constructed from the obtained calendar aging voltage and a reversible aging voltage data set is constructed from the reversible aging voltage.

[0101] S3, using the polarization curve data obtained by the polarization curve test of the fuel cell, a semi-mechanistic aging model is established based on the empirical polarization equation; using the calendar aging voltage data set, combined with the semi-mechanistic aging model with parameter identification processing based on the genetic algorithm, and a calendar aging voltage model is constructed based on the nonlinear filter algorithm;

[0102] like Figure 3 As shown, in one embodiment, the construction of the calendar aging voltage model specifically includes the following steps:

[0103] S311, using polarization curve data obtained by polarization curve testing of the fuel cell, a semi-mechanism aging model is established based on a semi-empirical formula of the polarization equation and an aging attenuation factor is introduced. The semi-empirical formula of the polarization equation can be expressed as follows:

[0104]

[0105] Among them, V c (t) is the calendar aging voltage at time t, t represents a certain moment in the operation of the fuel cell, N is the number of cells in the fuel cell stack, E ocv is the open circuit voltage, i(t) is the fuel cell output current at time t, Rohm is the ohmic internal resistance, a is the Tafel constant, B is the concentration polarization constant, i 0 (t) is the exchange current density at time t, i L (t) is the limiting current density at time t, and T is the stack operating temperature;

[0106] According to the meaning of the parameters of the semi-empirical formula of the polarization equation and the changes of the parameters under different aging conditions, the ohmic internal resistance R is selected. ohm and the limiting current density i L As the aging parameter, the aging factor α is introduced to represent the change of the aging parameter, and its expression is as follows:

[0107]

[0108] Among them, α(t) is used as the aging factor to characterize the aging state of the fuel cell at time t, R 0 is the initial value of the ohmic internal resistance, i L0 is the initial value of the limiting current; the functional relationship between the external voltage of the fuel cell and the aging time can be obtained by combining the above formulas, which is used as a semi-mechanism aging model and expressed as follows:

[0109]

[0110] Among them, V c (t) is the calendar aging voltage at time t, t represents a certain moment in the operation of the fuel cell, N is the number of cells in the fuel cell stack, E ocv is the open circuit voltage, i(t) is the output current of the fuel cell at time t, α(t) is the aging factor to characterize the aging state of the fuel cell at time t, R 0 is the initial value of the ohmic internal resistance, a is the Tafel constant, B is the concentration polarization constant, i 0 (t) is the exchange current density at time t, i L0 is the initial value of the limiting current, T is the operating temperature of the stack;

[0111] S312, performing parameter identification on the semi-mechanism aging model established in step S311 based on a genetic algorithm, and obtaining the optimal solution of the parameters through the fitness function in the genetic algorithm, as follows:

[0112] First, the parameter identification problem is transformed into an optimization problem, and the fitness function of the genetic algorithm is expressed as:

[0113]

[0114] Where k is the sampling point of the polarization curve, V c，k is the actual value of voltage, To identify the fitting value of the model; there are 6 dimensions of parameters that need to be identified: [Eocv , R 0 , a, b, i 0 ,i L0 ];

[0115] The genetic algorithm first randomly initializes the population, sets the crossover rate and mutation rate, and then calculates the fitness function f of the population. fitmess The size of the variables is used to perform selection, crossover, and mutation operations, and the optimal solution is selected after repeated operations to obtain the parameter identification results;

[0116] S313, based on the semi-mechanism aging model processed by parameter identification in step S312, a third-order aging attenuation factor is introduced, as follows:

[0117] The nonlinear system of the fuel cell aging process is expressed as follows:

[0118]

[0119] In the formula, x k-1 、x k They represent the aging state of the fuel cell at time k-1 and time k respectively, and u k-1 、u k They represent the external input at time k-1 and time k respectively, and y k is the system output at time k, w k-1 、v k represents the system process noise at time k-1 and the system measurement noise at time k, respectively. f(·) and g(·) represent the state transfer matrix and output matrix of the fuel cell aging model. The external input is the current load i k , the system output y k is the fuel cell calendar aging voltage V c，k ;

[0120] In order to better characterize the aging state of the fuel cell and improve the prediction accuracy of the model, the third-order aging factor x is defined k =[α k , β k , γ k ] T , the discrete time state space equation of the fuel cell aging process can be obtained:

[0121]

[0122] Among them, Δt is the sampling time interval, α k and α k-1 As the aging factor, β represents the aging state at time k and time k-1 respectively. k and β k-1 They represent the aging rates at time k and time k-1 respectively, γk and γ k-1 They represent the second-order derivatives of the aging state at time k and time k-1, respectively, and E ocv is the open circuit voltage, i k is the fuel cell load current at time k, R 0 is the initial value of the ohmic internal resistance, a is the Tafel constant, B is the concentration polarization constant, i 0 is the exchange current density, i L0 is the initial value of the limiting current density, T is the operating temperature of the battery stack, N is the number of battery stack cells, and w k-1 、v k They represent the system process noise at time k-1 and the system measurement noise at time k respectively; the external input is the current load i k , the system output y k is the fuel cell calendar aging voltage V c，k ;

[0123] By linearizing the nonlinear system using Taylor polynomials, the Jacobian matrix of the nonlinear system can be obtained and used for iterative calculation of the nonlinear filter algorithm. Here, the nonlinear filter algorithm uses the extended Kalman filter algorithm. In other embodiments, the nonlinear filter algorithm also includes a particle filter (PF) algorithm and an unscented Kalman filter (UKF) algorithm; the extended Kalman filter algorithm consists of three parts: initialization, prior estimation-prediction and a posteriori estimation-update. The algorithm formula is as follows:

[0124] (1) Initialization:

[0125]

[0126] (2) Prior estimation-prediction:

[0127] x k|k-1 =Ax k-1|k-1 , P k|k-1 =AP k-1|k-1 A T +Q;

[0128] (3) Posterior estimation-update:

[0129] x k|k =x k|k-1 +K k (y k -g(x k ,u k )), P k|k =(IK k H k ) k|k-1 ;

[0130] Among them, x 0|0 , P 0|0 They represent the initial system state and the initial system covariance matrix respectively, E represents the expectation, Var represents the covariance matrix, x(t 0 ) indicates t 0 The system state at the moment, H k represents the Jacobian matrix, g(·) represents the output matrix of the fuel cell aging model, x k represents the aging state of the fuel cell at time k, u k represents the external input of the system at time k, x k|k-1 represents the prior estimate of the system state at time k, x k-1|k-1 represents the posterior estimate of the system state at time k-1, A represents the system matrix, A T represents the transpose of the system matrix, P k|k-1 represents the prior estimate of the covariance matrix at time k, P k-1|k-1 represents the posterior estimate of the covariance matrix at time k-1, Q represents the process error covariance matrix, K k represents the Kalman gain at time k, represents the transpose of the Jacobian matrix, R represents the measurement error covariance matrix, x k|k represents the posterior estimate of the system state at time k, y k represents the measured output of the system, P k|k represents the posterior estimate of the covariance matrix at time k, and I represents the identity matrix;

[0131] The calendar aging voltage dataset was used for training, and a calendar aging voltage model of the fuel cell was constructed. Figure 3 The calendar aging voltage model is used to predict the aging test data under different training data sets. The curve predicted by EKF is the predicted calendar aging voltage result. It can be seen that the calendar aging voltage model can more accurately predict the voltage change trend of the fuel cell in the future.

[0132] Using the reversible aging voltage dataset and the start time of the polarization curve test of the fuel cell, combined with the automatic machine learning method, a reversible aging voltage model is constructed based on the long short-term memory neural network (LSTM neural network);

[0133] In one embodiment, the construction of the reversible aging voltage model in step S3 specifically includes the following steps:

[0134] S321, since the interruption of fuel cell operation during the polarization curve test will cause voltage recovery, the start time of the test can be used as prior information, and the start time of the fuel cell polarization curve test and the reversible aging voltage in the reversible aging voltage data set are integrated to construct a reversible aging data set, and the reversible aging data set is subjected to data standardization processing to improve the training speed of the model, and a standardized reversible aging data set is obtained; in this embodiment, the start time of the polarization curve test shutdown is as follows: t test =[48 185 348 515 658 823 991](h).

[0135] S322, using an automatic machine learning method to automatically optimize the hyperparameters of the long short-term memory neural network to obtain an optimized long short-term memory neural network;

[0136] Specifically, voltage reversible aging information has highly nonlinear characteristics, and recurrent neural networks (RNNs) are suitable for nonlinear modeling of time series data. Long short-term memory neural networks control the transmission of information by introducing three gate structures, avoiding the gradient vanishing or exploding phenomenon in traditional RNN training, thereby improving the prediction accuracy of the model.

[0137] The internal structure of the LSTM neural network unit includes input gate, forget gate, control gate and output gate. The input gate determines which information can pass through the neural unit and is defined as:

[0138] i t =sigm(W xi x t +W hi h t-1 +b i )

[0139] The forget gate decides to ignore the input information from the history memory and is defined as

[0140] f t =sigm(W xf x t +W hf h t-1 +b f )

[0141] The control door control unit status changes from C t-1 Update to C t , defined as

[0142] c t =tanh(W xc x t +W hc h t-1 +b c )

[0143] C t =f t ⊙C t-1 +i t ⊙c t

[0144] The output gate is responsible for the final output and updating the hidden state h t , defined as

[0145] o t =sigm(W xo x t +W ho h t-1 +b o )

[0146] h t =o t ⊙tanh(C t )

[0147] In the above formula, i t represents the input gate, f t represents the forget gate, c t Indicates control gate, o t represents the output gate, sigm is the sigmoid function, x t is the system input at time t, W xi , W hi , W xf , W hf , W xc , W hc , W xo , W ho is the weight matrix of each gate, B i , b f , b c , b o is the bias vector, h t-1 and h t is the hidden vector at time t-1 and time t, C t-1 and C t is the cell state at time t-1 and time t, ⊙ represents the element product, and tanh is used to scale the value to between -1 and +1.

[0148] Automatic machine learning methods are used to automatically optimize the hyperparameters of the long short-term memory neural network, such as the number of neurons, training rounds, and prediction sliding window length. This embodiment selects a genetic algorithm in the automatic machine learning (AutoML) method to perform adaptive optimization of the model's parameters.

[0149] S323, sending the standardized reversible aging data set in step S321 to the optimized long short-term memory neural network in step S322 for training, and obtaining the reversible aging voltage model optimized by the automatic machine learning method after the training is completed.

[0150] S4, using the calendar aging voltage model and the reversible aging voltage model to iteratively and rollingly predict the calendar aging voltage in the calendar aging voltage data set in step S2 and the reversible aging voltage in the reversible aging voltage data set, respectively, to obtain the calendar aging voltage estimate value and the reversible aging voltage estimate value at the future moment, and to superimpose the calendar aging voltage estimate value and the reversible aging voltage estimate value to obtain the final fuel cell stack prediction voltage, so as to realize the prediction of the remaining service life of the fuel cell. Among them, the calendar aging voltage estimate value, the reversible aging voltage estimate value, and the fuel cell stack prediction voltage are respectively used as the calendar aging prediction result, the reversible aging prediction result, and the long-term voltage prediction result.

[0151] Furthermore, the specific operations of step S4 are as follows:

[0152] S41, using the calendar aging voltage model and the reversible aging voltage model to perform iterative rolling prediction on the calendar aging voltage in the calendar aging voltage data set and the reversible aging voltage in the reversible aging voltage data set in step S2, respectively, wherein when performing iterative rolling prediction, we use the estimated value of the model in the previous prediction step As part of the input variables, get the estimated value of the model for the next forecast step By iterating and rolling forecasting in this way, voltage forecasts can be realized for multiple steps in the future, thereby obtaining calendar aging voltage estimates. and the estimated reversible aging voltage

[0153] S42, estimated value of calendar aging voltage and the estimated reversible aging voltage The final fuel cell stack predicted voltage is obtained by superposition. The calculation formula is as follows:

[0154]

[0155] The failure threshold point for predicting the remaining service life of the fuel cell is usually defined as the moment when the initial voltage drops by 5%. The remaining service life RUL of the fuel cell is the predicted voltage failure moment t f Subtract the current prediction time t k :

[0156] RUL=t f -t k

[0157] This enables prediction of the remaining service life of the fuel cell. Figure 4 This is the prediction result of the EKF-LSTM hybrid prediction model proposed in the present invention. It can be seen that the method of the present invention well describes the voltage calendar aging trend and the phenomenon of reversible aging, with high prediction accuracy and strong generalization ability.

[0158] In one embodiment, the RUL prediction results are compared and analyzed based on the relative error, RMSE, and MAPE evaluation indicators, where RMSE is the mean square error of the prediction results and MAPE is the mean absolute percentage error.

[0159] This fuel cell remaining service life prediction method based on a hybrid model establishes a hybrid prediction model for the remaining service life of the fuel cell based on the idea of ​​decomposition prediction. After the original voltage is filtered and decomposed, the calendar aging and reversible aging information of the fuel cell are predicted based on the extended Kalman filter algorithm and the long short-term memory neural network respectively, which combines the advantages of the two methods and improves the prediction accuracy and generalization ability of the model.

[0160] This hybrid model-based fuel cell remaining life prediction method is based on the semi-empirical formula of the polarization equation and introduces a third-order aging factor to characterize the aging information of the fuel cell, thereby more accurately describing the voltage calendar aging process; the prior time information of the polarization test is introduced to achieve more accurate modeling and analysis of the voltage recovery phenomenon.

[0161] This hybrid model-based fuel cell remaining life prediction method uses a genetic algorithm to identify the semi-empirical formula of the polarization equation, and uses an automatic machine learning algorithm (AutoML) to automatically optimize the hyperparameters of the long short-term memory neural network, thereby improving the prediction accuracy and training effect of the model.

[0162] This hybrid model-based fuel cell remaining life prediction method has developed an accurate and efficient fuel cell remaining life prediction method and device, thereby improving its durability, reducing maintenance and costs, and promoting the further popularization and development of proton exchange membrane fuel cell technology, which is of great significance.

[0163] Corresponding to the aforementioned embodiment of the method for predicting the remaining life of a fuel cell based on a hybrid model, the present invention also provides an embodiment of a device for predicting the remaining life of a fuel cell based on a hybrid model.

[0164] See also Figure 5 An embodiment of the present invention provides a fuel cell remaining life prediction device based on a hybrid model, including a memory and one or more processors. The memory stores executable code. When the processor executes the executable code, it is used to implement the fuel cell remaining life prediction method based on the hybrid model in the above embodiment.

[0165] The embodiment of the fuel cell remaining life prediction device based on the hybrid model of the present invention can be applied to any device with data processing capability, and the device with data processing capability can be a device or apparatus such as a computer. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of any device with data processing capability in which it is located reading the corresponding computer program instructions in the non-volatile memory into the internal memory for execution. From the hardware level, if Figure 5 As shown, it is a hardware structure diagram of any device with data processing capability where the fuel cell remaining life prediction device based on the hybrid model of the present invention is located, except Figure 5 In addition to the processor, memory, network interface, and non-volatile memory shown, any device with data processing capabilities in which the apparatus in the embodiments is located may also include other hardware, generally based on the actual functions of the device with data processing capabilities, which will not be described in detail.

[0166] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0167] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, in which the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of the present invention. Ordinary technicians in this field can understand and implement it without paying creative work.

[0168] An embodiment of the present invention further provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, the method for predicting the remaining life of a fuel cell based on a hybrid model in the above embodiment is implemented.

[0169] The computer-readable storage medium may be an internal storage unit of any device with data processing capability in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device of any device with data processing capability, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capability. The computer-readable storage medium is used to store computer programs and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.

[0170] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0171] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0172] The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a", "a", "an", and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0173] It should be understood that although the terms first, second, third, etc. may be used to describe various information in one or more embodiments of this specification, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0174] The above are only preferred embodiments of one or more embodiments of this specification and are not intended to limit one or more embodiments of this specification. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of this specification should be included in the scope of protection of one or more embodiments of this specification.

Claims

1. A method for predicting the remaining life of a fuel cell based on a hybrid model, characterized in that: The following steps are involved: S1, obtaining the operation monitoring data of the fuel cell, constructing the monitoring data set and performing preprocessing; S2, using a filtering algorithm to decompose the battery stack voltage in the monitoring data set to obtain a calendar aging voltage and a reversible aging voltage, and constructing a calendar aging voltage data set from the calendar aging voltage and a reversible aging voltage data set from the reversible aging voltage; S3, using polarization curve data obtained by polarization curve testing of the fuel cell, establishing a semi-mechanism aging model based on an empirical polarization equation; using a calendar aging voltage data set, combined with the semi-mechanism aging model with parameter identification processing based on a genetic algorithm, and constructing a calendar aging voltage model based on a nonlinear filter algorithm; Using the reversible aging voltage dataset and the start time of the polarization curve test of the fuel cell, combined with the automatic machine learning method, a reversible aging voltage model was constructed based on the long short-term memory neural network; S4, using the calendar aging voltage model and the reversible aging voltage model to iteratively and rollingly predict the calendar aging voltage in the calendar aging voltage data set in step S2 and the reversible aging voltage in the reversible aging voltage data set, respectively, to obtain the calendar aging voltage estimate value and the reversible aging voltage estimate value at the future moment, and to superimpose the calendar aging voltage estimate value and the reversible aging voltage estimate value to obtain the final fuel cell stack prediction voltage, so as to realize the remaining service life prediction of the fuel cell.

2. The method for predicting the remaining life of a fuel cell based on a hybrid model according to claim 1, characterized in that: In the step S1, the preprocessing is specifically to downsample the stack voltage data in the monitoring data set, and the downsampling interval is 1 hour.

3. The method for predicting the remaining life of a fuel cell based on a hybrid model according to claim 1, characterized in that: The step S2 specifically includes the following steps: S21, using a local weighted regression filtering algorithm to smooth the stack voltage data in the monitoring data set; The local weighted regression filtering algorithm includes weighted linear least squares method and second-order polynomial model. The window width of the filter is set to 300. The weight coefficient of the data point in the filter window is calculated by the weight function. The weight coefficient is defined as follows: Among them, w i is the weight coefficient, Δ(x) is the window width, x is the horizontal coordinate of the point to be fitted, indicating the measurement time of the data, and x i is the horizontal coordinate of the data point in the window, |·| is the absolute value, and the smoothing value of the fitting point (x, y) It can be obtained by weighted least square method; y is the ordinate of the point to be fitted, indicating the stack voltage; Represents the smoothed value of the stack voltage; the objective function J(a,b) minimized by the weighted least squares method in the local weighted regression filtering algorithm is defined as follows: Where m is the number of data points in the window, w i is the weight coefficient; y i is the function value corresponding to the data point, indicating the stack voltage; a and b are the parameters of the regression curve, x i is the horizontal coordinate of the data point in the window, indicating time; after obtaining the parameters a and b of the non-parametric regression curve of the stack voltage monitoring data, the smoothed value of the current voltage data point can be obtained S22, stack voltage V st (t) After filtering by S21, the calendar aging voltage V representing the aging trend of the fuel cell is obtained. c (t), t represents the measurement time of voltage, V st (t) minus V c (t) can get the reversible aging voltage V r (t), the final fuel cell aging voltage decomposition expression is as follows: V st (t)=V c (t)+V r (t) Therefore, a calendar aging voltage data set is constructed from the obtained calendar aging voltage and a reversible aging voltage data set is constructed from the reversible aging voltage.

4. The method for predicting the remaining life of a fuel cell based on a hybrid model according to claim 1, characterized in that: The construction of the calendar aging voltage model in step S3 specifically includes the following steps: S311, using the polarization curve data obtained by the polarization curve test of the fuel cell, a semi-mechanism aging model is established based on the semi-empirical formula of the polarization equation and the introduction of the aging attenuation factor. The semi-empirical formula of the polarization equation can be expressed as follows: Among them, V c (t) is the calendar aging voltage at time t, t represents a certain moment in the operation of the fuel cell, N is the number of cells in the fuel cell stack, E ocv is the open circuit voltage, i(t) is the fuel cell output current at time t, R ohm is the ohmic internal resistance, a is the Tafel constant, b is the concentration polarization constant, i0(t) is the exchange current density at time t, i L (t) is the limiting current density at time t, and T is the stack operating temperature; According to the meaning of the parameters of the semi-empirical formula of the polarization equation and the changes of the parameters under different aging conditions, the ohmic internal resistance R is selected. ohm and the limiting current density i L As the aging parameter, the aging factor α is introduced to represent the change of the aging parameter, and its expression is as follows: Among them, α(t) is the aging factor that characterizes the aging state of the fuel cell at time t, R0 is the initial value of the ohmic internal resistance, and i L0 is the initial value of the limiting current; the above equations can be combined to obtain the functional relationship between the external voltage of the fuel cell and the aging time, which is used as a semi-mechanism aging model and expressed as follows: Among them, V c (t) is the calendar aging voltage at time t, t represents a certain moment in the operation of the fuel cell, N is the number of cells in the fuel cell stack, E ocv is the open circuit voltage, i(t) is the fuel cell output current at time t, α(t) is the aging factor to characterize the aging state of the fuel cell at time t, R0 is the initial value of the ohmic internal resistance, a is the Tafel constant, b is the concentration polarization constant, i0(t) is the exchange current density at time t, i L0 is the initial value of the limiting current, T is the operating temperature of the stack; S312, performing parameter identification on the semi-mechanism aging model established in step S311 based on a genetic algorithm, and obtaining the optimal solution of the parameters through the fitness function in the genetic algorithm, as follows: First, the parameter identification problem is transformed into an optimization problem, and the fitness function of the genetic algorithm is expressed as: Where k is the sampling point of the polarization curve, V c,k is the actual value of voltage, To identify the fitting value of the model; there are 6 dimensions of parameters that need to be identified: [E ocv ,R0,a,b,i0,i L0 ]; The genetic algorithm first randomly initializes the population, sets the crossover rate and mutation rate, and then calculates the fitness function f of the population. fitmess The size is used to perform selection, crossover, and mutation operations on the variables, and the optimal solution is selected after repeated operations to obtain the parameter identification results; S313, based on the semi-mechanism aging model processed by parameter identification in step S312, a third-order aging attenuation factor is introduced, as follows: The nonlinear system of the fuel cell aging process is expressed as follows: In the formula, x k-1 、x k They represent the aging state of the fuel cell at time k-1 and time k respectively, and u k-1 、u k They represent the external input at time k-1 and time k respectively, y k is the system output at time k, w k-1 、v k represents the system process noise at time k-1 and the system measurement noise at time k, respectively. f(·) and g(·) represent the state transfer matrix and output matrix of the fuel cell aging model. The external input is the current load i k , the system output y k is the fuel cell calendar aging voltage V c,k ; Define the third-order aging factor x k =[α k ,β k ,γ k ] T , the discrete time state space equation of the fuel cell aging process can be obtained: Among them, Δt is the sampling time interval, α k and α k-1 As the aging factor, β represents the aging state at time k and time k-1 respectively. k and β k-1 They represent the aging rates at time k and time k-1 respectively, γ k and γ k-1 They represent the second-order derivatives of the aging state at time k and time k-1, respectively, and E ocv is the open circuit voltage, i k is the fuel cell load current at time k, R0 is the initial value of the ohmic internal resistance, a is the Tafel constant, b is the concentration polarization constant, i0 is the exchange current density, i L0 is the initial value of the limiting current density, T is the operating temperature of the battery stack, N is the number of battery stack cells, and w k-1 、v k They represent the system process noise at time k-1 and the system measurement noise at time k respectively; the external input is the current load i k , the system output y k is the fuel cell calendar aging voltage V c,k ; By linearizing the nonlinear system using Taylor polynomials, the Jacobian matrix of the nonlinear system can be obtained and used for the iterative calculation of the extended Kalman filter algorithm. The extended Kalman filter algorithm consists of three parts: initialization, prior estimation-prediction, and posterior estimation-update. The algorithm formula is as follows: (1) Initialization: x 0∣0 =E[x(t0)],P 0∣0 =There is[x(t0)], (2) Prior estimation-prediction: x k∣k-1 =Ax k-1∣k-1 ,P k∣k-1 =AP k-1∣k-1 A T +Q; (3) Posterior estimation-update: x k∣k =x k∣k-1 +K k (y k -g(x k ,u k )),P k∣k =(I-K k H k )P k∣k-1 ; Among them, x 0∣0 , P 0∣0 They represent the initial system state and the initial system covariance matrix respectively, E represents the expectation, Var represents the covariance matrix, x(t0) represents the system state at time t0, H k represents the Jacobian matrix, g(·) represents the output matrix of the fuel cell aging model, x k represents the aging state of the fuel cell at time k, u k represents the external input of the system at time k, x k∣k-1 represents the prior estimate of the system state at time k, x k-1∣k-1 represents the posterior estimate of the system state at time k-1, A represents the system matrix, A T represents the transpose of the system matrix, P k∣k-1 represents the prior estimate of the covariance matrix at time k, P k-1∣k-1 represents the posterior estimate of the covariance matrix at time k-1, Q represents the process error covariance matrix, K k represents the Kalman gain at time k, represents the transpose of the Jacobian matrix, R represents the measurement error covariance matrix, x k∣k represents the posterior estimate of the system state at time k, y k represents the measured output of the system, P k∣k represents the posterior estimate of the covariance matrix at time k, and I represents the identity matrix; The calendar aging voltage dataset was used for training, and a calendar aging voltage model of the fuel cell was constructed.

5. The method for predicting the remaining life of a fuel cell based on a hybrid model according to claim 1, characterized in that: In step S3, the reversible aging voltage data set and the start time of the polarization curve test of the fuel cell are used, combined with the automatic machine learning method, and the reversible aging voltage model is constructed based on the long short-term memory neural network, which specifically includes the following steps: S321, combining the reversible aging voltage in the reversible aging voltage data set with the start time of the polarization curve test of the fuel cell to construct a reversible aging data set, and performing data standardization processing on the reversible aging data set to obtain a standardized reversible aging data set; S322, using an automatic machine learning method to automatically optimize the hyperparameters of the long short-term memory neural network to obtain an optimized long short-term memory neural network; S323, sending the standardized reversible aging data set in step S321 to the optimized long short-term memory neural network in step S322 for training to construct a reversible aging voltage model.

6. The method for predicting the remaining life of a fuel cell based on a hybrid model according to claim 5, characterized in that: The specific operation of step S4 is as follows: S41, using the calendar aging voltage model and the reversible aging voltage model to perform iterative rolling prediction on the calendar aging voltage in the calendar aging voltage data set and the reversible aging voltage in the reversible aging voltage data set in step S2, respectively, wherein when performing iterative rolling prediction, we use the estimated value of the model in the previous prediction step As part of the input variables, get the estimated value of the model for the next forecast step By iterating and rolling forecasting in this way, voltage forecasts can be realized for multiple steps in the future, thereby obtaining calendar aging voltage estimates. and the estimated reversible aging voltage S42, estimated value of calendar aging voltage and the estimated reversible aging voltage The final fuel cell stack predicted voltage is obtained by superposition. The calculation formula is as follows: Among them, the failure threshold point of the remaining service life prediction of the fuel cell is selected, and the remaining service life RUL of the fuel cell is the voltage failure time t at which the predicted failure threshold point is located. f Subtract the current prediction time t k : RUL=t f -t k This enables prediction of the remaining service life of the fuel cell.

7. The method for predicting the remaining life of a fuel cell based on a hybrid model according to claim 6, characterized in that: The RUL prediction results are compared and analyzed based on the relative error, RMSE, and MAPE evaluation indicators, where RMSE is the mean square error of the prediction results and MAPE is the mean absolute percentage error.

8. A fuel cell remaining life prediction device based on a hybrid model, comprising a memory and one or more processors, wherein the memory stores executable code, characterized in that: When the processor executes the executable code, it is used to implement the fuel cell remaining life prediction method based on the hybrid model as described in any one of claims 1-7.

Images