Hydrogen fuel cell voltage degradation hybrid prediction method and system

Abstract

The application discloses a hydrogen fuel cell voltage degradation hybrid prediction method and system, and the method comprises the following steps: obtaining electrochemical impedance data and voltage historical operation data of a hydrogen fuel cell, and constructing an equivalent circuit model; based on the equivalent circuit model, performing parameter identification processing on the hydrogen fuel cell to obtain health state characteristic parameters of the hydrogen fuel cell; based on the health state characteristic parameters of the hydrogen fuel cell, performing voltage prediction processing on the hydrogen fuel cell by using a data driving method to obtain voltage degradation prediction results of the hydrogen fuel cell. By using the application, the accuracy of long-term prediction of hydrogen fuel cell voltage data can be improved while simplifying complex mechanism modeling. The application can be widely applied to the technical field of fuel cells as a hydrogen fuel cell voltage degradation hybrid prediction method and system.

Description

Technical Field

[0001] This invention relates to the field of fuel cell technology, and in particular to a hybrid method and system for predicting voltage degradation in hydrogen fuel cells. Background Technology

[0002] Hydrogen fuel cells are considered an ideal power source solution due to their outstanding advantages of high efficiency, low carbon emissions, and environmental friendliness. On the one hand, hydrogen fuel cells can be applied in industrial power generation, military equipment, manufacturing, and transportation. On the other hand, hydrogen energy has shown great development potential in recent years. Nevertheless, hydrogen fuel cells are still difficult to deploy and commercialize on a large scale. This is due not only to the high cost of fuel cells themselves, but also to the low durability and poor reliability of fuel cells. Currently, there are three main methods for predicting the degradation of hydrogen fuel cells: model-based, data-driven, and hybrid methods. Model-based methods are white-box methods, which have the advantage of requiring less data and can be used for long-term prediction problems. However, model-based methods generally involve complex mechanism modeling. Fuel cells are complex devices, and many of their degradation mechanisms are not yet clear. Data-driven methods are black-box methods, which do not require much understanding of the degradation process mechanism, but this method requires a large amount of data and performs poorly in long-term prediction problems. Summary of the Invention

[0003] To address the aforementioned technical problems, the present invention aims to provide a hybrid prediction method and system for voltage degradation of hydrogen fuel cells, which can improve the accuracy of long-term prediction of hydrogen fuel cell voltage data while simplifying complex mechanism modeling.

[0004] The first technical solution adopted in this invention is: a hybrid prediction method for voltage degradation of hydrogen fuel cells, comprising the following steps:

[0005] Obtain electrochemical impedance data and historical voltage operation data of hydrogen fuel cells, and construct an equivalent circuit model;

[0006] Based on the equivalent circuit model, parameter identification processing is performed on the hydrogen fuel cell to obtain the health status characteristic parameters of the hydrogen fuel cell.

[0007] Based on the health status characteristic parameters of hydrogen fuel cells, the voltage of hydrogen fuel cells is predicted using a data-driven method, resulting in a predicted voltage degradation result for hydrogen fuel cells.

[0008] Furthermore, the equivalent circuit model includes a cathode process semicircular arc, an anode process semicircular arc, and an ohmic impedance point, wherein:

[0009] The cathode process includes cathode activation polarization loss and concentration polarization loss, represented by a constant-phase element and a resistor in parallel;

[0010] The anodic process includes anodic activation polarization loss and concentration polarization loss, represented by a constant-phase element and a resistor connected in parallel;

[0011] The ohmic impedance point includes the loss caused by proton conduction resistance, expressed as a resistance.

[0012] Furthermore, the step of performing parameter identification processing on the hydrogen fuel cell based on the equivalent circuit model to obtain the health status characteristic parameters of the hydrogen fuel cell specifically includes:

[0013] The electrochemical impedance spectral parameters of the hydrogen fuel cell were obtained by fitting the electrochemical impedance data of the equivalent circuit model using the least squares method.

[0014] By selectively processing the electrochemical impedance spectroscopy parameters of hydrogen fuel cells, characteristic parameters of the hydrogen fuel cell's health status are obtained.

[0015] Furthermore, the specific electrochemical impedance spectroscopy parameter values ​​include the following parameters:

[0016] θ=(R ohm ,R A C dl,A ,n CPE,A ,R C C dl,C ,n CPE,C )

[0017] In the above formula, R ohm R represents the ohmic internal resistance. A C represents the parallel resistance of the anode. dl,A Indicates a constant phase angle element connected in parallel with the anode, n CPE,A R represents the constant of the parallel anode constant phase angle element. C C represents the parallel resistance of the cathode. dl,C Indicates a cathode connected in parallel with a constant phase angle element, n CPE,C This represents the constant of the cathode parallel constant phase angle element.

[0018] Furthermore, the step of predicting the voltage of the hydrogen fuel cell based on its health status characteristic parameters using a data-driven method to obtain the predicted voltage degradation result of the hydrogen fuel cell specifically includes:

[0019] Based on the health status characteristic parameters of hydrogen fuel cells, a functional relationship is constructed by combining the voltage and time changes of hydrogen fuel cells.

[0020] The self-recovery voltage prediction result is obtained by predicting the functional relationship using the self-recovery voltage prediction method.

[0021] Based on the historical operating data of hydrogen fuel cell voltage, the overall degradation trend of hydrogen fuel cell voltage is predicted by the particle filter prediction algorithm, and the trend prediction value is obtained.

[0022] Based on the trend prediction value, residual processing is performed on the historical voltage operation data to obtain the residual value;

[0023] The residual values ​​are predicted using the forest regression prediction algorithm to obtain the fluctuation prediction values;

[0024] The trend forecast and fluctuation forecast are added together to obtain the voltage degradation forecast result of the hydrogen fuel cell.

[0025] Furthermore, the step of predicting the overall degradation trend of the hydrogen fuel cell voltage using a particle filter prediction algorithm based on the historical operating data of the hydrogen fuel cell voltage to obtain the trend prediction value also includes adjusting the particle state within the hydrogen fuel cell through the self-recovering voltage prediction results during the overall voltage degradation trend prediction process.

[0026] Furthermore, the specific calculation process of the forest regression prediction algorithm is as follows:

[0027]

[0028] In the above formula, ρ k γ represents the autocorrelation coefficient, γ0 represents the covariance of the signal when the interval k is 0, i.e., the sample variance. k X represents the autocovariance, N represents the sequence length, and X represents the autocovariance. t Indicates a sequence segment.

[0029] The second technical solution adopted in this invention is: a hybrid prediction system for voltage degradation of hydrogen fuel cells, comprising:

[0030] A module is built to acquire electrochemical impedance data and historical voltage operating data of hydrogen fuel cells and construct an equivalent circuit model.

[0031] The identification module, based on the equivalent circuit model, performs parameter identification processing on the hydrogen fuel cell to obtain the health status characteristic parameters of the hydrogen fuel cell.

[0032] The prediction module, based on the health status characteristic parameters of the hydrogen fuel cell, uses a data-driven method to predict the voltage of the hydrogen fuel cell and obtain the voltage degradation prediction result.

[0033] The beneficial effects of the method and system of this invention are as follows: This invention acquires historical state data of hydrogen fuel cells and constructs an equivalent circuit model. Based on the model construction method, it identifies the parameters of hydrogen fuel cells, avoids complex mechanism modeling, and also gets rid of excessive dependence on data. The voltage of hydrogen fuel cells is predicted by a data-driven method, which improves the accuracy of long-term prediction and reduces prediction error. Attached Figure Description

[0034] Figure 1 This is a flowchart of the steps of a hybrid prediction method for voltage degradation in a hydrogen fuel cell according to the present invention.

[0035] Figure 2 This is a structural block diagram of a hybrid prediction system for voltage degradation in a hydrogen fuel cell according to the present invention.

[0036] Figure 3 This is a schematic diagram of the equivalent circuit model constructed in a specific embodiment of the present invention. Detailed Implementation

[0037] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only for ease of explanation and do not limit the order of the steps. The execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.

[0038] Reference Figure 1 This invention provides a hybrid prediction method for voltage degradation in hydrogen fuel cells, the method comprising the following steps:

[0039] S1. Obtain electrochemical impedance data and historical voltage operation data of hydrogen fuel cells, and construct an equivalent circuit model;

[0040] Specifically, refer to Figure 3 The equivalent circuit model includes a cathode process semicircular arc, an anode process semicircular arc, and an ohmic impedance point. The cathode process includes cathode activation polarization loss and concentration polarization loss, represented by a constant-phase element CPE. dl,C and a resistor R C Parallel connection; the anodic process includes anodic activation polarization loss and concentration polarization loss, with a constant phase element CPE. dl,A and a resistor R A Parallel connection; the ohmic impedance point includes the loss caused by proton conduction resistance, expressed as a resistance R. ohm express.

[0041] S2. Based on the equivalent circuit model, parameter identification processing is performed on the hydrogen fuel cell to obtain the health status characteristic parameters of the hydrogen fuel cell.

[0042] S21. The electrochemical impedance data of the equivalent circuit model is fitted using the least squares method to obtain the electrochemical impedance spectral parameter values ​​of the hydrogen fuel cell.

[0043] Specifically, the total impedance Z is known. tol The expression for impedance, which shows that impedance can be considered as a function of angular frequency ω and parameter θ, is shown below:

[0044] Z = f(ω) k ,θ)

[0045] In the above formula, Z represents impedance, ω k Represents the angular frequency, k = 1, 2, ..., n;

[0046] For EIS data measured at a certain moment, it is equivalent to having n sets of observation data {(ω1,Z1),(ω2,Z2),…(ω n Z n )}, based on this set of data, calculate the model parameters θ = (R) at the corresponding time. ohm ,R A C dl,A ,n CPE,A ,R C C dl,C ,n CPE,C Therefore, this work can be viewed as solving a nonlinear least squares problem.

[0047] The first-order Taylor expansion of the input function g(x) to be fitted is shown below:

[0048] g(x+Δx)≈g(x)+J(x) T Δx

[0049] In the above formula, J(x) represents the first derivative of g(x) with respect to x, i.e., the Jacobian matrix, and Δx represents the increment of x;

[0050] At this point, the nonlinear least squares problem can be expressed as finding the increment Δx such that ||g(x+Δx)|| 2 The smallest, its expression is:

[0051]

[0052] The above equation is the objective function, which can be expanded by square terms:

[0053]

[0054] Furthermore, by differentiating Δx and setting the derivative to 0, we obtain:

[0055] J(x) T J(x)Δx=-J(x)T g(x)

[0056] Let H = J T J, B = J T g, yielding the incremental equation:

[0057] HΔx=B

[0058] Therefore, solving the above equation yields the increment Δx;

[0059] In solving for the parameters of the equivalent circuit model, let the error be g(θ)=Zf(ω) k ,θ)

[0060] Then, a system of nonlinear equations can be obtained from n sets of observation data:

[0061]

[0062] Thus, the least squares problem is constructed:

[0063]

[0064] Thus far, the steps for solving the equivalent circuit parameters using the Gauss-Newton method are as follows: given initial parameter values ​​θ0; perform iterative calculations; for the l-th iteration, calculate the corresponding Jacobian matrix and error g(θ). l Solve the incremental equation HΔθ l =B; when Δθ l Small enough (e.g., Δθ) l / θ l <10 -4 If θ = 0, then stop iterating; otherwise, let θ = 0. l+1 =θ l +Δθ l Return to the iterative calculation steps;

[0065] There are a total of 7 electrochemical impedance spectroscopy parameters for the hydrogen fuel cell, as shown below:

[0066] θ=(R ohm ,R A C dl,A ,n CPE,A ,R C C dl,C ,n CPE,C )

[0067] In the above formula, R ohm R represents the ohmic internal resistance. A C represents the parallel resistance of the anode. dl,A Indicates a constant phase angle element connected in parallel with the anode, n CPE,A R represents the constant of the parallel anode constant phase angle element. C C represents the parallel resistance of the cathode. dl,CIndicates a cathode connected in parallel with a constant phase angle element, n CPE,C This represents the constant of the cathode parallel constant phase angle element;

[0068] S22. Selective processing of electrochemical impedance spectroscopy parameter values ​​of hydrogen fuel cells: Single parameters or combinations of multiple parameters (such as the sum of multiple parameters) whose values ​​change linearly with time are regarded as health status characteristic parameters of hydrogen fuel cells.

[0069] Specifically, characteristic parameters suitable for characterizing the health status of hydrogen fuel cells are selected from the electrochemical impedance spectroscopy parameter values. These characteristic parameters are as follows:

[0070] R pol =R ohm +R A +R C

[0071] In the above formula, R pol These are the characteristic parameters representing the health status of a hydrogen fuel cell.

[0072] S3. Based on the health status characteristic parameters of the hydrogen fuel cell, the voltage of the hydrogen fuel cell is predicted using a data-driven method to obtain the voltage degradation prediction result of the hydrogen fuel cell.

[0073] S31. Based on the health status characteristic parameters of hydrogen fuel cells, construct a functional relationship by combining the voltage and time changes of hydrogen fuel cells;

[0074] Specifically, the acquisition of the health status characteristic parameter R of the hydrogen fuel cell pol To understand the relationship between voltage and time, it is first necessary to establish the health state characteristic parameter R of the hydrogen fuel cell. pol A regression model with time was used, and then the health status characteristic parameters R of the hydrogen fuel cell were established. pol A regression model of the voltage at the self-recovery moment is used, and then the corresponding functional expression is constructed based on the relationship between the three.

[0075] The function expression is as follows:

[0076] R pol =2.035×10 -3 t+15.700

[0077] U rec =4.527-0.074R pol

[0078] In the above formula, U rec The voltage at the self-recovery moment is represented by t, and time is represented by t.

[0079] S32. The self-recovery voltage prediction method is used to predict the functional relationship and obtain the self-recovery voltage prediction result.

[0080] Specifically, comparing the identified impedance results with the characterized self-recovery voltage value revealed that the self-recovery voltage value is related to the health factor R. pol There is a very strong linear relationship:

[0081] U rec =F(R) pol ,t)

[0082] Among them, R pol =R ohm +R A +R C

[0083] Therefore, the voltage after self-recovery can be predicted based on the predicted health factors. For example, by linearly fitting the polarization resistance values ​​of the two fuel cells FC1 and FC2 515 hours ago with the recovered voltage values ​​at the corresponding times, the following linear models can be obtained:

[0084] R pol,FC1 =2.035×10 -3 t+15.700

[0085] R pol,FC2 =3.410×10 -3 t+16.001

[0086] U pol,FC1 =4.527-0.074R pol,FC1

[0087] U pol,FC2 =4.134-0.051R pol,FC2

[0088] Based on the above self-recovery voltage prediction model, the self-recovery voltage U after 515 hours can be predicted. rec Make predictions

[0089] S33. Based on the historical operating data of the hydrogen fuel cell voltage, the overall degradation trend of the hydrogen fuel cell voltage is predicted by the particle filter prediction algorithm to obtain the trend prediction value.

[0090] Specifically, the decreasing trend of voltage is first expressed using the following empirical formula:

[0091] u k =u k-1 -b·(t k -t k-1 )

[0092] In the above formula, u represents the voltage trend, and b is the rate of change of the trend over time;

[0093] In particle filtering, process noise v can be eliminated by the uncertainty of model parameters. Therefore, the above equation also serves as the state transition equation for particle filtering, assuming that the observation error follows a Gaussian distribution τ. k If ~N(0,σ), then the parameters involved are θ=(u,b,σ);

[0094] Based on the basic theory of particle filtering introduced above, the main idea of ​​particle filtering is to realize the sequential update of probability information through the changes in the state and weight of particles. Therefore, the main steps of particle filtering prediction include particle generation, initialization, and particle propagation, that is, particle state prediction followed by particle weight update, particle resampling, and finally, at the final observation time, the trained parameters are used to predict the trend.

[0095] First, generate N particles and set the initial distribution of the particles. Assuming that the posterior probability distribution of the last step is available, use it as the prior. Then, propagate all particles based on the state transition function.

[0096] Secondly, the particle weights are updated according to the likelihood function of the following formula, as shown below.

[0097]

[0098] Next, resampling is performed. This paper uses a multinomial resampling method, where high-weight particles are copied and low-weight particles are discarded. This process is repeated until the final observation point t is reached. ob Previously, the particles repeated the above-mentioned propagation, weight update, and resampling process;

[0099] Reaching the final observation point t ob Then, the voltage trend is predicted using the parameters obtained from training;

[0100] Since the particle filter prediction algorithm obtains the overall smooth downward trend of voltage from the historical operating data of hydrogen fuel cells, the self-recovering voltage prediction result U can be used in the process of predicting the overall voltage degradation trend. rec For hydrogen fuel cells at self-recovery time t c The particle state at that time is adjusted. Specifically, during the prediction phase, when the prediction time coincides with the characterization time, the self-recovery voltage prediction value U is used. rec Based on prior information, all particles are adjusted upwards as a whole.

[0101] S34. Based on the trend prediction value, perform residual processing on the historical voltage operation data to obtain the residual value;

[0102] S35. The residual values ​​are predicted using the forest regression prediction algorithm to obtain the fluctuation prediction values;

[0103] Specifically, the residual value information represents the periodic fluctuation information in the voltage degradation process. The solution method is to determine the periodicity p and divide the data by taking the remainder to train and predict the random forest regressor. The periodicity p can be determined by solving the autocorrelation coefficient ρ.

[0104] Solve for the autocorrelation function and obtain its periodicity;

[0105] The specific calculation process of the forest regression prediction algorithm is as follows:

[0106]

[0107] In the above formula, ρ k γ represents the autocorrelation coefficient, γ0 represents the covariance of the signal when the interval k is 0, i.e., the sample variance. k X represents the autocovariance, N represents the sequence length, and X represents the autocovariance. t Indicates a sequence segment;

[0108] The periodicity p of the fluctuation is obtained by arranging the function values ​​from largest to smallest.

[0109] The data partitioning method is as follows: for each value p in the periodic sequence i (i = 1, 2, ...), the training data is processed according to the following formula:

[0110]

[0111] In the above formula, t m Represents the sample, y m The label represents the value of the residual at time m, where the sample is at "time m". m ,m%p i Indicates the numerical value m relative to the period p i Find the remainder and use it as input for training;

[0112] Based on the trained model, the test set input for the data to be predicted for the next n hours is:

[0113] t m =n%p i

[0114] S36. Add the trend prediction value and the fluctuation prediction value together to obtain the voltage degradation prediction result of the hydrogen fuel cell.

[0115] Specifically, the final output of the prediction method is the sum of the prediction results of steps S33 and S35. That is, the final voltage degradation prediction result of the hydrogen fuel cell is the sum of the trend prediction value and the fluctuation prediction value. The reliable voltage prediction result provides effective information for the maintenance of the fuel cell. On the one hand, it helps maintenance personnel to take measures in advance to avoid further deterioration of the fuel cell stack and ensure the safety of the fuel cell stack. On the other hand, it can reduce unnecessary periodic maintenance and reduce maintenance costs.

[0116] Reference Figure 2 A hybrid voltage degradation prediction system for hydrogen fuel cells, comprising:

[0117] A module is built to acquire electrochemical impedance data and historical voltage operating data of hydrogen fuel cells and construct an equivalent circuit model.

[0118] The identification module, based on the equivalent circuit model, performs parameter identification processing on the hydrogen fuel cell to obtain the health status characteristic parameters of the hydrogen fuel cell.

[0119] The prediction module, based on the health status characteristic parameters of the hydrogen fuel cell, uses a data-driven method to predict the voltage of the hydrogen fuel cell and obtain the voltage degradation prediction result.

[0120] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0121] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A hybrid prediction method for voltage degradation in hydrogen fuel cells, characterized in that, Includes the following steps: Obtain electrochemical impedance data and historical voltage operating data of hydrogen fuel cells, and construct an equivalent circuit model. Based on the equivalent circuit model, parameter identification processing is performed on the hydrogen fuel cell to obtain the health status characteristic parameters of the hydrogen fuel cell. Based on the health status characteristic parameters of hydrogen fuel cells, a data-driven method is used to predict the voltage of hydrogen fuel cells, resulting in voltage degradation prediction results, specifically including: Based on the health status characteristic parameters of hydrogen fuel cells, a functional relationship is constructed by combining the voltage and time changes of hydrogen fuel cells. The self-recovery voltage prediction result is obtained by predicting the functional relationship using the self-recovery voltage prediction method. Based on the historical operating data of hydrogen fuel cell voltage, the overall degradation trend of hydrogen fuel cell voltage is predicted by the particle filter prediction algorithm, and the trend prediction value is obtained. Based on the trend prediction value, residual processing is performed on the historical voltage operation data to obtain the residual value; The residual value is predicted by the forest regression prediction algorithm to obtain the fluctuation prediction value. Specifically, the residual value represents the periodic fluctuation information in the voltage degradation process. The periodicity is obtained by solving the autocorrelation function and arranging the function values ​​from large to small to obtain the fluctuation periodicity. The data is divided by taking the remainder to train and predict the random forest regressor. The trend forecast and fluctuation forecast are added together to obtain the voltage degradation forecast result of the hydrogen fuel cell.

2. The hybrid prediction method for voltage degradation of a hydrogen fuel cell according to claim 1, characterized in that, The equivalent circuit model includes a cathode process semicircular arc, an anode process semicircular arc, and an ohmic impedance point, wherein: The cathode process includes cathode activation polarization loss and concentration polarization loss, represented by a constant-phase element and a resistor in parallel; The anodic process includes anodic activation polarization loss and concentration polarization loss, represented by a constant-phase element and a resistor connected in parallel; The ohmic impedance point includes the loss caused by proton conduction resistance, expressed as a resistance.

3. The hybrid prediction method for voltage degradation of a hydrogen fuel cell according to claim 2, characterized in that, The step of performing parameter identification processing on the hydrogen fuel cell based on the equivalent circuit model to obtain the health status characteristic parameters of the hydrogen fuel cell specifically includes: The electrochemical impedance spectral parameters of the hydrogen fuel cell were obtained by fitting the electrochemical impedance data of the equivalent circuit model using the least squares method. By selectively processing the electrochemical impedance spectroscopy parameters of hydrogen fuel cells, characteristic parameters of the hydrogen fuel cell's health status are obtained.

4. The hybrid prediction method for voltage degradation of a hydrogen fuel cell according to claim 3, characterized in that, The specific electrochemical impedance spectroscopy parameters include the following parameters: In the above formula, Indicates the internal resistance of the ohm. Indicates the parallel resistance of the anode. This indicates a constant phase angle element connected in parallel with the anode. This represents the constant of the parallel anode constant phase angle element. Indicates the parallel resistance of the cathode. This indicates a cathode connected in parallel with a constant phase angle element. This represents the constant of the cathode parallel constant phase angle element.

5. The hybrid prediction method for voltage degradation of a hydrogen fuel cell according to claim 4, characterized in that, The step of using historical operating data of hydrogen fuel cell voltage to predict the overall degradation trend of hydrogen fuel cell voltage through a particle filter prediction algorithm to obtain the trend prediction value also includes adjusting the particle state within the hydrogen fuel cell through the self-recovery voltage prediction results during the overall voltage degradation trend prediction process.

6. A hybrid voltage degradation prediction system for hydrogen fuel cells, characterized in that, Includes the following modules: A module is built to acquire electrochemical impedance data and historical voltage operating data of hydrogen fuel cells and construct an equivalent circuit model. The identification module is used to perform parameter identification processing on hydrogen fuel cells based on the equivalent circuit model to obtain the health status characteristic parameters of hydrogen fuel cells. The prediction module is used to predict the voltage of the hydrogen fuel cell based on its health status characteristic parameters using a data-driven method, obtaining the predicted voltage degradation result. Specifically, it includes: Based on the health status characteristic parameters of hydrogen fuel cells, a functional relationship is constructed by combining the voltage and time changes of hydrogen fuel cells. The self-recovery voltage prediction result is obtained by predicting the functional relationship using the self-recovery voltage prediction method. Based on the historical operating data of hydrogen fuel cell voltage, the overall degradation trend of hydrogen fuel cell voltage is predicted by the particle filter prediction algorithm, and the trend prediction value is obtained. Based on the trend prediction value, residual processing is performed on the historical voltage operation data to obtain the residual value; The residual value is predicted by the forest regression prediction algorithm to obtain the fluctuation prediction value. Specifically, the residual value represents the periodic fluctuation information in the voltage degradation process. The periodicity is obtained by solving the autocorrelation function and arranging the function values ​​from large to small to obtain the fluctuation periodicity. The data is divided by taking the remainder to train and predict the random forest regressor. The trend forecast and fluctuation forecast are added together to obtain the voltage degradation forecast result of the hydrogen fuel cell.

Images