Method and system for predicting performance degradation of proton exchange membrane fuel cell in dynamic operation

Abstract

The invention provides a dynamic operation proton exchange membrane fuel cell performance degradation prediction method and system. The method comprises the following steps: obtaining fuel cell aging data through a sensor measurement device, and carrying out noise removal and downsampling processing on the fuel cell aging data; fitting a fuel cell semi-empirical equation through polarization curve data, and calculating a relative voltage attenuation rate as a health index for dynamically operating the fuel cell; taking the relative voltage attenuation rate as a target value, calculating an SHAP value of each degradation parameter, and screening the degradation parameters as input parameters of a data driving model; the method comprises the following steps of: constructing a weighted-random forest fusion model through a single data driving model LSTM (Long Short Term Memory) and a Transform model; the LSTM and Transform model prediction values are input into the weighted-random forest fusion model, a final fuel cell degradation prediction result is obtained, the method can accurately capture the degradation trend of the fuel cell, meanwhile, prediction fluctuation is reduced, prediction stability is improved, and the method is suitable for achieving accurate prediction of fuel cell degradation under dynamic operation.

Description

Technical Field

[0001] The invention belongs to the field of proton exchange membrane fuel cell degradation prediction, and in particular relates to a method and system for predicting performance degradation of a dynamically operating proton exchange membrane fuel cell. Background Art

[0002] Proton Exchange Membrane Fuel Cell (PEMFC), as an efficient and clean energy conversion device, has received extensive attention in recent years, especially in the fields of electric vehicles and distributed power generation, showing great application potential. Among them, hydrogen fuel cell vehicles are the most effective way to efficiently utilize hydrogen energy, with an energy conversion efficiency of up to 60%. However, automotive fuel cell stacks will experience performance degradation during long-term operation, manifested as decreased output power and reduced efficiency, which is mainly related to factors such as its internal electrochemical reaction and material aging. Therefore, accurately predicting the health indicator (HI) of automotive fuel cell stacks is of great significance for extending their service life, improving system reliability, and reducing maintenance costs.

[0003] In recent years, data-driven degradation prediction methods have gradually become a hot topic of research. By utilizing historical data from the operation of fuel cells, machine learning algorithms can establish effective prediction models without having to deeply understand the internal mechanisms of the system. Among the many data-driven prediction methods, time series models such as the Long Short-term Memory (LSTM) model and the Transformer model based on the attention mechanism have performed well.

[0004] The LSTM model can effectively capture long-term dependencies in time series, but it does not work well when faced with longer sequences with long-distance dependencies. Transformer is superior in capturing global information with its self-attention mechanism and has the ability to process data in parallel, but Transformer does not directly embed time sequence information like LSTM, so its time series modeling ability is weak. Fuel cell degradation prediction with a single data-driven model often has problems with limited prediction accuracy and poor generalization. Summary of the invention

[0005] The purpose of the present invention is to provide a method for predicting the performance degradation of a dynamically operating proton exchange membrane fuel cell, which is implemented based on a fusion data-driven model. By combining the advantages of LSTM in processing long-term dependencies in time series and the excellent ability of Transformer in capturing global information and complex nonlinear relationships, it can accurately capture the degradation trend of the fuel cell while reducing the prediction fluctuation and improving the prediction stability. It is suitable for realizing accurate prediction of fuel cell degradation under dynamic operation.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A method for predicting performance degradation of a dynamically operated proton exchange membrane fuel cell comprises the following steps:

[0008] Step S1, obtaining aging parameter data of the fuel cell stack during operation; the aging parameters at least include stack temperature, cathode inlet pressure, anode inlet pressure, voltage, cathode intake flow rate, and anode intake flow rate;

[0009] Step S2, obtaining a relative voltage decay rate as a health indicator value according to the aging parameter data;

[0010] Step S3, taking the health index value as the target value and each aging parameter as the input value, calculating the SHAP (SHapley Additive exPlanation) value of each aging parameter, and sorting the importance of the aging parameter characteristics; sorting the importance to select the aging parameters that have a great impact on the fuel cell performance degradation as the input of the data-driven model;

[0011] Step S4, training a single data-driven LSTM model and a Transformer model according to the health index value and the screened aging parameters, and predicting the health index value according to the LSTM model and the Transformer model;

[0012] Step S5, inputting the observed values ​​and predicted health index values ​​into the weighted average fusion model and the random forest fusion model to obtain the predicted values;

[0013] Step S6, selecting a suitable fusion model according to the prediction error of the random forest model. If the error of the random forest model is greater than the deviation threshold, the weighted average fusion model is selected. Otherwise, the random forest model is selected for fusion to obtain the final fuel cell degradation prediction result.

[0014] As a more optimal technical solution of the present invention: the relative voltage attenuation rate calculation steps in step S2 are as follows:

[0015] Step S2.1: fitting the semi-empirical equation of fuel cell degradation through the initial polarization curve data measured in step S1;

[0016] Step S2.2: Calculate the starting voltage at different current values ​​by using the fitted polarization curve equation;

[0017] Step S2.3: For H time steps, the voltage E at the current time step t is t The starting voltage E0 is calculated by formula (2) to obtain RVLR,

[0018] RVLR=(E t -E0) / E0 (15)

[0019] As a more optimal technical solution of the present invention: the SHAP value of each aging parameter is calculated in step S3 and the importance of the aging parameter characteristics is sorted as follows:

[0020] Step S3.1: Taking the HI value calculated in step S2 as the target value and each aging feature i parameter as input, calculate the incremental contribution after feature i is added;

[0021] Step S3.2: Calculate the corresponding weight of each feature subset S;

[0022] Step S3.3: Compute the SHAP value φ of feature i by summing the weighted incremental contributions of all subsets S that do not contain i i ;

[0023] Step S3.4: Sort the aging characteristic parameters according to the SHAP values ​​calculated above, and select the parameters that have a great impact on the fuel cell performance degradation as the input of the data-driven model.

[0024] As a better technical solution of the present invention: the calculation formula of the LSTM model used in step S4 is:

[0025]

[0026] where f t is the output of the forget gate, i t is the output of the input gate, is the estimated cell state, t is the output of the output gate;

[0027] As a more optimal technical solution of the present invention: the self-attention mechanism of the Transformer model adopted in step S4 calculates the query (Query), key (Key) and value (Value) at each position in the input sequence:

[0028] Q=XW Q , K=XW K , V=XW V (17)

[0029] Where X is the input matrix, W Q , W K , W V is the weight matrix; the softmax function ensures that the scores are normalized to a probability distribution; then, the weighted values ​​are calculated by the attention weights to get the final output:

[0030]

[0031] where d k is the dimension of the key; Transformer uses a multi-head attention mechanism to calculate multiple attention heads in parallel, and the result is:

[0032]

[0033] Where W O , W i Q , W i K and W i V is the parameter matrix.

[0034] As a more optimal technical solution of the present invention: the step S5 comprises the following steps:

[0035] Step S5.1: The prediction results of LSTM and Transformer are used as input X = {x1, x2}. Define θ as a random variable to determine the structure and splitting method of the decision tree. X is the input vector of the random forest model with a domain of T. Then For every x∈T, there is only one leaf node l that satisfies x∈T.

[0036] Step S5.2: Use the bootstrap method to resample and generate the corresponding decision tree from m random training sets.

[0037] Step S5.3: For new data X i = {x1, x2}, the predicted value of a single decision tree T(θ) can be obtained by averaging the observed values ​​of the leaf nodes l(x, θ). Assuming that the observed value X i Belongs to the leaf node l(x,θ) and is not zero, so the weight is,

[0038]

[0039] Where R l (x,θ) represents the rectangular subspace corresponding to each leaf node l=1,2,...,M in the domain T, and the observation value y i (i=1,2,...,n) weighted average can calculate the prediction value of a single decision tree.

[0040]

[0041] Step S5.4: Use formula (12) to adjust the decision tree weight w i (h,θr)(r=1,2,...,m) to find the average value and get the weight w of each observation i (h),

[0042]

[0043] Step S5.5: The prediction value of the random forest fusion method is:

[0044]

[0045] Step S5.6: Assume that the root mean square error (RMSE) of the prediction results of the Transformer and LSTM models are R1 and R2 respectively. The weights ω1 and ω2 in formula (12) can be expressed as:

[0046]

[0047] The prediction results of each model are weighted according to their weights, and finally these weighted results are added together. The prediction value after weighted average fusion can be expressed as:

[0048]

[0049] As a more optimal technical solution of the present invention: the step S6 comprises the following steps:

[0050] Step S6.1: Input the prediction results X = {x1, x2} of LSTM and Transformer into the two fusion methods to obtain the RFR fusion model prediction value And weighted average fusion prediction value

[0051] Step S6.2: Use the weighted average fusion method prediction value to replace the disordered prediction value. The single data-driven model prediction average calculation formula is:

[0052]

[0053] The error threshold is defined as,

[0054]

[0055] Step S6.3: The prediction error of the random forest ensemble model is defined as,

[0056]

[0057] The error coefficient is set to 2.5, and the final fuel cell degradation prediction result is:

[0058]

[0059] Another object of the present invention is to provide a dynamic proton exchange membrane fuel cell performance degradation prediction system, comprising:

[0060] A parameter acquisition module, used to obtain aging parameter data of the fuel cell stack during operation;

[0061] A data calculation module, used for obtaining a relative voltage decay rate as a health index value according to the aging parameter data;

[0062] The data screening module is used to calculate the SHAP value of each aging parameter and sort the importance of the aging parameter characteristics; sort the importance to select the aging parameters that have a great impact on the performance degradation of the fuel cell as the input of the data-driven model;

[0063] A model training module, used for training a single data-driven LSTM model and a Transformer model according to the health indicator value and the screened aging parameter, and predicting the health indicator value according to the LSTM model and the Transformer model;

[0064] The result prediction module is used to obtain the predicted value by inputting the observed value and the predicted health index value into the weighted average fusion model and the random forest fusion model; the disordered predicted value appearing in the random forest fusion model is replaced by the weighted average fusion model predicted value to obtain the final fuel cell degradation prediction result.

[0065] The beneficial effects are as follows:

[0066] The existing research on fuel cell degradation prediction is mainly focused on a single data-driven model or its improvement, but a single data-driven model often has problems such as limited prediction accuracy and poor generalization. This paper proposes a multi-model joint prediction method based on the weighted-random forest fusion method, which combines the advantages of LSTM in processing long-term dependencies in time series and the excellent ability of Transformer in capturing global information and complex nonlinear relationships, and can effectively enhance the ability of fuel cell performance degradation modeling and improve the accuracy of fuel cell degradation prediction.

[0067] To address the problem of disordered early prediction values ​​of the random forest fusion method caused by the initial prediction value drift of a single data-driven model under a multi-input and multi-output prediction strategy, the weighted average fusion method prediction value is used to replace the disordered prediction values ​​in the random forest method. Since the weighted average method directly assigns weights according to the performance of the model on the training set, there will be no large fluctuations when the prediction value of the single data-driven model drifts.

[0068] Under dynamic operating conditions, voltage and power as health indicators cannot directly reflect the degradation of the fuel cell due to the uncertainty of load current changes. Extracting the relative voltage decay rate as a health indicator of fuel cell performance degradation can make a direct comparison of the degradation of fuel cell cells under different operating conditions, thereby more accurately evaluating the degradation trend.

[0069] The multi-model joint prediction method of weighted-random forest fusion proposed in the present invention can accurately capture the degradation trend of fuel cells while reducing prediction fluctuations and improving prediction stability, and is suitable for accurate prediction of fuel cell degradation under dynamic operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 The flowchart of extracting relative voltage decay rate of the present invention is as follows;

[0071] Figure 2 It is a block diagram of the fusion data driven model method of the present invention;

[0072] Figure 3 Schematic diagram of the multi-input multi-output prediction strategy of the present invention

[0073] Figure 4 This is a comparison chart of the prediction results of the fusion model of the present invention and the single data-driven model (TP=100, m=80);

[0074] Figure 5 This is a comparison chart of the evaluation indicators of the fusion model of the present invention and the single data-driven model. DETAILED DESCRIPTION

[0075] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be fully described below in combination with the drawings in the embodiments of the present invention through specific implementation methods. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work fall within the protection scope of the present invention.

[0076] Combined with the present invention Figure 1 and Figure 2This embodiment provides a dynamic operation fuel cell degradation prediction method based on a fusion data driven model. First, the fuel cell aging data is obtained through a sensor measurement device and the noise is removed and downsampled. The semi-empirical equation of the fuel cell is fitted through the polarization curve data, and the relative voltage decay rate is calculated as the health indicator of the dynamic operation fuel cell. The relative voltage decay rate is used as the target value, the SHAP value of each degradation parameter is calculated, and the degradation parameters are selected as the input parameters of the data model. A weighted-random forest fusion model is constructed through a single data driven model LSTM and Transformer model; the LSTM and Transformer model prediction values ​​are input into the weighted-random forest fusion model to obtain the final fuel cell degradation prediction result.

[0077] The multi-model joint prediction method of weighted-random forest fusion proposed in the present invention can accurately capture the fuel cell degradation trend while reducing the prediction fluctuation and improving the prediction stability, and is suitable for realizing accurate prediction of fuel cell degradation under dynamic operation. The specific steps are as follows:

[0078] Step S1, obtain the actual aging data of the fuel cell stack during the durability test experiment through the sensor measurement device. The collected aging parameters include stack temperature, cathode and anode inlet pressure, voltage, cathode and anode flow, etc. The obtained original fuel cell aging data is processed to remove noise and downsample the data.

[0079] Step S2, calculating the relative voltage decay rate (Relative Voltage-Loss Rate, RVLR) as a health indicator of fuel cell performance degradation.

[0080] Step S2.1: Fitting the semi-empirical equation of fuel cell degradation by the initial polarization curve data measured in step S1. The semi-empirical equation of fuel cell degradation is defined as:

[0081]

[0082] Where E OCV is the open circuit voltage, b, i loss , R, m and n are five fitting parameters.

[0083] Step S2.2: Calculate the starting voltage E0 at different current values ​​using the fitted polarization curve equation.

[0084] Step S2.3: For H time steps (H is the total number of data), the voltage E at the current time step t is calculated. t The starting voltage E0 is calculated by formula (2) to obtain RVLR. RVLR is defined as,

[0085] RVLR=(Et -E0) / E0 (31)

[0086] In step S3, the RVLR calculated in step 2 is used as the target value, and the values ​​of the various aging parameters of the fuel cell measured by the sensor measuring device in step 1 are used as input values. The SHAP value of each aging parameter is calculated, and the aging parameters are sorted by feature importance to select the aging parameters that have a greater impact on the performance degradation of the fuel cell as the input of the data-driven model.

[0087] The step S3 ranks the importance of the aging parameter features by calculating the SHAP value of each aging parameter, and includes the following steps:

[0088] Step S3.1: Take the HI value calculated in step S2 as the target value and the parameters of each aging feature i as input. Calculate the incremental contribution after adding feature i.

[0089] Δf(S,i)=f(S∪{i})-f(S) (32)

[0090] where f(S) represents the model output using only the feature subset S, and f(S∪{i}) represents the model output after adding feature i to S.

[0091] Step S3.2: Calculate the corresponding weight of each feature subset S. For each subset S, the weight of feature i added to the subset can be calculated as,

[0092]

[0093] Where |S| is the number of features in subset S, and |N| is the size of the full set of features.

[0094] Step S3.3: Compute the SHAP value φ of feature i by summing the weighted incremental contributions of all subsets S that do not contain i i ,

[0095]

[0096] Step S3.4: Sort the aging characteristic parameters according to the SHAP values ​​calculated above, and select the parameters that have a greater impact on the fuel cell performance degradation as the input of the data-driven model.

[0097] Step S4, the HI value calculated in step 2 and the aging parameter value selected in step 3 are combined into a fuel cell degradation data set. The data set is divided into a training set, a validation set, and a test set in a ratio of 6:1:3. The single data-driven model LSTM and Transformer model are trained through the training set and the validation set.

[0098] The calculation formula of the LSTM model used in step S4 is:

[0099]

[0100] where f t is the output of the forget gate, i t is the output of the input gate, is the estimated cell state, t is the output of the output gate.

[0101] The self-attention mechanism of the Transformer model calculates the query, key, and value for each position in the input sequence:

[0102] Q=XW Q , K=XW K , V=XW V (36)

[0103] Where X is the input matrix, W Q , W K , W V is the weight matrix. The softmax function ensures that the scores are normalized to a probability distribution. Then, the weighted values ​​are calculated by the attention weights to get the final output:

[0104]

[0105] where d k is the dimension of the key. Transformer uses a multi-head attention mechanism to calculate multiple attention heads in parallel, and the result is:

[0106]

[0107] Where W O , W i Q , W i K and W i V is the parameter matrix.

[0108] Step S5: Build a weighted-random forest fusion model and test the performance of the LSTM model and the Transformer model on the training set and the validation set. The weight factor of the weighted average fusion method and the random forest fusion model are obtained through the actual HI value and the HI value predicted by the LSTM and Transformer models.

[0109] The construction of the weighted-random forest fusion model includes the following steps:

[0110] Step S5.1: The prediction results of LSTM and Transformer are used as input X = {x1, x2}. Define θ as a random variable to determine the structure and splitting method of the decision tree. X is the input vector of the random forest model with a domain of T, then For every x∈T, there is only one leaf node l that satisfies x∈T.

[0111] Step S5.2: Resample using the bootstrap method and generate corresponding decision trees from m random training sets.

[0112] Step S5.3: For new data X i ={x1,x2}, a single decision tree T(θ) can be obtained by averaging the observed values ​​of leaf nodes l(x,θ)

[0113] The predicted value of . Assuming the observed value X i belongs to the leaf node l(x,θ) and is not zero. So the weight is,

[0114]

[0115] Where R l (x,θ) represents the rectangular subspace corresponding to each leaf node l=1,2,...,M in the domain T.

[0116] The observed value y i The prediction value of a single decision tree can be calculated by weighted averaging (i=1,2,...,n).

[0117]

[0118] Step S5.4: Use formula (12) to adjust the decision tree weight w i (h,θr)(r=1,2,...,m) to find the average value and get the weight w of each observation i (h).

[0119]

[0120] Step S5.5: The prediction value of the random forest fusion method is:

[0121]

[0122] Step S5.6: Assume that the root mean square error (RMSE) of the prediction results of the Transformer and LSTM models are R1 and R2 respectively. The weights ω1 and ω2 in formula (12) can be expressed as:

[0123]

[0124] The prediction results of each model are weighted according to their weights, and finally these weighted results are added together. The prediction value after weighted average fusion can be expressed as:

[0125]

[0126] Step S6: Input the prediction values ​​of the LSTM and Transformer models on the test set into the weighted-random forest fusion model to obtain the final fuel cell degradation prediction result. The weighted-random forest fusion model testing phase includes the following steps:

[0127] Step S6.1: Input the prediction results X = {x1, x2} of LSTM and Transformer into the two fusion methods to obtain the RFR fusion model prediction value And weighted average fusion prediction value

[0128] Due to the drift phenomenon of the initial prediction value of the single data-driven model in the multi-input multi-output prediction strategy, the random forest fusion method has disordered prediction values, which makes the prediction unstable and reduces the overall accuracy. The weighted average fusion method prediction value is used to replace these disordered values ​​to improve the stability and accuracy of the model. The single data-driven model prediction average calculation formula is:

[0129]

[0130] The error threshold is defined as,

[0131]

[0132] Step S6.3: The prediction error of the random forest ensemble model is defined as,

[0133]

[0134] The error coefficient is set to 2.5, and the final fuel cell degradation prediction result is:

[0135]

[0136] The effectiveness of the present invention is further illustrated by the simulation results on PyTorch1.13.1. In order to verify the effectiveness of the fusion model proposed in the present invention, the fuel cell aging data set provided by Xiangyang Daan Testing Center is used and this experiment is designed for verification. The multi-input multi-output prediction strategy used in this experiment is as follows: Figure 3As shown. By changing the input vector TP of different step lengths and predicting the RVLR of multiple future time steps m, the effectiveness of the fusion data driven model proposed in the present invention is verified and made more suitable for the needs of practical applications. The results of the fusion model are compared and analyzed with the results of the single data driven model, as shown in Figure 4 As shown. In the enlarged figure, it can be seen that when the LSTM and Transformer prediction values ​​have different deviations from the true values, the prediction results of the fusion method are closer to the true values. This is because the fusion model can obtain more accurate prediction values ​​in later predictions by "learning" the errors of the previous single data-driven model. The comparison of the prediction results of the fusion model of the present invention and the single data-driven model is shown in Table 1 (the optimal result is indicated in bold). Under different sliding window lengths and prediction lengths, the fusion method shows significant advantages in predicting the degradation trend of fuel cells, and has better prediction accuracy than the single Transformer and LSTM models.

[0137] Table 1

[0138]

[0139] In all experimental configurations, the RMSE and average MAE of the fusion method are significantly lower than those of the Transformer and LSTM models used separately, indicating that the fusion method can capture the degradation trend more accurately. Figure 5 As shown in the figure, the average RMSE of the fusion model at different prediction times is 0.088%, and the average MAE is 0.058%, which are 8% and 13% higher than those of the single model, respectively. Based on the above information, the present invention proposes a dynamic operation fuel cell degradation prediction method based on a fusion data-driven model. While accurately capturing the fuel cell degradation trend, it reduces the prediction fluctuation and improves the prediction stability, and is suitable for accurate prediction of fuel cell degradation under dynamic operation.

[0140] Based on the same idea, the present invention provides a dynamic proton exchange membrane fuel cell performance degradation prediction system in other implementations, including: a parameter acquisition module, used to obtain aging parameter data of the fuel cell stack during operation; a data calculation module, used to obtain the relative voltage decay rate as a health index value based on the aging parameter data; a data screening module, used to calculate the SHAP value of each aging parameter, and sort the importance of the aging parameter characteristics; sort the importance to screen out the aging parameters that have a great impact on the performance degradation of the fuel cell as the input of the data-driven model; a model training module, used to train a single data-driven LSTM model and a Transformer model according to the health index value and the screened aging parameters, and predict the health index value based on the LSTM model and the Transformer model; a result prediction module, used to obtain a predicted value by inputting the observed value and the predicted health index value into the weighted average fusion model and the random forest fusion model; the weighted average fusion model predicted value replaces the disordered predicted value that appears in the random forest fusion model to obtain the final fuel cell degradation prediction result.

[0141] The above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for predicting performance degradation of a dynamically operated proton exchange membrane fuel cell, characterized in that: The steps include: Step S1, obtaining aging parameter data of the fuel cell stack during operation; the aging parameters include stack temperature, cathode inlet pressure, anode inlet pressure, voltage, cathode intake flow rate, anode intake flow rate, etc.; Step S2, obtaining a relative voltage decay rate as a health indicator value according to the aging parameter data; Step S3, taking the health index value as the target value and each aging parameter as the input value, calculating the SHAP value of each aging parameter, and ranking the importance of the aging parameter features; The aging parameters with the greatest impact on fuel cell performance degradation are selected by importance sorting as the input of the data-driven model; Step S4, training a single data-driven LSTM model and a Transformer model according to the health index value and the screened aging parameters, and predicting the health index value according to the LSTM model and the Transformer model; Step S5, inputting the observed values ​​and predicted health index values ​​into the weighted average fusion model and the random forest fusion model to obtain the predicted values; Step S6, selecting a suitable fusion model according to the prediction error of the random forest model. If the error of the random forest model is greater than the deviation threshold, the weighted average fusion model is selected. Otherwise, the random forest model is selected for fusion to obtain the final fuel cell degradation prediction result.

2. The method for predicting performance degradation of a dynamically operated proton exchange membrane fuel cell according to claim 1, characterized in that: The step S2 is as follows: Fitting the initial polarization curve data measured in step S1 to the semi-empirical equation of fuel cell degradation; The starting voltage at different current values ​​is calculated by fitting the polarization curve equation; The voltage E at the current time step t t The starting voltage E0 is calculated by formula (2) to obtain RVLR, RVLR=(E t -E0) / E0 (1).

3. The method for predicting performance degradation of a dynamically operated proton exchange membrane fuel cell according to claim 1, characterized in that: The step S3 is specifically as follows: Taking the HI value calculated in step S2 as the target value and the parameters of each aging feature i as input, calculate the incremental contribution after the feature i is added; Calculate the corresponding weight of each feature subset S; The SHAP value φ of feature i is calculated by summing the weighted incremental contributions of all subsets S that do not contain i. i ; The various aging characteristic parameters are sorted according to the SHAP values ​​calculated above, and the parameters that have a great impact on the fuel cell performance degradation are selected as the input of the data-driven model.

4. The method for predicting performance degradation of a dynamically operated proton exchange membrane fuel cell according to claim 1, characterized in that: The calculation formula of the LSTM model used in step S4 is: where f t is the output of the forget gate, i t is the output of the input gate, is the estimated cell state, t is the output of the output gate.

5. The method for predicting performance degradation of a dynamically operated proton exchange membrane fuel cell according to claim 1, characterized in that: The self-attention mechanism of the Transformer model adopted in step S4 calculates the query, key and value of each position in the input sequence: Q=XW Q , K=XW K , V=XW V (3) Where X is the input matrix, W Q , W K , W V is the weight matrix; the softmax function ensures that the scores are normalized to a probability distribution; then, the weighted values ​​are calculated by the attention weights to get the final output: where d k is the dimension of the key; Transformer uses a multi-head attention mechanism to calculate multiple attention heads in parallel, and the result is: Where W O , W i Q , W i K and W i V is the parameter matrix.

6. The method for predicting performance degradation of a dynamically operated proton exchange membrane fuel cell according to claim 1, characterized in that: The step S5 comprises the following steps: The prediction results of LSTM and Transformer are used as input X = {x1, x2}. θ is defined as a random variable to determine the structure and splitting method of the decision tree. X is the input vector of the random forest model with a domain of T. Then For every x∈T, there is only one leaf node l that satisfies x∈T. Use the bootstrap method to resample and generate the corresponding decision tree from m random training sets. For new data X i = {x1, x2} The predicted value of a single decision tree T(θ) is obtained by averaging the observed values ​​of leaf nodes l(x, θ). Assuming that the observed value X i Belongs to the leaf node l(x,θ) and is not zero. The weight is, Where R l (x,θ) represents the rectangular subspace corresponding to each leaf node l=1,2,...,M in the domain T, and the observation value y i (i=1,2,...,n) weighted average can calculate the prediction value of a single decision tree. Use formula (12) to adjust the decision tree weight w i (h,θr)(r=1,2,...,m) to find the average value and get the weight w of each observation i (h), The predicted values ​​of the random forest ensemble method are: Assume that the root mean square errors of the prediction results of the Transformer and LSTM models are R1 and R2 respectively, and the weights ω1 and ω2 in formula (12) are expressed as, The prediction results of each model are weighted according to their weights, and finally these weighted results are added together to obtain the weighted average fusion prediction value.

7. The method for predicting performance degradation of a dynamically operated proton exchange membrane fuel cell according to claim 1, characterized in that: The step S6 comprises the following steps: The prediction results X = {x1, x2} of LSTM and Transformer are input into the two fusion methods to obtain the RFR fusion model prediction value And weighted average fusion prediction value The weighted average fusion method prediction value is used to replace the disordered prediction value. The single data-driven model prediction average calculation formula is: The error threshold is defined as, The prediction error of the random forest ensemble model is defined as, The error coefficient is set to 2.5, and the final fuel cell degradation prediction result is:

8. A dynamic proton exchange membrane fuel cell performance degradation prediction system, characterized in that: include: A parameter acquisition module, used to obtain aging parameter data of the fuel cell stack during operation; A data calculation module, used for obtaining a relative voltage decay rate as a health index value according to the aging parameter data; Data screening module, used to calculate the SHAP value of each aging parameter and sort the importance of aging parameter features; The aging parameters with the greatest impact on fuel cell performance degradation are selected by importance sorting as the input of the data-driven model; A model training module, used for training a single data-driven LSTM model and a Transformer model according to the health indicator value and the screened aging parameter, and predicting the health indicator value according to the LSTM model and the Transformer model; The result prediction module is used to obtain the predicted value by inputting the observed value and the predicted health index value into the weighted average fusion model and the random forest fusion model; the disordered predicted value appearing in the random forest fusion model is replaced by the weighted average fusion model predicted value to obtain the final fuel cell degradation prediction result.

Images