Vehicle fuel cell performance prediction method, device and electronic equipment

Abstract

The present invention relates to the field of fuel cell technology, and provides a method, device and electronic device for predicting the performance of a vehicle fuel cell. The method comprises: extracting a time series of aging performance indicators of a vehicle fuel cell under dynamic conditions; extracting enhanced features for characterizing the regularity and / or trend of the fuel cell aging process based on the aging performance indicator time series; inputting the aging performance indicator time series and the enhanced features for characterizing the regularity and / or trend of the fuel cell aging process as training data into a preset multi-input convolutional neural network to train a vehicle fuel cell performance prediction model; and predicting the fuel cell performance degradation based on the trained vehicle fuel cell performance prediction model. The present invention can improve the accuracy and reliability of performance prediction when the fuel cell aging data is limited, and is of great significance to the health management of vehicle fuel cells.

Description

Technical Field

[0001] The present invention relates to the technical field of fuel cells, and in particular to a method, device and electronic equipment for predicting the performance of a vehicle fuel cell. Background Art

[0002] Using fuel cells as the main power source for new energy vehicles has the advantages of high energy density and fast fuel replenishment. It can make up for the shortcomings of pure electric vehicles in the field of long-distance heavy-duty commercial vehicles and has great development prospects. However, the service life of automotive fuel cells is greatly affected by actual operating conditions. Complex on-board operating conditions make the operating conditions of fuel cells more stringent, which accelerates the performance degradation of fuel cells. Due to the high maintenance and replacement costs of automotive fuel cells, how to accurately and reliably predict fuel cell aging can help reduce fuel cell costs and improve operating efficiency. It is the basis for building a fault prediction and health management system for fuel cell vehicles, and it is also the key to improving the durability and reliability of automotive fuel cells.

[0003] During the operation of automotive fuel cells, it is very difficult to accurately judge the aging state of the fuel cell at the current moment because the aging phenomenon of the fuel cell involves complex physical and chemical mechanisms, such as catalyst corrosion and dissolution, gas diffusion layer structure damage, proton exchange membrane polymer decomposition, etc. The aging mechanisms of various parts are very different and the aging rates of various parts are also different. For example, voltage fluctuations under dynamic cycle conditions will cause corrosion of the catalyst carbon carrier, leading to catalyst loss; high potential under idling conditions will cause polymer decomposition and aggravate proton membrane aging. Using the equivalent extrapolation method to predict fuel cell aging using a single aging index is not suitable for aging prediction under actual dynamic conditions.

[0004] In the process of implementing the present invention, the applicant discovered that the prior art has at least the following technical disadvantages:

[0005] 1) The existing non-invasive aging characteristics description methods for automotive fuel cells are inaccurate and have poor generalization capabilities. For example, currently, most fuel cell parameters that are easy to measure, such as voltage and power, are used to describe the aging characteristics of fuel cells. However, the aging of automotive fuel cells is greatly affected by operating conditions and has highly nonlinear characteristics. In particular, variable load conditions cause the output voltage of the fuel cell to change with the load and exhibit a response lag characteristic. The existing aging estimation methods that use voltage and power as aging indicators have obvious limitations. In addition, the fuel cell aging index constructed based on the strong assumption of the linear change trend of the limiting current and ohmic internal resistance lacks an explanation of the attenuation mechanism of the ohmic internal resistance and the limiting current density, and has limited generalization capabilities.

[0006] 2) Existing methods for estimating the aging characteristics of automotive fuel cells have a large demand for data and limited early-stage aging estimation capabilities. For example, existing methods for estimating the aging characteristics of fuel cells generally use data-based multi-step advance predictions to achieve online tracking of fuel cell aging. Data-based methods rely on large amounts of data training to achieve feature grasp and trend tracking of time series. However, in the early stages of fuel cell aging, the amount of aging data available for training prediction algorithms is limited, and the aging characteristics are not obvious, resulting in insufficient training of the prediction algorithm. At the same time, the aging characteristic information mixed in the data will interfere with the algorithm's ability to track the aging status of the fuel cell, leading to overfitting. Summary of the invention

[0007] In view of the above problems, the present invention is proposed to provide a method, device and electronic device for predicting the performance of a vehicle fuel cell that overcome the above problems.

[0008] One aspect of the present invention provides a method for predicting the performance of a vehicle fuel cell, the method comprising:

[0009] Extract the time series of aging performance indicators of automotive fuel cells under dynamic conditions;

[0010] Extracting enhanced features for characterizing the regularity and / or trend of the fuel cell aging process based on the aging performance index time series;

[0011] Inputting the aging performance index time series and the reinforcement features for characterizing the regularity and / or trend of the fuel cell aging process as training data into a preset multi-input convolutional neural network to train a vehicle fuel cell performance prediction model;

[0012] The fuel cell performance degradation is predicted based on the trained automotive fuel cell performance prediction model.

[0013] Furthermore, the time series of aging performance indicators of automotive fuel cells under dynamic conditions are extracted, including:

[0014] An analytical formula for exchange current density decay is constructed based on a preset exchange current density calculation model and an empirical model for electrochemical surface area aging of the catalyst layer;

[0015] Obtain the changing trend of ohmic internal resistance during dynamic loading of fuel cells;

[0016] A linear assumption is made for the ohmic internal resistance according to the variation trend, and a comprehensive aging index model of a fuel cell for a vehicle is established in combination with the exchange current density decay analytical expression, which is recorded as α(t).

[0017] R int =R0(1+kα(t))

[0018]

[0019] Where: R int is the ohmic internal resistance, R0 is the initial value of the ohmic internal resistance, i0 is the exchange current density, is the intrinsic exchange current density, is the reference oxygen concentration, is the oxygen concentration, Δb act is the activation energy of oxygen reduction reaction, T ref is the reference temperature, γ0 is the reaction order, t is the aging test time, t cyc is the duration of a single cycle, K is the proportional constant, R is the molar gas constant, T is the thermodynamic temperature, k is the normalization factor, S min is the normalized minimum electrochemically active surface area.

[0020] Furthermore, the variation trend of the ohmic internal resistance during the dynamic loading process of the fuel cell is obtained, including:

[0021] The polarization model of fuel cell is used to characterize the output characteristics of fuel cell;

[0022] Select the ohmic internal resistance R in the polarization model int , exchange current density i0 as the time-varying physical parameter of fuel cell aging to be extracted, durability test of fuel cell is carried out, the voltage of single cell of fuel cell is measured, and the polarization model of fuel cell is fitted to obtain the changing trend of each variable physical parameter during the dynamic loading process of fuel cell.

[0023] Furthermore, after establishing the comprehensive aging index of the vehicle fuel cell, the method further includes:

[0024] The Levenberg-Marquardt optimization algorithm is used to fit and solve the comprehensive aging index of automotive fuel cells, and the obtained aging index fitting results are used as the aging performance index time series of automotive fuel cells.

[0025] Furthermore, the Levenberg-Marquardt optimization algorithm is used to fit and solve the comprehensive aging index of the vehicle fuel cell, including:

[0026] Define the initial parameter set p0 as the initial guess value of the model function f(x,p), f(x,p) is the fitting function of the aging index α(t), and set the search step λ;

[0027] Calculate the residual vector r(p), where each element r i (p) is the difference between the model prediction and the actual data: r i (p) = y i -f(x i ,p), and calculate the Jacobian matrix J, whose elements Jij is the residual r i For parameter p j The partial derivative of

[0028] Update the parameter p using the preset update formula new , calculate the new parameter p new The error sum of squares S(p new ), if S(p new ) is less than the historical S(p), and the new parameters are determined to be better, so the new parameters are accepted and λ is reduced. If S(p new ) is greater than or equal to S(p), then the new parameter is rejected and λ is increased;

[0029] The update formula is as follows:

[0030] p new =p+(J T J+λD) -1 J T r(p)

[0031] Where D is a diagonal matrix whose diagonal elements are J T Diagonal elements of J;

[0032] Repeat the above steps until the convergence condition is met or the preset number of iterations is reached, and the final parameter p is output as the fitting result.

[0033] Furthermore, based on the aging performance index time series, the enhanced features for characterizing the regularity and / or trend of the fuel cell aging process are extracted, including:

[0034] The three-parameter exponential smoothing method Holt-Winters is used to perform feature decomposition on the aging performance index time series to obtain the level feature component, increase and decrease trend component and period feature component corresponding to the aging performance index time series;

[0035] The level characteristic component is selected as the first enhanced feature for characterizing the trend of the fuel cell aging process, and the periodic characteristic component is selected as the second enhanced feature for characterizing the regularity of the fuel cell aging process.

[0036] Furthermore, the network structure of the multi-input convolutional neural network includes a padding layer, a convolution layer, a pooling layer, a fully connected layer and an output layer arranged in sequence;

[0037] The padding layer is used to pad and arrange the input data of the multi-input convolutional neural network, the convolution layer is used to identify the feature information of the input data after padding and arrangement, the pooling layer is used to compress and reduce the dimension of the identified feature information, the fully connected layer is used to interpret the feature information after dimensionality reduction, and the output layer is used to realize time series prediction output.

[0038] Another aspect of the present invention provides a vehicle fuel cell performance prediction device, the device comprising:

[0039] Performance index extraction module, used to extract the time series of aging performance index of vehicle fuel cells under dynamic conditions;

[0040] A feature enhancement module, for extracting enhanced features for characterizing the regularity and / or trend of the fuel cell aging process based on the aging performance index time series;

[0041] A model training module, for inputting the aging performance index time series and the reinforcement features for characterizing the regularity and / or trend of the fuel cell aging process as training data into a preset multi-input convolutional neural network to train a vehicle fuel cell performance prediction model;

[0042] The prediction module is used to predict the fuel cell performance degradation based on the trained vehicle fuel cell performance prediction model.

[0043] Furthermore, the performance indicator extraction module includes:

[0044] The first construction unit is used to construct an exchange current density decay analytical formula according to a preset exchange current density calculation model and an empirical model of electrochemical surface area aging of the catalyst layer;

[0045] An acquisition unit, used to acquire a change trend of the ohmic internal resistance during a dynamic loading process of the fuel cell;

[0046] The second construction unit is used to make a linear assumption on the ohmic internal resistance according to the change trend, and to construct a comprehensive aging index model of the vehicle fuel cell in combination with the exchange current density decay analytical expression, which is recorded as α(t):

[0047] R int =R0(1+kα(t))

[0048]

[0049] Where: R int is the ohmic internal resistance, R0 is the initial value of the ohmic internal resistance, i0 is the exchange current density, is the intrinsic exchange current density, is the reference oxygen concentration, is the oxygen concentration, Δb act is the activation energy of oxygen reduction reaction, T ref is the reference temperature, γ0 is the reaction order, t is the aging test time, t cycis the duration of a single cycle, K is the proportional constant, R is the molar gas constant, T is the thermodynamic temperature, k is the normalization factor, S min is the normalized minimum electrochemically active surface area.

[0050] Another aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor; when the computer program is executed by the processor, the steps of the vehicle fuel cell performance prediction method as described above are implemented.

[0051] The vehicle fuel cell performance prediction method, device and electronic device provided by the embodiment of the present invention proposes an aging feature enhancement method for the situation where the data of the vehicle fuel cell is incomplete and the performance degradation trend is not obvious in the early stage of performance degradation, and is used to extract the regular phenomena in the fuel cell aging process as the input for training the fuel cell performance prediction algorithm, and enhance the learning ability of the prediction algorithm. It can improve the accuracy and reliability of fuel cell performance prediction when the fuel cell aging data is limited. At the same time, it improves computing efficiency and avoids resource waste, which is of great significance to the construction of a health management system for vehicle fuel cells.

[0052] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Various other advantages and benefits will become apparent to those skilled in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only used for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. In the accompanying drawings:

[0054] Figure 1 A flow chart of a method for predicting performance of a vehicle fuel cell according to an embodiment of the present invention;

[0055] FIG2( a ) is a schematic diagram showing a change trend of the ohmic internal resistance during dynamic loading of a fuel cell according to an embodiment of the present invention;

[0056] FIG2( b ) is a schematic diagram showing a change trend of the exchange current density during the dynamic loading process of a fuel cell according to an embodiment of the present invention;

[0057] Figure 3 A schematic diagram showing the fitting effect of the exchange current density decay analytical formula proposed in an embodiment of the present invention;

[0058] Figure 4A schematic diagram of the fitting results of the comprehensive aging index proposed in an embodiment of the present invention;

[0059] Figure 5 A schematic diagram of the decomposition of the aging performance index of a fuel cell for a vehicle proposed in an embodiment of the present invention;

[0060] Figure 6 A schematic diagram of a fuel cell performance prediction model based on enhanced features proposed in an embodiment of the present invention;

[0061] Figure 7 A specific implementation flow chart of the vehicle fuel cell performance prediction based on enhanced feature extraction proposed in an embodiment of the present invention;

[0062] Figure 8 The figure is a schematic diagram of the structure of a vehicle fuel cell performance prediction system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0064] Those skilled in the art will appreciate that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined.

[0065] The embodiment of the present invention provides a method for predicting the performance of a vehicle fuel cell, which can accurately monitor and predict the working state of the fuel cell by extracting and correcting the aging characteristics of the vehicle fuel cell and using feature enhancement technology to enhance feature extraction, and improve the accuracy of performance prediction when the data of the early stage of fuel cell performance degradation is incomplete. Figure 1 As shown, the vehicle fuel cell performance prediction method proposed by the present invention comprises the following steps:

[0066] S1. Extract the time series of aging performance indicators of automotive fuel cells under dynamic conditions.

[0067] Among them, the aging performance index time series is obtained by fitting the aging characteristic curve of the automotive fuel cell under dynamic conditions, which can reflect the multi-component aging parameters.

[0068] S2. Extracting enhanced features for characterizing the regularity and / or trend of the fuel cell aging process based on the aging performance indicator time series.

[0069] S3. Input the aging performance index time series and the reinforcement features for characterizing the regularity and / or trend of the fuel cell aging process as training data into a preset multi-input convolutional neural network to train a vehicle fuel cell performance prediction model.

[0070] S4. Predict fuel cell performance degradation based on the trained automotive fuel cell performance prediction model.

[0071] The vehicle fuel cell performance prediction method provided by the embodiment of the present invention proposes an aging feature enhancement method for the situation where the data of the vehicle fuel cell is incomplete and the performance degradation trend is not obvious in the early stage of performance degradation. The method is used to extract the regular phenomena in the fuel cell aging process as the input for training the fuel cell performance prediction algorithm, and enhance the learning ability of the prediction algorithm. It can improve the accuracy and reliability of fuel cell performance prediction when the fuel cell aging data is limited. At the same time, it improves computing efficiency and avoids resource waste, which is of great significance to the construction of a health management system for vehicle fuel cells.

[0072] In the embodiment of the present invention, the extraction of the time series of the aging performance index of the vehicle fuel cell under dynamic conditions in step S1 specifically includes the following steps which are not shown in the accompanying drawings:

[0073] S11. Construct an analytical formula for exchange current density decay based on a preset exchange current density calculation model and an empirical model for electrochemical surface area aging of the catalyst layer.

[0074] S12, obtaining the changing trend of the ohmic internal resistance during the dynamic loading process of the fuel cell. Further, obtaining the changing trend of the ohmic internal resistance during the dynamic loading process of the fuel cell includes: using the polarization model of the fuel cell to characterize the output characteristics of the fuel cell; selecting the ohmic internal resistance and exchange current density in the polarization model as the time-varying physical parameters of the fuel cell aging to be extracted, conducting a durability test on the fuel cell, measuring the single-chip cell voltage of the fuel cell, and fitting the polarization model of the fuel cell to obtain the changing trend of each variable physical parameter during the dynamic loading process of the fuel cell.

[0075] S13. A linear assumption is made for the ohmic internal resistance according to the change trend, and a comprehensive aging index model of a fuel cell for a vehicle is established in combination with the exchange current density decay analytical expression, which is recorded as α(t).

[0076] In this embodiment, the semi-mechanistic polarization model of a single fuel cell is as follows:

[0077] V cell=E Nerst -E ohm -E act -E conc (1)

[0078] Among them, V cell E is the voltage of a single cell of the fuel cell. Nerst is the reversible Nerst potential, obtained from the Nerst equation; the ohmic overpotential E ohm It indicates the ohmic loss of the stack during operation, which is related to the thickness of the proton exchange membrane, conductivity, etc.; the activation overpotential E act It represents the activation energy used to overcome the electrochemical reaction in the low potential region. It is affected by the catalyst activity and is derived from the Butler-Volmar equation. It is related to the exchange current density and is affected by the fuel cell temperature, the catalyst active surface area, etc.; the concentration difference overpotential E conc The concentration of reactants on the electrode surface in the high potential area decreases rapidly, reflecting the mass transfer capacity of the fuel cell. It can be derived from Fick's law and is related to the limiting current density. The polarization model of the fuel cell is used to characterize the output characteristics of the fuel cell. The specific expression is:

[0079]

[0080] Where i is the total current density (A / cm 2 ), R int is the ohmic internal resistance (Ω·cm 2 ), R is the molar gas constant (8.3145 J / mol / K), T is the thermodynamic temperature (K), α is the charge transfer coefficient, F is the Faraday constant (96485 C / mol), i0 is the exchange current density (A / cm 2 ), B is the concentration coefficient (V), i L is the limiting current density (A / cm 2 ).

[0081] The present invention selects R in the polarization model int , i0 as the time-varying physical parameters of fuel cell aging to be extracted, through durability experiments, the fuel cell voltage is measured, the above polarization model is fitted, and the change trend of the parameters during the dynamic loading of the fuel cell aging parameters is obtained, as shown in Figure 2, Figure 2 (a) is the change trend of the ohmic internal resistance during the dynamic loading of the fuel cell, and (b) is the change trend of the exchange current density during the dynamic loading of the fuel cell. It shows a certain degree of stable upward trend, and the overall upward trend is about 23%. The exchange current density shows a typical nonlinear change trend, with an overall change range of 87%, and it shows a rapid downward trend in the first 200 hours of aging test, and then the downward trend tends to be gentle, with an overall exponential function change trend.

[0082] Based on the exchange current density calculation model, see Formula 3, and the catalyst layer electrochemical surface area aging empirical model, see Formula 4:

[0083]

[0084]

[0085] K=K SC ×K T ×K RH ×K V (5)

[0086] The exchange current density decay analytical formula is obtained to characterize the performance aging of the fuel cell:

[0087]

[0088] Combined with the analysis results of the ohmic internal resistance aging trend, a linear assumption is made on the ohmic internal resistance aging trend, and the exchange current density decay analytical formula is combined to establish a comprehensive aging index model for fuel cells, which is denoted as α(t).

[0089] R int =R0(1+kα(t)) (7)

[0090]

[0091] Where: R int is the ohmic internal resistance, R0 is the initial value of the ohmic internal resistance, i0 is the exchange current density, is the intrinsic exchange current density (A / cm 2 ), is the reference oxygen concentration, is the oxygen concentration, is the effective area of ​​catalyst Pt (cm 2 ), Δb act is the activation energy of oxygen reduction reaction (J / mol), T ref is the reference temperature (K), γ0 is the reaction order, N is the number of dynamic cycle loading, t is the aging test time, t cyc is the duration of a single cycle (h), is the effective area of ​​the catalyst (A / cm 2 ), ECSA is the electrochemically active specific surface area (cm 2 / cm 2 ), S N is the normalized electrochemical active surface area, K is the proportionality constant, K SC is the proportional constant under standard working conditions, K T is the temperature correction factor, K RH is the relative humidity correction factor, KV is the cycle condition correction coefficient, R is the molar gas constant, T is the thermodynamic temperature, k is the normalization factor, S min is the normalized minimum electrochemically active surface area.

[0092] The fitting effect of the exchange current density decay analytical model is as follows: Figure 3 shown.

[0093] In an embodiment of the present invention, after establishing the comprehensive aging index of the vehicle fuel cell, the method further includes: using the Levenberg-Marquardt optimization algorithm to fit and solve the comprehensive aging index of the vehicle fuel cell, and using the obtained aging index fitting result as the aging performance index time series of the vehicle fuel cell.

[0094] In the actual operation of automotive fuel cells, they will gradually age as the usage time increases, affecting the output performance and life of the battery. When extracting the performance indicators of automotive fuel cells, it is assumed that the aging state of the fuel cell over a period of time is constant, that is, during this period of time, the impact of aging on performance is uniform and stable. On this basis, the Levenberg-Marquardt optimization Levenberg-Marquardt algorithm, a numerical solution method for nonlinear least squares problems, is used to fit and solve the above-mentioned fuel cell comprehensive aging indicator model.

[0095] Furthermore, the specific steps of fitting and solving the comprehensive aging index of the vehicle fuel cell using the Levenberg-Marquardt optimization algorithm are as follows:

[0096] Step 1: Initialize parameters

[0097] Define the initial parameter set p0 as the initial guess value of the model function f(x,p), f(x,p) is the fitting function of the aging index α(t), and set the search step λ;

[0098] Step 2: Calculate the error and Jacobian matrix

[0099] Calculate the residual vector r(p), where each element r i (p) is the difference between the model prediction and the actual data: r i (p) = y i -f(x i ,p), and calculate the Jacobian matrix J, whose elements J ij is the residual r i For parameter p j The partial derivative of

[0100] Step 3: Update parameters

[0101] Update the parameter p using the preset update formulanew , the update formula is as follows:

[0102] p new =p+(J T J+λD) -1 J T r(p)

[0103] Where D is a diagonal matrix whose diagonal elements are J T Diagonal elements of J;

[0104] Step 4: Evaluate the new parameters

[0105] Calculate the new parameter p new The error sum of squares S(p new ), if S(p new ) is less than the historical S(p), and the new parameters are determined to be better, so the new parameters are accepted and λ is reduced. If S(p new ) is greater than or equal to S(p), then the new parameter is rejected and λ is increased;

[0106] Step 5: Iteration

[0107] Repeat the above steps until the convergence condition is met or the preset number of iterations is reached, and the final parameter p is output as the fitting result.

[0108] The above fitting method is used to fit the comprehensive aging index of the pre-obtained vehicle fuel cell durability data set. The fitting results are shown in Figure 4 shown.

[0109] In an embodiment of the present invention, extracting the enhanced features for characterizing the regularity and / or trend of the fuel cell aging process based on the aging performance index time series in step S2 specifically includes the following steps:

[0110] S21. Use the three-parameter exponential smoothing method Holt-Winters to perform feature decomposition on the aging performance index time series to obtain the level feature component, increase and decrease trend component and period feature component corresponding to the aging performance index time series.

[0111] In order to reduce the interference of nonlinear factors in the aging process of automotive fuel cells, the Holt-Winters algorithm is used to decompose the extracted fuel cell aging performance index time series to enhance the fuel cell performance index. The decomposition model of the Holt-Winters method is as follows:

[0112]

[0113] b t =β(l t -l t-1 )+(1-β)bt-1 (11)

[0114]

[0115] The decomposed performance index is expressed as:

[0116]

[0117] Formulas (10)-(12) are the update formulas for the level component, the increase and decrease trend component, and the period component, and Formula 13 is the update formula for the recursive time series data.

[0118] Among them, l t is the grade component, b t is the increasing or decreasing trend component, s t is the periodic component, α, β, γ are smoothing coefficients, h is the time period, t is the aging test time, y t is the predicted value of the time series at time t+h, and m is the cycle length in the time series.

[0119] The optimization function for calibrating the α, β, and γ smoothing coefficients is obtained by using Matlab fminsearch to calculate the minimum value of the multivariate function. By decomposing the aging performance indicators, the level trend, increase and decrease trend, and periodic characteristics of the indicators are obtained to strengthen the aging characteristics of the fuel cell. The results are as follows: Figure 5 The present invention is based on feature enhancement technology to solve the problem of aging feature estimation when the aging features of vehicle fuel cells are not significant in the early stage of operation and the aging data is insufficient. It lays the foundation for establishing a fuel cell vehicle fault prediction and health management system, and is the key to improving the durability and reliability of vehicle fuel cells.

[0120] S22, selecting the level feature component as a first enhanced feature for characterizing the trend of the fuel cell aging process, and selecting the periodic feature component as a second enhanced feature for characterizing the regularity of the fuel cell aging process.

[0121] Furthermore, the network structure of the multi-input convolutional neural network in step S3 includes a padding layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer arranged in sequence; the padding layer is used to pad and arrange the input data of the multi-input convolutional neural network, the convolutional layer is used to identify feature information of the input data after padding and arrangement, the pooling layer is used to compress and reduce the dimension of the identified feature information, the fully connected layer is used to interpret the feature information after dimensionality reduction, and the output layer is used to realize time series prediction output.

[0122] The present invention uses a multi-input convolutional neural network to train the processed aging data and predict and track the performance degradation of the fuel cell. The input of the multi-input convolutional neural network module is a multivariate time series with a certain time step, including the decomposed periodic characteristics, grade characteristics and aging index time series. The input data is filled and arranged, multi-convolution layer feature recognition, pooling layer feature information compression and dimension reduction, and fully connected layer feature information interpretation, and finally the time series prediction output is realized. The prediction process is shown. The convolution layer calculation rule is:

[0123]

[0124] Among them, f is the activation function, w s is the weight value of the convolution filter, * represents the convolution operator, b s is the bias parameter, x is the convolution layer input, s is the number of convolution layers, q and u are the sizes of the convolution kernel matrix, and the activation function between neuron nodes uses the ReLU function. Figure 6 The schematic diagram of the fuel cell performance prediction model based on the enhanced characteristics is shown. The evaluation results of the fuel cell performance prediction predicted by the model are shown in Table 1:

[0125] Table 1 Rating results of multi-input performance prediction model

[0126]

[0127] In Table 1, MAPE is the mean absolute percentage error and RMSE is the root mean square error.

[0128] The vehicle fuel cell performance prediction method provided by the present invention solves the problem of how to accurately judge the current aging state of the fuel cell in the vehicle fuel cell power system and predict the aging trend of the fuel cell, so as to formulate corresponding control and maintenance strategies and extend the operating life of the vehicle fuel cell. Figure 7 As shown:

[0129] 1. Through solving the aging parameters of automotive fuel cells, combining the variation law of ohmic internal resistance and the empirical formula of exchange current density aging suitable for dynamic working conditions, the comprehensive aging index of automotive fuel cells under dynamic working conditions is obtained by segmented fitting of the input signal.

[0130] 2. By strengthening the early aging data trend and reversible aging characteristics of fuel cells, the short-term prediction accuracy of multivariate convolutional neural networks for fuel cell aging trends is improved.

[0131] The present invention provides a non-invasive method for representing the aging characteristics of automotive fuel cells. By fitting the aging characteristic curve of automotive fuel cells under dynamic conditions, it is proposed to introduce the empirical formula for the attenuation of the electrochemically active surface area of ​​the catalyst layer into the analytical formula for the exchange current density. Different from the original single aging index, the automotive fuel cell aging index proposed by the present invention can reflect multi-component aging parameters and effectively improve the accuracy of the original aging characteristic extraction method.

[0132] Furthermore, the present invention provides a method for predicting the performance of automotive fuel cells based on reinforcement feature learning, which fully considers the nonlinear characteristics of aging indicators, decomposes them, and obtains regular trends for training prediction algorithms. This method can improve the accuracy and reliability of performance prediction when fuel cell aging data is limited, and is of great significance to the construction of a health management system for automotive fuel cells.

[0133] For the method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the embodiments of the present invention are not limited by the order of the actions described, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0134] Another embodiment of the present invention further provides a vehicle fuel cell performance prediction device, the device comprising a functional module for implementing any of the vehicle fuel cell performance prediction methods described above. Figure 8 The schematic diagram of the structure of a vehicle fuel cell performance prediction device provided by an embodiment of the present invention is shown schematically. Figure 8 The vehicle fuel cell performance prediction device of the embodiment of the present invention specifically includes a performance index extraction module 201, a feature enhancement module 202, a model training module 203 and a prediction module 204, wherein:

[0135] The performance index extraction module 201 is used to extract the aging performance index time series of the vehicle fuel cell under dynamic conditions;

[0136] A feature enhancement module 202, for extracting enhanced features for characterizing the regularity and / or trend of the fuel cell aging process based on the aging performance index time series;

[0137] A model training module 203 is used to input the aging performance index time series and the reinforcement features used to characterize the regularity and / or trend of the fuel cell aging process as training data into a preset multi-input convolutional neural network to train a vehicle fuel cell performance prediction model;

[0138] The prediction module 204 is used to predict the fuel cell performance degradation based on the trained vehicle fuel cell performance prediction model.

[0139] In the embodiment of the present invention, the performance indicator extraction module 201 includes a first construction unit, an acquisition unit and a second construction unit, wherein:

[0140] The first construction unit is used to construct an exchange current density decay analytical formula according to a preset exchange current density calculation model and an empirical model of electrochemical surface area aging of the catalyst layer;

[0141] An acquisition unit, used to acquire a change trend of the ohmic internal resistance during a dynamic loading process of the fuel cell;

[0142] The second construction unit is used to make a linear assumption on the ohmic internal resistance according to the change trend, and to construct a comprehensive aging index of the vehicle fuel cell in combination with the exchange current density decay analytical expression, which is recorded as α(t):

[0143] R int =R0(1+kα(t))

[0144]

[0145] Where: R int is the ohmic internal resistance, R0 is the initial value of the ohmic internal resistance, i0 is the exchange current density, is the intrinsic exchange current density, is the reference oxygen concentration, c O2 is the oxygen concentration, Δb act is the activation energy of oxygen reduction reaction, T ref is the reference temperature, γ0 is the reaction order, t is the aging test time, t cyc is the duration of a single cycle, K is the proportional constant, R is the molar gas constant, T is the thermodynamic temperature, k is the normalization factor, S min is the normalized minimum electrochemically active surface area.

[0146] In this embodiment, the acquisition unit is specifically used to characterize the output characteristics of the fuel cell using the polarization model of the fuel cell; the ohmic internal resistance and exchange current density in the polarization model are selected as the time-varying physical parameters of the fuel cell to be extracted during aging, a durability test is performed on the fuel cell, the single-chip cell voltage of the fuel cell is measured, and the polarization model of the fuel cell is fitted to obtain the changing trends of various variable physical parameters during the dynamic loading process of the fuel cell.

[0147] In this embodiment, the performance index extraction module 201 also includes a calculation unit, which is used to fit and solve the comprehensive aging index of the vehicle fuel cell using the Levenberg-Marquardt optimization algorithm after establishing the comprehensive aging index of the vehicle fuel cell, and use the obtained aging index fitting result as the aging performance index time series of the vehicle fuel cell.

[0148] Furthermore, the calculation unit is specifically used to execute the following fitting solution method:

[0149] Define the initial parameter set p0 as the initial guess value of the model function f(x,p), f(x,p) is the fitting function of the aging index α(t), and set the search step λ;

[0150] Calculate the residual vector r(p), where each element r i (p) is the difference between the model prediction and the actual data: r i (p) = y i -f(x i ,p), and calculate the Jacobian matrix J, whose elements J ij is the residual r i For parameter p j The partial derivative of

[0151] Update the parameter p using the preset update formula new , calculate the new parameter p new The error sum of squares S(p new ), if S(p new ) is less than the historical S(p), and the new parameters are determined to be better, so the new parameters are accepted and λ is reduced. If S(p new ) is greater than or equal to S(p), then the new parameter is rejected and λ is increased;

[0152] The update formula is as follows:

[0153] p new =p+(J T J+λD) -1 J T r(p)

[0154] Where D is a diagonal matrix whose diagonal elements are J T Diagonal elements of J;

[0155] Repeat the above steps until the convergence condition is met or the preset number of iterations is reached, and the final parameter p is output as the fitting result.

[0156] In the embodiment of the present invention, the feature enhancement module 202 specifically includes a feature decomposition unit and a feature selection unit, wherein:

[0157] A feature decomposition unit, used for performing feature decomposition on the aging performance index time series by using a three-parameter exponential smoothing method Holt-Winters to obtain a level feature component, an increase and decrease trend component and a period feature component corresponding to the aging performance index time series;

[0158] The feature selection unit is used to select the level feature component as a first enhanced feature for characterizing the trend of the fuel cell aging process, and select the periodic feature component as a second enhanced feature for characterizing the regularity of the fuel cell aging process.

[0159] In the embodiment of the present invention, the network structure of the multi-input convolutional neural network trained by the model training module 203 includes a padding layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer arranged in sequence;

[0160] The padding layer is used to pad and arrange the input data of the multi-input convolutional neural network, the convolution layer is used to identify the feature information of the input data after padding and arrangement, the pooling layer is used to compress and reduce the dimension of the identified feature information, the fully connected layer is used to interpret the feature information after dimensionality reduction, and the output layer is used to realize time series prediction output.

[0161] In the specific implementation process of the present device embodiment, reference may be made to the above method embodiment, which has corresponding technical effects.

[0162] Another aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor; when the computer program is executed by the processor, the steps of the above-mentioned method for predicting the performance of a vehicle fuel cell are implemented. Figure 1 Alternatively, when the processor executes the computer program, the functions of each module in the above-mentioned vehicle fuel cell performance prediction device embodiments are realized, for example Figure 8 The shown modules include a performance indicator extraction module 201 , a feature enhancement module 202 , a model training module 203 and a prediction module 204 .

[0163] In the specific implementation process of this device embodiment, the above method embodiment can be referred to, and the corresponding technical effects are achieved.

[0164] In addition, those skilled in the art will appreciate that, although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present invention and form different embodiments. For example, any one of the claimed embodiments may be used in any combination.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be used to modify the technical solutions or to perform equivalent replacements on some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the performance of a fuel cell for a vehicle, characterized in that: The method comprises: Extract the time series of aging performance indicators of automotive fuel cells under dynamic conditions; Extracting enhanced features for characterizing the regularity and / or trend of the fuel cell aging process based on the aging performance index time series; Inputting the aging performance index time series and the reinforcement features for characterizing the regularity and / or trend of the fuel cell aging process as training data into a preset multi-input convolutional neural network to train a vehicle fuel cell performance prediction model; Predict fuel cell performance degradation based on the trained vehicle fuel cell performance prediction model; The step of extracting the time series of aging performance indicators of a vehicle fuel cell under dynamic conditions includes: An analytical formula for exchange current density decay is constructed based on a preset exchange current density calculation model and an empirical model for electrochemical surface area aging of the catalyst layer; Obtain the changing trend of ohmic internal resistance during dynamic loading of fuel cells; According to the change trend, a linear assumption is made for the ohmic internal resistance, and the comprehensive aging index model of the vehicle fuel cell is established in combination with the exchange current density decay analytical formula, which is recorded as : in: is the ohmic internal resistance, is the initial value of the ohmic internal resistance, is the exchange current density, is the intrinsic exchange current density, is the reference oxygen concentration, is the oxygen concentration, is the activation energy of the oxygen reduction reaction, is the reference temperature, is the reaction order, is the aging test time, is the duration of a single cycle operation, is the proportionality constant, is the molar gas constant, is the thermodynamic temperature, k is the normalization factor, S min is the normalized minimum electrochemically active surface area.

2. The method according to claim 1, characterized in that Obtain the changing trend of ohmic internal resistance during dynamic loading of fuel cells, including: The polarization model of fuel cell is used to characterize the output characteristics of fuel cell; Select the ohmic internal resistance in the polarization model , exchange current density As the time-varying physical parameters of fuel cell aging to be extracted, durability experiments are carried out on the fuel cell, the voltage of the single cell of the fuel cell is measured, and the polarization model of the fuel cell is fitted to obtain the changing trends of various variable physical parameters during the dynamic loading process of the fuel cell.

3. The method according to claim 1, characterized in that After establishing the comprehensive aging index of the vehicle fuel cell, the method further includes: The Levenberg-Marquardt optimization algorithm is used to fit and solve the comprehensive aging index of automotive fuel cells, and the obtained aging index fitting results are used as the aging performance index time series of automotive fuel cells.

4. The method according to claim 3, characterized in that The Levenberg-Marquardt optimization algorithm is used to fit and solve the comprehensive aging index of the vehicle fuel cell, including: Defining the initial parameter set , as the model function The initial guess value of Aging index The fitting function and set the search step size ; Calculate the residual vector , where each element is the difference between the model prediction and the actual data: , and calculate the Jacobian matrix , whose elements is the residual Parameters The partial derivative of Update parameters using preset update formulas , calculate the new parameters The sum of squared errors under ,if Less than history , determine that the new parameters are better, then accept the new parameters and reduce ,if Greater than or equal to , then reject the new parameter and increase ; The update formula is as follows: in, is a diagonal matrix whose diagonal elements are The diagonal elements of Repeat the above steps until the convergence condition is met or the preset number of iterations is reached, and the final parameters are obtained. Output as fitting result.

5. The method according to claim 1, characterized in that: Extracting enhanced features for characterizing the regularity and / or trend of the fuel cell aging process based on the aging performance index time series includes: The three-parameter exponential smoothing method Holt-Winters is used to perform feature decomposition on the aging performance index time series to obtain the level feature component, increase and decrease trend component and period feature component corresponding to the aging performance index time series; The level characteristic component is selected as the first enhanced feature for characterizing the trend of the fuel cell aging process, and the periodic characteristic component is selected as the second enhanced feature for characterizing the regularity of the fuel cell aging process.

6. The method according to claim 1, characterized in that The network structure of the multi-input convolutional neural network includes a padding layer, a convolution layer, a pooling layer, a fully connected layer and an output layer arranged in sequence; The padding layer is used to pad and arrange the input data of the multi-input convolutional neural network, the convolution layer is used to identify the feature information of the input data after padding and arrangement, the pooling layer is used to compress and reduce the dimension of the identified feature information, the fully connected layer is used to interpret the feature information after dimensionality reduction, and the output layer is used to realize time series prediction output.

7. A vehicle fuel cell performance prediction device, characterized in that: The device comprises: Performance index extraction module, used to extract the time series of aging performance index of vehicle fuel cells under dynamic conditions; A feature enhancement module, for extracting enhanced features for characterizing the regularity and / or trend of the fuel cell aging process based on the aging performance index time series; A model training module, for inputting the aging performance index time series and the reinforcement features for characterizing the regularity and / or trend of the fuel cell aging process as training data into a preset multi-input convolutional neural network to train a vehicle fuel cell performance prediction model; A prediction module is used to predict the fuel cell performance degradation based on the trained vehicle fuel cell performance prediction model; wherein the performance index extraction module includes: The first construction unit is used to construct an exchange current density decay analytical formula according to a preset exchange current density calculation model and an empirical model of electrochemical surface area aging of the catalyst layer; An acquisition unit, used to acquire a change trend of the ohmic internal resistance during a dynamic loading process of the fuel cell; The second construction unit is used to make a linear assumption on the ohmic internal resistance according to the change trend, and to construct a comprehensive aging index of the vehicle fuel cell in combination with the exchange current density decay analytical expression, which is recorded as ; in: is the ohmic internal resistance, is the initial value of the ohmic internal resistance, is the exchange current density, is the intrinsic exchange current density, is the reference oxygen concentration, is the oxygen concentration, is the activation energy of the oxygen reduction reaction, is the reference temperature, is the reaction order, is the aging test time, is the duration of a single cycle operation, is the proportionality constant, is the molar gas constant, is the thermodynamic temperature, k is the normalization factor, S min is the normalized minimum electrochemically active surface area.

8. An electronic device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor; when the computer program is executed by the processor, the steps of the vehicle fuel cell performance prediction method described in any one of claims 1 to 6 are implemented.

Images