CN115098999A - Multi-mode fusion fuel cell system performance attenuation prediction method - Google Patents

Multi-mode fusion fuel cell system performance attenuation prediction method Download PDF

Info

Publication number
CN115098999A
CN115098999A CN202210579312.7A CN202210579312A CN115098999A CN 115098999 A CN115098999 A CN 115098999A CN 202210579312 A CN202210579312 A CN 202210579312A CN 115098999 A CN115098999 A CN 115098999A
Authority
CN
China
Prior art keywords
modal
fuel cell
performance index
decomposition
time
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210579312.7A
Other languages
Chinese (zh)
Inventor
袁浩
戴海峰
魏学哲
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tongji University
Original Assignee
Tongji University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tongji University filed Critical Tongji University
Priority to CN202210579312.7A priority Critical patent/CN115098999A/en
Publication of CN115098999A publication Critical patent/CN115098999A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/04—Ageing analysis or optimisation against ageing

Abstract

The invention relates to a multi-mode fused fuel cell system performance attenuation prediction method, which comprises the following steps: s1, acquiring performance index time sequence data capable of describing the performance attenuation of the fuel cell system; s2, performing modal decomposition on the performance index time sequence data by adopting a self-adaptive data decomposition method to obtain a plurality of modal quantities; s3, respectively constructing prediction models for predicting the modal quantities for each modal quantity, and training the prediction models; s4, performing modal decomposition on the fuel cell historical actual measurement performance index time sequence by adopting the self-adaptive data decomposition method, and inputting each modal quantity to the corresponding prediction model to obtain each modal predicted value; and S5, fusing the modal predicted values to obtain a performance index predicted value, and determining the attenuation condition of the fuel cell based on the performance index predicted value. Compared with the prior art, the method can effectively learn the influence of various disturbances on the dynamic attenuation in the operation process of the fuel cell system, and has higher prediction precision.

Description

Multi-mode fusion fuel cell system performance attenuation prediction method
Technical Field
The invention relates to the technical field of fuel cells, in particular to a multi-mode fusion fuel cell system performance attenuation prediction method.
Background
The fuel cell automobile is an important branch of new energy automobiles, is considered as one of the final solutions of future automobiles due to the advantages of high filling speed, high efficiency, low noise, zero emission and the like, but the large-scale commercial application of the fuel cell automobile is still limited by durability. The fuel cell system is subjected to attenuation prediction, so that references can be provided for system control and health management, the service life of the fuel cell system is prolonged, and large-scale commercial application of the fuel cell system is promoted.
The chinese patent document CN114373965A adopts a model-based method to predict the attenuation, but the fuel cell system is a system with strong nonlinearity and strong time-varying property, and it is very challenging to accurately model the attenuation, and the model established in this patent is a linear attenuation model, and has a certain difference from the actual nonlinear attenuation of the fuel cell.
Chinese patent document CN114137431A uses a linear function, a quadratic function and a double-exponential function to fit the graphs of the fuel cell with voltage varying with time under different current densities, and selects the function with the highest fitting accuracy as the final prediction function, and obtains the attenuation trend and the termination time of the fuel cell according to the prediction function and the cut-off voltage. This method clearly only qualitatively predicts the degradation of the fuel cell.
Chinese patent document CN113657024A utilizes a depth projection coding echo state network to construct a fuel cell life prediction model, and adopts a genetic algorithm to optimize parameters of the fuel cell life prediction model. However, the traditional echo state network can only carry out shallow learning on the fuel cell and cannot fully dig the deep nonlinear attenuation characteristic.
The chinese patent document CN110059377A adopts a deep convolutional neural network to perform attenuation prediction. As is well known, there are many performance recovery stages in the process of fuel cell performance degradation, and the degradation is also affected by external operating conditions, such as temperature, humidity, gas flow, pressure, etc., and the traditional deep convolutional neural network cannot separately consider these external influencing factors.
For this reason, it is necessary to propose a fuel cell system degradation prediction method with better accuracy.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a multi-mode fused fuel cell system performance attenuation prediction method.
The purpose of the invention can be realized by the following technical scheme:
a method for multi-modal fused fuel cell system performance decay prediction, the method comprising:
s1, acquiring performance index time sequence data capable of describing the performance attenuation of the fuel cell system;
s2, performing modal decomposition on the performance index time sequence data by adopting a self-adaptive data decomposition method to obtain a plurality of modal quantities;
s3, respectively constructing prediction models for predicting the modal quantities for each modal quantity, and training the prediction models;
s4, performing modal decomposition on the fuel cell historical actual measurement performance index time sequence by adopting the self-adaptive data decomposition method, and inputting each modal quantity to the corresponding prediction model to obtain each modal predicted value;
and S5, fusing the modal predicted values of the fuel cell to obtain a performance index predicted value, and determining the attenuation condition of the fuel cell based on the performance index predicted value.
Preferably, the performance index includes any one of voltage, active area, impedance and limiting current density.
Preferably, the adaptive data decomposition method comprises a complementary set empirical mode decomposition algorithm.
Preferably, the complementary set empirical mode decomposition algorithm comprises the following steps:
1) respectively adding positive white noise and negative white noise to an original signal to be decomposed;
2) performing empirical mode decomposition on the two newly-noised signals;
3) repeating the process of addition and decomposition until the requirement of cycle times is met, and finally averaging the obtained decomposition results:
Figure BDA0003661726670000021
Figure BDA0003661726670000022
wherein, IMF k (t) is the kth order natural mode function at time t, r K (t) is the residual quantity at time t, IMF k,j (t) adding a white noise signal, and performing IMF (intrinsic mode function) on the kth order at t moment in the j-th cyclic calculation k,-j (t) adding a negative white noise signal, and in the j-th cyclic calculation, performing a kth-order intrinsic mode function at t moment K,j (t) is the residual amount at t in the j-th loop calculation by adding a white noise signal, r K,-j (t) is the residual quantity at t time in the j-th cycle calculation by adding the negative white noise signal, and M is the total cycle calculation frequency.
Preferably, the empirical mode decomposition in step 2) specifically includes:
21) obtaining all local maximum value points and local minimum value points of the time sequence data p (t) to be decomposed, and obtaining an upper envelope line e by a cubic spline interpolation method 1,1 (t) and the lower envelope e 1,2 (t) calculating the envelope mean curve a 1,1 (t)；
22) Calculating an envelope mean curve a 1,1 (t) curve h of the difference between the time-series data p (t) to be decomposed and the time-series data to be decomposed 1,1 (t)；
23) If h 1,1 (t) if IMF condition is not satisfied, then h is used 1,1 (t) replacing p (t), returning to the step 21) and repeating, and setting the difference curve obtained at the ith time as h 1,i (t)：
h 1,i (t)＝h 1,(i-1) (t)-a 1,i (t)
24) When the difference curve obtained at the ith time meets the stopping criterion, the first intrinsic mode function IMF is regarded as the first intrinsic mode function 1 (t) and calculating the corresponding residual r 1 (t) using r 1 (t) substituting p (t), going back to step 21) and repeating until the k-th time IMF is obtained k (t) is a monotonic function or k reaches a desired number of times, the decomposition stops.
Preferably, the IMF conditions are: the number of the zero points and the extreme points is equal to or different from 1; the average value of the upper envelope curve determined by the local maximum value points and the lower envelope curve determined by the local minimum value points is constant 0.
Preferably, the predictive model comprises a gated loop element network.
Preferably, in the step S3, when the prediction model is trained, normalization processing is performed on each of the modal quantities obtained in the step S2, and network training is performed using the normalized modal quantity data.
Preferably, in step S5, the mode prediction values of the fuel cell are fused to obtain the performance index prediction value as:
Figure BDA0003661726670000031
wherein the content of the first and second substances,
Figure BDA0003661726670000032
is a predicted value of the performance index at time t,
Figure BDA0003661726670000033
is the predicted value of the k-th order intrinsic mode function at the time t of the performance index,
Figure BDA0003661726670000034
and K is the predicted value of the residual error quantity at the time of the performance index t, and is the total order of the inherent modal function.
Preferably, the determination of the fuel cell degradation based on the predicted value of the performance index in step S5 is embodied as calculating a degradation rate η (t) of the performance index:
Figure BDA0003661726670000035
Figure BDA0003661726670000036
is a predicted value of the performance index at time t, p new Is the performance index of the fuel cell before the fuel cell begins to age. Compared with the prior art, the invention has the following advantagesThe advantages are that:
(1) the method firstly carries out modal decomposition on the performance index time series data which can describe the performance attenuation of the fuel cell system, predicts each modal component through a prediction model, and further fuses the prediction values of each modal component to obtain the prediction value of the performance index, so that the performance attenuation of the fuel cell is predicted, a complex or empirical fuel cell performance attenuation model does not need to be established, and the method has better precision compared with other popular diagnostic methods.
(2) The method adopts a complementary set empirical mode decomposition algorithm to carry out mode decomposition, and can effectively separate the nonlinear non-stationary time sequence data of short-term random fluctuation and long-term attenuation of the fuel cell, so that the model can better learn the influence of external dynamic disturbance on the performance attenuation of the fuel cell, and the prediction precision is further improved.
Drawings
FIG. 1 is a block flow diagram of a method for predicting degradation in performance of a multi-modal integrated fuel cell system in accordance with the present invention;
FIG. 2 is an example of the decay voltage data selection of the present invention;
FIG. 3 is a flow chart of empirical mode decomposition according to the present invention;
FIG. 4 is a flow chart of complementary set Empirical Mode Decomposition (EMD) according to the present invention;
FIG. 5 is an example of the decaying voltage multi-modal decomposition of the present invention;
FIG. 6 is a multi-modal fusion model attenuation prediction framework of the present invention;
FIG. 7 is a structure of a gated loop cell employed in the present invention;
FIG. 8 shows the predicted results of an example of the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. Note that the following description of the embodiment is merely an example of the nature, and the present invention is not intended to be limited to the application or the use thereof, and is not limited to the following embodiment.
Examples
As shown in fig. 1, the present embodiment provides a method for predicting performance degradation of a multi-modal fused fuel cell system, including:
s1, acquiring performance index time sequence data capable of describing the performance attenuation of the fuel cell system;
s2, performing modal decomposition on the performance index time sequence data by adopting a self-adaptive data decomposition method to obtain a plurality of modal quantities;
s3, respectively constructing prediction models for predicting the modal quantities for each modal quantity, and training the prediction models;
s4, performing modal decomposition on the fuel cell historical actual measurement performance index time sequence by adopting the self-adaptive data decomposition method, and inputting each modal quantity into a corresponding prediction model to obtain each modal prediction value;
and S5, fusing the modal predicted values of the fuel cell to obtain a performance index predicted value, and determining the attenuation condition of the fuel cell based on the performance index predicted value.
The performance index includes any one of voltage, active area, impedance, and limiting current density, wherein the voltage during the operation of the fuel cell is easy to measure, so the present embodiment is the most preferable embodiment, and the operation voltage of the fuel cell is used as the performance index describing the performance degradation of the fuel cell system, which will be described in detail below.
S1, obtaining a data set capable of describing the performance attenuation of the fuel cell system
And inputting working condition data of the fuel cell commercial vehicle running for a long time, determining voltage capable of describing the attenuation trend of the fuel cell system as a health state index, and taking the voltage data sequence as attenuation model training and verification data.
In this example, taking a certain fuel cell commercial vehicle as an example, the number of occurrences of different current densities is shown in fig. 2 (a), and it can be seen that the number of occurrences of load currents of 135A, 250A, and 255A is 500 or more. And sampling the corresponding cell stack voltage once per hour according to the operation duration, taking a first sampling value when the time is many hours, and performing linear interpolation processing on missing hours. As shown in fig. 2(b) and (c), it can be seen that the distribution of the time points of the current interval of only 135A is relatively uniform, and the other current intervals are similar to 250A, and there are time periods with extremely sparse distribution, which is not of research value. The stack voltage at a particular current value 135A is therefore selected, i.e. as the study data set.
S2, performing modal decomposition on the performance index time series data (in this embodiment, the fuel cell voltage time series data is specifically selected) by using an adaptive data decomposition method to obtain a plurality of modal quantities.
Complex operating conditions and frequent load changes during operation of a fuel cell commercial vehicle are the primary causes of fuel cell performance degradation, including short term stochastic and long term intrinsic degradation. In fact, such short-term randomly fluctuating, long-term decaying nonlinear non-stationary time series data may be considered as a mixed signal with multiple characteristic time scale components, suitable for decoupling the individual components using an adaptive data decomposition method.
Empirical Mode Decomposition (EMD) may decompose an input non-stationary nonlinear signal into a plurality of Intrinsic Mode Functions (IMFs) components and a residual error (Res) component. The IMF components obtained by decomposition have different characteristic time scales and two conditions are met: the number of the zero points and the extreme points is equal to or different from 1; the average value of the upper envelope curve determined by the local maximum value point and the lower envelope curve determined by the local minimum value point is constant 0. Compared with Fourier transform and wavelet transform, the EMD does not need to determine a basis function in advance, but carries out scale transform on the signal to obtain the IMF, and the processing efficiency is higher. As shown in fig. 3, the EMD includes the following steps:
1) obtaining all local maximum value points and local minimum value points of the original voltage time sequence data p (t), and obtaining an upper envelope line e by a cubic spline interpolation method 1,1 (t) and the lower envelope e 1,2 (t) calculating the envelope mean curve a 1,1 (t)：
Figure BDA0003661726670000061
2) Difference curve h of averaging curve and original data 1,1 (t)：
h 1,1 (t)＝p(t)-a 1,1 (t) (2)
3) If h 1,1 (t) if the above-mentioned two IMF conditions are not satisfied, then h is used 1,1 (t) replacing p (t), returning to the step 1) and repeating. Let the difference curve obtained at the ith time be:
h 1,i (t)＝h 1,(i-1) (t)-a 1,i (t) (3)
in the actual decomposition process, proper and loose Cauchy-type screening criterion SD is generally defined i 0.2-0.3 to avoid obtaining a frequency modulated signal with almost constant amplitude or an alias signal containing too many frequency components, including:
Figure BDA0003661726670000062
4) when the difference curve obtained at the ith time meets the stopping criterion, the first intrinsic mode function IMF is regarded as the first intrinsic mode function 1 (t), let the residual:
r 1 (t)＝p(t)-IMF 1 (t) (5)
by r 1 (t) replacing p (t), returning to the step 1) and repeating. Until the k-th acquisition of IMF k (t) is a monotonic function or k reaches a desired number of times, the decomposition stops. The original signal is now decomposed into:
Figure BDA0003661726670000063
however, the modal aliasing problem that EMD is difficult to avoid will reduce the effectiveness of signal decomposition, and to improve this problem, fully integrated empirical mode decomposition (CEEMD) has been introduced. As shown in fig. 4, positive white noise and negative white noise are added to the original signal, respectively. The EMD decomposition is then performed on the two new signals and the process of addition and decomposition is repeated until the cycle number requirement is met. And finally, averaging the obtained decomposition results. On this basis, the resulting IMF component and residual component are expressed as:
Figure BDA0003661726670000064
Figure BDA0003661726670000065
the voltage data in fig. 2(b) are respectively subjected to EMD and CEEMD decomposition to obtain the result in fig. 5, where (a) to (f) in fig. 5 are multi-modes subjected to EMD decomposition, and (g) to (m) in fig. 5 are multi-modes subjected to CEEMD decomposition, it can be seen that a plurality of IMF components fluctuating up and down around 0, which can be called fluctuation components, are obtained through EMD and CEEMD, and according to the sequence of decomposition output, the characteristic time scales of the IMF components are sequentially increased, the characteristic frequencies are sequentially decreased, and from the operating condition of the fuel cell commercial vehicle, the components can be regarded as recoverable stack voltage fluctuation caused by accidental external factors such as temperature and humidity change, gas impurities, vibration and impact, and the like, and can not cause long-term voltage degradation. However, the EMD and CEEMD decompositions yield IMFs of different numbers and morphologies, the latter decompositions yield IMFs of greater numbers, and the IMFs 1 of the two morphologies differ significantly. In FIG. 5, (n) and (o) are the IMFs obtained from EMD and CEEMD, respectively 1 Frequency spectrum obtained by fast Fourier transform, IMF 1 EMD is more uniformly distributed over the entire frequency band, meaning that it contains more frequency components, which can be considered to be significant spectral aliasing; IMF 1 CEEMD, which contains mainly higher frequency components, effectively improves the spectral aliasing of EMD and thus resolves a greater number of IMF components.
S3, constructing a corresponding Gated Recurrent Unit (GRU) network for training the obtained plurality of inherent modal functions and a residual component
The attenuation prediction diagnosis framework of the invention is shown in fig. 6, and comprises the following specific steps:
1) after obtaining each IMF component and a residual Res component, carrying out normalization processing to improve the training speed and the training precision of the network, and carrying out normalization by adopting the following equation;
Figure BDA0003661726670000071
in the formula, z std Is a normalized data sequence, z is the original data sequence,
Figure BDA0003661726670000072
is the mean and SD is the standard deviation.
2) Selecting a split point, determining a training set and a test set, and constructing a corresponding GRU network for each IMF component and Res signal;
3) setting model parameters of each GRU network, including the number of hidden layers, learning rate, descent rate, training times, gradient threshold value and the like;
4) and after setting model parameters and a penalty function, performing model training by using an Adam algorithm.
At this point, the training process of the model is completed. The GRU network used therein is shown in FIG. 7, where the GRU combines the internal state vector and the output vector, the output h through the previous hidden layer t-1 And input x of the current cell t Get updated door z t And a reset gate r t The gate signal output of (c):
z t ＝σ(W z ·[h t-1 ,x t ]+b z ) (10)
r t ＝σ(W r ·[h t-1 ,x t ]+b r ) (11)
in the formula, W z And W r The weights of the update gate and the reset gate respectively; b z And b r Bias for the update gate and the reset gate, respectively; sigma is sigmoid function. And processing the data by utilizing the reset gate to realize the storage of the data. the function of the tanh function is to scale the data directly to 0 and 1:
Figure BDA0003661726670000073
in the formula W n An element of weight,. is a matrix dot product. And finally, realizing data forgetting and selective storage by using the same door:
Figure BDA0003661726670000074
s4, performing modal decomposition on the fuel cell historical actual measurement performance index time sequence by adopting the self-adaptive data decomposition method, and inputting each modal quantity into a corresponding prediction model to obtain each modal prediction value;
and S5, fusing the modal predicted values of the fuel cell to obtain a performance index predicted value, and determining the attenuation condition of the fuel cell based on the performance index predicted value.
The method comprises the following steps of fusing the modal predicted values of the fuel cell to obtain a performance index predicted value, wherein the performance index predicted value is represented as:
Figure BDA0003661726670000081
wherein the content of the first and second substances,
Figure BDA0003661726670000082
is a predicted value of the performance index at time t,
Figure BDA0003661726670000083
is the predicted value of the k-th order intrinsic mode function at the time t of the performance index,
Figure BDA0003661726670000084
the predicted value of the residual error quantity at the time of the performance index t is shown, and K is the total order of the inherent mode function.
Then, the attenuation situation of the fuel cell is determined based on the predicted value of the performance index, specifically, the attenuation rate eta (t) of the performance index is calculated:
Figure BDA0003661726670000085
Figure BDA0003661726670000086
is a predicted value of the performance index at time t, p new Is the performance index of the fuel cell before the fuel cell starts to age.
In this example, the voltage data in fig. 2(b) is trained and test-verified, and the first 80% of the data is set as a training set, and the last 20% of the data is set as a test set. The voltage results predicted by the different models are shown in FIG. 8 and Table 1, where LSTM represents the long and short term memory network, EMD-LSTM represents that the input of LSTM is multi-modal by EMD decomposition, EMD-GRU represents that the input of GRU is multi-modal by EMD decomposition, CEEMD-LSTM represents that the input of LSTM is multi-modal by CEEMD decomposition, and CEEMD-GRU represents that the input of GRU is multi-modal by CEEMD decomposition. Compared with direct LSTM and GRU models, the voltage attenuation prediction precision can be improved by adopting multi-mode input; in addition, the CEEMD-GRU has the highest voltage prediction accuracy, and the training time and the testing time are less compared with those of CEEMD-LSTM, so that the method provided by the invention is very suitable for online application and is suitable for a large amount of commercial vehicle data samples in the future.
TABLE 1 calculation accuracy and calculation time for different models
Figure BDA0003661726670000087
In table 1, MAE: average absolute error; MAPE: average percent error; RMSE: root mean square error.
The method firstly carries out modal decomposition on the performance index time sequence data which can describe the performance attenuation of the fuel cell system, and predicts each modal component through a prediction model, and further fuses the predicted values of each modal component to obtain the predicted value of the performance index, so that the performance attenuation of the fuel cell is predicted, a complex or empirical fuel cell performance attenuation model does not need to be established, and the method has better precision compared with other popular diagnosis methods.
The above embodiments are merely examples and do not limit the scope of the present invention. These embodiments may be implemented in other various manners, and various omissions, substitutions, and changes may be made without departing from the technical spirit of the present invention.

Claims (10)

1. A method for predicting degradation in the performance of a multi-modal fused fuel cell system, the method comprising:
s1, acquiring performance index time sequence data capable of describing the performance attenuation of the fuel cell system;
s2, performing modal decomposition on the performance index time sequence data by adopting a self-adaptive data decomposition method to obtain a plurality of modal quantities;
s3, respectively constructing prediction models for predicting the modal quantities for each modal quantity, and training the prediction models;
s4, performing modal decomposition on the fuel cell historical actual measurement performance index time sequence by adopting the self-adaptive data decomposition method, and inputting each modal quantity to the corresponding prediction model to obtain each modal predicted value;
and S5, fusing the modal predicted values of the fuel cell to obtain a performance index predicted value, and determining the attenuation condition of the fuel cell based on the performance index predicted value.
2. The method of claim 1, wherein the performance index includes any one of voltage, active area, impedance, and limiting current density.
3. The method of claim 1, wherein the adaptive data decomposition method comprises a complementary set empirical mode decomposition algorithm.
4. The method of claim 2, wherein the complementary ensemble empirical mode decomposition algorithm comprises the steps of:
1) respectively adding positive white noise and negative white noise to an original signal to be decomposed;
2) performing empirical mode decomposition on the two newly-noisy signals;
3) repeating the process of addition and decomposition until the requirement of cycle times is met, and finally averaging the obtained decomposition results:
Figure FDA0003661726660000011
Figure FDA0003661726660000012
wherein, IMF k (t) is the kth order natural mode function at time t, r K (t) is the residual quantity at time t, IMF k,j (t) adding a white noise signal, and performing IMF (intrinsic mode function) on the kth order at t moment in the j-th cyclic calculation k,-j (t) adding a negative white noise signal, and performing a kth order intrinsic mode function at t moment in the j-th cycle calculation, r K,j (t) is the residual amount at t in the j-th loop calculation by adding a white noise signal, r K,-j (t) is the residual quantity at the time t in the j-th cycle calculation by adding a negative white noise signal, and M is the total cycle calculation frequency.
5. The method of claim 4, wherein the empirical mode decomposition in step 2) specifically comprises:
21) obtaining all local maximum value points and local minimum value points of the time sequence data p (t) to be decomposed, and obtaining an upper envelope line e by a cubic spline interpolation method 1,1 (t) and the lower envelope e 1,2 (t) calculating the envelope mean curve a 1,1 (t)；
22) Calculating an envelope mean curve a 1,1 (t) curve of the difference between the time series data p (t) to be decomposed and the time series data h (t) 1,1 (t)；
23) If h 1,1 (t) if IMF conditions are not satisfied, the useh 1,1 (t) replacing p (t), returning to the step 21) and repeating, and setting the difference curve obtained at the ith time as h 1,i (t)：
h 1,i (t)＝h 1,(i-1) (t)-a 1,i (t)
24) When the difference curve obtained at the ith time meets the stopping criterion, the first intrinsic mode function IMF is regarded as the first intrinsic mode function 1 (t) and calculating the corresponding residual r 1 (t) using r 1 (t) substituting p (t), going back to step 21) and repeating until the k-th time IMF is obtained k (t) is a monotonic function or k reaches a desired number of times, the decomposition stops.
6. The method of claim 5 wherein the IMF conditions are: the number of the zero points and the extreme points is equal to or different from 1; the average value of the upper envelope curve determined by the local maximum value points and the lower envelope curve determined by the local minimum value points is constant 0.
7. The method of claim 1, wherein the predictive model comprises a gated cyclic unit network.
8. The method according to claim 1, wherein the normalization process is performed on each of the modal quantities obtained in step S2 during the training of the prediction model in step S3, and the network training is performed using the normalized modal quantity data.
9. The method according to claim 3, wherein the step S5 is to fuse the modal predicted values of the fuel cell to obtain the predicted value of the performance index, wherein the predicted value of the performance index is represented by:
Figure FDA0003661726660000021
wherein the content of the first and second substances,
Figure FDA0003661726660000022
is a predicted value of the performance index at time t,
Figure FDA0003661726660000023
is the predicted value of the k-th order intrinsic mode function at the time t of the performance index,
Figure FDA0003661726660000024
the predicted value of the residual error quantity at the time of the performance index t is shown, and K is the total order of the inherent mode function.
10. The method according to claim 1, wherein the step S5 of determining the fuel cell degradation based on the predicted value of the performance index is to calculate the degradation rate η (t) of the performance index:
Figure FDA0003661726660000031
Figure FDA0003661726660000032
is a predicted value of the performance index at time t, p new Is the performance index of the fuel cell before the fuel cell starts to age.
CN202210579312.7A 2022-05-25 2022-05-25 Multi-mode fusion fuel cell system performance attenuation prediction method Pending CN115098999A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210579312.7A CN115098999A (en) 2022-05-25 2022-05-25 Multi-mode fusion fuel cell system performance attenuation prediction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210579312.7A CN115098999A (en) 2022-05-25 2022-05-25 Multi-mode fusion fuel cell system performance attenuation prediction method

Publications (1)

Publication Number Publication Date
CN115098999A true CN115098999A (en) 2022-09-23

Family

ID=83289625

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210579312.7A Pending CN115098999A (en) 2022-05-25 2022-05-25 Multi-mode fusion fuel cell system performance attenuation prediction method

Country Status (1)

Country Link
CN (1) CN115098999A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116130721A (en) * 2023-04-11 2023-05-16 杭州鄂达精密机电科技有限公司 Status diagnostic system and method for hydrogen fuel cell
CN117236082A (en) * 2023-11-15 2023-12-15 中汽研新能源汽车检验中心(天津)有限公司 Fuel cell performance decay prediction method and system based on big data platform

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116130721A (en) * 2023-04-11 2023-05-16 杭州鄂达精密机电科技有限公司 Status diagnostic system and method for hydrogen fuel cell
CN117236082A (en) * 2023-11-15 2023-12-15 中汽研新能源汽车检验中心(天津)有限公司 Fuel cell performance decay prediction method and system based on big data platform
CN117236082B (en) * 2023-11-15 2024-01-23 中汽研新能源汽车检验中心(天津)有限公司 Fuel cell performance decay prediction method and system based on big data platform

Similar Documents

Publication Publication Date Title
CN109886498B (en) EMD-GRU short-term power load prediction method based on feature selection
CN110674604B (en) Transformer DGA data prediction method based on multi-dimensional time sequence frame convolution LSTM
CN111199270B (en) Regional wave height forecasting method and terminal based on deep learning
CN115098999A (en) Multi-mode fusion fuel cell system performance attenuation prediction method
CN110163433B (en) Ship flow prediction method
CN112966432A (en) Method and device for predicting remaining effective life of lithium ion battery
CN111931983B (en) Precipitation prediction method and system
CN112949164A (en) Lithium battery health state prediction method
CN112734002B (en) Service life prediction method based on data layer and model layer joint transfer learning
CN114912077B (en) Sea wave forecasting method integrating random search and mixed decomposition error correction
CN116316591A (en) Short-term photovoltaic power prediction method and system based on hybrid bidirectional gating cycle
CN114707712A (en) Method for predicting requirement of generator set spare parts
CN111985719A (en) Power load prediction method based on improved long-term and short-term memory network
CN113468796A (en) Voltage missing data identification method based on improved random forest algorithm
CN113409072A (en) Empirical mode decomposition and distributed GRU neural network and price prediction method
CN115511162A (en) Short-term power load prediction method based on CVMD-GRU-DenseNet hybrid model
CN111141879A (en) Deep learning air quality monitoring method, device and equipment
CN114239990A (en) Time series data prediction method based on time series decomposition and LSTM
CN115860277B (en) Data center energy consumption prediction method and system
Khaleghi et al. Towards machine-learning driven prognostics and health management of Li-ion batteries. A comprehensive review
CN116933152B (en) Wave information prediction method and system based on multidimensional EMD-PSO-LSTM neural network
CN113722308B (en) Acceleration response data complement method and device based on EEMD-MultiCNN-LSTM
CN116665483A (en) Novel method for predicting residual parking space
CN116127833A (en) Wind power prediction method, system, device and medium based on VMD and LSTM fusion model
CN115883424A (en) Method and system for predicting traffic data between high-speed backbone networks

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination