CN114371416A - Method for predicting service life of fuel cell - Google Patents
Method for predicting service life of fuel cell Download PDFInfo
- Publication number
- CN114371416A CN114371416A CN202111616100.3A CN202111616100A CN114371416A CN 114371416 A CN114371416 A CN 114371416A CN 202111616100 A CN202111616100 A CN 202111616100A CN 114371416 A CN114371416 A CN 114371416A
- Authority
- CN
- China
- Prior art keywords
- fuel cell
- data
- prediction
- predicting
- lifetime
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
Abstract
The invention discloses a life prediction method of a fuel cell, which comprises the following steps: s1, reading original life degradation data of the fuel cell, and preprocessing the original life degradation data to obtain preprocessed data; s2, carrying out eigenmode decomposition on the preprocessed data by using an empirical mode decomposition algorithm to obtain a series of eigenmode components; s3, randomly initializing the weight parameters of the recurrent neural network; s4, determining the training time length, the size of a storage pool and the spectrum radius; s5, inputting the eigenmode components into the recurrent neural network respectively for training to obtain a training model; s6, predicting the prediction step size of the eigenmode component through the training model to obtain prediction data of the eigenmode component corresponding to the prediction step size; and S7, fusing all the prediction data in the step S6 to obtain the prediction data of the original service life of the fuel cell. The method has higher prediction precision and is suitable for on-line real-time prediction.
Description
Technical Field
The application belongs to the technical field of fuel cells, and particularly relates to a life prediction method of a fuel cell.
Background
Fuel cells are power generation devices that directly convert chemical energy in fuel into electrical energy through electrochemical reactions. Compared with the traditional energy, the fuel cell is an efficient and clean electrochemical power generation device, and has gained common attention at home and abroad in recent years.
The proton exchange membrane fuel cell power generation technology is a technology with great application potential and industrial prospect, and is widely applied to the fields of automobile power, distributed power generation, portable power supply and the like due to the characteristics of no pollution, high efficiency, low noise and the like. Currently, the operating life of a fuel cell is one of the main factors that restrict its further commercialization. Therefore, accurate prediction of the performance degradation curve of the fuel cell is an important ring for commercial popularization and application. The purpose of the PEMFC life prediction is to enable the system to learn the aging trend (FDT) of the fuel cell by studying historical empirical data of the fuel cell, and then predict the remaining life of the fuel cell, thereby playing a role in prediction and prevention.
Currently, there are a model-based prediction method and a data-based prediction method for the life prediction method of a fuel cell in academic and industrial fields. The model-based prediction method needs to accurately model each component of the fuel cell, and has large error and high difficulty, so that the current academic majority adopts the data-based prediction method. Schwann et al, in his paper, "prediction of remaining service life of a hydrogen fuel cell based on particle filtering and genetic algorithm", predicts the degradation curve of the fuel cell using the particle filtering algorithm optimized by the genetic algorithm and shows good prediction accuracy, but it can only predict one unit backwards each time, so the method is not very practical. Most of the existing prediction methods for fuel cells can only realize single-step prediction or a small number of multi-step predictions, and the accuracy of the multi-step prediction is not high, so that the existing prediction technologies are yet to be further improved and enhanced.
Disclosure of Invention
In view of the above, it is necessary to provide a method for predicting the life of a fuel cell to solve the problems that the existing prediction method can only predict a single step or the accuracy of multi-step prediction is not high. The fuel cell is mathematically modeled by combining the recurrent neural network with the empirical mode decomposition technology, known time series data are learned by using the recurrent neural network, the unknown performance degradation trend is predicted in multiple steps, higher prediction accuracy can be achieved, and the problem that the multi-step service life prediction accuracy of the existing fuel cell is low is effectively solved.
The embodiment of the invention provides a method for predicting the service life of a fuel cell, which comprises the following steps:
s1, reading original life degradation data of the fuel cell, and preprocessing the original life degradation data to obtain preprocessed data;
s2, carrying out eigenmode decomposition on the preprocessed data by using an Empirical Mode Decomposition (EMD) algorithm to obtain a series of eigenmode components (IMF);
s3, randomly initializing the weight parameters of the recurrent neural network;
s4, determining the training time length, the size of a storage pool and the spectrum radius;
s5, respectively inputting a series of intrinsic mode components (IMF) of the step S2 into the recurrent neural network for training to obtain a training model;
s6, predicting the prediction step length of the series of intrinsic mode components (IMF) in the step S2 through the training model in the step S5 to obtain prediction data of the intrinsic mode components (IMF) corresponding to the prediction step length
S7, all the prediction data in the step S6
And fusing to obtain the prediction data of the original service life of the fuel cell.
In the step S1, in the step S,
the read is read by Python.
The raw life degradation data refers to curve data of the output voltage of the fuel cell as a function of the use time during long-term use.
The curve data is collected and drawn by the following method: in the long-term use process of the fuel cell, the output voltage value of the fuel cell is measured every 0.5s on average, and then a curve is drawn by taking the use time as an abscissa and the output voltage value as an ordinate to obtain the curve data.
The pretreatment is realized by the following method: sampling a data point at intervals of 30 data points in the curve data, then drawing a curve by taking the sampled data point as a data set, taking the service time as an abscissa and taking the output voltage value as an ordinate, and obtaining the preprocessing data. Because the amount of original data is too large, redundant data can be simplified through preprocessing, and the computational efficiency of a computer can be improved while the sampling uniformity is not influenced.
In the step S2, in the step S,
the specific calculation process of the empirical mode decomposition algorithm is as follows:
(1) finding all maximum value points of the preprocessed data x (t), and fitting a maximum value envelope line e through a cubic spline function+(t); similarly, finding all minimum value points of the preprocessed data x (t), and fitting a minimum value envelope line e through a cubic spline function_(t); taking the average value of the sum of the maximum envelope and the minimum envelope as the mean envelope m of the preprocessed data x (t)1(t) satisfies:
(2) subtracting m from the preprocessed data x (t)1(t) obtaining a first order modal component imf1(t) satisfies: imf1(t)＝x(t)-m1(t)；
(3) Assume imf1(t) if the IMF condition is satisfied, repeating the above steps (1) and (2) to obtain IMFn(t), wherein n is 1, 2, 3 …;
the IMF conditions are as follows:
firstly, on the signal of the whole preprocessed data, the difference between the number of extreme points (including a maximum point and a minimum point) and the number of zero crossing points is not more than 1;
and secondly, at any point, the mean value of the envelope curve of the maximum value and the mean value of the envelope curve of the minimum value are both 0.
(4) Decomposing the preprocessed data x (t) into a series of imf (t) and a residual signal r (t) by the decomposition, satisfying:
preferably, the number of decomposition layers of the empirical mode decomposition algorithm is n ═ 10 layers, that is, the preprocessed data x (t) satisfy:
preferably, in the step (1),
the cubic spline function S (x) satisfies:
S(ti)＝x(ti) (formula I) is shown in the specification,
wherein, tiThe maximum value point or the minimum value point;
according to the continuity conditions:
S(ti-0)＝S(ti+0)
S(ti-0)′＝S(ti+0)′
S(ti-0)″＝S(ti+0) "(formula two),
add boundary assumptions:
S(t0)′＝x(t0)′＝0
S(tn)′＝x(tn) ' -0 (formula three);
and calculating to obtain the maximum value envelope line or the minimum value envelope line by combining the first formula, the second formula and the third formula.
In the application, as the number of decomposition layers increases, the signal gradually fades away the nonlinear factor owned by the signal until the signal is degenerated to be close to a linear model.
In the step S3, in the step S,
the recurrent neural network is preferably an Echo State Network (ESN).
The weight parameters are preferably an output layer weight parameter matrix Y and a pool weight parameter matrix W.
The initialization is realized by an initialization code, and the initialization code is as follows:
Y＝np.random.rand(predLen，1)；
W＝np.random.rand(resSize，resSize)-0.5；
where predLen is the one-time prediction length, resSize is the size of the pool, and np is python third-party library numpy.
In the step S4, in the step S,
preferably, the duration of the training is 100-300h, and more preferably 200 h. Preferably, the size of the reservoir is 50 to 150, more preferably 100.
Preferably, the spectrum radius is a maximum eigenvalue of the storage pool weight parameter matrix W.
The spectral radius is preferably 0.35 to 1.25, more preferably 0.7.
In the step S5, in the step S,
the training is to construct a loss function Lossfunction and continuously update the weight parameter matrix W of the storage pool by using a gradient descent methodn+1. In the present application, the training process is a process in which the loss function is continuously reduced.
The loss function is preferably a mean square error function of the predicted voltage value of the fuel cell and the actual voltage value of the fuel cell, and specifically includes the following steps:
where predLen represents the one-time prediction length, yiRepresenting the voltage of the fuel cell in fact,
represents with respect to yiIs measured (i.e., the predicted value of the voltage of the fuel cell).
The weight parameter matrix Wn+1Calculated by the following formula:
wherein, WnThe weight parameter matrix is the nth time; wn+1The obtained weight parameter matrix of the (n + 1) th time is updated; β is the learning rate, preferably 0.001 in this example;
is the partial derivative of the loss function to the weight parameter matrix.
The training model is an echo state network model constructed by using a weight parameter matrix of a storage pool corresponding to the minimum value of the loss function.
In the step S6, in the step S,
the prediction step size is 5-30 steps, preferably 20 steps.
In the step S7, in the step S,
the fusing is performed according to an inverse process of the Empirical Mode Decomposition (EMD) algorithm; prediction data of the original life x (t) of the fuel cell
Satisfies the following conditions:
the invention constructs an echo state network model fused with an empirical mode decomposition algorithm, performs effective mathematical modeling on the service life data of the fuel cell, optimizes the parameters of the reserve pool by learning partial time series data, realizes the fitting of known data and the prediction of unknown data, and can ensure higher prediction precision. In the prediction process, the training time required by the echo state network model fused with the empirical mode decomposition algorithm is short, the prediction step length is greatly increased compared with the prior art, the algorithm is simple and clear, the running time of the program is not more than 5 minutes each time, and the method is very suitable for online real-time prediction.
Drawings
The foregoing and other objects, features and advantages of the invention will be apparent from the following more particular description of preferred embodiments of the invention, as illustrated in the accompanying drawings. Like reference numerals refer to like parts throughout the drawings, and the drawings are not intended to be drawn to scale in actual dimensions, emphasis instead being placed upon illustrating the principles of the invention.
Fig. 1 is a schematic flow chart of a method for predicting the life of a fuel cell according to an embodiment of the present invention;
FIG. 2 is a graph of raw life degradation data and pre-processing data for a fuel cell used in an embodiment of the present invention;
FIG. 3 is a diagram illustrating the fitting and prediction effects of an echo state network on IMF components according to an embodiment of the present invention;
FIG. 4 is a diagram showing the effect of the life prediction method of a fuel cell according to the embodiment of the present invention on the life prediction of the fuel cell used;
fig. 5 is a graph showing the predicted voltage error (left) and the predicted voltage error distribution (right) for the fuel cell used in the embodiment of the present invention.
Detailed Description
To facilitate an understanding of the invention, the invention will now be described more fully with reference to the accompanying drawings. Preferred embodiments of the present invention are shown in the drawings. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.
It will be understood that when an element is referred to as being "connected" to another element, it can be directly connected to the other element and be integral therewith, or intervening elements may also be present. The terms "mounted," "one end," "the other end," and the like are used herein for illustrative purposes only.
It should be noted that if directional indications (such as up, down, left, right, front, and back … …) are referred to in the embodiments of the present application, the directional indications are only used to explain the relative positional relationship between the components, the movement situation, and the like in a specific posture (as shown in the drawings), and if the specific posture is changed, the directional indications are changed accordingly.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
At present, most of the existing prediction methods for the fuel cell can only realize single-step prediction or a small number of multi-step predictions, and the accuracy of the multi-step prediction is not high. Based on this, it is necessary to provide a method for predicting the lifetime of a fuel cell to solve the above technical problems.
As shown in fig. 1, an embodiment of the present invention provides a method for predicting a lifetime of a fuel cell, including the following steps:
s1, reading original life degradation data of the fuel cell, and preprocessing the original life degradation data to obtain preprocessed data;
s2, carrying out eigenmode decomposition on the preprocessed data by using an Empirical Mode Decomposition (EMD) algorithm to obtain a series of eigenmode components (IMF);
s3, randomly initializing the weight parameters of the recurrent neural network;
s4, determining the training time length, the size of a storage pool and the spectrum radius;
s5, respectively inputting a series of intrinsic mode components (IMF) of the step S2 into the recurrent neural network for training to obtain a training model;
s6, predicting the prediction step length of the series of intrinsic mode components (IMF) in the step S2 through the training model in the step S5 to obtain prediction data of the intrinsic mode components (IMF) corresponding to the prediction step length
S7, all the prediction data in the step S6
And fusing to obtain the prediction data of the original service life of the fuel cell.
In the step S1, in the step S,
the read is read by Python. Namely, the existing python can realize reading by using a data reading program written by a python programming language.
The raw life degradation data refers to curve data of the output voltage of the fuel cell as a function of the use time during long-term use.
The curve data is collected and drawn by the following method: in the long-term use process of the fuel cell, the output voltage value of the fuel cell is measured every 0.5s on average, and then a curve is drawn by taking the use time as an abscissa and the output voltage value as an ordinate to obtain the curve data.
The pretreatment is realized by the following method: sampling a data point at intervals of 30 data points in the curve data, then drawing a curve by taking the sampled data point as a data set, taking the service time as an abscissa and taking the output voltage value as an ordinate, and obtaining the preprocessing data. Because the amount of original data is too large, redundant data can be simplified through preprocessing, and the computational efficiency of a computer can be improved while the sampling uniformity is not influenced. FIG. 2 is a graph of raw life degradation data and pre-processing data for a fuel cell used in an embodiment of the present invention; as can be seen from fig. 2, the raw data is effectively simplified by preprocessing without affecting the sampling uniformity. In the step S2, in the step S,
the specific calculation process of the empirical mode decomposition algorithm is as follows:
(1) finding all maximum value points of the preprocessed data x (t), and fitting a maximum value envelope line e through a cubic spline function+(t); similarly, finding all minimum value points of the preprocessed data x (t), and fitting a minimum value envelope line e through a cubic spline function-(t); taking the average value of the sum of the maximum envelope and the minimum envelope as the mean envelope m of the preprocessed data x (t)1(t) satisfies:
(2) subtracting m from the preprocessed data x (t)1(t) obtaining a first order modal component imf1(t) satisfies: imf1(t)＝x(t)-m1(t)；
(3) Assume imf1(t) if the IMF condition is satisfied, repeating the above steps (1) and (2) to obtain IMFn(t), wherein n is 1, 2, 3 …;
the IMF conditions are as follows:
firstly, on the signal of the whole preprocessed data, the difference between the number of extreme points (including a maximum point and a minimum point) and the number of zero crossing points is not more than 1;
and secondly, at any point, the mean value of the envelope curve of the maximum value and the mean value of the envelope curve of the minimum value are both 0.
(4) Decomposing the preprocessed data x (t) into a series of imf (t) and a residual signal r (t) by the decomposition, satisfying:
preferably, the number of decomposition layers of the empirical mode decomposition algorithm is n ═ 10 layers, that is, the preprocessed data x (t) satisfy:
preferably, in the step (1),
the cubic spline function S (x) satisfies:
S(ti)＝x(ti) (formula I) is shown in the specification,
wherein, tiThe maximum value point or the minimum value point;
according to the continuity conditions:
S(ti-0)＝S(ti+0)
S(ti-0)′＝S(ti+0)′
S(ti-0)″＝S(ti+0) "(formula two),
add boundary assumptions:
S(t0)′＝x(t0)′＝0
S(tn)′＝x(tn) ' -0 (formula three);
and calculating to obtain the maximum value envelope line or the minimum value envelope line by combining the first formula, the second formula and the third formula.
In the application, as the number of decomposition layers increases, the signal gradually fades away the nonlinear factor owned by the signal until the signal is degenerated to be close to a linear model.
In the step S3, in the step S,
the recurrent neural network is preferably an Echo State Network (ESN). In the present application, an Echo State Network (ESN) is a new type of recurrent neural network, which consists of an input layer, a reservoir, and an output layer. The Echo State Network (ESN) designs a reserve pool into a sparse network consisting of a plurality of neurons, and achieves the functions of memorizing data and predicting by adjusting the characteristics of internal weights of the network.
The weight parameters are preferably an output layer weight parameter matrix Y and a pool weight parameter matrix W.
The initialization is realized by an initialization code, and the initialization code is as follows:
Y＝np.random.rand(predLen，1)；
W＝np.random.rand(resSize，resSize)-0.5；
where predLen is the one-time prediction length, resSize is the size of the pool, and np is python third-party library numpy.
In the step S4, in the step S,
preferably, the duration of the training is 100-300h, and more preferably 200 h. Preferably, the size of the reservoir is 50 to 150, more preferably 100.
Preferably, the spectrum radius is a maximum eigenvalue of the storage pool weight parameter matrix W.
The spectral radius is preferably 0.35 to 1.25, more preferably 0.7.
In the step S5, in the step S,
the training is to construct a Loss function and continuously update the weight parameter matrix W of the storage pool by using a gradient descent methodn+1. In the present application, the training process is a process in which the loss function is continuously reduced.
The loss function is preferably a mean square error function of the predicted voltage value of the fuel cell and the actual voltage value of the fuel cell, and specifically includes the following steps:
where predLen represents the one-time prediction length, yiRepresenting the voltage of the fuel cell in fact,
represents with respect to yiIs measured (i.e., the predicted value of the voltage of the fuel cell).
The weight parameter matrix Wn+1Calculated by the following formula:
wherein, WnThe weight parameter matrix is the nth time; wn+1The obtained weight parameter matrix of the (n + 1) th time is updated; β is the learning rate, preferably 0.001 in this example;
is the partial derivative of the loss function to the weight parameter matrix.
The training model is an echo state network model constructed by using a weight parameter matrix of a storage pool corresponding to the minimum value of the loss function (in the application, the construction is performed by replacing a hidden layer of a general neural network with a reserve pool).
Fig. 3 is a diagram illustrating the fitting and prediction effects of the echo state network on each IMF component according to an embodiment of the present invention.
In the step S6, in the step S,
the prediction step length is 5-30 steps; specifically, in the present embodiment, the prediction step size is preferably 20 steps.
In the step S7, in the step S,
the fusing is performed according to an inverse process of the Empirical Mode Decomposition (EMD) algorithm; prediction data of the original life x (t) of the fuel cell
Satisfies the following conditions:
the life prediction effect of the fuel cell life prediction method according to the embodiment of the present invention on the used fuel cell is shown in fig. 4, and the predicted voltage error (left) and the predicted voltage error distribution diagram (right) thereof are shown in fig. 5. As can be seen from the graphs in FIGS. 4 and 5, the prediction accuracy can be ensured in the multi-step prediction, the training time required by the prediction model is short, and the prediction speed is high.
The invention constructs an echo state network model fused with an empirical mode decomposition algorithm, performs effective mathematical modeling on the service life data of the fuel cell, optimizes the parameters of the reserve pool by learning partial time series data, realizes the fitting of known data and the prediction of unknown data, and can ensure higher prediction precision. In the prediction process, the training time required by the echo state network model fused with the empirical mode decomposition algorithm is short, the prediction step length is greatly increased compared with the prior art, the algorithm is simple and clear, the running time of the program is not more than 5 minutes each time, and the method is very suitable for online real-time prediction. The invention can ensure the prediction precision in multi-step prediction, and the prediction model requires less training time and has high prediction speed.
The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (11)
1. A method for predicting a lifetime of a fuel cell, comprising the steps of:
s1, reading original life degradation data of the fuel cell, and preprocessing the original life degradation data to obtain preprocessed data;
s2, carrying out eigenmode decomposition on the preprocessed data by using an empirical mode decomposition algorithm to obtain a series of eigenmode components;
s3, randomly initializing the weight parameters of the recurrent neural network;
s4, determining the training time length, the size of a storage pool and the spectrum radius;
s5, inputting the series of eigenmode components of the step S2 into the recurrent neural network respectively for training to obtain a training model;
s6, predicting the prediction step length of the series of intrinsic mode components in the step S2 through the training model in the step S5 to obtain prediction data of the intrinsic mode components corresponding to the prediction step length
S7, all the prediction data in the step S6
And fusing to obtain the prediction data of the original service life of the fuel cell.
2. The method of predicting the lifetime of a fuel cell according to claim 1, wherein in step S1, the raw lifetime degradation data is curve data of an output voltage of the fuel cell with respect to a usage time during a long-term use;
the curve data is collected and drawn by the following method: in the long-term use process of the fuel cell, the output voltage value of the fuel cell is measured every 0.5s on average, and then a curve is drawn by taking the use time as an abscissa and the output voltage value as an ordinate to obtain the curve data.
3. The method of predicting the lifetime of a fuel cell according to claim 2, wherein in step S1, the preprocessing is performed by: sampling a data point at intervals of 30 data points in the curve data, then drawing a curve by taking the sampled data point as a data set, taking the service time as an abscissa and taking the output voltage value as an ordinate, and obtaining the preprocessing data.
4. The method of predicting the lifetime of a fuel cell according to claim 1, wherein in step S2,
the specific calculation process of the empirical mode decomposition algorithm is as follows:
(1) finding all maximum value points of the preprocessed data x (t), and fitting a maximum value envelope line e through a cubic spline function+(t); similarly, finding all minimum value points of the preprocessed data x (t), and fitting a minimum value envelope line e through a cubic spline function-(t); taking the average value of the sum of the maximum envelope and the minimum envelope as the mean envelope m of the preprocessed data x (t)1(t) satisfies:
(2) subtracting m from the preprocessed data x (t)1(t) obtaining a first order modal component imf1(t) satisfies:
imf1(t)＝x(t)-m1(t)；
(3) assume imf1(t) if the IMF condition is satisfied, repeating the above steps (1) and (2) to obtain IMFn(t), wherein n is 1, 2, 3 …;
(4) decomposing the preprocessed data x (t) into a series of imf (t) and a residual signal r (t) by the decomposition, satisfying:
5. the method of predicting the lifetime of a fuel cell according to claim 4, wherein in the step (1),
the cubic spline function S (x) satisfies:
S(ti)＝x(ti) (formula I) is shown in the specification,
wherein, tiThe maximum value point or the minimum value point;
according to the continuity conditions:
S(ti-0)＝S(ti+0)
S(ti-0)′＝S(ti+0)′
S(ti-0)″＝S(ti+0) "(formula two),
add boundary assumptions:
S(t0)′＝x(t0)′＝0
S(tn)′＝x(tn) ' -0 (formula three);
and calculating to obtain the maximum value envelope line or the minimum value envelope line by combining the first formula, the second formula and the third formula.
6. The method of predicting the lifetime of a fuel cell according to claim 1, wherein in step S3,
the recurrent neural network is an Echo State Network (ESN);
the weight parameters are an output layer weight parameter matrix Y and a storage pool weight parameter matrix W.
7. The method of predicting the life of a fuel cell according to claim 6, wherein the initialization is performed by an initialization code as follows:
Y＝np.random.rand(predLen，1)；
W＝np.random.rand(resSize，resSize)-0.5；
where predLen is the one-time prediction length, resSize is the size of the pool, and np is python third-party library numpy.
8. The method of predicting the lifetime of a fuel cell according to claim 1, wherein in step S4,
the training time is 100-300 h; the size of the storage pool is 50-150;
and the spectrum radius is the maximum eigenvalue of the weight parameter matrix W of the storage pool.
9. The method of predicting the lifetime of a fuel cell according to claim 1, wherein in step S5, the training is to construct a Loss function, which is continuously performed by using a gradient descent methodUpdating the weight parameter matrix W of the storage pooln+1；
The loss function is a mean square error function of the predicted voltage value of the fuel cell and the actual voltage value of the fuel cell, and specifically includes the following steps:
where predLen represents the one-time prediction length, yiRepresenting the voltage of the fuel cell in fact,
represents with respect to yiThe predicted value of (2).
10. The method of predicting the lifetime of a fuel cell according to claim 9, wherein the weight parameter matrix Wn+1Calculated by the following formula:
wherein, WnThe weight parameter matrix is the nth time; wn+1The obtained weight parameter matrix of the (n + 1) th time is updated; beta is the learning rate;
is the partial derivative of the loss function to the weight parameter matrix;
the training model is an echo state network model constructed by using a weight parameter matrix of a storage pool corresponding to the minimum value of the loss function.
11. The method of predicting the lifetime of a fuel cell according to claim 1, wherein in step S7,
the fusion is carried out according to the inverse process of the empirical mode decomposition algorithm; the fuel electricityPrediction data of the original life of the pool x (t)
Satisfies the following conditions:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111616100.3A CN114371416A (en) | 2021-12-27 | 2021-12-27 | Method for predicting service life of fuel cell |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111616100.3A CN114371416A (en) | 2021-12-27 | 2021-12-27 | Method for predicting service life of fuel cell |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114371416A true CN114371416A (en) | 2022-04-19 |
Family
ID=81142491
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111616100.3A Pending CN114371416A (en) | 2021-12-27 | 2021-12-27 | Method for predicting service life of fuel cell |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114371416A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116027199A (en) * | 2022-12-08 | 2023-04-28 | 帕诺(常熟)新能源科技有限公司 | Method for detecting short circuit in whole service life of battery cell based on electrochemical model parameter identification |
-
2021
- 2021-12-27 CN CN202111616100.3A patent/CN114371416A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116027199A (en) * | 2022-12-08 | 2023-04-28 | 帕诺(常熟)新能源科技有限公司 | Method for detecting short circuit in whole service life of battery cell based on electrochemical model parameter identification |
| CN116027199B (en) * | 2022-12-08 | 2023-09-29 | 帕诺(常熟)新能源科技有限公司 | Method for detecting short circuit in whole service life of battery cell based on electrochemical model parameter identification |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112614009B (en) | Power grid energy management method and system based on deep expectation Q-learning | |
| Xiao et al. | Accurate state-of-charge estimation approach for lithium-ion batteries by gated recurrent unit with ensemble optimizer | |
| CN110187290B (en) | Lithium ion battery residual life prediction method based on fusion algorithm | |
| CN111260030B (en) | A-TCN-based power load prediction method and device, computer equipment and storage medium | |
| CN112990556A (en) | User power consumption prediction method based on Prophet-LSTM model | |
| CN111563611A (en) | Cloud data center renewable energy space-time prediction method for orientation graph convolutional network | |
| CN112116144B (en) | Regional power distribution network short-term load prediction method | |
| CN108764588A (en) | A kind of temperature influence power prediction method based on deep learning | |
| Qi et al. | Energyboost: Learning-based control of home batteries | |
| CN112365056A (en) | Electrical load joint prediction method and device, terminal and storage medium | |
| CN115983485A (en) | Prediction method and system for carbon emission of power plant | |
| Chen et al. | Optimal control strategy for solid oxide fuel cell‐based hybrid energy system using deep reinforcement learning | |
| CN114371416A (en) | Method for predicting service life of fuel cell | |
| CN115759336A (en) | Prediction method and storage medium for short-term power load prediction | |
| Wu et al. | Explainable temporal dependence in multi-step wind power forecast via decomposition based chain echo state networks | |
| Wen et al. | Lithium battery health state assessment based on vehicle-to-grid (V2G) real-world data and natural gradient boosting model | |
| Gang et al. | Dynamic economic emission dispatch with wind power based on improved multi‐objective brain storm optimisation algorithm | |
| CN112731183B (en) | Improved ELM-based lithium ion battery life prediction method | |
| CN112330044A (en) | Support vector regression model based on iterative aggregation grid search algorithm | |
| CN117494931A (en) | Marginal carbon emission factor determining method, system and equipment based on power grid node | |
| CN117252288A (en) | Regional resource active support capacity prediction method and system | |
| Xu et al. | Short-term electricity consumption forecasting method for residential users based on cluster classification and backpropagation neural network | |
| CN113779861B (en) | Photovoltaic Power Prediction Method and Terminal Equipment | |
| Ren et al. | Predicting spacecraft telemetry data by using Grey–Markov model with sliding window and particle swarm optimization | |
| CN114139783A (en) | Wind power short-term power prediction method and device based on nonlinear weighted combination |
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 |