CN117312797A - Composite denoising super capacitor backup power performance degradation rule and residual life LSTM prediction method - Google Patents
Composite denoising super capacitor backup power performance degradation rule and residual life LSTM prediction method Download PDFInfo
- Publication number
- CN117312797A CN117312797A CN202311328326.2A CN202311328326A CN117312797A CN 117312797 A CN117312797 A CN 117312797A CN 202311328326 A CN202311328326 A CN 202311328326A CN 117312797 A CN117312797 A CN 117312797A
- Authority
- CN
- China
- Prior art keywords
- super capacitor
- capacity
- data
- residual
- lstm
- 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
- 239000003990 capacitor Substances 0.000 title claims abstract description 90
- 238000000034 method Methods 0.000 title claims abstract description 75
- 230000015556 catabolic process Effects 0.000 title claims abstract description 33
- 238000006731 degradation reaction Methods 0.000 title claims abstract description 33
- 239000002131 composite material Substances 0.000 title claims abstract description 16
- 238000009499 grossing Methods 0.000 claims abstract description 18
- 230000008569 process Effects 0.000 claims abstract description 17
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 15
- 230000009467 reduction Effects 0.000 claims abstract description 15
- 238000007599 discharging Methods 0.000 claims abstract description 8
- 230000008859 change Effects 0.000 claims abstract description 6
- 244000062645 predators Species 0.000 claims abstract description 5
- 238000012545 processing Methods 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 18
- 238000012360 testing method Methods 0.000 claims description 12
- 230000032683 aging Effects 0.000 claims description 10
- 238000013528 artificial neural network Methods 0.000 claims description 5
- 238000005457 optimization Methods 0.000 claims description 4
- 244000062662 apex predator Species 0.000 claims description 3
- 230000000694 effects Effects 0.000 claims description 2
- 238000001914 filtration Methods 0.000 abstract description 7
- 238000006243 chemical reaction Methods 0.000 abstract description 2
- 238000011156 evaluation Methods 0.000 abstract description 2
- 230000009286 beneficial effect Effects 0.000 abstract 1
- 230000008929 regeneration Effects 0.000 abstract 1
- 238000011069 regeneration method Methods 0.000 abstract 1
- 208000028659 discharge Diseases 0.000 description 23
- 238000010586 diagram Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000004146 energy storage Methods 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000013178 mathematical model Methods 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 101000597273 Homo sapiens PHD finger protein 11 Proteins 0.000 description 1
- 101000935642 Homo sapiens Phosphoinositide 3-kinase adapter protein 1 Proteins 0.000 description 1
- 102100025312 Protein BCAP Human genes 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 239000002253 acid Substances 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 230000010485 coping Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 238000005485 electric heating Methods 0.000 description 1
- 238000003487 electrochemical reaction Methods 0.000 description 1
- 238000003411 electrode reaction Methods 0.000 description 1
- 239000003792 electrolyte Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000000178 monomer Substances 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000035802 rapid maturation Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
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
- G01R31/003—Environmental or reliability tests
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- Environmental & Geological Engineering (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Testing Electric Properties And Detecting Electric Faults (AREA)
Abstract
The invention provides a composite denoising super capacitor backup power performance degradation rule and residual service life LSTM prediction method, which comprises the following five steps: (1) data acquisition, (2) SG smoothing filtering, (3) MPA-VMD decomposition noise reduction, (4) LSTM prediction, and (5) evaluation. The invention has the beneficial effects that: the capacity is selected as a performance index, noise generated by capacity reduction, rebound and the like in the charging and discharging processes of the super capacitor is eliminated by using an SG smoothing filtering method (Savitzky-Golay Smoothing method, SG), then parameters of variation modal decomposition (Variational Mode Decomposition, VMD) are optimized by using a marine predator algorithm (Marine Predators Algorithm), noise reduction processing is carried out on data, a denoised capacity sequence is reconstructed, finally, the performance degradation rule (PDL) and the residual service life (RUL) of the super capacitor are predicted by using an LSTM, the influence of the capacity regeneration phenomenon, the charging and discharging rate difference, the internal temperature change of the super capacitor, the chemical reaction, the external electromagnetic interference and other factors of the super capacitor backup power supply on the prediction accuracy in the service process is effectively solved, the RMSE is greatly reduced, the R is improved, the prediction accuracy is high, the safety and reliability of the super capacitor backup power supply performance PDL and RUL can be accurately predicted, and the safety and reliability of the wind power generator are improved under severe wind conditions.
Description
Technical Field
The invention belongs to the technical field of supercapacitors, and particularly relates to a composite denoising method for predicting a performance degradation rule and a residual life LSTM of a backup power supply of a supercapacitor.
Background
Wind power generation has become an important means for coping with energy challenges, and plays a key role in environmental protection and resource conservation. The pitch system is an important component of the wind generating set, and can absorb wind energy and output rated power by adjusting the angle of the blades so as to maintain the safe operation of the fan. And the backup power source is used as a key component for supplying electric energy for the pitch system. The super capacitor has the advantages of high power density, high charge and discharge speed, long cycle life, green and environment-friendly scrapping treatment, capability of bearing instantaneous heavy current charge and discharge and the like as an energy storage device with excellent performance, and gradually replaces the traditional batteries such as a lead-acid storage battery to become a standby power supply of the pitch system of the wind driven generator. With rapid maturation and large-scale application of supercapacitor energy storage technologies, supercapacitor as an independent or auxiliary energy storage system has received increasing attention as to the operational safety of the supercapacitor. The backup power supply is typically operated intermittently. In general, the pitch system is powered by a power grid, and the standby power supply is in a standing state. However, in extreme conditions such as severe wind weather or grid outages, the backup power supply may be enabled to provide the pitch system with the power required for emergency feathering to ensure safe and stable operation of the wind turbine. Therefore, how to accurately predict the residual service life of the super capacitor and replace the super capacitor in time before reaching the service life threshold value plays an important role in ensuring the normal operation of the fan and improving the adaptability and the operation safety of the wind generating set.
Existing methods for predicting the residual service life of the super capacitor are divided into two main types: model-based methods and data-driven based methods. The method based on the model aims at utilizing parameter data such as voltage, current, temperature and the like, adopting a corresponding estimation method to estimate the resistance, the capacitance value and the residual allowance of the equivalent circuit of the super capacitor, taking the parameters as indexes, then establishing a physical mathematical model for describing the aging behavior of the super capacitor, and predicting the residual service life by utilizing the model; however, because the electrochemical system in the super capacitor has the characteristics of high complexity, multiple variables and strong coupling, an accurate mathematical model of aging behavior adapting to a real running environment is difficult to build, and the prediction precision is low; the data driving method does not need to consider a specific physical mechanism, directly learns the common law behind a large amount of data, and updates the data driving model in real time on the basis of a training data set, and has the advantages of simple structure, low complexity, small estimation error and the like.
However, due to the fact that the super capacitor is influenced by factors such as electromagnetic interference, charge-discharge multiplying power difference, self heat release, chemical reaction in the capacitor and the like in the full life cycle circulation charge-discharge process, abnormal capacity fluctuation rising and fluctuation falling phenomena exist in the aging curve of the capacitor, irregular noise signals are mixed, and the influence of the interference on the life prediction accuracy of the super capacitor is caused; in the initial stage of the service life of the super capacitor, the capacity of the super capacitor is greatly fluctuated due to the instability of electrochemical reaction and charge transmission, charging activation phenomenon and the like; in the middle of the service life of the super capacitor, the temperature change of the super capacitor can cause the diffusion of electrolyte and the change of the electrode reaction rate, so that the capacity is greatly fluctuated, and small fluctuation and high-frequency noise are also mixed; when the service life of the super capacitor is close to the threshold value, the performance of the super capacitor is seriously degraded, the capacity of the super capacitor is rapidly reduced, and the fluctuation amplitude is large; therefore, in order to improve the performance degradation and the residual service life prediction precision of the super capacitor, the invention provides a super capacitor performance degradation rule (PLD) and residual service life (RUL) LSTM prediction method based on composite denoising, according to different characteristics of data of each service life period of the whole service life period of the super capacitor, the effective denoising is performed by adopting the composite denoising method, the performance degradation rule and the residual service life of the super capacitor are accurately predicted by adopting the LSTM, and the super capacitor is replaced in time before reaching a failure service life threshold value, so that the safe reliability of operation of the wind driven generator under severe wind conditions is very necessary.
Disclosure of Invention
Aiming at the defects existing in the existing super capacitor performance degradation rule and residual service life prediction method, the invention discloses a super capacitor backup power performance degradation rule and residual service life LSTM prediction method for composite denoising.
The invention adopts the following technical scheme:
simulating an intermittent working mode of a standby power supply of the super capacitor, and performing ageing cycle test on the super capacitor to obtain capacity data;
the aging cycle test steps of the super capacitor are as follows:
1) Adopting a battery tester model EBC-A10H to perform charge-discharge cycle life test, and performing an accelerated aging experiment on the super capacitor at a constant temperature of 65 ℃;
2) Charging with constant voltage until the charging current is reduced to the charging cut-off current, and indicating that the charging is finished;
3) Discharging with constant current until the discharge process is stopped when the voltage of the super capacitor is respectively reduced to the discharge cut-off voltage;
4) And setting standing time after the charging and discharging processes are finished, so that terminal voltage is stabilized.
Step two, carrying out noise reduction treatment on small fluctuation and high-frequency noise data segments in the original capacity data in the middle of the super capacitor life by adopting an SG smoothing filter algorithm, wherein the method comprises the following steps of:
selecting raw datax i Left and right M sample points respectively and byx i As the origin, an array with a window size of 2M+1 is constructed so thatpOrder polynomial de-noisingq(n) Fitting the polynomial to the numberGroup:
wherein a is k Is the firstkFitting coefficients of the order;
obtaining residual C as a residual C through least square fitting
Wherein x (n) is a data set to be fitted; when residual isCThe smallest time indicates the best fitting effect;
step three, denoising data segments with large fluctuation amplitude and complex noise in the original capacity data of the initial, middle and near life threshold values of the super capacitor by adopting an MPA-VMD ocean predator algorithm optimization variation modal decomposition method, wherein the denoising processing is specifically as follows:
(1) Establishing an MPA optimization algorithm fitness functionfThe method is characterized by comprising the following steps:
weighted mean square error MSE and correlation coefficientrConstructing MPA-optimized fitness functionsfIs that
In the method, in the process of the invention,w 1 andw 2 is a weight coefficient, and the value range is [0,1 ]]And (2) andw 1 + w 2 =1；、whereiny i The sequence of original capacities is represented and,x k (i) Representing the K-th modal component,nrepresenting the number of samples; (2) optimizing VMD parameters using MPA algorithm; firstly, initializing MPA parameters, calculating and comparing the current adaptation degreeFAnd previous fitness value in iterationF 0 Is of a size of (a) and (b). If it isF<F 0 ThenF 0 =FAnd calculating and updating predator position, and setting the current iteration times to avoid causing the problem of local optimal solutioniLess than the maximum number of iterationsi max Up toi>i max Obtaining a global optimal fitness value and a top predator position, thereby obtaining a parameter optimal solutionK * 、α * 、β * 、w 1 * 、w 2 * ；
(3) According to the obtainedK * 、α * 、β * Performing VMD variation modal decomposition and noise reduction to obtainK * A modal component;
fourthly, the maximum value of the modal component is larger than a threshold valueβIs added up to obtain a reconstruction capacity sequence of
In the method, in the process of the invention,u k (t) Is the firstkModal component max%u k ) Represent the firstkMaximum value of each modal component, (-) is an indication function, when the condition in brackets is established, the indication function value is 1, otherwise, the indication function value is 0;
and fifthly, taking the data of 70% of the front reconstructed capacity sequence as a training set and 30% of the rear data as a prediction set, training and predicting by using an LSTM neural network to obtain a super capacitor residual capacity predicted value and a residual service life RUL, wherein the change rule of the residual capacity predicted value reflects the super capacitor performance degradation rule.
Drawings
FIG. 1 is a flow chart of a method for predicting the performance degradation rule and the residual life LSTM of a super capacitor backup power supply by composite denoising according to the invention
FIG. 2 is a graph of supercapacitor capacity data according to the invention
FIG. 3 is a smoothing filter diagram of a segment IV SG of the present invention
FIG. 4 is a smoothing filter of SG at V-th stage of the present invention
FIG. 5 is a diagram of the modal components of the VMD of the present invention after decomposition
FIG. 6 is a comparison of the sequences before and after noise reduction according to the present invention
FIG. 7 is a block diagram of the LSTM of the invention
FIG. 8 is a graph of SG-VMD-LSTM prediction results of the present invention
FIG. 9 is a graph showing the prediction contrast of the different combining methods of the present invention
Detailed description of the preferred embodiments
The following is a further detailed description of the aspects of the invention, taken in conjunction with the accompanying drawings and examples.
Referring to the drawings, fig. 1 is a flow chart of a method for predicting the performance degradation rule and the residual life LSTM of a composite denoising super capacitor backup power supply, and the method for predicting the performance degradation rule and the residual life LSTM of the composite denoising super capacitor backup power supply comprises the following five steps: (1) data acquisition, (2) SG smoothing filtering, (3) MPA-VMD decomposition noise reduction, (4) LSTM prediction, and (5) evaluation.
Step one: and (3) obtaining data, namely selecting 1 super capacitor with excellent new factory performance as a calibration capacitor, wherein the model is BCAP 0350E 270T 11 350F manufactured by Maxwell company, and the rated voltage is 2.7V.
Constructing a super-capacitor charge-discharge test platform, wherein the super-capacitor charge-discharge test platform comprises a battery tester and a data acquisition upper computer, the model of the battery tester is EBC-A10H, and the data acquisition upper computer is a PC; before testing, setting an electric heating constant temperature box to enable the experimental temperature to be constant at 65 ℃, firstly discharging each super capacitor monomer to 0.1V, and then testing, wherein the testing process comprises four stages of 'charging-standing-discharging-standing': the first stage is a charging stage, wherein the charging is firstly carried out at a constant voltage of 2.7V, the charging cut-off current is set to be 0.05A, and the charging is ended until the charging current is reduced to the charging cut-off current; the second stage is a standing stage, and the capacitor is kept standing for 5 minutes until the terminal voltage is stable; the third stage is a discharge stage, wherein the discharge is carried out by using 3A constant current, the discharge cut-off voltage is set to be 0.1V, and the discharge process is stopped when the super capacitor voltage is respectively reduced to the discharge cut-off voltage; the fourth stage is a standing stage, and standing is performed for 5 minutes until the terminal voltage is stable. According to the four-stage test conditions of charging, standing, discharging and standing, an accelerated aging test is carried out on the super capacitor on a test platform to obtain attribute data, wherein the data comprise sampling total time(s), charging current (A), discharging current (A), terminal voltage (V), capacity (Ah), cycle number and the like, and the sampling period is 2s.
Referring to the drawings, FIG. 2 is a graph of super capacitor capacity data, wherein the ratio of the actual capacity value to the rated capacity value of the super capacitor is used for measuring the capacity of the super capacitor and is used as the super capacitor capacity data for subsequent denoising and prediction, and the expression is that
；
In the method, in the process of the invention,C(t) Is super capacitor at the firsttThe capacity value at the time of the secondary charge-discharge cycle,C(0) The capacity value is rated for the super capacitor.
Step two: SG smoothing filtering; referring to the drawings, fig. 3 is a smoothing filter diagram of an iv-th segment SG of the present invention, and fig. 4 is a smoothing filter diagram of a v-th segment SG of the present invention, and noise reduction processing is performed on small fluctuation and high-frequency noise data segments in raw capacity data of a super capacitor in a mid-life period by adopting an SG smoothing filter algorithm. And selecting a window 2M+1 as 50, smoothing the window 2M+1 to obtain a smoothing order of 4, removing the fluctuation descending and fluctuation ascending phenomena generated by the capacity in the original data by using an SG smoothing method, and finally obtaining a capacity sequence with relatively stable degradation trend.
Step three: MPA-VMD decomposition noise reduction, for the data segment with large fluctuation amplitude and complex noise in the original capacity data of the super capacitor near the life threshold in the early, middle and middle stages of life, referring to the I, II and III segments in figure 2 as VMD decomposition area, adopting MPA to optimize VMD decomposition layer number K and penalty factorαDetermining a reconstruction threshold for the reconstructed signalβIs the optimal solution of the correlation numberrCombining with MSE as fitness function to obtain global optimal fitness value and top predator position, thereby obtaining parametersK、α、β、w 1 、w 2 8, 620, 11, 0.3, 0.7, respectively. Decomposing capacity sequence into 8 solids by VMDWith modal components (IMFs), the VMD is decomposed to obtain IMF 1-IMF 8 components, the decomposed components are shown in figure 5, figure 5 is a modal component diagram of the VMD decomposed according to the invention, and the optimal threshold is reusedβEffective components are selected for reconstruction, and a denoising signal reflecting degradation trend is obtainedS(t) Referring to fig. 6, fig. 6 is a graph showing the capacity sequence before and after noise reduction according to the present invention.
Step four: LSTM prediction; referring to the drawings, fig. 7 is a structural diagram of LSTM, and training and predicting are performed by using an LSTM neural network model to obtain a predicted value of residual capacity of the super capacitor and a residual service life RUL, where a change rule of the predicted value of residual capacity reflects a degradation rule of performance of the super capacitor.
Referring to the drawings, FIG. 8 is a graph showing the SG-VMD-LSTM prediction result of the present invention, and the capacity prediction curve selects the noise-reduced response degradation trend signalS(t) And forming a prediction training set, taking the data of 70% of the front data of the capacity time sequence as the training set and the data of 30% of the rear data as the prediction set, and performing training and prediction by using the LSTM neural network to obtain a capacity prediction curve. LSTM neural network prediction model parameters are set as follows: the input layer is 20; the hidden layer is 10; the Relu activation layer is 1; the full-connection layer has a1 learning rate of 0.005; the training period is 300 rounds, and each round of iteration is 12 times. After training iteration, predicting the capacity of the corresponding super capacitor after obtaining the target position, and starting the charge and discharge cycle times T corresponding to the starting point when the prediction is started 0 When the capacity predictive value reaches the failure threshold value (80%), recording the corresponding super capacitor charge-discharge cycle times T 0 =2201, and the supercapacitor RUL is calculated to be 597 charge-discharge cycles.
Step five: and evaluating the prediction of the performance degradation rule (PDL) and the residual service life (RUL). The advantages and effectiveness of the methods herein are verified by various combinations of algorithms and by comparative analysis with other predictive methods.
The effectiveness verification of the performance degradation rule and the residual life LSTM prediction method of the super capacitor backup power supply with composite denoising is provided in the patent. The capacity data used is obtained by carrying out an accelerated aging experiment on the super capacitor at a constant temperature of 65 ℃. Failure thresholds are considered to be reached when the supercapacitor capacity is below 80%. The prediction target is the number of charge and discharge times when the capacity curve drops to 80%. The experimental scene is set to select 70% of data before the capacity time sequence as a training set for LSTM neural network training, 30% of data after the capacity time sequence as a prediction set for LSTM prediction of PDL and RUL.
Referring to the drawings, fig. 9 is a graph of prediction comparison of different combination methods of the present invention, in which the SG-LSTM curve is located at the top and deviates from the experimental curve maximally, the SG-LSTM method pre-processes the whole original data by using SG smoothing filtering, and the SG smoothing filtering is adapted to remove local small fluctuation and high frequency noise, so that the capacity fluctuation caused by the charge activation phenomenon in the early charge-discharge cycle of the super capacitor is too smooth, which seriously affects the LSTM network training precision, so that the performance degradation prediction precision is the worst. The VMD-LSTM curve is positioned at the top, deviates from the experimental curve by a certain degree and has larger fluctuation, especially when approaching to the failure threshold value, because the VMD-LSTM method adopts VMD mode decomposition noise reduction to preprocess the whole original data, the VMD mode decomposition noise reduction is suitable for the nonlinear and non-stationary noise of the capacity data in the cyclic charge and discharge process of the super capacitor, and excessively decomposes and denoises the medium-term charge and discharge cyclic stable region of the super capacitor, thereby influencing the training precision of the LSTM network and leading the prediction precision of performance degradation to be poor. The LSTM curve is positioned below the experimental curve, has smaller deviation from the experimental curve, but presents sawtooth ripple, mainly because the LSTM method uses original data which is not subjected to any denoising treatment in the network training and predicting process, and the noise of the LSTM curve influences the LSTM network training precision, so that the performance degradation predicting precision is not high. The curve of the composite denoising super capacitor backup power performance degradation rule and residual life LSTM prediction method is positioned below an experimental curve, and the curve is smooth and deviates from the experimental curve to the minimum, mainly because the method disclosed by the invention is mainly aimed at noise generated by different stage characteristics in the aging process of the super capacitor, and adopts SG smoothing filtering and VMD modal decomposition denoising methods to segment and denoise the original data of the super capacitor in early, middle and near life threshold stages, so that noise interference is effectively removed, and the LSTM network training precision and performance degradation prediction precision are greatly improved. As can be seen from Table 2, the RMSE of the SG-VMD-LSTM, SG-LSTM, LSTM methods throughout the prediction phase was 1.655, 2.898, 3.850, 3.547, respectively, the RMSE of SG-VMD-LSTM was minimal, and the RMSE of the other 3 methods was 1.75, 2.33, 2.14 times that of the methods herein, respectively; r is 0.955, 0.863, 0.759 and 0.796 respectively, R of SG-VMD-LSTM is more than 0.95 and is closest to 1, and R of other 3 methods is reduced by 9.6%, 20.5% and 16.6% compared with the method. Therefore, the method has the highest prediction precision on the performance degradation trend of the super capacitor.
As shown in Table 4, the actual RUL of the super capacitor is 597 charge-discharge cycles, and the RUL of SG-VMD-LSTM, SG-LSTM and LSTM methods are 603, 620, 622 and 623 respectively, the RUL errors of the methods are 0.84%, 3.85%, 4.19% and 4.35%, the RUL errors of the methods are the smallest, and the RUL errors of other methods are about 5 times of the methods. The standby power supply belongs to typical intermittent operation, and in a normal case, the pitch system is powered by a power grid, and the standby power supply is in a standing state; the standby power supply is started only when the emergency power supply encounters extreme conditions such as severe wind weather or power grid outage, and the standby power supply is used less frequently. In addition, the closer the supercapacitor is to the failure threshold, the longer the time span required for its charge-discharge cycle. And because the charge and power generation cycle times of the super capacitor are used as RUL, a slight prediction error of the RUL can bring a lot of errors for taking time as the residual service life. Therefore, according to the method for predicting the performance degradation rule and the residual life LSTM of the super capacitor backup power supply by composite denoising, which is provided by the patent, aiming at different types of noise caused by different characteristics at different stages in the service process of the super capacitor backup power supply, the performance degradation rule and the residual life of the super capacitor backup power supply can be accurately predicted in time, and the performance degradation rule and the residual life of the super capacitor backup power supply can be accurately mastered in time, so that the replacement can be timely carried out before the backup power supply of a wind turbine generator reaches a service life threshold, and the safety and reliability of operation of the wind turbine generator under severe wind conditions are improved.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, alternatives, and improvements that fall within the spirit and scope of the invention.
Claims (1)
1. A super capacitor performance degradation rule and residual life LSTM prediction method based on composite denoising comprises the steps of firstly obtaining capacity data in a super capacitor aging cycle process through a super capacitor charging and discharging test platform; respectively adopting different noise reduction methods to carry out composite denoising on the data of the super capacitor in different service life periods, and reconstructing a capacity data sequence; then, LSTM is used for predicting the performance degradation rule and the residual service life (RUL), and the RUL predicted value isWherein->Predicting the number of cycles at the end of life for the supercapacitor, < >>The number of cycles at which the starting point of the super capacitor is predicted is set; and by combiningRULError->Root mean square error->、/>Evaluating the predicted result, whereinnFor predicting cycle number,/->Mean of the capacity experimental values、x i The capacity experimental value is a capacity prediction value; the super capacitor performance degradation rule and residual life LSTM prediction method based on composite denoising is characterized by comprising the following steps of:
step one, carrying out noise reduction treatment on small fluctuation and high-frequency noise data segments in original capacity data in the middle of super capacitor life by adopting an SG smoothing filter algorithm, wherein the method comprises the following steps of:
selecting raw datax i Left and right M sample points respectively and byx i As the origin, an array with a window size of 2M+1 is constructed so thatpOrder polynomial de-noisingq(n) Polynomial fitting this array:
；
wherein a is k Is the firstkFitting coefficients of the order;
obtaining residual C as a residual C through least square fitting
；
Wherein x (n) is a data set to be fitted; when residual isCThe smallest time indicates the best fitting effect;
step two, denoising data segments with large fluctuation amplitude and complex noise in the original capacity data of the initial, middle and near life threshold values of the super capacitor by adopting an MPA-VMD ocean predator algorithm optimization variation modal decomposition method, wherein the denoising processing is specifically as follows:
(1) Establishing an MPA optimization algorithm fitness functionfThe method is characterized by comprising the following steps:
weighted mean square error MSE and correlation coefficientrConstructing MPA-optimized fitness functionsfIs that
；
In the method, in the process of the invention,w 1 andw 2 is a weight coefficient, and the value range is [0,1 ]]And (2) andw 1 + w 2 =1；、wherein the method comprises the steps ofy i The sequence of original capacities is represented and,x k (i) Representing the K-th modal component,nrepresenting the number of samples;
(2) Optimizing VMD parameters by MPA algorithm to obtain global optimal fitness value and top predator position, thereby obtaining parameter optimal solutionK * 、α * 、β * 、w 1 * 、w 2 * ；
(3) According to the obtainedK * 、α * 、β * Performing VMD variation modal decomposition and noise reduction to obtainK * A modal component;
step three, the maximum value of the modal component is larger than a threshold valueβIs added up to obtain a reconstruction capacity sequence of
；
In the method, in the process of the invention,u k (t) Is the firstkModal component max%u k ) Represent the firstkMaximum value of each modal component, (-) is an indication function, when the condition in brackets is established, the indication function value is 1, otherwise, the indication function value is 0;
and fourthly, taking the data of 70% of the front reconstructed capacity sequence as a training set and the data of 30% of the rear reconstructed capacity sequence as a prediction set, training and predicting by using an LSTM neural network to obtain a super capacitor residual capacity predicted value and a residual service life RUL, wherein the change rule of the residual capacity predicted value reflects the super capacitor performance degradation rule.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311328326.2A CN117312797A (en) | 2023-10-13 | 2023-10-13 | Composite denoising super capacitor backup power performance degradation rule and residual life LSTM prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311328326.2A CN117312797A (en) | 2023-10-13 | 2023-10-13 | Composite denoising super capacitor backup power performance degradation rule and residual life LSTM prediction method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117312797A true CN117312797A (en) | 2023-12-29 |
Family
ID=89261878
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311328326.2A Pending CN117312797A (en) | 2023-10-13 | 2023-10-13 | Composite denoising super capacitor backup power performance degradation rule and residual life LSTM prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117312797A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117825859A (en) * | 2024-01-31 | 2024-04-05 | 华锐风电科技(集团)股份有限公司 | Health detection method and device for super capacitor of pitch system |
-
2023
- 2023-10-13 CN CN202311328326.2A patent/CN117312797A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117825859A (en) * | 2024-01-31 | 2024-04-05 | 华锐风电科技(集团)股份有限公司 | Health detection method and device for super capacitor of pitch system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107957562B (en) | Online prediction method for residual life of lithium ion battery | |
| CN111398833B (en) | Battery health state assessment method | |
| CN112034356B (en) | GP-UKF-based online SOH estimation method for electric vehicle power battery | |
| CN113049962B (en) | LSTM-based energy storage device operation situation deduction method | |
| CN109543317B (en) | Method and device for predicting remaining service life of PEMFC | |
| CN112327188A (en) | Model-data hybrid-driven lithium ion battery residual life prediction method | |
| CN107918103B (en) | Lithium ion battery residual life prediction method based on gray particle filtering | |
| CN113406521B (en) | Lithium battery health state online estimation method based on feature analysis | |
| CN117312797A (en) | Composite denoising super capacitor backup power performance degradation rule and residual life LSTM prediction method | |
| CN112415414A (en) | Method for predicting remaining service life of lithium ion battery | |
| CN113093020B (en) | Method for predicting remaining service life of lithium ion battery based on LSTM neural network | |
| CN113064093A (en) | Energy storage battery state of charge and state of health joint estimation method and system | |
| CN111143973B (en) | Valve-regulated lead-acid storage battery degradation trend prediction method based on Gauss process regression | |
| CN109444598B (en) | Performance test method of super capacitor | |
| CN111414709A (en) | Early prediction method for cycle life and capacity attenuation of super capacitor based on neural network | |
| CN112881916A (en) | Method and system for predicting health state and remaining usable life of lithium battery | |
| CN114740388A (en) | Lithium battery residual life state evaluation method based on improved TCN | |
| CN113484771A (en) | Method for estimating wide-temperature full-life SOC and capacity of lithium ion battery | |
| CN114062993A (en) | CVT error state prediction method based on time convolution network | |
| CN117554844A (en) | Battery monomer fault detection method and device of energy storage system | |
| CN113791351B (en) | Lithium battery life prediction method based on transfer learning and difference probability distribution | |
| CN115219918A (en) | Lithium ion battery life prediction method based on capacity decline combined model | |
| CN114740360A (en) | Lithium ion battery residual service life prediction method based on SAE-CEEMDAN-LSTM | |
| CN110728103B (en) | Method for identifying parameters of equivalent circuit model of super capacitor module through segmented optimization | |
| CN113759258A (en) | Power battery SOC estimation method and device and pure electric vehicle |
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 |