CN113743008A - Fuel cell health prediction method and system - Google Patents

Fuel cell health prediction method and system Download PDF

Info

Publication number
CN113743008A
CN113743008A CN202111014710.6A CN202111014710A CN113743008A CN 113743008 A CN113743008 A CN 113743008A CN 202111014710 A CN202111014710 A CN 202111014710A CN 113743008 A CN113743008 A CN 113743008A
Authority
CN
China
Prior art keywords
fuel cell
data
health
target domain
model
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.)
Withdrawn
Application number
CN202111014710.6A
Other languages
Chinese (zh)
Inventor
陈睿杨
徐瑞龙
孙震东
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.)
Anhui Li'anji Technology Co ltd
Original Assignee
Anhui Li'anji Technology Co ltd
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 Anhui Li'anji Technology Co ltd filed Critical Anhui Li'anji Technology Co ltd
Priority to CN202111014710.6A priority Critical patent/CN113743008A/en
Publication of CN113743008A publication Critical patent/CN113743008A/en
Withdrawn legal-status Critical Current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
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • 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
    • 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
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H01—ELECTRIC ELEMENTS
    • H01M—PROCESSES OR MEANS, e.g. BATTERIES, FOR THE DIRECT CONVERSION OF CHEMICAL ENERGY INTO ELECTRICAL ENERGY
    • H01M8/00—Fuel cells; Manufacture thereof
    • H01M8/04—Auxiliary arrangements, e.g. for control of pressure or for circulation of fluids
    • H01M8/04298—Processes for controlling fuel cells or fuel cell systems
    • H01M8/04305—Modeling, demonstration models of fuel cells, e.g. for training purposes
    • H—ELECTRICITY
    • H01—ELECTRIC ELEMENTS
    • H01M—PROCESSES OR MEANS, e.g. BATTERIES, FOR THE DIRECT CONVERSION OF CHEMICAL ENERGY INTO ELECTRICAL ENERGY
    • H01M8/00—Fuel cells; Manufacture thereof
    • H01M8/04—Auxiliary arrangements, e.g. for control of pressure or for circulation of fluids
    • H01M8/04298—Processes for controlling fuel cells or fuel cell systems
    • H01M8/04992—Processes for controlling fuel cells or fuel cell systems characterised by the implementation of mathematical or computational algorithms, e.g. feedback control loops, fuzzy logic, neural networks or artificial intelligence
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
    • Y02E60/30—Hydrogen technology
    • Y02E60/50—Fuel cells

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Sustainable Development (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • General Chemical & Material Sciences (AREA)
  • Electrochemistry (AREA)
  • Medical Informatics (AREA)
  • Manufacturing & Machinery (AREA)
  • Molecular Biology (AREA)
  • Sustainable Energy (AREA)
  • Chemical & Material Sciences (AREA)
  • Chemical Kinetics & Catalysis (AREA)
  • Fuzzy Systems (AREA)
  • Automation & Control Theory (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Fuel Cell (AREA)

Abstract

The invention discloses a fuel cell health prediction method and a system, belonging to the technical field of fuel cells, and comprising the steps of obtaining fuel cell aging data including source domain data and target domain data; extracting consistent health characteristic data from the source domain data and the target domain data, wherein the consistent health characteristic data are respectively the source domain health characteristic data and the target domain health characteristic data; pre-training the basic model by utilizing the source domain health characteristic data, and determining parameters of the basic model; transferring the parameters of the basic model to a fuel cell degradation model, and training the fuel cell degradation model by using the target domain health characteristic data to obtain a transfer model; performing health prediction on the target domain fuel cell by using the migration model and the target domain data; and smoothing the health prediction result of the fuel cell by adopting a particle filter algorithm, and realizing the probabilistic prediction of the residual life of the fuel cell. The invention applies the source domain knowledge to the target domain, and can accurately predict the health state of the fuel cell by using a small amount of data of the target domain.

Description

Fuel cell health prediction method and system
Technical Field
The invention relates to the technical field of fuel cells, in particular to a fuel cell health prediction method and a fuel cell health prediction system.
Background
Energy and ecology are two major topics in the current society, and energy transformation becomes a great strategic direction in order to realize 'carbon neutralization'. The fuel cell directly converts chemical energy into electric energy in an electrochemical reaction mode, and has the characteristics of high conversion efficiency, environmental protection and the like. At present, the domestic fuel cell is not commercialized in a large scale and still in a starting stage, and the durability problem of the domestic fuel cell becomes a bottleneck for restricting the large-scale application of the fuel cell. Therefore, it is necessary to develop effective health assessment and management and control research of the fuel cell to realize safe and efficient operation of the fuel cell, and the health prediction of the fuel cell is an important link of the health assessment and control of the fuel cell. However, since the fuel cell system has a complex system structure and reaction mechanism, and environmental factors affect a lot, the health decay thereof has non-linear and uncertain characteristics, and accurate health prediction of the fuel cell still has a challenge.
At present, the health prediction of fuel cells mainly comprises model-based methods, data-driven methods and fusion methods. Model-based methods include mechanistic-based aging models and empirical-based aging models, the prediction accuracy of which depends on the accuracy of the model. The data-driven based method utilizes a machine learning approach to achieve non-linear mapping of externally measurable features to internal aging states with accuracy dependent on the number and quality of experimental data samples.
With the development of artificial intelligence, people pay more and more attention to a data-driven method, but the training data of the method is in large demand, and the current fuel cell is expensive in cost and insufficient in data volume.
Disclosure of Invention
The present invention is directed to overcoming the above-mentioned deficiencies in the background art, and to enabling fuel cell health prediction based on a small amount of data.
To achieve the above object, in one aspect, the present invention employs a fuel cell health prediction method, including the steps of:
acquiring fuel cell aging data, wherein the fuel cell aging data comprises source domain data and target domain data;
extracting consistent health characteristic data from the source domain data and the target domain data, wherein the consistent health characteristic data are respectively the source domain health characteristic data and the target domain health characteristic data, and the health characteristic data are used for representing life attenuation characteristics of the fuel cell;
pre-training the constructed fuel cell degradation basic model by using the source domain health characteristic data, and determining parameters of the fuel cell degradation basic model;
transferring the parameters of the trained fuel cell degradation basic model to a fuel cell degradation model of a target domain, and training the fuel cell degradation model by using the target domain health characteristic data to obtain a fuel cell health prediction transfer model;
performing health prediction on the target domain fuel cell by using the fuel cell health prediction migration model and the target domain data;
and smoothing the health prediction result of the fuel cell by adopting a particle filter algorithm to realize probabilistic prediction of the residual life of the fuel cell.
Further, the acquiring fuel cell aging data, which includes source domain data and target domain data, includes:
performing a durability test of the fuel cell by using a WLTC (wafer level temperature test) cycle working condition to obtain test data of a fuel cell aging test as the source domain data;
and acquiring actual aging data of the fuel cell in engineering application as the target domain data.
Further, after the acquiring the fuel cell aging data, the fuel cell aging data includes source domain data and target domain data, the method further includes:
and processing the source domain data and the target domain data by adopting a filtering algorithm to obtain the source domain data and the target domain data after noise reduction.
Further, the health characteristics for characterizing the life decay of the fuel cell include output power, output voltage and limiting current;
after the extracting the consistent health feature data from the source domain data and the target domain data, further comprising:
the effectiveness and rationality of the extracted health features were analyzed using Pearson and Spearman correlation analysis methods.
Further, the fuel cell degradation basic model is a fuel cell degradation basic model based on a deep GRU neural network, and comprises an input layer, a GRU hiding layer, a full connection layer and an output layer; the GRU hidden layer introduces an updating mechanism, and the full connection layer adopts a linear rectifying unit as an activation function.
Further, the hidden unit in the GRU hidden layer includes an update gate, a reset gate and an output mechanism, wherein:
the update gate function is:
zt＝σ(Wz·[ht-1,xt])
the reset gate function is:
rt＝σ(Wr·[ht-1,xt])
the output mechanism is as follows:
Figure BDA0003239534210000031
wherein the content of the first and second substances,
Figure BDA0003239534210000032
ztindicating the output of the update gate, WzWeight matrix, h, representing the updated gatet-1Hidden state representing stored time t-1, xtAn input vector representing time t, rtRepresents the output of a reset gate, WrA weight matrix representing the reset gates is shown,
Figure BDA0003239534210000041
representing memory weight, sigma () representing sigmoid gatingFunction, htThe hidden state representing the time t contains the information of the previous node,
Figure BDA0003239534210000042
the history information indicating the reset and the current input weighted selective memory information, represents a Hadamard product, and represents a matrix multiplication.
Further, the training the constructed fuel cell degradation base model by using the source domain health characteristic data to determine parameters of the fuel cell degradation base model includes:
establishing a loss function based on a difference between an output of the fuel cell degradation base model and actual data;
normalizing the source domain health characteristic data, and determining a hyper-parameter of the fuel cell degradation basic model, wherein the hyper-parameter comprises neuron data, the number of network layers, network weight and rejection rate;
training, verifying and testing the fuel cell degradation basic model according to the source domain health characteristic data to obtain network model parameters;
and optimizing the network model parameters by adopting an Adam algorithm, and determining the optimized network model parameters as the parameters of the fuel cell degradation basic model.
Further, the predicting the health of the target domain fuel cell by using the fuel cell health prediction migration model and the target domain data includes:
performing single-step prediction on the target domain health characteristic data by using the target domain data based on the fuel cell health prediction migration model to obtain a single-step prediction result;
taking the single-step prediction result as the input of the fuel cell health prediction migration model at the next moment, and performing multi-step prediction on the fuel cell health in a rolling iteration mode;
and determining the end-of-life condition of the fuel cell, and calculating the remaining service life of the fuel cell according to the multi-step prediction result.
Further, still include:
and performing iterative optimization on the fuel cell health prediction migration model by using the newly-added data in the target domain data.
In another aspect, the present invention employs a fuel cell health prediction system comprising: data acquisition module, characteristic extraction module, training module, optimization training module and health prediction module in advance, wherein:
the data acquisition module is used for acquiring fuel cell aging data, and the fuel cell aging data comprises source domain data and target domain data;
the characteristic extraction module is used for extracting consistent health characteristic data from the source domain data and the target domain data, wherein the health characteristic data are respectively the source domain health characteristic data and the target domain health characteristic data, and the health characteristic data are used for representing life attenuation characteristics of the fuel cell;
the pre-training module is used for pre-training the constructed fuel cell degradation basic model by utilizing the source domain health characteristic data and determining the parameters of the fuel cell degradation basic model;
the optimization training module is used for transferring the parameters of the trained fuel cell degradation basic model to a fuel cell degradation model of a target domain, and training the fuel cell degradation model by using the target domain health characteristic data to obtain a fuel cell health prediction transfer model;
and the health prediction module is used for performing health prediction on the target domain fuel cell by using the fuel cell health prediction migration model and the target domain data, and smoothing the prediction result of the fuel cell by using a particle filter algorithm to realize probabilistic prediction of the residual life of the fuel cell.
Compared with the prior art, the invention has the following technical effects: according to the method, the model parameters are obtained by pre-training the source domain data stack fuel cell degradation basic model, the model parameters are transferred to the fuel cell degradation model of the target domain, the fuel cell degradation model is retrained by using a small amount of early data in the target domain, and the model parameters are adjusted to adapt to the health degradation mode of the fuel cell in the target domain, so that the health prediction of the fuel cell is realized, the prediction precision is high, the required data amount is small, and the method has guiding significance for the performance evaluation and the health management of the fuel cell.
Drawings
The following detailed description of embodiments of the invention refers to the accompanying drawings in which:
FIG. 1 is a flow chart of a fuel cell health prediction method;
FIG. 2 is a block diagram of a fuel cell health prediction system;
FIG. 3 is a diagram of a fuel cell degradation model architecture based on a deep GRU neural network;
FIG. 4 is a diagram of a GRU neural network element architecture;
fig. 5 is a flow chart of a particle filtering algorithm.
Detailed Description
To further illustrate the features of the present invention, refer to the following detailed description of the invention and the accompanying drawings. The drawings are for reference and illustration purposes only and are not intended to limit the scope of the present disclosure.
As shown in fig. 1, the present embodiment discloses a fuel cell health prediction method, including the following steps S1 to S6:
s1, acquiring fuel cell aging data, wherein the fuel cell aging data comprises source domain data and target domain data;
s2, extracting consistent health characteristic data from the source domain data and the target domain data, wherein the health characteristic data are respectively the source domain health characteristic data and the target domain health characteristic data, and the health characteristic data are used for representing life attenuation characteristics of the fuel cell;
s3, pre-training the constructed fuel cell degradation basic model by using the source domain health characteristic data, and determining parameters of the fuel cell degradation basic model;
s4, transferring the parameters of the trained fuel cell degradation basic model to a fuel cell degradation model, and training the fuel cell degradation model by using the target domain health characteristic data to obtain a fuel cell health prediction transfer model;
s5, utilizing the fuel cell health prediction migration model and the target domain data to perform health prediction on the target domain fuel cell;
and S6, smoothing the prediction result of the fuel cell by using a particle filter algorithm, and realizing the probabilistic prediction of the residual life of the fuel cell.
It should be noted that both the fuel cell degradation model of the target domain and the fuel cell degradation base model of the source domain are designed based on a deep GRU neural network, the fuel cell degradation model of the target domain is a model for describing the health degradation of the fuel cell, the degradation model pre-trained by using the data of the source domain in the source domain is called a base model, the parameters of the base model are migrated into the fuel cell degradation model of the target domain, the data of the target domain is used for performing model parameter adjustment, and the adjusted model is called a migration model.
Because the traditional method for predicting the service life by utilizing deep learning has large required training data amount, the embodiment adopts the deep migration learning idea, trains the fuel cell degradation basic model by utilizing source domain data, and migrates model parameters obtained by training into the fuel cell degradation model of a target domain, thereby realizing the health prediction of the fuel cell of the target domain. According to the scheme, a health prediction result with higher precision of the fuel cell can be obtained only by a small amount of early target domain data.
As a more preferable embodiment, in step S1: acquiring fuel cell aging data including source domain data and target domain data, including the following subdivision steps S11 to S12:
s11, using WLTC circulation working condition to carry out durability test of the fuel cell, and obtaining aging test data of the fuel cell as the source domain data;
and S12, acquiring the actual aging data of the fuel cell in engineering application as the target domain data.
It should be noted that, in this embodiment, a world light vehicle test schedule (WLTC) is used as an aging cycle condition of the fuel cell to perform a durability test of the fuel cell, and obtain test data of the aging test of the fuel cell, which includes the following specific processes:
(a) starting the fuel cell engine, carrying out a polarization curve test, and then closing the fuel cell engine;
(b) starting a fuel cell engine platform, performing a fuel cell pre-starting task, importing a WLTC (wafer level temperature coefficient) aging experiment circulation working condition, and performing a fuel cell durability test;
(c) operating for 72 hours according to the WLTC working condition, storing aging experimental data, and then closing the fuel cell engine;
(d) restarting the fuel cell engine, performing a polarization curve test, storing experimental data of the polarization curve, and then closing the fuel cell engine;
(e) and (c) judging whether the fuel cell reaches the end-of-life condition or not according to the polarization curve, if so, stopping the aging test, otherwise, returning to the step (b) and continuing to perform the durability test.
As a more preferable embodiment, in step S1: after acquiring fuel cell aging data including source domain data and target domain data, the method further includes:
and processing the source domain data and the target domain data by adopting a filtering algorithm to obtain the source domain data and the target domain data after noise reduction.
It should be noted that, in this embodiment, the SG filtering algorithm is used to perform data preprocessing, the weighted filtering is performed on the data in the window, the weight is obtained by local least square polynomial fitting, the smoothing of the sample point can be realized, and the original data change information is retained, so that smoother experimental data can be obtained, and the filtering window is determined according to the actual data.
Different from the traditional moving average filtering, the SG filtering method adopted in the embodiment can more effectively retain transient information of signal change while smoothing filtering, and is suitable for scenes with large data change. The moving average filtering method is only suitable for scenes with small data change or linear change, and is not suitable for use if the data has nonlinear mutation, such as impulsive interference; for data with larger variation, a larger sliding window needs to be set, which causes larger hysteresis.
As a further preferable technical solution, in the step S2, the health characteristics for characterizing the life decay of the fuel cell include output power, output voltage, and limiting current.
Specifically, the extraction process of the health features is as follows:
the experimental data including output power, output voltage and limiting current are standardized to eliminate dimensional difference, the standardized data are used as extracted health characteristics, and the standardization of the data is realized by using the following Min-Max method:
Figure BDA0003239534210000091
wherein x is experimental measurement data (such as limiting current), xmax,xminThe maximum and minimum values (e.g., maximum limiting current and minimum limiting current) for the data, respectively.
As a more preferable embodiment, in step S2: after extracting the consistent health characteristic data from the source domain data and the target domain data, the method further comprises the following steps:
the effectiveness and rationality of the extracted health features were analyzed using Pearson and Spearman correlation analysis methods.
It should be noted that, by performing reasonableness analysis on the source domain health characteristic data, the correlation between the extracted health characteristic and the life of the fuel cell can be determined, and if the correlation is large, the health characteristic is reasonable. And (3) directly utilizing the correlation analysis result of the source domain in the target domain, namely directly adopting the extracted health characteristics to carry out health prediction without carrying out correlation analysis.
As a further preferred technical solution, as shown in fig. 3 to 4, the fuel cell degradation basic model is a fuel cell degradation basic model based on a deep GRU neural network, and includes an input layer, a GRU hidden layer, a full connection layer, and an output layer; the GRU hidden layer introduces an updating mechanism, and the full connection layer adopts a linear rectifying unit as an activation function.
It should be noted that, in this embodiment, the gated neural network GRU is used as a basic model, so that the model is simpler. And by introducing a forgetting mechanism, the disappearance of network gradients and gradient explosion are prevented, and meanwhile, a fuel cell degradation basic model is established by utilizing a GRU deep neural network, so that the nonlinear mapping of the multidimensional health characteristics and the health state of the fuel cell in source domain data is realized.
As a further preferred technical solution, the hidden unit in the GRU hidden layer includes an update gate, a reset gate and an output mechanism, wherein:
the update gate function is:
zt＝σ(Wz·[ht-1,xt])
the reset gate function is:
rt＝σ(Wr·[ht-1,xt])
the output mechanism is as follows:
Figure BDA0003239534210000101
wherein the content of the first and second substances,
Figure BDA0003239534210000102
ztindicating the output of the update gate, WzWeight matrix, h, representing the updated gatet-1Hidden state representing stored time t-1, xtAn input vector representing time t, rtRepresents the output of a reset gate, WrA weight matrix representing the reset gates is shown,
Figure BDA0003239534210000103
representing memory weight, (. sigma.). represents sigmoid gating function, htThe hidden state representing the time t contains the information of the previous node,
Figure BDA0003239534210000104
the history information indicating the reset and the current input weighted selective memory information, represents a Hadamard product, and represents a matrix multiplication.
As a more preferable embodiment, in step S3: pre-training the constructed fuel cell degradation basic model by using the source domain health characteristic data, and determining parameters of the fuel cell degradation basic model, wherein the steps from S31 to S34 are subdivided as follows:
s31, establishing a loss function based on the difference value between the output of the fuel cell degradation basic model and actual data;
note that, in the present embodiment, the loss function is a logarithmic loss function.
S32, carrying out normalization processing on the source domain health characteristic data, and determining hyper-parameters of the fuel cell degradation basic model, wherein the hyper-parameters comprise neuron data, network layer number, network weight and discarding rate;
it should be noted that, in order to avoid overfitting, the present embodiment employs a discarding mechanism, and neurons are discarded at a certain discarding rate in the training stage, where the discarding rate is specifically set to 0.05.
S33, training, verifying and testing the fuel cell degradation basic model according to the source domain health characteristic data to obtain network model parameters;
s34, adopting an Adam algorithm, comprehensively considering the historical gradient mean value and the square sum, adaptively adjusting the learning rate of the network parameters, optimizing the network model parameters, and determining the optimized network model parameters as the parameters of the fuel cell degradation basic model.
As a more preferable embodiment, in step S5: and performing health prediction on the target domain fuel cell by using the fuel cell health prediction migration model and the target domain data, wherein the method comprises the following steps of subdividing S51 to S53:
s51, performing single-step prediction on the target domain health characteristic data based on the fuel cell health prediction migration model by using the target domain data to obtain a single-step prediction result;
s52, taking the single-step prediction result as the input of the next moment of the fuel cell health prediction migration model, and performing multi-step prediction on the fuel cell health in a rolling iteration mode;
and S53, determining the end-of-life condition of the fuel cell, and calculating the remaining service life of the fuel cell according to the multi-step prediction result.
As a more preferable technical solution, as shown in fig. 5, the step S6: smoothing the prediction result by using a particle filter algorithm to realize the probabilistic prediction of the residual life of the fuel cell, wherein the method comprises the following subdivision steps S61 to S62:
s61, initialization: randomly initializing particles to form a particle set, and distributing weight to each particle point;
s62, prediction: predicting the predicted value of the particles in the particle set at the moment k, namely the predicted result of the fuel cell, by using a state equation, and calculating an observation value corresponding to the predicted value of each particle by using an observation equation;
s63, weight updating: at the moment k, updating the weight of the particle;
s64, resampling: to avoid particle depletion, the present embodiment employs a systematic resampling mechanism, which copies high-weight particles multiple times according to the weight of the particles, discards low-weight particles, and ensures that the total number of particles remains unchanged. In order to reduce the computation time cost of resampling, the embodiment adopts a self-adaptive resampling mechanism, and sets a weight threshold of a valid particle, where the valid particle is a valid particle when the weight of the particle is greater than the value, and the invalid particle is an invalid particle when the weight of the particle is less than the value; and when the number of invalid particles is larger than the resampling threshold value, resampling is carried out, otherwise, resampling is not carried out.
S65, state estimation: and calculating an estimated value of the health state of the fuel cell according to the resampled particle set, namely taking the weighted average of all the particles as the estimated value of the current health state, further obtaining a plurality of health prediction curves with probabilistic distribution, and obtaining the residual life time of the fuel cell when the health prediction curves fall to the end-of-life condition, thereby realizing the probabilistic prediction of the residual life of the fuel cell.
By adopting the particle filter algorithm, the prediction result of the depth migration learning algorithm can be smoothed on one hand, and probabilistic prediction can be realized on the other hand. Meanwhile, the resampling mechanism can avoid particle depletion, but is time-consuming, and the adaptive resampling mechanism is adopted, so that the time cost is reduced.
As a further preferred technical solution, the method further comprises:
and performing iterative optimization on the fuel cell health prediction migration model by using newly-added data in the target domain data so as to improve the adaptability of the model to the target domain fuel cell degradation mode and improve the fuel cell health prediction precision.
It should be noted that, in this embodiment, the basic model obtained through source domain data training is migrated into the target domain, a small amount of data of the target domain is used for model adjustment to adapt to the aging tendency of the fuel cell under different aging conditions, and health prediction of the target domain fuel cell is realized through iterative optimization, so that the health prediction capability is greatly improved, and the aging experiment time is saved; and the health prediction result is smoother by combining with a particle filter algorithm, and the probabilistic prediction of the residual life of the fuel cell is realized.
As shown in fig. 2, the present embodiment discloses a fuel cell health prediction system including: a data acquisition module 10, a feature extraction module 20, a pre-training module 30, an optimization training module 40, and a health prediction module 50, wherein:
the data acquisition module 10 is configured to acquire fuel cell aging data, where the fuel cell aging data includes source domain data and target domain data;
the feature extraction module 20 is configured to extract consistent health feature data from the source domain data and the target domain data, where the health feature data are used to characterize life decay features of the fuel cell and are respectively source domain health feature data and target domain health feature data;
the pre-training module 30 is configured to pre-train the constructed fuel cell degradation base model by using the source domain health characteristic data, and determine parameters of the fuel cell degradation base model;
the optimization training module 40 is used for transferring the parameters of the trained fuel cell degradation basic model to a target domain fuel cell degradation model, and training the fuel cell degradation model by using the target domain health characteristic data to obtain a fuel cell health prediction transfer model;
the health prediction module 50 is configured to perform health prediction on the target domain fuel cell by using the fuel cell health prediction migration model and the target domain data, smooth a prediction result by using a particle filter algorithm, and implement probabilistic prediction of the remaining life of the fuel cell.
As a further preferable technical scheme, the source domain data is obtained by performing a durability test on the fuel cell by using a WLTC cycle condition to obtain test data of an aging test of the fuel cell; the target domain data is actual aging data of the fuel cell in engineering application.
As a further preferred technical solution, the system further includes a data processing module, configured to process the source domain data and the target domain data by using a filtering algorithm, so as to obtain smoothed source domain data and target domain data.
As a further preferred technical solution, the system further comprises an analysis module for analyzing the validity and rationality of the extracted health features by using Pearson and Spearman correlation analysis methods.
The system provided by the embodiment of the present invention is used for executing the above method embodiments, and for details of the process and the details, reference is made to the above embodiments, which are not described herein again.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (10)

1. A fuel cell health prediction method, comprising:
acquiring fuel cell aging data, wherein the fuel cell aging data comprises source domain data and target domain data;
extracting consistent health characteristic data from the source domain data and the target domain data, wherein the consistent health characteristic data are respectively the source domain health characteristic data and the target domain health characteristic data, and the health characteristic data are used for representing life attenuation characteristics of the fuel cell;
pre-training the constructed fuel cell degradation basic model by using the source domain health characteristic data, and determining parameters of the fuel cell degradation basic model;
transferring the parameters of the trained fuel cell degradation basic model to a fuel cell degradation model of a target domain, retraining the model by using the health characteristic data of the target domain, and adjusting the parameters to obtain a fuel cell health prediction transfer model;
the method comprises the steps that a fuel cell health prediction migration model and target domain data are utilized to conduct health prediction on a target domain fuel cell to obtain a fuel cell health prediction result;
and smoothing the health prediction result of the fuel cell by adopting a particle filter algorithm to realize probabilistic prediction of the residual life of the fuel cell.
2. The fuel cell health prediction method of claim 1, where the obtaining fuel cell aging data, the fuel cell aging data including source domain data and target domain data, comprises:
performing a durability test of the fuel cell by using a WLTC (wafer level temperature test) cycle working condition to obtain test data of a fuel cell aging test as the source domain data;
and acquiring actual aging data of the fuel cell in engineering application as the target domain data.
3. The fuel cell health prediction method of claim 2, after the obtaining fuel cell aging data, the fuel cell aging data including source domain data and target domain data, further comprising:
and processing the source domain data and the target domain data by adopting a filtering algorithm to obtain the source domain data and the target domain data after noise reduction.
4. The fuel cell health prediction method of claim 1, where the health features characterizing the decay in fuel cell life include output power, output voltage, and limiting current;
after the extracting the consistent health feature data from the source domain data and the target domain data, further comprising:
the effectiveness and rationality of the extracted health features were analyzed using Pearson and Spearman correlation analysis methods.
5. The fuel cell health prediction method of claim 1, wherein the fuel cell degradation base model is a deep GRU neural network based fuel cell degradation base model comprising an input layer, a GRU hidden layer, a full connection layer, and an output layer; the GRU hidden layer introduces an updating mechanism, and the full connection layer adopts a linear rectifying unit as an activation function.
6. The fuel cell health prediction method of claim 5, wherein the hidden units in the GRU hidden layer comprise an update gate, a reset gate, and an output mechanism, wherein:
the update gate function is:
zt＝σ(Wz·[ht-1,xt])
the reset gate function is:
rt＝σ(Wr·[ht-1,xt])
the output mechanism is as follows:
Figure FDA0003239534200000031
wherein the content of the first and second substances,
Figure FDA0003239534200000032
ztindicating the output of the update gate, WzWeight matrix, h, representing the updated gatet-1Hidden state representing stored time t-1, xtAn input vector representing time t, rtRepresents the output of a reset gate, WrA weight matrix representing the reset gates is shown,
Figure FDA0003239534200000033
representing memory weight, (. sigma.). represents sigmoid gating function, htThe hidden state representing the time t contains the information of the previous node,
Figure FDA0003239534200000034
the history information indicating the reset and the current input weighted selective memory information, represents a Hadamard product, and represents a matrix multiplication.
7. The method for predicting the health of a fuel cell according to claim 1, wherein the training the constructed fuel cell degradation base model by using the source domain health characteristic data to determine the parameters of the fuel cell degradation base model comprises:
establishing a loss function based on a difference between an output of the fuel cell degradation base model and actual data;
normalizing the source domain health characteristic data, and determining a hyper-parameter of the fuel cell degradation basic model, wherein the hyper-parameter comprises neuron data, the number of network layers, network weight and rejection rate;
training, verifying and testing the fuel cell degradation basic model according to the source domain health characteristic data to obtain network model parameters;
and optimizing the network model parameters by adopting an Adam algorithm, and determining the optimized network model parameters as the parameters of the fuel cell degradation basic model.
8. The fuel cell health prediction method of claim 1, wherein the performing the health prediction for the target domain fuel cell using the fuel cell health prediction migration model and the target domain data comprises:
performing single-step prediction on the target domain health characteristic data by using the target domain data based on the fuel cell health prediction migration model to obtain a single-step prediction result;
taking the single-step prediction result as the input of the fuel cell health prediction migration model at the next moment, and performing multi-step prediction on the fuel cell health in a rolling iteration mode;
and determining the end-of-life condition of the fuel cell, and calculating the remaining service life of the fuel cell according to the multi-step prediction result.
9. The fuel cell health prediction method according to claim 8, further comprising:
and performing iterative optimization on the fuel cell health prediction migration model by using the newly-added data in the target domain data.
10. A fuel cell health prediction system, comprising: data acquisition module, characteristic extraction module, training module, optimization training module and health prediction module in advance, wherein:
the data acquisition module is used for acquiring fuel cell aging data, and the fuel cell aging data comprises source domain data and target domain data;
the characteristic extraction module is used for extracting consistent health characteristic data from the source domain data and the target domain data, wherein the health characteristic data are respectively the source domain health characteristic data and the target domain health characteristic data, and the health characteristic data are used for representing life attenuation characteristics of the fuel cell;
the pre-training module is used for pre-training the constructed fuel cell degradation basic model by utilizing the source domain health characteristic data and determining the parameters of the fuel cell degradation basic model;
the optimization training module is used for transferring the parameters of the trained fuel cell degradation basic model to a fuel cell degradation model of a target domain, and training the fuel cell degradation model by using the target domain health characteristic data to obtain a fuel cell health prediction transfer model;
and the health prediction module is used for performing health prediction on the target domain fuel cell by using the fuel cell health prediction migration model and the target domain data, and smoothing the prediction result of the fuel cell by using a particle filter algorithm to realize probabilistic prediction of the residual life of the fuel cell.
CN202111014710.6A 2021-08-31 2021-08-31 Fuel cell health prediction method and system Withdrawn CN113743008A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111014710.6A CN113743008A (en) 2021-08-31 2021-08-31 Fuel cell health prediction method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111014710.6A CN113743008A (en) 2021-08-31 2021-08-31 Fuel cell health prediction method and system

Publications (1)

Publication Number Publication Date
CN113743008A true CN113743008A (en) 2021-12-03

Family

ID=78734398

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111014710.6A Withdrawn CN113743008A (en) 2021-08-31 2021-08-31 Fuel cell health prediction method and system

Country Status (1)

Country Link
CN (1) CN113743008A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116381541A (en) * 2023-06-05 2023-07-04 苏州时代华景新能源有限公司 Health assessment method and system for energy storage lithium battery system
CN116609692A (en) * 2023-05-05 2023-08-18 深圳职业技术学院 Battery health state diagnosis method and system based on charger detection

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116609692A (en) * 2023-05-05 2023-08-18 深圳职业技术学院 Battery health state diagnosis method and system based on charger detection
CN116609692B (en) * 2023-05-05 2024-02-13 深圳职业技术学院 Battery health state diagnosis method and system based on charger detection
CN116381541A (en) * 2023-06-05 2023-07-04 苏州时代华景新能源有限公司 Health assessment method and system for energy storage lithium battery system
CN116381541B (en) * 2023-06-05 2023-10-20 苏州时代华景新能源有限公司 Health assessment method and system for energy storage lithium battery system

Similar Documents

Publication Publication Date Title
CN112241608B (en) Lithium battery life prediction method based on LSTM network and transfer learning
CN111339712A (en) Method for predicting residual life of proton exchange membrane fuel cell
CN112990556A (en) User power consumption prediction method based on Prophet-LSTM model
CN113743008A (en) Fuel cell health prediction method and system
CN113554466B (en) Short-term electricity consumption prediction model construction method, prediction method and device
CN112734002B (en) Service life prediction method based on data layer and model layer joint transfer learning
CN109726517B (en) Equipment residual life prediction method based on multivariate associated data
CN114596726B (en) Parking berth prediction method based on interpretable space-time attention mechanism
CN113687242A (en) Lithium ion battery SOH estimation method for optimizing and improving GRU neural network based on GA algorithm
CN115236522A (en) End-to-end capacity estimation method of energy storage battery based on hybrid deep neural network
CN116930609A (en) Electric energy metering error analysis method based on ResNet-LSTM model
CN115147645A (en) Membrane module membrane pollution detection method based on multi-feature information fusion
CN114580262A (en) Lithium ion battery health state estimation method
CN113850320A (en) Transformer fault detection method based on improved support vector machine regression algorithm
CN117406100A (en) Lithium ion battery remaining life prediction method and system
CN116303786B (en) Block chain financial big data management system based on multidimensional data fusion algorithm
CN112014757A (en) Battery SOH estimation method integrating capacity increment analysis and genetic wavelet neural network
CN117312835A (en) Fuel cell performance prediction method based on mixed probability data driving model
CN116933025A (en) Transformer top layer oil temperature prediction method based on VMD and DBO-LSTM-AT
CN117035464A (en) Enterprise electricity consumption carbon emission prediction method based on time sequence network improved circulation network
CN117034762A (en) Composite model lithium battery life prediction method based on multi-algorithm weighted sum
CN116432535A (en) Performance verification error analysis method for electric power metering device considering multiple environmental elements
CN115456287A (en) Long-and-short-term memory network-based multi-element load prediction method for comprehensive energy system
CN115840917A (en) Motor temperature rise prediction method
CN114462208A (en) Effluent BOD online soft measurement method based on self-organizing RBFNN

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
WW01 Invention patent application withdrawn after publication
WW01 Invention patent application withdrawn after publication

Application publication date: 20211203