Long term performance prediction of proton exchange membrane fuel cells using machine learning method

Abstract : The environmental issues, especially the global warming due to greenhouse effect, has become more and morecritical in recent decades. As one potential candidate among different alternative "green energy" solutions forsustainable development, the Proton Exchange Membrane Fuel Cell (PEMFC) has been received extensiveresearch attention since many years for energy and transportation applications. The PEMFC stacks, can produceelectricity directly from electrochemical reaction between hydrogen and oxygen in the air, with the only by-productsof water and heat. If the hydrogen is produced from renewable energy sources, this energy conversion is 100% ecofriendly.However, the relatively short lifespan of PEMFCs operating under non-steady-state conditions (for vehicles forexample) impedes its massive use. The accurate prediction of their aging mechanisms can thus help to designproper maintenance patterns of PEMFCs by providing foreseeable performance degradation information. In addition,the prediction could also help to avoid or mitigate the unwanted degradation of PEMFC systems during operation.This thesis proposes a novel data driven approach to predict the performance degradation of the PEMFC using animproved relevance vector machine method.Firstly, the theoretical description of the PEMFC during operation will be presented followed by an extensivelydetailed illustration on impacts of operational conditions on PEMFC performance, along with the degradationmechanisms on each component of PEMFC. Moreover, different approaches of PEMFC performance prediction inthe literature will also be briefly introduced.Further, a performance prediction method using an improved Relevance Vector Machine (RVM) would be proposedand demonstrated. The prediction results based on different training zones from historical data will also bediscussed and compared with the prediction results using conventional Support Vector Machine (SVM).Moreover, a self-adaptive kernel RVM prediction method will be introduced. At the meantime, the design matrix ofthe RVM training will also be modified in order to acquire higher precision during prediction. The prediction resultswill be illustrated and discussed thoroughly in the end.In summary, this dissertation mainly discusses the analysis of the PEMFC performance prediction using advancedmachine learning methods.
Document type :
Theses
Complete list of metadatas

Cited literature [37 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-01872917
Contributor : Abes Star <>
Submitted on : Wednesday, September 12, 2018 - 4:09:06 PM
Last modification on : Thursday, September 13, 2018 - 1:04:24 AM
Long-term archiving on : Thursday, December 13, 2018 - 3:21:46 PM

File

These_WU_UTBM.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-01872917, version 1

Collections

Citation

Yiming Wu. Long term performance prediction of proton exchange membrane fuel cells using machine learning method. Electric power. Université de Technologie de Belfort-Montbeliard, 2016. English. ⟨NNT : 2016BELF0308⟩. ⟨tel-01872917⟩

Share

Metrics

Record views

202

Files downloads

68