| Summary: | 碩士 === 華梵大學 === 工業工程與經營資訊學系碩士班 === 93 === Energy problem has always been taken as the most important issue to the mankind. Facing to the energy deficiency and beneficial utilization, many of current researches on fuel cells related paid much attention on the parameter design and conducted various experiments for different parameter settings. In the aspect of economically producing fuel cells, the parameter design problem is a stern challenge for the researchers, the development engineers as well as the fabricators.
This research, first of all, reviews relevant literatures and organizes all the possible essential parameters and quality characteristics of fuel cells. To develop a quality prediction system based on some popular neural network paradigms, the information of PEM fuel cells about the relations between parameters and working power are collected in the early stage. After the comparison of the neural network paradigms, the most appropriate one is selected to predict the current density under a certain working voltage on the performance curve of fuel cell. When the working voltage is the same, a fuel cell with higher current density in general possesses higher efficiency of energy transformation and output power. That means it has higher performance.
Through comprehensive evaluations using practical data set, this research compares back-propagation (BP), radial basis function (RBF) and general regression (GR) neural network paradigms and finally it is verified that the BP paradigm performs the best prediction with the lowest error rate. Furthermore, linear regression analysis is also conducted to investigate the relations between the parameters and performance curve. From the result, it is found that voltage and oxidizer influence performance curve significantly. According to the results from this research, the fuel cell related organizations may be interested in developing a system for predicting the relations between parameters and performance curve.
|
|---|
