0
Research Papers

Study on Nonlinear Identification SOFC Temperature Model Based on Particle Swarm Optimization–Least-Squares Support Vector Regression

[+] Author and Article Information
Jinwei Chen

Gas Turbine Research Institute,
Shanghai Jiao Tong University,
Shanghai 200240, China
e-mail: chenjinweituihou@sjtu.edu.cn

Huisheng Zhang

Gas Turbine Research Institute,
Shanghai Jiao Tong University,
Shanghai 200240, China
e-mail: zhslm@sjtu.edu.cn

Shilie Weng

Gas Turbine Research Institute,
Shanghai Jiao Tong University,
Shanghai 200240, China
e-mail: slweng@sjtu.edu.cn

1Corresponding author.

Manuscript received September 23, 2016; final manuscript received May 16, 2017; published online June 21, 2017. Assoc. Editor: Robert J. Braun.

J. Electrochem. En. Conv. Stor. 14(3), 031003 (Jun 21, 2017) (10 pages) Paper No: JEECS-16-1130; doi: 10.1115/1.4036805 History: Received September 23, 2016; Revised May 16, 2017

In order to facilitate valid solid oxide fuel cell (SOFC) temperature control scheme, a nonlinear identification method of SOFC temperature dynamic behaviors is proposed using an autoregressive network with exogenous inputs (NARX) model, whose nonlinear function is described by a least-squares support vector regression (LSSVR) method with radial basis kernel function (RBF). During the identifying process, a particle swarm optimization (PSO) algorithm is introduced to optimize the parameters of LSSVR. On the other hand, a mechanism model is developed to sample the training data to regress the NARX model. Investigations are conducted to analyze the effects of training data size and PSO fitness function on the accuracy of the NARX model. The results demonstrate that the NARX model with tenfold cross-validation fitness function and large size data is precise enough in predicting the SOFC temperature dynamic behaviors. The maximum errors of cathode and anode outlet temperature are 0.3081 K and 0.3293 K, respectively. Furthermore, the simulation speed of NARX model is much faster than the mechanism model because NARX model avoids the internal complex computation process. The training time of the NARX model with large size data is about 1.2 s. For a 20,000 s simulation, the predicting time of the NARX model is about 0.2 s, while the mechanism model is about 36 s. In consideration of its high computational speed and accuracy, NARX model is a powerful candidate for valid multivariable model predictive control (MPC) schemes.

FIGURES IN THIS ARTICLE
<>
Copyright © 2017 by ASME
Your Session has timed out. Please sign back in to continue.

References

Figures

Grahic Jump Location
Fig. 1

Schematic of SOFC-GT hybrid system

Grahic Jump Location
Fig. 2

The structure diagram of NARX model

Grahic Jump Location
Fig. 3

Training process of NARX model of SOFC temperature

Grahic Jump Location
Fig. 4

Optimization process of LSSVR parameters

Grahic Jump Location
Fig. 5

Training data of input variables

Grahic Jump Location
Fig. 6

Training data of output variables

Grahic Jump Location
Fig. 7

Effect of fitness function on cathode outlet temperature

Grahic Jump Location
Fig. 8

Effect of fitness function on anode outlet temperature

Grahic Jump Location
Fig. 9

Large size training data of input variables

Grahic Jump Location
Fig. 10

Large size training data of output variables

Grahic Jump Location
Fig. 11

Effect of training data size on cathode outlet temperature

Grahic Jump Location
Fig. 12

Effect of training data size on anode outlet temperature

Grahic Jump Location
Fig. 13

Running time of NARX model and mechanism model

Tables

Errata

Discussions

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Related eBook Content
Topic Collections

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In