Premium
Levenberg‐Marquardt backpropagation algorithm for parameter identification of solid oxide fuel cells
Author(s) -
Yang Bo,
Chen Yijun,
Guo Zhengxun,
Wang Jingbo,
Zeng Chunyuan,
Li Danyang,
Shu Hongchun,
Shan Jieshan,
Fu Ting,
Zhang Xiaoshun
Publication year - 2021
Publication title -
international journal of energy research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.808
H-Index - 95
eISSN - 1099-114X
pISSN - 0363-907X
DOI - 10.1002/er.6929
Subject(s) - levenberg–marquardt algorithm , backpropagation , matlab , artificial neural network , identification (biology) , algorithm , stability (learning theory) , computer science , nonlinear system , heuristic , engineering , artificial intelligence , machine learning , botany , biology , physics , quantum mechanics , operating system
Summary Fast and precise identification of several unknown parameters for solid oxide fuel cell (SOFC) models play a critical role in modeling analysis, optimal control, and behavior prediction. However, inherent high‐nonlinear, multi‐variable, and strongly coupled features usually lead to thorny obstacles that hinder conventional methods to identify them with a high speed, high accuracy, and reliable stability. Hence, a Levenberg‐Marquardt backpropagation (LMBP) algorithm‐based parameter identification technique is proposed in this study, which is applied to efficiently train designed artificial neural networks (ANNs) to implement the identification task. Furthermore, two typical models, for example, electrochemical model (ECM) and steady‐state model (SSM), are taken into account to validate the identification performance of the LMBP algorithm under different operation conditions. Simulation results based on MATLAB demonstrate that the LMBP algorithm can extremely improve the accuracy, speed, and stability for estimating these unknown parameters via a comprehensive comparison with four mainstream meta‐heuristic algorithms, that is, artificial ecosystem‐based optimization (AEO), equilibrium optimizer (EO), grey wolf optimization (GWO), and moth‐flame optimization (MFO).