Open AccessOnline LSSVM model to predict temperatures of furnace system in Chinese Space Station
Author(s) -
Ziyun Huang,
Yuhui Qiao,
Gangao Li
Publication year - 2021
Publication title -
journal of instrumentation
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.741
H-Index - 84
ISSN - 1748-0221
DOI - 10.1088/1748-0221/16/07/p07005
Subject(s) - hyperparameter , hyperparameter optimization , computer science , identification (biology) , controller (irrigation) , set (abstract data type) , process (computing) , cluster analysis , system identification , sample (material) , data mining , support vector machine , algorithm , artificial intelligence , chemistry , botany , chromatography , agronomy , biology , programming language , measure (data warehouse) , operating system
Crystal growth furnace is the platform of material melting experiment in space station. The precision of the temperature controller of the furnace determines whether the experiment is successful or not. Generally, system identification contributes to the design of a better controller. However, many identification methods are only effective with large datasets and cannot track the changing features of the system characteristics when the environment changes. In this article, we present an online method to identify the time-varying and large time delay furnace system. Inspired by clustering theory, to track the latest characteristics of the system, the training set is dynamically updated based on sample similarities. Grid search and grey wolf optimizer are used respectively for hyperparameter optimization in a two-phase tuning process. The presented identification method is validated using Tiangong-2 furnace data set. The results show the established recursive least-squares SVMs can successfully predict the temperatures of the furnace with different experiment environment.