Nonlinear System Identification: Prediction Error Method vs Neural Network

Jinming Sun, Yanqiu Huang*, Wanli Yu, Alberto Garcia-Ortiz

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review

1 Citation (Scopus)
200 Downloads (Pure)


System identification has been used in various domains for analyzing system properties and carrying out filtering, prediction and automatic control. Prediction error method (PEM) is one of the classic methods to estimate system parameters and exploit dynamical structure of the studied system; while neural network (NN) is favorable for black-box systems with unknown structures. As the popularity of Internet of Things (IoT) and Cyber-physical systems (CPS) increases, the identification tasks are moving more towards resource-constrained devices. Accordingly, some studies incorporate system prior knowledge into NN to improve its efficiency. However, it is unclear whether the adapted NN outperforms the classic PEM.
This paper provides a fair comparison between two techniques in terms of estimation accuracy and speed on several common
nonlinear systems. The results indicate that NN is wider applicable and accurate, but more expensive from computational perspective; whereas PEM is more lightweight, but has limitations when the system input has frequent abrupt changes.
Original languageEnglish
Title of host publication2021 10th International Conference on Modern Circuits and Systems Technologies (MOCAST)
Place of PublicationPiscataway, NJ
Number of pages4
ISBN (Print)978-1-6654-1847-8
Publication statusPublished - 27 Jul 2021
Event10th International Conference on Modern Circuits and Systems Technologies, MOCAST 2021 - Virtual Conference
Duration: 5 Jul 20217 Jul 2021
Conference number: 10


Conference10th International Conference on Modern Circuits and Systems Technologies, MOCAST 2021
Abbreviated titleMOCAST 2021
CityVirtual Conference

Cite this