Parameter Magnitude-Based Information Criterion For Optimum Model Structure Selection In System Identification

Model structure selection is among one of the steps in system identification and in order to carry out this, information criterion is developed. It plays an important role in determining an optimum model structure with the aim of selecting an adequate model to represent a real system. A good informa...

全面介绍

Saved in:
书目详细资料
主要作者: Mohd Nasir, Abdul Rahman
格式: Thesis
语言:English
English
出版: 2020
主题:
在线阅读:http://eprints.utem.edu.my/id/eprint/25448/1/Parameter%20Magnitude-Based%20Information%20Criterion%20For%20Optimum%20Model%20Structure%20Selection%20In%20System%20Identification.pdf
http://eprints.utem.edu.my/id/eprint/25448/2/Parameter%20Magnitude-Based%20Information%20Criterion%20For%20Optimum%20Model%20Structure%20Selection%20In%20System%20Identification.pdf
标签: 添加标签
没有标签, 成为第一个标记此记录!
实物特征
总结:Model structure selection is among one of the steps in system identification and in order to carry out this, information criterion is developed. It plays an important role in determining an optimum model structure with the aim of selecting an adequate model to represent a real system. A good information criterion not only evaluates predictive accuracy but also the parsimony of model. There are many information criteria those are widely used such as Akaike information criterion (AIC) and Bayesian information criterion (BIC). On bias evaluation, these criteria only tackle on the number of parameters in a model. There scarcely have been any information criterion that evaluates parsimony of model structures (bias contribution) based on the magnitude of parameter or coefficient. The magnitude of parameter could have a big role in choosing whether a term is significant enough to be included in a model and justifies one’s judgement in choosing or discarding a term/variable. This study presents the comparison between parameter-magnitude based information criterion 2 (PMIC2), PMIC (an earlier version of its kind), AIC and BIC in selecting a correct model on simulated data and real data. For simulated data, PMIC2 was compared to AIC and BIC using enumerative approach and genetic algorithm. The test were made to a number of simulated systems in the form of discrete-time models of various linearity, lag orders and number of terms/variables. Then, PMIC2 was tested in selecting a good model to represent a real system based on gas furnace data and the results is compared to PMIC. The selected model was then tested using correlation test for model validation. Overall conclusion, it is shown that PMIC2 is able to select a more parsimonious model, yet adequately accurate, than AIC, BIC and PMIC.