SON-EM

An algorithm for hybrid system identification of Switched affine AutoRegressive model with eXogenous inputs (SARX) models using Sum of Norms regularization and Expectation Maximization

Cite this software

If you use this software please, cite the following paper:

Hartmann, A., Lemos, J. M., Costa, R. S., Xavier, J., & Vinga, S. (2015). Identification of Switched ARX Models via Convex Optimization and Expectation Maximization. Journal of Process Control, (28), 9–16.

Available at: http://dx.doi.org/10.1016/j.jprocont.2015.02.003

Installation

$tar -xzf son-em.tar.gz

Make the directory son-em accessible to MATLAB

Prerequisites:

• MATLAB, tested on versions:
  • 7.14.0.739 (R2012a)
  • 8.6.0.267246 (R2015b)
• Statistics Toolbox
• System Identification Toolbox
• CVX MATLAB package for convex optimization, available at http://cvxr.com/cvx/download/

Content

• runtests.m -> Run this to execute all tests (runs in about 1 minute on 2.4gHz cpu)
• LICENSE -> License file (GPL v3)
• README.md -> This file
• generateHMM.m -> generates a hidden markov model
• son_EM_plot.m -> plots an example time-series
• son_EM_sensitivity_test.m -> tests the sensitivity of the son_em method to the lambda parameter
• son_EM_son.m -> son_em method
• Theta.mat -> example time-varying parameter