ME 298, Section 28 - Spring 2004

ME 298 Subspace-based Model Identification
Lec 028

Spring 2004

Instructor
Dr. Rolf Johansson
5141 Etcheverry Hall, Phone (510) 642-6460
johansson@me.berkeley.edu

Lectures
MW 9:30-11, 3110 Etcheverry

Schedule

Introduction: Why system identification? Why subspace-based model identification? Autonomous systems. Input-output analysis. Disturbances. Relationship to modeling.
Time Series Analysis: Difference equations. Autoregressive models (AR). Moving average models (MA). Spectral interpretation. Autoregressive Moving Average Models (ARMA). Linear estimation. Prediction and reconstruction. Optimal d-step prediction. Kalman filter;
Classical System Identification: Heuristic and approximate methods: Impulse response analysis, Step response analysis. Frequency response analysis; Signals and systems. Discretization. Effects of finite measurement time. z-transform. Discrete Fourier transform (DFT). Spectrum analysis. Linear regression, Least-squares identification, Maximum-likelihood identification, Prediction-error methods. Output-error methods. Inverse model identification;
Model Reduction: Why heuristic methods fail. Balanced realization. Balanced model reduction. Balanced truncation. Balanced model reduction. Padé approximation. Continued fraction approximation;
Realization Theory: State-space realization of impulse response. Ho-Kalman method, Kung’s algorithm, Covariance analysis, Stochastic realization theory of cross covariance functions and autocovariance functions. Realization-based system identification based on input-output data;
Subspace-based Model Identification I: Multivariable Output-error State-space (MOESP), Canonical variance analysis (CVA), 4SID, Relationship to realization-based methods;
Subspace-based Model Identification II: Frequency-domain methods, Continuous-time model identification methods, Statistical consistency, Statistical Validation Methods;
Application and Case Studies: Cardiology: Atrial Fibrillation, Neurophysiology: Human postural dynamics; Combustion Engine Control: Homogenous Charge Compression Ignition (HCCI) Control.

Certain pages in this document may require the Acrobat Reader. Click on the link to download a free copy and then setup your browser to associate .pdf files with the Acrobat reader.