Identification of Switching ARX Models for Hybrid Systems

  • Author / Creator
    Nazari, Sohail
  • This dissertation explores the development of a system identification method for Switching ARX (SARX) models for Off-line and Online applications. The switching sequence in SARX models converts the model parameter's estimation into a mix-integer optimization problem. To cope with complexity of the problem, an existing approach that provides an alternative formulation for multi-mode switching models is adopted. The Algebraic Geometric approach addresses the aforementioned problem by executing the identification procedure via two steps. The first step estimates the parameters of the linear ARX model, which is constructed through embedding all the sub-models. The second step retrieves parameters of sub-models from the estimated model obtained in the first step. Although the AG method delivers exact estimation in the deterministic situation, it suffers from a lack of accuracy in the presence of noise.

    This dissertation investigates the root cause of the mentioned drawback in the AG method and provides a systematic approach to deal with the measurement noise so the identification performance is improved. The proposed Stochastic Algebraic Geometric (SAG) approach reformulates the SARX parameters estimation problem into a "lifted" error-in-variable (EIV) model. Moreover, the characteristics of the proposed EIV model along with the estimation of its parameters are closely investigated. The requirements of a consistent estimation are derived through statistical analysis. In order to calculate the parameters of the sub-models improved retrieving procedures are proposed.

    In order to extend the application of the SAG into the online parameter estimation, a recursive version of the SAG approach is developed. To achieve this goal, a recursive algorithm for a class of EIV models is derived. Also, a parameter retrieving procedure independent of the data points is developed to determine parameters of the sub-models.

    To demonstrate a potential application of the proposed approach, a novel fault detection method is developed for linear switching systems. This approach is independent of estimating the sub-model's parameters. By using a residual evaluation method, the incipient changes of the sub-models' parameters can be detected and isolated. The applicability of this approach is demonstrated via simulation examples.

  • Subjects / Keywords
  • Graduation date
    Fall 2013
  • Type of Item
  • Degree
    Doctor of Philosophy
  • DOI
  • License
    This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.