Multivariable predictive control design with application to a gas turbine power plant

Omar, Abdallah Ali

Thesis

Multivariable predictive control design with application to a gas turbine power plant

Download PDF

Creator

Omar, Abdallah Ali

Rights statement

Strathclyde Thesis Copyright

Awarding institution

University of Strathclyde

Date of award

1997

Thesis identifier

T9179

Qualification Level

Masters (Postgraduate)

Qualification Name

Master of Philosophy (MPhil)

Department, School or Faculty

Abstract

The application of Model-Based Predictive Control (MBPC) to Gas Turbine Power Plants is investigated. This begins with the discussion of the advantages and disadvantages of (MBPC). Three well-known algorithms namely, Model Algorithmic Control (MAC), Dynamic Matrix Control (DMC) and Generalised Predictive Control (GPC) are presented and compared. A multivariable state space GPC algorithm is then presented and discussed. The general optimisation problem is addressed. Four types of constraints namely, constraints on control signal increments, constraints on control signal amplitude, constraints on process outputs and constraints on overshoot are extended to the multivariable state space GPC algorithm and the constrained GPC problem is solved as a Quadratic Programming (QP) optimisation problem. Analysis and design aspects of GPC control strategy are discussed. A method for multivariable state space GPC closed-loop formulation is developed to derive conditions for performance and stability robustness assessment of the control design. Simulation of a GPC controlled multivariable industrial process is presented and the results are analysed. The operation of a single unit gas turbine power plant is discussed and a mathematical model is developed. The nonlinear model is implemented on SIMULINK and stabilised using PID controllers. A linear model is then derived and reduced using MATLAB and SIMULINK toolboxes for control design purposes. A supervisory control system for the nonlinear power plant model is designed using the multivariable GPC algorithm. Closed-loop analysis is carried out and conditions for performance and the stability robustness are derived. Compromise among various design objectives is made for the final controller design. The performance of PID and GPC controllers is compared. A constrained supervisory control system is designed to handle process and actuator constraints. A predictive controller in an adaptive (gain scheduling) manner is then proposed. The technique is applied to optimise the start-up and shut-down operations of the power turbine and to design a constrained adaptive predictive controller. For this purpose an identification experiment is designed and a recently developed subspace-based identification algorithm is applied off-line to identify several linear models.

Advisor / supervisor