Note: This content is accessible to all versions of every browser. However, this browser does not seem to support current Web standards, preventing the display of our site's design details.


Application of Model Predictive Control to a Cascade of River Power Plants


C. Setz, A. Heinrich

Diploma/Master Thesis, WS 05/06

Hydroelectric power plants are constructions that are built into the natural course of a river to generate electrical energy. For environmental and navigational reasons, the water flow through their facilities (turbines and weirs) has to be manipulated such that a specified water level upstream each power plant is kept close to a predefined reference value and within prescribed bounds. For the same reasons, the turbine discharge variations shall be small. In order to decelerate turbine wear out, a minimal amount of control moves is desired. The currently employed control concept uses local PI controllers with feed-forward. To reduce the amount of control moves a hysteresis is applied to the output of the controller. The tuning of the PI controllers is demanding and constraints can not be explicitly handled. This may lead to an amplification of the natural discharge fluctuations. In this thesis a supervisory Model Predictive Controller is developed for a cascade of four hydroelectric power plants in a river containing locks. Lock operations induce significant disturbances rendering the control task challenging. A linear discrete-time state space model of the power plant cascade is derived from the Saint Venant equations and used as internal controller model. In order to reduce computation time, balanced model reduction is employed. The states are estimated by a Kalman filter from the four available water level measurements. The control objectives are to minimize the concession level deviations and the turbine discharge variations as well as to reduce the amount of control moves. The first two control objectives are expressed in a quadratic cost function subject to constraints. To account for the third control objective of reducing the amount of control moves, a hysteresis is included in the controller which introduces boolean states. The control problem is expressed in the mixed logical dynamical (MLD) formulation leading to a Mixed Integer Quadratic Program (MIQP). Since implementing the hysteresis for the entire prediction horizon renders the optimization problem too complex to be solved within the sampling time of the particular application, several heuristics are developed and tested for different scenarios in closed-loop simulations with the hydraulics software FLORIS.

Supervisors: P. Rostalski, Dr. G. Papafotiou, Prof. M. Morari


Type of Publication:

(12)Diploma/Master Thesis

M. Morari

No Files for download available.
% Autogenerated BibTeX entry
@PhdThesis { SetHei:2006:IFA_2813
Permanent link