Adaptive hybrid predictive control for a combined cycle power plant optimization

Abstract

The design and development of an adaptive hybrid predictive controller for the optimization of a real combined cycle power plant (CCPP) are presented. The real plant is modeled as a hybrid system, i.e. logical conditions and dynamic behavior are used in one single modeling framework. Start modes, minimum up/down times and other logical features are represented using mixed integer equations, and dynamic behavior is represented using special linear models: adaptive fuzzy models. This approach allows the tackling of special non-linear characteristics, such as ambient temperature dependence on electrical power production (combined cycle) and gas exhaust temperature (gas turbine) properly to fit into a mixed integer dynamic (MLD) model. After defining the MLD model, an adaptive predictive control strategy is developed in order to economically optimize the operation of a real CCPP of the Central Interconnected System in Chile. The economic results obtained by simulation tests provide a 3% fuel consumption saving compared to conventional strategies at regulatory level. Copyright (c) 2007 John Wiley & Sons, Ltd.