Intelligent multi-objective model predictive control applied to steam turbine start-up

Abstract

This study proposes an intelligent multi-objective model predictive control method in which an artificial neural network and a genetic algorithm are used to realize satisficing decision-making, which is an interactive multi-objective programming technique. We considered model predictive optimization under a dynamic environment with multiple objectives. To predict nonlinear function forms with dynamic plant characteristics, we applied a recurrent radial basis function network, which is a type of artificial neural network. For optimization with multiple objectives, we applied a satisficing trade-off method along with metaheuristic optimization in the form of genetic algorithms. The features of this control method are as follows. (1) Several conflicting control objectives can be optimized in online control based on multi-objective evaluation through human-computer interaction and (2) an optimal and flexible plant control can be performed within a restrained practical computing time for real-time applications, with acceptable control quality using online adaptive model prediction. This study demonstrates the success of model prediction using computational intelligence combined with an interactive optimization technique for multi-objective model predictive control problems by applying the proposed method to steam turbine start-up control with multiple objectives consisting of the start-up time and rotor thermal stress of the steam turbine. The dynamic simulation results showed an effective control performance within a reasonable computing time.

Keywords

• genetic algorithm
• model predictive control
• multi-objective optimization
• power plant
• radial basis function network
• recurrent neural network
• satisficing trade-off method
• steam turbine