Application of Single- and Multi-Objective Evolutionary Algorithms for Optimal Nonlinear Controller Design in Boiler–Turbine System

Abstract

This paper presents an application of evolutionary algorithm techniques for optimal nonlinear controller design in drum-type boiler–turbine system. Designing of controller for third-order boiler–turbine system is always a complicated task due to the presence of highly interactive nonlinearities. The present work is the first one which attempts to design and implement recently developed finite time convergent controller in third-order boiler–turbine dynamics, and it is described as multi-input–multi-output (MIMO) nonlinear system. The present work explores the possibility of application of single-objective and multi-objective evolutionary algorithm techniques toward optimal tuning of finite time convergent controller to achieve the desired performance for third-order boiler–turbine system. The single-objective evolutionary algorithm techniques such as real-coded genetic algorithm (RGA), modified particle swarm optimization (MPSO), differential evolution (DE), and self-adaptive differential evolution (SADE) are implemented with the minimization of integral square error (ISE) as an objective to obtain optimal tuning parameters. Also, the present paper explores the possibility of simultaneous minimization of conflicting objectives such as ISE and computational cost of the proposed controller using multi-objective evolutionary algorithms such as non-dominated sorting genetic algorithm (NSGA) and modified non-dominated sorting genetic algorithm-II (MNSGA-II). The performance of the proposed optimal finite time convergent controller is validated by simulating different kinds of set point changes, and the obtained results are presented as various case studies. The adaptability of the proposed controller during parameter variations is also examined. The performance of the single-objective and multi-objective evolutionary algorithms has been statistically analyzed, and the results are reported. The results reveal that among the four single-objective EA techniques, SADE offers better performance due to its inherit self-adaptive capability. Also, during multi-objective optimization, MNSGA-II has provided better solution due the presence of dynamic crowding distance (DCD) and control elitism (CE) strategies.