Multi-objective evolutionary algorithm for the optimization of noisy combustion problems


D. Bueche, P. Stoll, R. Dornberger, P. Koumoutsakos, IEEE Transactions on Systems, Man and Cybernetics, Part C, 32(4), 460-473, 2002



Evolutionary Algorithms have been applied to single and multiple bjectives optimization problems, with a strong emphasis on problems, olved through numerical simulations. However in several engineering roblems, there is limited availability of suitable models and there is need or optimization of realistic or experimental configurations. The multiobjective ptimization of an experimental set-up is addressed in this work. xperimental setups present a number of challenges to any optimization technique including: availability only of pointwise information, experimental oise in the objective function, uncontrolled changing of environmental onditions and measurement failure. his work introduces a multi-objective evolutionary algorithm capable f handling noisy problems with a particular emphasis on robustness gainst unexpected measurements (outliers). The algorithm is based on the trength Pareto Evolutionary Algorithm (SPEA) of Zitzler and Thiele and ncludes the new concepts of domination dependent lifetime, re-evaluation f solutions and modifications in the update of the archive population. Several ests on prototypical functions underline the improvements in convergence peed and robustness of the extended algorithm. he proposed algorithm is implemented to the Pareto optimization of the ombustion process of a stationary gas turbine in an industrial setup. The areto front is constructed for the objectives of minimization of NOx emissions nd reduction of the pressure fluctuations (pulsation) of the flame. oth objectives are conflicting affecting the environment and the lifetime
of the turbine, respectively. The optimization leads a Pareto front corresponding o reduced emissions and pulsation of the burner. The physical mplications of the solutions are discussed and the algorithm is evaluated