New nature-inspired metaheuristics applied to the constrained optimization of a heavy-duty gas turbine operation

Abstract

Resumo: Os codigos computacionais complexos das mais diversas areas, tais como industria 4.0 e energia, apresentam caracteristicas como nao-linearidade, escala, multimodalidade e presenca de restricoes. Por este motivo, as tecnicas classicas Newtonianas e baseadas em gradiente nao sao recomendadas para problemas de otimizacao global, os quais contem inumeras variaveis de projeto, restricoes e simulacoes incorporadas. Isso incentivou novas pesquisas em metaheuristicas baseadas em fenomenos naturais, principalmente comportamentos de animais com caracteristicas cooperativas ou colaborativas. Entretanto, nao existe um algoritmo unico capaz de ter bom desempenho para todos os tipos de problemas de otimizacao, o que justifica a busca recorrente por novas abordagens para solucionar esses problemas. Portanto, a presente tese introduz duas metaheuristicas com estruturas inovadoras inspiradas na natureza e nunca propostas. A primeira e baseada na especie Canis latrans e denominada Algoritmo de Otimizacao dos Coiotes (do ingles Coyote Optimization Algorithm, COA). A segunda, por sua vez, e inspirada na especie Cebus capucinus e denominada Otimizador dos Macacos-prego-da-cara-branca (do ingles Whitefaced Capuchin Monkeys Optimizer, WfCMO). Os algoritmos propostos sao avaliados sob um conjunto de funcoes de benchmarks empregadas nas competicoes do Congresso de Computacao Evolutiva (do ingles Congress on Evolutionary Computation, CEC) organizado pelo Instituto de Engenheiros Eletricistas e Eletronicos (do ingles Institute of Electrical and Electronics Engineers, IEEE) e comparadas a outras metaheuristicas inspiradas na natureza. Alem disso, a modelagem de um problema de otimizacao com restricoes de uma turbina a gas do tipo heavy-duty de uma termeletrica brasileira tambem e proposto nesta pesquisa. Para soluciona-lo, uma versao cultural do COA e proposta e seu desempenho e avaliado e comparado com outros algoritmos do estado-da-arte. Os resultados mostram que as metaheuristicas propostos nesta pesquisa alcancaram desempenho satisfatorio e superaram os outros algoritmos com 95% de confianca estatistica com base no teste nao-parametrico deWilcoxon-Mann-Whitney e tambem nos criterios do IEEE CEC 2017. Ainda, os resultados conquistados para problems multimodais e de alta dimensao mostram que as tecnicas sao promissoras para estes tipos de problema, que sao usuais em problemas reais. Ademais, as analises de curva de convergencia e de diversidade da populacao indicam um balanco adequado entre exploracao e aproveitamento. Por fim, a versao cultural do COA, que se demonstrou capaz de evitar convergencia prematura, superou os demais algoritmos do estado-da-arte para o problema de otimizacao da operacao da turbina. Palavras-chave: Industria 4.0, Inteligencia Computacional, Otimizacao Global, Metaheuristicas inspiradas na natureza.

Abstract: The real-world applications from the most diverse fields such as industry 4.0 and energy have been formulated into complex computational codes with features as non-linearity, scale, multimodality, and the presence of constraints. Because of that, the classic Newtonians and gradient-based techniques are not recommended for global optimization applications with many design variables, constraints, and simulations embedded. It has encouraged new researches on metaheuristics based on natural phenomena, mainly animal behaviors with cooperative or collaborative features. However, there is not a unique algorithm able to perform well for all types of optimization problems, which justifies the recurrent search for new approaches. Hence, this thesis presents two never-proposed nature-inspired metaheuristics with innovative structures. The first one is based on the Canis latrans species and it is denoted Coyote Optimization Algorithm (COA). The second one is inspired by the Cebus capucinus species and receives the name of White-faced Capuchin Monkeys Optimizer (WfCMO). The proposed algorithms are evaluated under a set of benchmark functions employed in the Institute of Electrical and Electronics Engineers (IEEE) Congress on Evolutionary Computation (CEC) competitions and compared to other state-of-the-art nature-inspired metaheuristics. Besides, the design of a constrained optimization problem of a heavy-duty gas turbine operation from a Brazilian thermoelectric power plant is proposed in this research. To solve it, a cultural version of the COA is proposed and its performance is evaluated and compared to other state-of-the-art algorithms. The results show that the proposed metaheuristics achieve profitable performance and outperform some state-of-the-art algorithms with 95% of statistical confidence based on the Wilcoxon-Mann- Whitney non-parametric test and the criteria of the IEEE CEC of 2017. Also, these algorithms present promising results for multimodal and high dimensional problems, which are the most usual features of real-world problems. Moreover, the convergence and diversity curves indicate a suitable balance between exploration and exploitation. Further, the proposed cultural version of the COA outperforms other state-of-the-art algorithms for the gas turbine operation problem. Its ability to avoid premature convergence is also demonstrated. Keywords: Industry 4.0, Computational Intelligence, Global Optimization, Nature-Inspired Metaheuristics.