Um algoritmo genético paralelo autoadaptável misto para a otimização multimodal com e sem restrições

Abstract

Genetic algorithms are search algorithms that use a set of parameters to find the best solution to a given problem. These parameter’s choice is a challenge because different problems can demand different sets. Most of the times, this choice is made using the trial and error method that demands time and may not lead to the best possible setting. Other option is to use the experiment design (factor n k ) that further on limited to the amount of k factors, demands that multiple simulations are done before the definition of the best setting, which is also dependent of the problem that is being solved. In this context, the self-adaptive algorithms arise, allowing the best selection of rates and genetic operators within execution time. Thereby, this work develops a self-adaptive algorithm that in addition of choosing between four different kinds of crossover and four mutation operators also adapts in execution’s time the rates of crossover and mutation. Having overcome the challenge of finding the best configuration, remains increasing the execution time, because as more parameters are adjusted a higher computational power is required. To solve this problem, it is used the parallel computing, which became acessible and popular after the criation and cost reduction of multicore processors. In this work, the parallelism is achieved using the OpenMp, which is an application programming interface (API) that supplies a portable and scalable model that allows programmers to create parallel applications. To evaluate the self-adaptive algorithm performance standard benchmark functions were used, such as Schwefel, Griewank, Rastringin, Rosenbrock, variations of the last two ones (Shifted Rastringin and Shifted Rosenbrock) and a real application called Dynamic Economic Dispacth. The evaluation was made in terms of the solution’s quality using ANOVA and Tukey’s test. In terms of parallelism, the evaluation was done using speedup and eficiency metrics