Proposal of a Memory-Based Ensemble Particle Swarm Optimizer

Besides classical issues such as scheduling and traveling salesman, optimization problems can be found in several research areas: process industries, agriculture, electric power systems, and medical engineering. In those scenarios, the focus is to find the best possible solutions to a computational problem. To do this search, estimating an objective function’s minimum or maximum points is necessary. Depending on the function, assigning a specific algorithm to obtain the best solutions for all optimization problems is difficult. One option to solve this is to select dynamically a set of algorithms based on particle swarm optimization (PSO) according to the types of problems. In many cases, PSO approaches have a simpler implementation than genetic algorithms. However, they do not store the optimal solutions during the evolution of the particles. This fragility can cause the loss of good generations during the evolution of the particles. To solve this weakness, the article presents a PSO approach based on a sliding memory that stores these generations and applies them in the dynamic selection of algorithms, making a choice even more efficient. We compare the proposal with other particle swarm techniques using the CEC2017 benchmark of 29 optimization problems to evaluate that. Numerical results show that the memory-based approach performs best in approximately \(90\%\) of the problems, and the runtime is about \(23\%\) smaller on average than other optimizers used in the tests.