A neighbor-based learning particle swarm optimizer with short-term and long-term memory for dynamic optimization problems

This paper presents a novel Particle Swarm Optimization algorithm to address Dynamic Optimization Problems. The algorithm incorporates a neighbor-based learning strategy into the velocity update of Particle Swarm Optimization, in order to enhance the exploration and exploitation capabilities of particles. Unlike the traditional swarm update scheme, a “worst replacement” strategy is used to update the swarm, whereby the position of the worst particle in the swarm is replaced by a better newly generated position. The short-term memory is employed to store solutions with intermediate fitnesses from the most recent environment, and the long-term memory is to store the historical best solutions found in all previous environments. After an environmental change is detected, some particles’ positions in the swarm are replaced by the members of the short-term memory, and the best member in the long-term memory under the current environment is re-introduced to the active swarm along with its Gaussian neighborhood, then the remaining particles’ positions are re-initialized. The performance of the proposed algorithm is compared with six state-of-the-art dynamic algorithms over the Moving Peaks Benchmark problems and Dynamic Rotation Peak Benchmark Generator. Experimental results indicate that out algorithm obtains superior performance compared with the competitors.