Multi-sample learning particle swarm optimization with adaptive crossover operation

Abstract

Particle swarm optimization (PSO) is a well-known optimization method used for solving various optimization problems. However, PSO suffers from premature convergence and is ineffective in balancing exploration and exploitation when solving complex optimization problems. To overcome these drawbacks of PSO, a multi-sample learning particle swarm optimization with adaptive crossover operation (MLPSO) is proposed. In MLPSO, two novel strategies, multi-sample selecting strategy (MSS) and adaptive sample crossover strategy (ASC), are used to select proper learning samples for the whole population. Firstly, in MSS, two sample pools, namely elite pool and improver pool, are used to save elites and improvers. Elites are the particles with preferable fitness, while improvers denote the particles whose fitness have been improved largely in recent consecutive generations. In each generation, two particles are randomly selected from the two sample pools respectively to breed a learning sample through crossover operation for the whole population. Therefore, the generated learning sample by MSS strategy contains more diversity information. Secondly, in ASC, various crossover operations are conducted for breeding a learning sample according to the evolutionary states. Therefore, the ASC strategy proposed in this paper can realize a better trade-off between exploration and exploitation. Finally, the performance of MLPSO is evaluated using CEC2013, CEC2017 test suites and three engineering optimization problems. Experimental results show that MLPSO outperforms compared seven competitive PSO variants and 19 meta-heuristics algorithms in most functions.