Adaptive comprehensive learning particle swarm optimization with cooperative archive

Abstract

Comprehensive learning particle swarm optimization (CLPSO) enhances its exploration capability by exploiting all other particles’ historical information to update each particle’s velocity. However, CLPSO adopts a set of fixed comprehensive learning (CL) probabilities to learn from other particles, which may impair its performance on complex optimization problems. To improve the performance and adaptability of CLPSO, an adaptive mechanism for adjusting CL probability and a cooperative archive (CA) are combined with CLPSO, and the resultant algorithm is referred to as adaptive comprehensive learning particle swarm optimization with cooperative archive (ACLPSO-CA). The adaptive mechanism dividing the CL probability into three levels and adjusting the individual particle’s CL probability level dynamically according to the performance of the particles during the optimization process. The cooperative archive is employed to provide additional promising information for ACLPO-CA and itself is updated by the cooperative operation of the current swarm and archive. To evaluate the performance of ACLPSO-CA, ACLPSO-CA is tested on CEC2013 test suite and CEC2017 test suite and compared with seven popular PSO variants. The test results show that ACLPSO-CA outperforms other comparative PSO variants on the two CEC test suites. ACLPSO-CA achieves high performance on different types of benchmark functions and exhibits high adaptability as well. In the end, ACLPSO-CA is further applied to a radar system design problem to demonstrate its potential in real-life optimization.