Adaptive heterogeneous comprehensive learning particle swarm optimization with history information and dimensional mutation

In many multimedia systems, optimization problems tend to be multimodal, complex and high-dimensional. Although particle swarm optimization (PSO) algorithm has excellent performance in solving optimization problems, how to avoid premature convergence in complex and multimodal situations is the problem that need to be solved urgently. To overcome this problem, an adaptive heterogeneous comprehensive learning particle swarm optimization with history information and dimensional mutation (AHPSO) is proposed in this paper. In order to keep the population diversity, the whole population is divided into two subpopulations and particles' information and knowledge are mined to provide adaptive strategy in both subpopulations. In exploitation subpopulation, an adaptive inertia weight (AIW) method is proposed according to the particles' historical information. In exploration subpopulation, adaptive dimension mutation strategy (ADM) is introduced to improve the ability of the method to solve multimodal and complex problems in multimedia systems. Meanwhile, in order to increase particle diversity, dynamic-opposite learning (DOL) is used in exploration subpopulation. The exploration subpopulation does not learn from any particles in the exploitation subpopulation, so the information passing between subpopulations is one-way. The diversity in the exploration subpopulation can be maintained even if the exploitation subpopulation converges prematurely. In CEC 2013 test suite, in terms of Friedman test result, compared with traditional two swarm method, the solution accuracy of the proposed AHPSO in this paper is improved by 22.4 percentage points. The performance of AHPSO is compared with 8 peer variants and 8 other evolutionary algorithms on CEC2013 and CEC2017 test suites. Experimental results verify that AHPSO has a remarkable performance in complex and multimodal conditions.