Adaptive multistrategy ensemble particle swarm optimization with Signal-to-Noise ratio distance metric

• Introduced the signal-to-noise ratio distance metric in metric learning to the PSO community. • A new adaptive strategy selection framework is proposed, named ESE-SNR. • Nonlinear acceleration coefficient based on singer mapping is used to better balance diversity and convergence. • A global best perturbation mechanism is employed to help the population escape from the local optimum. • The results show the algorithm outperforms or is comparable to other PSO variants and meta-heuristic evolutionary algorithms. This paper proposes an adaptive multistrategy ensemble particle swarm optimization (PSO) with signal-to-noise ratio (SNR) distance metric called AMSEPSO, which aims to solve the problems of a single learning mode of PSO and easy premature convergence when solving complex problems. In AMSEPSO, an evolutionary state estimation (ESE) strategy selection framework is proposed based on the SNR distance metric, named ESE-SNR. The appropriate learning strategy is adaptively selected through the ESE-SNR framework. To balances diversity and convergence better, nonlinear acceleration coefficient based on Singer mapping is adopted. Finally, a global best perturbation mechanism is employed to help the population escape from the local optimum. On the CEC2017 benchmarks, comparison with other advanced PSO variants and meta-heuristic algorithms show that AMSEPSO achieves remarkable performance in solving functions with different characteristics, ranking first in the results. The results show that the ESE-SNR framework can effectively evaluate the search state of the population and can greatly save the computational time of the evaluation. The ESE-SNR framework proposed in this paper provides an innovative idea for the development of multistrategy ensemble learning, and the introduction of metric learning into the PSO community helps further promote the organic integration of machine learning and swarm intelligence.