CACO-LD: Parallel Continuous Ant Colony Optimization with Linear Decrease Strategy for Solving CNOP

Abstract

Increasing intelligence algorithms have been applied to solve conditional nonlinear optimal perturbation (CNOP), which is proposed to study the predictability of numerical weather and climate prediction. Currently, swarm intelligence algorithms have much lower stability and efficiency than single individual intelligence algorithms, and the validity of CNOP (in terms of CNOP magnitude and CNOP pattern) obtained by swarm intelligence algorithms is not as good as that obtained by single individual intelligence algorithms. In this paper, we propose an improved parallel swarm intelligence algorithm, continuous ant colony optimization with linear decrease strategy (CACO-LD), to solve CNOP. To verify the validity of the CACO-LD, we apply it to study EI Niño-Southern Oscillation (ENSO) event with Zebiak-Cane (ZC) model. Experimental results show that the CACO-LD can achieve better CNOP magnitude, better CNOP pattern with much higher stability than the modified artificial bee colony algorithm (MABC) and the continuous tabu search algorithm with sine maps and staged strategy (CTS-SS), which respectively are the latest best swarm intelligence algorithm and single individual intelligence algorithm for solving CNOP. Moreover, when using 32 processes, the parallel CACO-LD runs 3.9 times faster than the parallel MABC, and is competitive with the parallel CTS-SS in efficiency.