Parallel Modified Artificial Bee Colony Algorithm for Solving Conditional Nonlinear Optimal Perturbation

Intelligent algorithms have been applied to solving conditional nonlinear optimal perturbation (CNOP), which plays an important role in the study of weather and climate predictability. Single particle intelligent optimization algorithms can get similar CNOP to adjoint method, and show higher time efficiency in solving CNOP. However, swarm intelligent optimization algorithms can only get similar CNOP, and still show lower time efficiency than adjoint method. In this paper, we proposed a modified artificial bee colony algorithm (MABC) to solve CNOP, and to accelerate the computation speed, we parallelize the MABC algorithm with MPI technology. In order to demonstrate its validity and efficiency, we apply MABC algorithm to solving CNOP in Zebiak-Cane model, which is a medium-complexity numerical model. The results obtained are compared with those from standard bee colony algorithm (ABC) algorithm, generic algorithm (GA) and adjoint method which is the benchmark. The MABC algorithm can get better results than the standard ABC and GA algorithm in solving CNOP, and can obtain similar results with adjoint method in CNOP magnitude and pattern aspects. The parallel MABC with MPI also shows a higher efficiency than adjoint method. All the experimental results show that it is feasible and efficient to solve CNOP with the proposed parallel modified artificial bee colony algorithm.