Paralleled Continuous Tabu Search Algorithm with Sine Maps and Staged Strategy for Solving CNOP

Abstract

Intelligent algorithms have been extensively applied in scientific computing. Recently, some researchers apply variable intelligent algorithms to solve conditional nonlinear optimal perturbation (CNOP) which is proposed to study the predictability of numerical weather and climate prediction. Among all the methods that have been studied, the principal components-based great deluge method (PCGD) showed remarkable effect and achieved the best result from the perspectives of CNOP magnitudes and patterns and efficiency. However, compared with adjoint-based method which is referred to as a benchmark, PCGD gets the smaller CNOP magnitude and cannot always get stable solutions. This paper proposes continuous tabu search algorithm with sine map and staged strategy (CTS-SS) to solve CNOP, then parallels CTS-SS with MPI. Based on continuous tabu search algorithm, CTS-SS uses sine chaotic maps to generate the initial candidates to avoid trapping in local optimum and then uses staged search strategy to accelerate the solving speed. To demonstrate the validity of CTS-SS, we take Zebiak-Cane model as a case to compare CTS-SS with the adjoint-based method and PCGD. Experimental results show that CTS-SS can efficiently obtain a satisfactory CNOP magnitude which is more close to the one computed with the adjoint-based method and larger than PCGD. Besides, CTS-SS can get more stable result than PCGD. In Addition, CTS-SS consumes similar time to PCGD and the adjoint-based method with 15 initial guess fields.