PPSO: PCA based particle swarm optimization for solving conditional nonlinear optimal perturbation. (October 2015)

• Record Type: Journal Article
• Authors: Mu, Bin; Wen, Shicheng; Yuan, Shijin; Li, Hongyu
• Is Part Of: Computers & geosciences. Volume 83(2015)
• Page Start: 65
• Page End: 71
• Publication Date: 2015-10
• Subjects: Particle swarm optimization -- PCA -- CNOP -- ZC model; Environmental policy -- Periodicals; 550.5
• DOI: 10.1016/j.cageo.2015.06.016
• Languages: English
• ISSNs: 0098-3004

Abstract

Abstract: Conditional nonlinear optimal perturbation (CNOP) 1 has been widely applied to predictability and sensitivity studies of nonlinear models in meteorology and oceanography. The popular solution of CNOP is based on adjoint models, which is also treated as the benchmark. However, the development of adjoint models is time-consuming, especially for the large-scale air-sea coupled model. Intelligent algorithms are another solution of CNOP, but they are just confined to some simple ideal models due to dimensionality. To avoid adjoint models, this paper proposes a principal component analysis (PCA) based particle swarm optimization method to solve the CNOP of complicated practical models. Through dimension reduction, the original problem can be mapped into a low space so that the particle swarm optimization method can search results in it. To demonstrate the validity, the proposed method is applied to the Zebiak–Cane model and compared with the adjoint based method. Experimental results show that the proposed method can approximately solve the CNOP of complicated models without the need of computing adjoint models, and achieve similar results with the adjoint based method. Highlights: We propose a PCA based PSO method to solve CNOP of complicated practical models. The proposed method is free of adjoint models. We improve the PSO algorithm to find the globally optimal position. The proposed method can achieve similar results with the adjoint based method. The proposed method can be transferred to other complicated models easily.