標題: Computational optimization for S-type biological systems: Cockroach genetic algorithm
作者: Wu, Shinq-Jen
Wu, Cheng-Tao
電控工程研究所
Institute of Electrical and Control Engineering
關鍵字: Inverse problem;S-system;Memetic algorithm;Cockroach swarm evolution;Structure identification
公開日期: 1-Oct-2013
摘要: S-type biological systems (S-systems) are demonstrated to be universal approximations of continuous biological systems. S-systems are easy to be generalized to large systems. The systems are identified through data-driven identification techniques (cluster-based algorithms or computational methods). However, S-systems' identification is challenging because multiple attractors exist in such highly nonlinear systems. Moreover, in some biological systems the interactive effect cannot be neglected even the interaction order is small. Therefore, learning should be focused on increasing the gap between the true and redundant interaction. In addition, a wide searching space is necessary because no prior information is provided. The used technologies should have the ability to achieve convergence enhancement and diversity preservation. Cockroaches live in nearly all habitats and survive for more than 300 million years. In this paper, we mimic cockroaches' competitive swarm behavior and integrated it with advanced evolutionary operations. The proposed cockroach genetic algorithm (CGA) possesses strong snatching-food ability to rush forward to a target and high migration ability to escape from local minimum. CGA was tested with three small-scale systems, a twenty-state medium-scale system and a thirty-state large-scale system. A wide search space ([0,100] for rate constants and [-100,100] for kinetic orders) with random or bad initial starts are used to show the high exploration performance. (C) 2013 Elsevier Inc. All rights reserved.
URI: http://dx.doi.org/10.1016/j.mbs.2013.07.019
http://hdl.handle.net/11536/23013
ISSN: 0025-5564
DOI: 10.1016/j.mbs.2013.07.019
期刊: MATHEMATICAL BIOSCIENCES
Volume: 245
Issue: 2
起始頁: 299
結束頁: 313
Appears in Collections:Articles


Files in This Item:

  1. 000325590900018.pdf

If it is a zip file, please download the file and unzip it, then open index.html in a browser to view the full text content.