Forecasting chaotic time series with genetic algorithms

Physical Review E - Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary Topics

Volumn 55, Issue 3 SUPPL. A, 1997, Pages 2557-2568

  Szpiro, George G ^a,b  

^a Israeli Center for Academic Studies   (Israel)

^b NONE   (Israel)

Author keywords

[No Author keywords available]

Indexed keywords

ASTROPHYSICS; CHAOS THEORY; DIFFERENTIAL EQUATIONS; FORECASTING; POPULATION STATISTICS; STATISTICAL MECHANICS; THEOREM PROVING; TIME SERIES ANALYSIS;

BELOUSOV-ZHABOTINSKII REACTION; DARWINIAN THEORY OF NATURAL SELECTION; HENON ATTRACTOR; MACKEY GLASS PROCESS; ROSSLER ATTRACTOR; SUNSPOTS; SURVIVAL OF THE FITTEST THEORY;

GENETIC ALGORITHMS;

EID: 0031100934     PISSN: 1063651X     EISSN: None     Source Type: Journal    
DOI: 10.1103/physreve.55.2557     Document Type: Article

References (45)

1. P. Grassberger and I. Procaccia, Phys. Rev. Lett. 50, 448 (1983).
2. A. Ben Mizrachi, I. Procaccia, and P. Grassberger, Phys. Rev. A 29, 975 (1984).
3. P. Grassberger, R. Hegger, H. Kantz, C. Schaffrath, and T. Schreiber, Chaos 3, 127 (1993).
4. G. G. Szpiro, Physica D 65, 289 (1993).
5. K. Judd and A. Mees, Physica D 82, 426 (1995).
6. t-λ the lag.
7. J. D. Farmer and J. J. Sidorowich, Phys. Rev. Lett. 59, 845 (1987).
8. R. Doerner, B. Hübinger, and W. Martienssen, Phys. Rev. E 50, 12 (1994).
9. M. Casdagli, Physica D 35, 335 (1989).
10. J. Moody and C. J. Darken, Neural Comput. 1, 281 (1989).
11. A. Lapedes and R. Farber, Los Alamos National Laboratory Technical Report No. LA-UR-97-2662, 1987 (unpublished).
12. A. M. Albano, A. Passamante, T. Hediger, and M. E. Farrell, Physica D 58, 1 (1992).
13. Nonlinear Modeling and Forecasting, edited by M. Casdagli and S. Eubanks, SFI Studies in the Sciences of Complexity (Addison-Wesley, New York, 1992).
14. A. S. Weigend and N. A. Gershenfeld, Time Series Prediction, SFI Studies in the Sciences of Complexity (Addison-Wesley, New York, 1994).
15. U. Parlitz and G. Mayer-Kress, Phys. Rev. E 51, 2709 (1995).
16. J. B. Ramsey and Z. Zhang, in Proceedings of the First International Conference on High Frequency Data in Finance (Olsen and Associates, Zurich, Switzerland, 1995).
17. T. Meyer and N. H. Packard, in Nonlinear Modeling and Forecasting (Ref. [13]), pp. 249-264.
18. Jie Lin, Y. Bartal, and R. E. Uhrig, Nucl. Technol. 111, 46 (1995).
19. Cao Liangyue, Hong Yiguang, Fang Haiping, and He Guowei, Physica D 85, 225 (1995).
20. X. Guan, R. J. Mural, and E. C. Uberbacher, in Proceedings of the Tenth Conference on Artificial Intelligence for Applications (IEEE Computer Society Press, Morgantown, 1994).
21. J. Bauer, Genetic Algorithms and Investment Strategies (Wiley, New York, 1994).
22. E. M. Azoff, Neural Networks Time Series Forecasting of Financial Markets (Wiley, New York, 1994).
23. J. H. Holland, Adaptation in Natural and Artificial Systems, 2nd ed. (University of Michigan Press, Ann Arbor, 1992).
24. D. E. Goldberg, Genetic Algorithms in Search, Optimization and Machine Learning (Addison-Wesley, Reading, MA, 1989).
25. T. Brodmeier and E. Pretsch, J. Comput. Chem. 15, 588 (1994).
26. Y. Zeiri, Phys. Rev. E 51, 2769 (1995).
27. P. Gerstoft, J. Acoust. Soc. Am. 97, 2181 (1995).
28. J. Piper, Patt. Recog. Lett. 16, 857 (1995).
29. R. G. Palmer, W. B. Arthur, J. H. Holland, B. LeBaron, and P. Taylor, Physica D 75, 264 (1994).
30. A. Prügel-Bennett and J. L. Shapiro, Phys. Rev. Lett, 72, 1305 (1994).
31. J. R. Koza, Genetic Programming (MIT Press, Cambridge, 1992).
32. In the language of search and optimization algorithms, mutation ensures that the population of agents does not get stuck on a local maximum in the "landscape" of the search-space.
33. In a manner of speech, "heavy" components of the equation may hide the "light" ones during the first run of the algorithm. Once the effect of the heavy components has been isolated, the genetic algorithm may - in the second stage - discover remaining parts of the formula.
34. O. E. Rössler, Phys. Lett. A 57, 397 (1976).
35. t-z when reproducing equation strings.
36. M. C. Mackey and L. Glass, Science 197, 287 (1977).
37. When referring to one-period forecasts we simply mean the next entry in the time series, and similarly for n-period predictions. The period does not necessarily refer to the characteristic time of the chaotic process.
38. K. Matsumoto and I. Tsuda, in The Theory of Dynamical Systems and its Application to Nonlinear Problems, edited by Hiroshi Kawakami (World Scientific, Singapore, 1984).
39. M. Hénon, Commun. Math. Phys. 50, 69 (1976).
40. The ability to predict such explosions would be a desirable feature of the algorithm. However, a training period could include at most one explosion which may not suffice to train the genetic algorithm for this type of occurrence. Further investigation is needed in this matter.
41. See Neural Networks Time Series Forecasting of Financial Markets (Ref. [22]), pp. 104-105.
42. The 11-year cycle that is generally believed to underlie the sunspot series, is recreated by Eq. (23), even when random initial data are used to produce different realizations of the series. The formulas bred by the algorithm could therefore provide an indication that the 11-cycle may actually be produced by a combination of 3- and 9-cycles.
43. L. Cao, Y. Hong, H. Fang, and G. He, Physica D 85, 225 (1995).
44. T. Subba Rao, in Nonlinear Modeling and Forecasting (Ref. [13]), pp. 199-228. in particular Table 2.
45. A. S. Weigend, B. A. Huberman, and D. E. Rumelhart, in Nonlinear Modeling and Forecasting (Ref. [13]), pp. 395-432.