Chaotic analysis of seismic time series and short term
Chaotic analysis of seismic time series and short term
forecasting using neural networks.
V.P. Plagianakos and E. Tzanaki
Department of Mathematics,
University of Patras,
University of Patras Artificial Intelligence Research Center (UPAIRC).
GR-261.10 Patras, Greece.
e-mail:{vpp,helen}@math.upatras.gr
Abstract
In this study, a chaotic analysis approach was appliedto a time series composed of seismic events occurred inGreece. The dynamics of the earthquakes belong to thecategory of dissipative systems, which exhibit chaoticbehavior. After the chaotic analysis, short termforecasting using an Artificial Neural Network has beenperformed. Neural networks, under appropriateconditions, are known to be universal functionapproximators, thus they have been investigated as toolsfor time series forecasting. Here, a neural network istrained to make short term earthquake predictions. Thenetwork architecture has been dictated by thecalculated characteristics of the time series itself.Preliminary results indicate that this is a promisingapproach.
the predictions made by mathematical models tomeasurements made in the field. The ultimate goal fortime-series analysis might be to construct a computerprogram that, without any knowledge of the physicalsystem from which the data come, can take themeasured data as input and provide as output amathematical model describing the data [5].
The rest of the paper is organized as follows: the nextSection, describes the fundamentals of chaotic timeseries analysis theory. In Section 3 the proposedmethodology is introduced, while in Section 4, theneural network architecture employed is highlighted.Section 5 concludes the paper.
2 Chaotic Time Series Analysis
When dealing with data, we need to introduce two newconcepts: measurement and noise. In conducting anexperiment we can measure only a limited set ofquantities and we are able to make those measurementswith limited precision. Measurement noise refers tofluctuations in measurements that arise from chance.Dynamical noise is another important source of noise indata. Real-world systems do not exist in isolation; theyare affected by outside influences [5].
Since there is neither dynamical nor measurement noise,the model is completely deterministic. This means that,in principle, if we have the initial condition we canaccurately calculate the future values. One the otherhand, if the model is chaotic there may be practicallimitations on our ability to predict it. Most techniquesfor nonlinear data analysis involve two steps:
1. In the first step, the data are used to reconstruct the
dynamics of the system.
2. The second step involves characterization of the
reconstructed dynamics.More generally (for the first step), the time series isembedded in an m-dimensional space, by taking mconsecutive data points,x(ti) to define the reconstructedtime series:
1 Introduction
One of the central problems of science is forecasting.Given the past, how can we predict the future. If wehave a system that doesn't change with time then it isvery easy to predict the next value. We take the lastobservation as a forecast for the next one. If the systemchanges periodically with time all we have to do is toobserve a full cycle. On the other it is very difficult tomake a prediction for random numbers. The systemsthat we are interested in are something in between. Theyare not periodic but they contain some kind of structure,which can be exploited to obtain better predictions. Ifthe predictions are perfect, then the system iscompletely deterministic. If the predictions are good,but not perfect, then the system has a deterministiccomponent. If the predictions are terrible, then thesystem is not deterministic at all [4].
Starting with a sequence of measurements, x(ti), whichformulate the time series, our purpose is to determinewhat the data themselves can reveal about the dynamics.In particular, we will discuss some tools from timeseries analysis that can sometimes be used to suggestwhat types of equations are appropriate, or to compare
Lauder Department of Mathematics, Royal Holloway, University of London, Egham, Surrey, TW20 0EX, United Kingdom September 29, 1997 The author is an ...
DIMACS Technical Report 97-29 July 1997 The Chromatic Numbers of Random Hypergraphs by Michael Krivelevich 1 Department of Mathematics, Raymond and Beverly ...
Reme Department of Mathematics, University of Bergen M. Espedal Department of Mathematics, University of Bergen G. A. ye Department of Mathematics, ...
Department of Mathematics, We provide some new necessary and sucient conditions which guarantee arbitrary pole placement of a particular linear system over ...
Absolutely continuous invariant measures for piecewise real-analytic expanding maps on the plane Masato TSUJII Department of Mathematics, Hokkaido University ...
FOUR SHORT QUESTIONS ON BIRKHOFF POLYTOPE Igor Pak Department of Mathematics, Yale University, New Haven, CT 06520 paki@math.yale.edu January 8, 1999 ...
Large and Moderate Deviations for the Local Time of a Recurrent Markov Chain on Z 2 Z N. Gantert Department of Mathematics, TU Berlin, Strasse des ...
Optimal realizations of generic 5-point metrics Jack Koolen Department of Mathematics, POSTECH, Pohang, South Korea. email: koolen@postech.ac.kr Alice ...
We construct numerous examples of the above types such that隐藏>> NONEXCELLENCE OF CERTAIN FIELD EXTENSIONS A.S. Sivatski Department of Mathematics-2 St....
ON SOME ELEMENTS OF THE BRAUER GROUP OF A CONIC A.S. Sivatski Department of Mathematics-2 St.Petersburg Electrotechnical University 197376, St.Petersburg...
热门文档