Wednesday , June 20 2018

Fuzzy Time Series Estimation and Prediction: Criticism, Suitable New Methods and Experimental Evidence

Vasile GEORGESCU
University of Craiova
13 A.I.Cuza, Craiova, 200585, Romania

Abstract: This paper is devoted to exploring suitable methods for modelling, estimating and forecasting fuzzy time series, when facing the problem of non-invertibility of the standard Minkovsky addition and multiplication in a fuzzy framework. Some generalized versions of Hukuhara difference, which allow the fuzzy estimation problem to be handled in some L2-type metric space, are first examined from a critical viewpoint. This leads us to propose a new estimation procedure, where the monolithic fuzzy model is broken in several more tractable crisp estimation sub-problems, based upon a partial decoupling principle. Our aim is to produce fuzzy estimations with non-negative spreads, capable not only to help decomposing, but also to make the process invertible, by recomposing a non-stationary fuzzy time series from its components, such as trend, cycle, seasonality and the simulated residuals, all of them properly defined as LR-fuzzy sets. Computational Intelligence techniques such as wavelet decomposition and de-noising or nonlinear model fitting with wavelet networks are also addressed. Finally, the proposed methods are exemplified for a fuzzy time series with fuzzy daily temperatures (minimum, average and maximum values).

Keywords: Fuzzy time series estimation and prediction, Generalized Hukuhara difference, Projection cones vs. projection subspaces, Wavelet decomposition and de-noising, Nonlinear fitting with wavelet networks.

>>Full text
CITE THIS PAPER AS:
Vasile, GEORGESCU, Fuzzy Time Series Estimation and Prediction: Criticism, Suitable New Methods and Experimental Evidence, Studies in Informatics and Control, ISSN 1220-1766, vol. 19 (3), pp. 229-242, 2010.

1. Introduction

The study of fuzzy time series has attracted great interest and is expected to expand rapidly. Fuzzy time series have an inherent fuzzy and random nature.

We consider an extension of the probability space image002-2010,3,3 by the dimension of fuzziness, i.e., by introducing a membership scale. This enables the consideration of imprecise observations as fuzzy realizations image004-2010,3,3 of each elementary event image006-2010,3,3. We will restrict attention to the class image008-2010,3,3 of normal convex fuzzy sets on image010-2010,3,3, whose image012-2010,3,3-level sets are in the class image014-2010,3,3 of nonempty compact real intervals.

A fuzzy random variable image016-2010,3,3 is the fuzzy result of the uncertain mapping image018-2010,3,3, such that for each image020-2010,3,3 and image006-2010,3,3, the image012-2010,3,3-level intervals image025-2010,3,3, generated by the mapping image027-2010,3,3, are random sets. In other words, image029-2010,3,3 are Borel-measurable w.r.t. the Borel image031-2010,3,3-field generated by the topology associated with a suitable metric on image014-2010,3,3, usually the Hausdorff metric image034-2010,3,3.

A fuzzy random process image036-2010,3,3 is defined as a family of fuzzy random variables image038-2010,3,3 over the space image040-2010,3,3 of the time coordinate image042-2010,3,3. A fuzzy time series image044-2010,3,3 is a realization of a fuzzy random process image036-2010,3,3image046-2010,3,3 and consists of a temporally ordered sequence of fuzzy variables image046-2010,3,3, each one assigned to each discrete observation time.

For extending both the classical estimation theory and some computational intelligence techniques like wavelet analysis and wavelet networks to time series models with fuzzy data, appropriate assumptions should be stated and suitable methods should be developed.

Square-integrable random variables are assumed, defined on a Hilbert space equipped with a suitable image048-2010,3,3-metric that allows the projection theorem to be still valid. However, it cannot be properly applied as usually onto a subspace, but rather onto cones (i.e., subject to some constraints), due to the lack of a general additive inverse in the space of fuzzy variables, which is only a semi-linear space.

This may lead to distorted results such as obtaining fuzzy least squares estimates with negative spreads. Using Hukukara difference instead of fuzzy subtraction has been proposed to overcome the problem. Unfortunately, it does not always exist, and even if it exists, some distortions may still appear when applying least square estimation.

A criticism of the existing fuzzy estimation methods in the literature is first addressed and suitable new methods are then proposed, based upon a partial decoupling principle. It allows decomposing the monolithic fuzzy model into several crisp models, starting from that one corresponding to modal values (image050-2010,3,3) in fuzzy data, and then proceeding in a decremental way for left and right image012-2010,3,3-level bounds, with image012-2010,3,3 progressively decreasing towards image054-2010,3,3. The estimates of modal values are not subject to any constraints, thus being obtained by applying the Hilbert space projection theorem directly onto the corresponding subspace. However, the estimates for the left and right image012-2010,3,3-level bounds can only be obtained by applying the projection theorem onto cones, in such a way to obtain least squares estimates without negative spreads. This leads to constrained quadratic programs, conveniently defined.

As an alternative to fuzzy estimation methods, computational intelligence techniques, based on wavelet decomposition and wavelet networks for nonlinear model fitting have been proposed to address fuzzy time series estimation and prediction.

References:

  1. DIAMOND, P., Fuzzy Least Squares, Inform. Sci. 46, 1988, p. 141-157.
  2. DIAMOND, P., P. KLOEDEN, Metric Spaces of Fuzzy Sets: Theory and Applications, World Scientific, Singapore, 1994.
  3. DIAMOND, P., R. KÖRNER, Extended Fuzzy Linear Models and Least Squares Estimates, Comput. Math. Appl. 33, 1997, pp. 15-32.
  4. DONOHO, D., Nonlinear Wavelet Methods for Recovery of Signals, Densities, and Spectra from Indirect and Noisy Data. In Different Perspectives on Wavelets, I. Daubechies (eds). Amer. Math. Soc., Providence, R.I., 1993, pp. 173-205.
  5. EL ABIDINE SKHIRI, M. Z., CHTOUROU, M., Synthesis of Denoising Wavelet Neural Networks, Studies in Informatics and Control, 17/4, 2008
  6. GEORGESCU, V., Estimation of Fuzzy Regression Models using Quadratic Programming, Economic Computation and Economic Cybernetics Studies and Research, 4, 1997, pp. 105-124.
  7. GEORGESCU, V., New Estimation Methods in Fuzzy Regression Analysis, Based on Projection Theorem and Decoupling Principle, Fuzzy Economic Review, III/1, 1998, pp. 21-38.
  8. GEORGESCU, V., Fuzzy Control Applied to Economic Stabilization Policies, Studies in Informatics and Control, 10/1, 2001, pp. 37-60
  9. GEORGESCU, V., Multivariate Fuzzy-termed Data Analysis: Issues and Methods, Fuzzy Economic Review, VI/.1, 2001, pp. 19-48.
  10. GEORGESCU, V., On the Foundations of Granular Computing Paradigm, Fuzzy Economic Review, VIII/2, 2003, pp. 73-105.
  11. GEORGESCU, V., A Generalization of Symbolic Data Analysis Allowing the Processing of Fuzzy Granules, Lecture Notes in Artificial Intelligence, 3131, Springer 2004, pp. 215-226
  12. GEORGESCU, V., Granular vs. Point-Wise Metrics and Statistics to Accommodate Machine Learning Algorithms in Granular Feature Spaces, Fuzzy Economic Review, XII/2, 2007, pp. 45-74.
  13. GEORGESCU, V., A Time Series Knowledge Mining Framework Exploiting the Synergy between Subsequence Clustering and Predictive Markovian Models, Fuzzy Economic Review, XIV/1, 2009, pp. 41-66
  14. HUKUHARA, M., Integration des applications measurables dont la valeur est un compact convexe, Funkcialaj Ekvacioj 10, 1967, pp. 205-223.
  15. KÖRNER, R., W. NÄTHER, Linear Regression with Random Fuzzy Variables: Extended Classical Estimates, Best Linear Estimates, Least Squares Estimates, Inform. Sci. 109, 1998, pp. 95-118.
  16. MALLAT, S. G., A Wavelet Tour of Signal Processing: The Sparse Way, Academic Press, 3rd Edition, 2009.
  17. MÖLLER, B., U. REUTER, Uncertainty Forecasting in Engineering, Springer-Verlag Berlin Heidelberg, 2007.
  18. PURI, M.L., D.A. RALESCU, Fuzzy random variables, J. Math. Anal. Appl. 114, 1986, pp. 409-422.
  19. RĂDULESCU F., Cyclic Hilbert Spaces, Studies in Informatics and Control, 18/1, 2009
  20. STEFANINI, L., A Generalization of Hukuhara Difference and Division for Interval and Fuzzy Arithmetic, Fuzzy Sets and Systems (2009), doi: 10.1016/j.fss.2009.06.009.

https://doi.org/10.24846/v19i3y201003