Fuzzy control for forecasting and pattern recognition in a time series.

摘要

Starting with pioneering work of Lapedes and Farber, the neural networks have been advantageously applied to prediction and modeling of time series, in domains as distinct as chaotic dynamics and corporate bond rating prediction. For most real-world control and signal processing problems, the information concerning design and evaluation can be classified into two kinds: numerical information obtained from sensor measurements, and linguistic information obtained from human experts. Generally, neural control is suited for using numerical data pairs (input-output pairs), whereas fuzzy control is an effective approach to utilizing linguistic rules. When fuzzy rules are generated from numerical data pairs, the two kinds of information are combined into a common framework. As compared to neural networks, the fuzzy controllers can operate in real time; their Teaming process does not require many iterations to converge. For this reason fuzzy controllers deserve their legitimacy in time series forecasting, where the real time detection and identification of trends is sought. The paper describes an object-oriented implementation of the algorithm advanced by Wang and Mendel. Numerical results are presented both for time series with seasonal changes and time series corresponding to chaotic situations, such as encountered in the context of strange attractors. In the latter case, the effect of noise on predictive power of the fuzzy controller are explored. In addition, by introducing a distance between an observed and predicted data, one can apply the results of this study to a pattern recognition of temporal signatures.