Semiqualitative Temporal Patterns in Time-Series Databases

- 1 1 1' 1 Martín. M. Martín and N. Piera. Infonnation Processing and Management of Uncertainty in .Knowledge Based Systems. Vol. l, pp. 250252 (extended abstract), 1990. [ Aguilar 96] Data Based lnductive Knowledge Representation, J. Aguilar, J. C. Aguado, N. Piera. pp. 1-6, proceedings of the International Panel Conference on Soft and lntelligent Computing, edited by the Tecbnical University ofBudapest, 1996. [Broomhead 88] Multivariable Functional lnterpolation ami Adaptive Network. Broomhead, D.S., and D. Lowe. Complex Systems, 2, 321-355, 1988. [Costes 84] Application de la classification automatique avec apprentissage a la surveillance d'une centrale solaire. V. Costes-Albrespique, J. Aguilar-Martín. RAlR.O Automatique/Systems Analysis and Control, Vol. 18, nº 4, 1984. [Desroches 87] Syclaire: Systeme de classification avec apprentissage et recoinassaince de formes. Manuel d'utilisation. P. Desroches ... Research report" 87/9 del Centre d'estudis Avan�ts de Blanes. November 1987. [Desroches 90] Variation points in incremental conceptual clustering. P. Desroches, N. Piera and J. Aguilar-Martín. "Technical report" of the LAAS nº 90254. July 1990. [Mangasarian 90JCancer diagnosis via linear programming, O. L. Mangasarian and W. H. Wolberg. SIAM News, Volume 23, Number 5,pp 1 -18, September 1990. [Musavi 92) 0n the training of radial basis function classifiers. Musavi, M.T., W. Ahmed, K.H.Chan, K.B.Faris, and D.M.Hummels. Neural Networks, 5(4), 595-603, 1992. [Prechelt 94) PROBENJ -Set of Neural Network Benchmark Problems and Benchmarlcing Rules. L. Prechelt. Tecbnical Report 21194, Department of Computer Science, University of Karlsruhe, 1994. [Piera 87] Connectius de logiques no estandard com a operadors d' agregació en classificació multivariable i reconeixement de fonnes. N. Piera. Doctorate dissertation in the Universitat Politecnica de Catalunya,1987. [Piera 88) Mixed Connectives of Lineal Compensation. N. Piera and J. Aguilar-Martín. "Technical report" of the LAAS nº 88301, 1988. [Piera 89) IAMDA: An Incremental Conceptual Clustering Metlwd. N. Piera, P. Desroches and J. Aguilar-Martín. ''Technical report" of the LAAS nº 89420,1989. [Piera 90) Variation points in p,attem recognition. N. Piera, P. Desrocbes �md J. Aguilar-Martín. Pattem Recognition Letters. Vol. 11 pp. 519-524, 1990. [Piera 91) Controlling Selectivity in Nonstandard Pattem Recognition Algo1ithms. N. Piera � J. Aguilar-Martín. IEEE Transactions on S, M & C. Vol. 21, nº 1, pp. 71-82,1991. [Poggio 90) Networks for Approximatio1z and Leaming. Poggio, T. and F.Girosi. Proceedings ofthe IEEE, 78, 1481-1497, 1990. [Sklar 77) Random variables, joint distributions and copulas. A Sklar. Kybemetica Vol. 9 pp. 449-460, 1977. [Schweizer 83) Probabilistic metric s¡xrces. B. Schweiz.er, A Sklar. Elsevier North-Holland, 1983. [Vrckonovik 90) Radial basis function classification of impulse radar waveforms Vrckovnik, G., C.R. Carter and S. Haykin .. Proceedings of the International Joint Conference on Neural Netwmks, l, 45-50, 1990. [Z.adeh 65) Fuzzy sets. Information and Control. Vol. 8, pp. 338-353, 1965. [Zadeh 71) Fuzzy sets as a base for a theory of possibility. Fuzzy Sets and Systems. Vol. 1 (1), pp. 3-28, 1971. 10 Semiqualitative Temporal Patterns in Time-Series Databases J.A. Ortega, R.M. Gasea, M. Toro, F.J. Galán and J.M. Cañete Departamento de Lenguajes y Sistemas Informáticos. University of Sevilla Avda. Reina Mercedes s/n. Sevilla (Spain) { ortega, gasca,mtoro,galanm,canete }©lsi.us.es Abstract A way to. obt� behaviour pattems of semiqualitative models of dynamic systems automatically is Posed m this pa�er. The temporal evolution of these models is stored into a database. This is a time d�tabase. This database may be obtained by means of sensor data or by means of semiqualitative mula�1ons. In any _way, the data� con� the values of state variables and parameters. Searching tdmilar �tterns ":1 �uch database IS essential, because it help.5 in predictions, hypothesis testing and n eral, m data mmmg and rule discovery. ' llnguage to carry out qu�es abo�t the qualitative and temporal properties of this time-series database propo�. T�e language IS � mtended to classify the different qualitative behaviours of a model. lt claastficat1?n may be carn� o�t �cording with a specific criterion or automatically by means of ring_algonthms. The senuqualitat1ve behaviour of a system is expressed by means of hierarchical ul obtaíned by means of machine leaming algorithms. ' 1 m thodology is applied to a logistics growth model with a delay. Introduction systems studied in science and engineer1 lt le difficult to find mathematical models present them in an appropriate way. The lt Id llng techniques should obviate certain as1 C tbe system. The simulation of these 11 helps us to study the evolution of the iy tem. A way to carry out these simu03 Is described in [11] in depth. However, n t always possible to obtain a mathema m del of a system. Thus, it is necessary 1 pply other techniques in order to carry out •udy. A possibility may be placing sensors In h r al system. The analysis of these data ltlw1 to study the system evolution. ti �b other hand, knowledge about dynamic systems may be quantitative, qualitative, and semiqualitative. When these models are stud ied all this knowledge should be taken into ac count. Different levels of numeric abstraction hav� � <:°nsidered: purely qualitative [8], serruqualitat1ve [6] [10], and quantitative. In this paper, a technique to carry out the analysis of dynamic systems with qualitative and quantitative knowledge is proposed. Toe idea follows: the quantitative behaviours o/ a real system are stored into a database and tech niques o/ Knowledge Discovery in Databases (KDD) are applied to study the system. The way to obtain the behaviours does not matter: by means of the simulation of a model or by means of the data sensors. The tenn KDD is used to refer to the over11