Advertisement

DGOR / ÖGOR pp 95-102 | Cite as

Ein konnexionistischer Ansatz zur Prognose und Steuerung von Ablaufschadstoffen einer Kläranlage

  • Martin Natter
  • Martin Lukanowicz
Conference paper
Part of the Operations Research Proceedings 1992 book series (ORP, volume 1992)

Zusammenfassung

Diese Arbeit beschreibt ein konnexionistisches Modell zur Prognose und Steuerung am Beispiel des Ablaufschadstoffs Nitrat (NO3-N). In einer empirischen Analyse werden Kläranlagenmeßwerte aus einer Betriebsprotokolldatenbank zur Modellschätzung verwendet. Besondere Aufmerksamkeit wird der Modellierung und der Modellselektion gewidmet. Backpropagation, eine Kleinstquadratprozedur, die Gradientenverfahren mit dynamischer Rückkoppelung kombiniert, dient zur Bestimmung der Kantengewichte des Netzwerks. Um die Gefahr der Überschreitung von gesetzlichen Grenzwerten (z.B. Nitrat) zu reduzieren wurden Warn- und Alarmvariable modelliert.

Abstract

This paper presents a connectionist model for forcasting and controlling of nitrate (NO3-N) which is a typical pollutant of a waste water treatment. In an empirical study we use the operating database of a purification plant for estimating the model. We pay special attention to the process of problem representation and model selection. Backpropagation — a least square procedure that combines the method of gradient decent with dynamic feedback — is applied for estimating the model. To prevent violations of legal bounds (e.g. NO3−N < 15mg/l) warning and alarm variables are added to the model.

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

Literatur

  1. [CYBENKO 89]
    Cybenko, G. (1989): Continuous Value Neural Networks with Two Hidden Layers are Sufficient, in: Mathematics of Control, Signal and Systems, pp. 303–314Google Scholar
  2. [HRUSCHKA 92]
    Hruschka, H. (1992): Determining Market Response Functions by Neural Network Modeling. A comparison to Econometric Techniques. Erscheint in European Journal of Operational ResearchGoogle Scholar
  3. [LlTKANOWICZ 92]
    Lukanowicz, M. (1992): Integration eines multivariaten Logitmodells in das Kläranlagenexpertensystem KLEX als induktive Lernkomponente, Operations- Research Proceedings 1991, Springer-Verlag, Berl in HeidelbergGoogle Scholar
  4. [MOODY 92]
    Moody, J.E. (1992): The Effective Number of Parameters: An Analysis of Generalization and Regularization in Nonlinear Learning Systems; Appears in J.E. Moody, S.J. Hanson, and R.P. Lippmann (editors): Advances in Neural Information Processing Systems 4, Morgan Kaufmann Publishers, San Mateo CAGoogle Scholar
  5. [OEWWV 89]
    ÖWWV (1989): Vortragsunterlagen zum Klärwärter Grundkurs des Österreichischen WasserwirtschaftsverbandesGoogle Scholar
  6. [RUMELHART 86]
    Rumelhart. D.E.; McClelland, J. (1986): The PDP Research Group: Parallel Distributed Processing: Explorations in the Microstructure of Cognition; Volume 1: Foundations, MIT Press, Cambridge MAGoogle Scholar
  7. [STONE 74]
    Stone,M. (1974): Cross-Validatory Choice and Assessment of Statistical Predictions; in: Journal of the Royal Statistical Society, Series B, pp. 111–133Google Scholar
  8. [UTANS 91]
    Utans. J.; Moody, J. (1991): Selecting Neural Network Architectures via the Prediction Risk: Application to Corporate Bond Rating Prediction: First International Conference on Artificial Intelligence Applications on Wall Street; IEEE Computer Society Press, Los Alamitos CAGoogle Scholar
  9. [WERBOS 74]
    Werbos, P. (1974): Beyond Regression: New tools for prediction and analysis in the Behavioral Sciences, Ph.D. thesis Harvard UniversityGoogle Scholar
  10. [WHITE 89]
    White, H. (1989): Connectionist Nonparametric Regression: Multilayer Feed-forward Networks Can Learn Arbitrary Mappings, Working Paper, Department of Economics, University of California San DiegoGoogle Scholar

Copyright information

© Springer-Verlag Heidelberg 1993

Authors and Affiliations

  • Martin Natter
    • 1
  • Martin Lukanowicz
    • 1
  1. 1.WienAustria

Personalised recommendations