coupling neural networks and fuzzy logic for forecasting purposes

From: Giorgio Corani (corani@elet.polimi.it)
Date: Tue Nov 27 2001 - 12:42:12 MET

  • Next message: S. F. Thomas: "Re: Stupid question"

    If we think about a forecast issued by a neural network, we can argue
    that it doesn't know about the state of the dynamic system: it get the
    values of input to the system and returns the predicted output of the
    system. However, there are applications where the state of the system
    has a key role. I'm dealing with flood forecasting: the inputs are past
    measurements of rainfall and water level (this is the autoregressive
    part of the input). In this case, the increase in water level caused by
    an identical rainfall event depends heavily on the state of the basin.
    Roughly speaking: an identical rainfall event causes a small water
    level increase if the soil is dry, and a very larger increase is the
    soil is wet and does not infiltrate water.

    The problem i would to address is that my knowledge about the state of
    the system is quite fuzzy (i can say for example the basin is more or
    less wet).
    My idea is to configure different neural predictors (i.e. for dry,
    medium and wet soil conditions), and then in real time operations, to
    switch in a fuzzy way from each to other.
    However, it implies several problems.
    For example every data of the training set belongs, with a different
    rate of membership, to the dry, to the medium and to the wet situation.
    How is then possible to calibrate the model?
    Maybe it can be convenient to include all the data with membership
    different from 0, and then, when evaluating the objective function, to
    weight the related errors on the base of their membership.

    I apologize for this long posting.
    I welcome suggestion of papers which deal with such problems.

    best regards, Giorgio Corani

    ############################################################################
    This message was posted through the fuzzy mailing list.
    (1) To subscribe to this mailing list, send a message body of
    "SUB FUZZY-MAIL myFirstName mySurname" to listproc@dbai.tuwien.ac.at
    (2) To unsubscribe from this mailing list, send a message body of
    "UNSUB FUZZY-MAIL" or "UNSUB FUZZY-MAIL yoursubscription@email.address.com"
    to listproc@dbai.tuwien.ac.at
    (3) To reach the human who maintains the list, send mail to
    fuzzy-owner@dbai.tuwien.ac.at
    (4) WWW access and other information on Fuzzy Sets and Logic see
    http://www.dbai.tuwien.ac.at/ftp/mlowner/fuzzy-mail.info
    (5) WWW archive: http://www.dbai.tuwien.ac.at/marchives/fuzzy-mail/index.html



    This archive was generated by hypermail 2b30 : Tue Nov 27 2001 - 12:44:41 MET