![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Assessment of a fuzzy based flood forecasting system optimized by simulated annealing |
VerfasserIn |
Aida Reyhani Masouleh, Sabine Pakosch, Markus Disse |
Konferenz |
EGU General Assembly 2010
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 12 (2010) |
Datensatznummer |
250032222
|
|
|
|
Zusammenfassung |
Flood forecasting is an important tool to mitigate harmful effects of floods. Among the many
different approaches for forecasting, Fuzzy Logic (FL) is one that has been increasingly
applied over the last decade. This method is principally based on the linguistic description of
Rule Systems (RS). A RS is a specific combination of membership functions of input and
output variables. Setting up the RS can be implemented either automatically or
manually, the choice of which can strongly influence the resulting rule systems. It is
therefore the objective of this study to assess the influence that the parameters of an
automated rule generation based on Simulated Annealing (SA) have on the resulting
RS.
The study area is the upper Main River area, located in the northern part of Bavaria,
Germany. The data of Mainleus gauge with area of 1165 km2 was investigated in the whole
period of 1984 and 2004. The highest observed discharge of 357 m3/s was recorded in 1995.
The input arguments of the FL model were daily precipitation, forecasted precipitation,
antecedent precipitation index, temperature and melting rate. The FL model of this study has
one output variable, daily discharge and was independently set up for three different forecast
lead times, namely one-, two- and three-days ahead. In total, each RS comprised 55 rules and
all input and output variables were represented by five sets of trapezoidal and triangular fuzzy
numbers.
Simulated Annealing, which is a converging optimum solution algorithm, was applied for
optimizing the RSs in this study. In order to assess the influence of its parameters
(number of iterations, temperature decrease rate, initial value for generating random
numbers, initial temperature and two other parameters), they were individually
varied while keeping the others fixed. With each of the resulting parameter sets,
a full-automatic SA was applied to gain optimized fuzzy rule systems for flood
forecasting.
Evaluation of the performance of the resulting fuzzy rule forecasting systems (with the
intention to draw conclusions on the best SA parameters) was carried out in two
steps:
a) Evaluation of objective functions such as Nash-Sutcliffe and RMSE for all
RSs.
b) Manual evaluation of the preselected results from the first step. The evaluation was
based on visual inspection (scatter plots, time-series and Degree Of Fulfilment (DOF) graphs)
as well as logical interpretation of the rule systems.
Comparing the results showed that there were SA parameter sets which lead to
forecast systems of equally high quality (with respect to objective criteria such as
Nash-Sutcliffe), however the underlying rule systems significantly varied from each
other. Therefore, manual inspection played a key role in finding the overall best
results.
In the presentation, the procedure of preparing different sets of SA parameters, the
evaluation process of different results and the performance of the optimal RS will be
explained and presented by an example. |
|
|
|
|
|