![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Challenges and Benefits of Direct Policy Search in Advancing Multiobjective Reservoir Management |
VerfasserIn |
Andrea Castelletti, Matteo Giuliani, Jazmin Zatarain-Salazar, John Hermann, Francesca Pianosi, Patrick Reed |
Konferenz |
EGU General Assembly 2015
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 17 (2015) |
Datensatznummer |
250106073
|
Publikation (Nr.) |
EGU/EGU2015-5726.pdf |
|
|
|
Zusammenfassung |
Optimal management policies for water reservoir operation are generally designed
via stochastic dynamic programming (SDP). Yet, the adoption of SDP in complex
real-world problems is challenged by the three curses of dimensionality, of modeling,
and of multiple objectives. These three curses considerably limit SDP’s practical
application.
Alternatively, in this study, we focus on the use of evolutionary multi-objective direct
policy search (EMODPS), a simulation-based optimization approach that combines direct
policy search, nonlinear approximating networks and multi-objective evolutionary algorithms
to design Pareto approximate operating policies for multi-purpose water reservoirs. Our
analysis explores the technical and practical implications of using EMODPS through a
careful diagnostic assessment of the EMODPS Pareto approximate solutions attained and the
overall reliability of the policy design process. A key choice in the EMODPS approach is the
selection of alternative formulations of the operating policies. In this study, we distinguish the
relative performance of two widely used nonlinear approximating networks, namely
Artificial Neural Networks and Radial Basis Functions, and we further compare
them with SDP. Besides, we comparatively assess state-of-the-art multi-objective
evolutionary algorithms (MOEAs) in terms of efficiency, effectiveness, reliability, and
controllability.
Our diagnostic results show that RBFs solutions are more effective that ANNs in
designing Pareto approximate policies for several water reservoir systems. They also
highlight that EMODPS is very challenging for modern MOEAs and that epsilon dominance
is critical for attaining high levels of performance. Epsilon dominance algorithms
epsilon-MOEA, epsilon-NSGAII and the auto adaptive Borg MOEA, are statistically superior
for the class of problems considered. |
|
|
|
|
|