![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Experiments with models committees for flow forecasting |
VerfasserIn |
J. Ye, N. Kayastha, S. J. van Andel, F. Fenicia, D. P. Solomatine |
Konferenz |
EGU General Assembly 2012
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 14 (2012) |
Datensatznummer |
250064097
|
|
|
|
Zusammenfassung |
In hydrological modelling typically a single model accounting for all possible hydrological loads, seasons and regimes is used. We argue however, that if a model is not complex enough (and this is the case if conceptual or semi-distributed models are used), then a single model can hardly capture all facets of a complex process, and hence more flexible modelling architectures are required. One possibility here is building several specialized models and making them responsible for various sub-processes. An output would be then a combination of outputs of individual models.
In machine learning this approach is widely applied: several learning models are combined in a committee (where each model has a “voting” right with a particular weight). In this presentation we concentrate on optimising the above mentioned process of building a model committee, and on various ways of (a) building individual specialized models (mainly concentrating on calibrating them on various subsets of data and regimes corresponding to hydrological sub-processes), and (b) on various ways of combining their outputs (using the ideas of a fuzzy committee with various parameterisations). In doing so, we extend the approaches developed in [1, 2] and present new results.
We consider this problem in multi-objective optimization setting (where objective functions correspond to different hydrological regimes) – leading to a number of Pareto-optimal model combinations from which the most appropriate for a given task can be chosen. Applications of the presented approach to flow forecasting are presented.
References
[1] D.P. Solomatine. Optimal modularization of learning models in forecasting environmental variables. Proc. of the iEMSs 3rd Biennial Meeting: "Summit on Environmental Modelling and Software" (A. Voinov, A. Jakeman, A. Rizzoli, eds.), Burlington, USA, July 2006.
[2] Fenicia, F., Solomatine, D. P., Savenije, H. H. G. and Matgen, P. Soft combination of local models in a multi-objective framework. Hydrol. Earth Syst. Sci., 11, 1797-1809, Special Issue “Data-driven approaches, optimization and model integration: hydrological applications”, R. Abrahart, L. See, D. Solomatine, and E. Toth (eds.), 2007. |
|
|
|
|
|