An evaluation of the potential of adaptive neuro-fuzzy inference system in hydrological modelling and prediction
The use of data-driven modelling (DDM) in hydrological forecasting has been in practice since decades ago. Nevertheless, despite the ease of use, DDM approach has also been criticised due its 'black box' nature where the physical insights of the modelled processes are far from reach. Whil...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2012
|
| Subjects: | |
| Online Access: | http://umpir.ump.edu.my/id/eprint/8774/1/NGAHZAIFA%20AB%20GHANI.PDF |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The use of data-driven modelling (DDM) in hydrological forecasting has been in practice since decades ago. Nevertheless, despite the ease of use, DDM approach has also been criticised due its 'black box' nature where the physical insights of the modelled processes are far from reach. Whilst hydrologists are craving for the insight, the operational modellers are and will always prefer an easily applicable method regardless of whether the model is able to deliver knowledge. Hence, a method that could fulfil the need of both would be a perfect solution, ANFIS (Adaptive Neuro Fuzzy Inference System), for its advantages of having linguistic representation of models has been the interests of both groups and have been successfully tested on a number of international catchments. It is however still unclear as to what extent is ANFIS able to deliver the required knowledge; how capable is ANFIS in modelling sediment-discharge; and what are the advantages and disadvantages of using ANFIS as a modelling tool. This thesis explores ANFIS capability in
order to address these issues. The methods involved include creating synthetic datasets that mimic the sediment-discharge relationships; experimenting with different ANFIS
parameter settings; and observing and analysing the behaviour of models with the help of statistical and graphical evaluations. The results highlight that ANFIS is capable to model most of the tested relationships, but the produced model is very dependent on the parameters applied when training the model. Wrong choice of parameters may lead to the production of models with good metrics but low transferability, or even worse, not transferable at all. As a conclusion, ANFIS can be used in sediment-discharge modelling with certain restrictions on training parameters but this is mostly applicable to the simple and common rating curves. More studies are needed in order to explore the potential of ANFIS to model complex sediment-discharge processes. |
|---|