Interpretable fuzzy systems for monthly rainfall spatial interpolation and time series prediction
Kajornrit, Jesada (2014) Interpretable fuzzy systems for monthly rainfall spatial interpolation and time series prediction. PhD thesis, Murdoch University.
This thesis proposes methodologies to analyze and establish interpretable fuzzy systems for monthly rainfall spatial interpolation and time series prediction. A fuzzy system has been selected due to its capability of handling the uncertainty in the data and due to its interpretability characteristic.
In the first part, this thesis proposes a methodology to analyze and establish interpretable fuzzy models for monthly rainfall spatial interpolation using global and local methods. In the global method, the proposed methodology begins with clustering analysis to de-termine the appropriate number of clusters, and fuzzy modeling and a genetic algorithm are then used to establish the fuzzy interpretation model. In the local method, the modu-lar technique has been applied to improve the accuracy of the global models while the interpretability capability of the model is maintained.
In the second part, this thesis proposes a methodology to establish single and modular interpretable fuzzy models for monthly rainfall time series predictions. In the single model, the cooperative neuro-fuzzy technique and a genetic algorithm have been used. In the modular model, the modular technique has been applied to simplify the complexi-ty of the single model. The whole system is decomposed into twelve sub-modules ac-cording to the calendar months. The proposed modular model consists of two function-ally consecutive layers, the prediction layer and the aggregation layer. In the aggregation layer, Bayesian reasoning has been applied.
The case study used in this thesis is located in the northeast region of Thailand. The proposed models were compared with commonly-used conventional and intelligent methods in the hydrological discipline. The experimental results showed that, in the quantitative aspect, the proposed models can provide good prediction accuracy and, in the qualitative aspect, the proposed models can also meet the criteria used for model in-terpretability assessment.
|Publication Type:||Thesis (PhD)|
|Murdoch Affiliation:||School of Engineering and Information Technology|
|Supervisor:||Wong, Kevin and , Lance|
|Item Control Page|
Downloads per month over past year