Author(s): Nahm-Chung Jung; Roland. K. Price
Linked Author(s):
Keywords: No Keywords
Abstract: Nowadays the two most widely used types of models are physically-based and data-driven model. However, physical models based on, e. g. natural laws and mass-balance relationships inevitably have variations between the model output and the observed data according to the uncertainty in model parameters and structure, or measurement errors. In that case, successful uncertainty analysis is the key to successful modelling. Data-driven models are an approach to manage uncertainty of model structure, input data and parameters. Artificial neural networks including fuzzy logic and model trees are a state-of-the-art technology in this area, because both models are more explicit than any other data mining techniques. Although they need a considerable amount of historical data, these kinds of data-driven models have been growing tendency to complement or even replace physically based models that need uncertainty analysis. This paper is supposed to analyze both data-driven models one of which will be a component of decision support system for reservoir water quality management. Until now the cases of water quality and ecological modelling using data mining techniques such as ANFIS and MTs have not nearly been reported. In this paper, MTs show better performance and comprehension than ANFIS. However, ANFIS shows higher non-linear performance than MT, allowing tracing of peaks in time series modelling. Whereas MTs show disadvantage in tracing the peaks, it takes the mean of variables which lie inside two standard deviations from the mean. The choice of the better solution between ANFIS and MT for DSS (Decision Support System) should be decided after further research, for instance a time series CCF (cross correlation function) analysis between each sampling station and all sampling stations of tributaries.
Year: 2005