A new fuzzy linear regression approach for dissolved oxygen prediction |
| |
Authors: | Usman T. Khan |
| |
Affiliation: | Mechanical Engineering, University of Victoria, Victoria, British Columbia V8W 3P6, Canada |
| |
Abstract: | AbstractA new method for fuzzy linear regression is proposed to predict dissolved oxygen using abiotic factors in a riverine environment, in Calgary, Canada. The proposed method is designed to accommodate fuzzy regressors, regressand and coefficients, i.e. representing full system uncertainty. The regression equation is built to minimize the distance between fuzzy numbers, and generalizes to crisp regression when crisp parameters are used. The method is compared to two existing fuzzy linear regression techniques: the Tanaka method and the Diamond method. The proposed new method outperforms the existing methods with higher Nash-Sutcliffe efficiency, and lower RMSE, AIC and total fuzzy distance. The new method demonstrates that nonlinear membership functions are more suitable for representing uncertain environmental data than the typical triangular representations. A result of this research is that low DO prediction is improved and consequently the approach can be used for risk analysis by water resource managers. Editor D. Koutsoyiannis; Associate editor T. Okruszko |
| |
Keywords: | dissolved oxygen fuzzy numbers linear regression fuzzy linear regression uncertainty risk water quality |
|
|