Forecasting dam water inflows with fuzzy support vector regression using a rainy season-based membership function

Abstract

Establishing proper management of dam water inflows is essential for successful planning and irrigation. This research aimed to forecast the water inflow into the Queen Sirikit Dam, located in Pha Leud Sub-district, Tha Pla District, Uttaradit Province, Thailand. The dam was constructed across the Nan River. The research proposes a new method that uses fuzzy support vector regression (FSVR) based on a generalized bell-shaped membership function that depends on the rainy season. Moreover, a comparison was made between this method and FSVR based on the linear function of time, which is commonly used, as well as classical support vector regression (SVR). The predictive ability of the models was evaluated using 10-fold cross-validation on 3,700 samples. The results indicate that FSVR based on a generalized bell-shaped membership function is more effective than the other two methods, with a mean absolute error (MAE) of 15.6247 m3/s and R-square (R2) of 0.7984. This shows an improvement over the linear membership function, which had an MAE of 17.2566 m3/s and an R2 of 0.7555. Meanwhile, the classical SVR model had an MAE of 23.6997 m3/s and an R2 of 0.6741. These findings suggest that FSVR based on a generalized bell-shaped membership function is an effective approach for forecasting dam water inflows. Consequently, it will be highly beneficial for the efficient management of the dam, including drainage or warning announcements to downstream populations.

Keywords

• water inflow
• fuzzy support vector regression
• forecasting
• membership function