Wavelet kernel support vector machines forecasting techniques: Case study on water-level predictions during typhoons

Abstract

This paper presents a novel algorithm, wavelet support vector machines (wavelet SVMs), for forecasting the hourly water levels at gauging stations. These stations are under strong precipitations and affected by tidal effects during typhoons. An admissible wavelet kernel SVMs implements the combination of wavelet technique with SVMs. The wavelet is a multi-dimension wavelet function that can approximate arbitrary nonlinear functions. Using both classical Gaussian and wavelet SVMs, this study constructed the channel level models for forecasting downstream water levels. The developed models were then applied to the Tanshui River Basin in Taiwan and the water levels at various lag times predicted by both Gaussian and wavelet SVMs were compared. Analysis results showed that the optimal situation occurred at the lag time of 3 h with relative mean square errors (RMSEs) of 0.205 and 0.160 m obtained by the Gaussian and wavelet SVMs, respectively at Taipei Bridge station and RMSEs of 0.154 and 0.092 m at Tudigong station, respectively. As seen in the comparison, wavelet SVMs yielded more accurate predictions than Gaussian SVMs and offered a practical solution to the problem of water-level predictions during typhoon attacks.