An Ensemble Real-Time Tidal Level Prediction Mechanism Using Multiresolution Wavelet Decomposition Method

Precise real-time tidal prediction is essential for management of marine activities. Note that the tidal change is a complex time-varying nonlinear process, which is not only generated by periodic configurations of celestial bodies but also influenced by various time-varying meteorological factors. To achieve precise real-time tidal prediction, an ensemble tidal prediction mechanism is established by combining harmonic analysis and variable neural networks which are constructed by discrete wavelet transform (DWT). In the ensemble prediction mechanism, a conventional harmonic analysis method is used for representing the effects of celestial factors, while a DWT-based variable neural network is used for representing the nonlinear time-varying influences of meteorological factors and other unmodeled factors. The decomposition of tidal residual time series enables the precise prediction of time-varying dynamics by using a variable neural network whose dimensions and parameters are both adaptively tuned online. High accuracy of the proposed ensemble real-time tidal prediction mechanism is demonstrated by simulation studies on the actual tidal measurements collected from the Old Port Tampa tidal station and other four tidal stations in the USA.