A sequential coastal current prediction approach based on hierarchical decomposition

Precise prediction of coastal tidal current is essential for the efficient operation of tidal power generation, coastal engineering and maritime activities. To excavate the useful information in coastal current movement thus improving the accuracy of coastal current prediction, a real-time sequential mechanism for coastal current prediction is proposed based on a data reconstruction scheme. The reconstruction decomposes the coastal current time series by taking both advantage of the autonomy of the empirical mode decomposition and the arbitrariness of the discrete wavelet transformation, and the decomposed components are identified and predicted respectively by radial basis function networks with variable structure whose hidden units’ locations can be adjusted in real-time. To improve the adaptivity and rapidity of the prediction mechanism, the Lipschitz quotients method is employed to determine the prediction system structure, with a sliding data window serving as system dynamics observer. Coastal current prediction simulation is conducted using the measurement data of the tidal gauge of Cumberland Sound, USA and the results validated the effectiveness of the proposed mechanism in respect of prediction accuracy and processing speed.