Abstract: The processing and prediction of tidal current signal are significant and valuable in various aspects. This study introduced a signal sparse representation theory, constructed a sparse AR model in order to find out the connection among the tidal currents, and to carry out the forecast analysis. Firstly, we established the conventional AR model by measuring current signal, so we could get a set of complete sparse matrix. Secondly, we randomly extracted parts of this to create underdetermined system of equations. We acquired the sparse AR coefficients by using sparse optimization algorithm. After repeating above steps for many times, the stability increased and reached to average sparse AR coefficients. Finally, the last step was to reconstruct or predict the tidal current signal by using the last sparse AR coefficients. By using the measured observations of tidal current, especially the observations with many peaks or rotary current phenomenon. Many experiments had been done to establish the sparse AR model and prediction. Comparing this AR model with flow harmonic method, the sparse AR model is much better than the traditional tidal harmonic method especially in analysing the region with changeable tidal current. Moreover, the variance of the sparse AR model is less than that of the traditional tidal harmonic analysis.

Key words: AR model, tidal current forecast, sparse optimization, Orthogonal Matching Pursuit algorithms