Modeling and prediction for the Buoy motion characteristics

To ensure the stability of buoy systems, it is necessary to predict the buoy motion characteristics in different environment conditions. Unfortunately, an accurate model is still not available for some unknown interaction between waves, currents and winds in oceans. Modeling and designing a suitable buoy for further manufacture often encounters some challenges in practice. In this work, the least squares support vector regression (LSSVR) method is proposed to predict buoy motion characteristics. First, some modeling data are collected from the data processing system of a buoy. Additionally, an evaluation criterion is designed to automatically divide training data into two subsets, corresponding to the normal and extreme conditions. Moreover, two local LSSVR models are constructed using the subsets, respectively. Finally, a suitable model is automatically selected for each new sample. Consequently, with limited modeling samples, different property in the normal and extreme conditions can be effectively captured. Experimental results show the superiority of the proposed method.