A Twin LS-SVM-based maneuvering predictor for marine crafts

Abstract

In this paper, dedicated to maneuvering prediction for marine crafts with a limited amount of data, a Twin LS-SVM-based error-compensated maneuvering prediction strategy for the parametric model is presented. The strategy used one LS-SVM to pre-predict marine craft motion and obtain training data errors. Another LS-SVM is designed for the prediction and compensation of testing errors generated by the former LS-SVM. The new strategy can render accurate and robust prediction improvements in data-limited situations. Then, a comparative analysis of prediction performance between the Twin LS-SVM strategy and the conventional method based on LS-SVM is contributed to demonstrate the effectiveness of the proposed strategy. Both simulation test results and experimental study results indicate that the developed strategy is a powerful and practical prediction tool for marine crafts, which shows that it reduces the root mean square error of speeds by up to 64.7% in simulation tests, and 49.2% in experiments.