一种新型的船舶横摇运动实时预报方法

本文在阐述强化学习的基本原理，方法的基础上，提出了一种强化学习的TD算法与BP算法相结合的BPTD方法，并将基用于对角回归神经网络的在线训练，最后以在船舶横摇运动时预报技术上的应用为例，说明这种算法有很强的实时多步预报能力。

New Method of Ship Rolling Real Time Forecast

The ship rolling real time forecast is carried out using the diagonal recurrent neural network, the traing training of which uses the BPTD algorithm proposed in this paper. The time delay (TD) algorithm based on the basic principle of reinforcement learning (BL) is presented. Then, an algorithm named BPTD based on the time delay algorithm in reinforcement learning and the error back propagation algorithm for the train of multi layer feed forward neural networks are proposed to train the diagonal recurrent neural networks. The diagonal recurrent neural network using the BPTD algorithm is applied to the real time forecast of ship rolling. Experiments based on the measured data of full scale ship roll motion indicated the results of 1 step, 2 steps, 3 steps and 4 steps forecast have the very good precision, but the results of real time forecast after 5 steps begin to deteriorate. These results demonstrate the effectiveness and, efficiency of the methods proposed in this paper.