船舶操纵性神经计算

利用人工神经网络完成从船舶参数空间到船舶操纵性指标空间的映射。作为输入参量的船舶参数具有一定的自由度，比较符合船舶设计的循序渐进过程。本文采用３层９隐节点ＢＰ网络，９艘次实船资料的拟合结果的精度为０．６１３７（单项平均平方误差）。对船舶操纵性预报进行了初步研究。利用８艘散货船的训练结果，对另１艘同类船舶的回转试验的进距、横距和战术直径进行了预报。

ARTIFICIAL NEURAL NETWORKS

Computation of ship maneuverability using ANN is based on ANN's nonlinear mapping function to perform the project from ship particular space to ship maneuverability space. It is more flexible to some extent to select input particulars and convenient to be used in ship design. BP(Back Propagation) networks and the neuron function are used in this paper. The learning algorithm is gradient descending combined with momentum factor. The sample data were condensed to 0. 1-0. 9 in advance. According to the result of 9 various ships' computation, the RMS is 0. 6137. Furthermore, 9 bulk ships are selected for ship's maneuverability prediction. After 12, 000 times training based on data of turning test of 8 bulk ships, the relative prediction precision of turning test of the 9th bulk ship is - 7. 16% (advance), 7. 16 % (transfer) and 0. 03% (tactical diameter).