基于神经网络的船舶阻力计算数值实验研究

Experimental Research on Neural Network for Ship Resistance Computation

3层以上的前向神经网络能以任意精度逼近任何非线性函数,采用误差反传算法,利用系列60船模原始试验数据作为训练和检验样本,应用输出与目标的线性回归、相关系数和相对误差,以及利用该神经网络绘制的曲线作为评价指标,通过大量的数值实验对最佳训练函数、性能函数、传递函数、隐层神经元数的选用进行了研究,建立了最佳的系列60船舶阻力计算3层和4层BP神经网络系统。

Applying back propagation arithmetic,the original experiment data of series 60 ship models are as training and test sample,the regression line,the correlation coefficient,the relative error and the performance curve plotted by neural network are as appraise index,a great deal of experiments are made to optimize training function,performance function,transfer function and the neuron number of latent layer.At last,the optimal three-layer and four-layer back propagation neural network for ship resistance computation is founded.