摘要
建立了基于系列60船模原始实验数据的船舶阻力计算3层BP神经网络系统,利用随机选取的检验样本和插值样本作为输入向量,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,验证了该神经网络的可靠性.在该神经网络的建立过程中,对训练函数、性能函数、传递函数、隐层神经元数和神经网络绘制的性能曲线进行了实验,并通过数据预处理方式,确定了最佳的船舶阻力计算3层BP神经网络系统.
Applying the original experiment data of series 60 ship models, three layers backpropagation neural network is founded. By using random experiment data and insert vectors as input sample, the linear regression ,the linear correlation value and the relative error between the output of the neural network and the target are obtained. The results and the performance curve constructed by the neural network indicate that the implementation is creditable. In the course of founding the neural network, by making plenty of experiments to optimize training function, performance function, transfer func- tion, latent layer neuron number and the performance curve constructed by the neural network, and by applying data pretreatment manner, the optimal three layers backpropagation neural network of ship resistance comoutation is founded.
出处
《武汉理工大学学报(交通科学与工程版)》
2010年第1期209-212,216,共5页
Journal of Wuhan University of Technology(Transportation Science & Engineering)
基金
广东省自然科学基金项目(批准号:06026442)
广州航海高等专科学校科研项目(批准号:200712B04)资助
关键词
船舶阻力
BP神经网络
系列60
ship resistance
backpropagation neural network
series 60