改进PSO训练的BPNN方法的舰船主尺度建模

Modeling of the principal dimensions of large vessels based on a BPNN trained by an improved PSO

利用改进粒子群优化算法训练的BP神经网络(BPNN)对以航母为代表的大型舰船主尺度进行了回归分析.对粒子群优化算法(PSO)的学习因子进行了关于迭代进程的自适应调整,并将改进后的PSO算法对BPNN训练过程进行优化.利用新型BPNN对典型航母主尺度(总长、总宽、设计水线长、设计水线宽、吃水与满载排水量)进行数学建模,与基于传统多项式回归的结果进行对比分析.结果表明经改进PSO训练的BPNN具有更高的输出精度且具有良好的分段光滑特性,这对于大型舰船方案论证与总体设计可起到重要的指导性作用.

A back propagation neural network （BPNN） trained by an improved particle swarm optimization （PSO） was applied to the principal dimensions of large vessels by regression analysis. First, the improved PSO learning factors conceming the iterative process were adjusted adaptively and the BPNN trained process was optimized by using an improved PSO. Secondly, a new BPNN method was applied to establish a mathematic model of an aircraft＇s principal dimensions （including overall length, breadth moulded, length of the design waterline, breadth of the design waterline, draft, and full load displacement）. Finally, compared with the results of traditional polynomial regression, a BPNN trained by an improved PSO has higher accuracy and fine characteristics of smooth at every subsection. Therefore, the mathematic model has a guidance effect on the scheme demonstration and overall design of large vessels.