基于NARX神经网络的FPSO系泊缆张力预报

NARX Neural Network Based FPSO Mooring Lines Tension Prediction

深海浮式结构系泊系统的时域耦合运动分析涉及复杂的外部环境荷载、浮式结构物的水动力性能与系泊系统的非线性动态特性等因素的数值仿真。为了减小时域耦合分析过程的计算时间,将带有外部输入的非线性自回归神经网络(NARX)模型应用于某一分布式浮式生产储卸油平台(FPSO)系泊系统数值求解中,在此过程中选取NARX神经网络中的并行模式(NARX-P)进行训练与预报。此方法基于OrcaFlex软件短时的时域仿真计算结果,对NARX神经网络进行训练,使其具有预报之后时域张力响应的能力。结果表明系泊线张力的最大相对误差为7.61%,在可接受范围内,同时该替代模型能够大幅减小数值仿真的时间。

The coupled motion analysis of floating structure mooring system involves numerical simulation of complex external environmental loads,hydrodynamic performance of floating structures and nonlinear dynamic characteristics of mooring systems in deep and ultra-deep waters.In order to reduce the computation time of the time domain analysis process,a nonlinear autoregressive neural network(NARX)with external input is applied to the numerical solution of a spread floating production storage and offloading platform(FPSO)mooring system.In this process,the parallel mode(NARX-P)is selected in the NARX neural network.Based on the short-term time domain simulation results of OrcaFlex software,the NARX neural network is trained to predict the time domain tension response.The results show that the maximum relative error is of morning line tension 7.61%,which is within the acceptable range.At the same time,this alternative model can greatly reduce the time of numerical simulation.