基于神经网络的系泊系统主动式截断试验仿真

Simulation of Active Truncated System Based on Neural Network

摘要:

系泊系统主动式截断试验技术对缆索数值模型的求解速度有着较高的要求,而基于集中质量模型或有限元模型的缆索数值模型在计算时需要在每个时间步内迭代求解,因此求解速度缓慢；为解决此问题,采用神经网络辨识得到缆索上端点与截断点运动的时域等价近似模型；该等价模型在每个时间步内无需迭代求解,因此求解速度快于传统的缆索数值模型；同时利用商软OrcaFlex以及Matlab搭建了系泊系统主动式截断试验的仿真平台,将离线辨识得到的神经网络模型代替集中质量模型对缆索截断点运动进行求解；通过对比截断水深以及全水深下张力以及运动结果,说明了神经网络模型代替传统数值模型进行截断点运动计算的可行性。

Abstract:

A cable numerical model with fast solution speed is required in active truncated system. Because of intermediate equilibrium iterations, numerical model based on lumped-mass method’s solution speed is slow. To solve this problem, neural network was used to get equivalent model in time domain which reflects the motion relationship between upper point and truncated point of cable. This network model do not need iterative process so its solution speed can be more faster than lumped-mass method. A simulation platform was built by using Orcaflex and Matlab. In the process of simulation, lumped-mass method was replaced by neural network which is got off-line to do on-line computation. Comparison with results between full-depth system and truncated-depth system, it shows neural network is feasible in active truncated system.