基于LightGBM的拖曳系统动力响应预报方法

针对传统数值仿真计算方法耗时长、占用计算机资源多等缺点,提出基于LightGBM算法的拖曳系统动力响应进行评估的回归预测模型,以已有的OrcaFlex数值模拟得到的数据为样本,以拖曳系统上的海洋环境条件、拖船航速和下放缆长为特征,以动力响应为目标,引入LightGBM算法,对拖曳缆顶端张力最大值等动力响应进行预测分析。与传统数值模拟方法相比,LightGBM算法在保证结果准确性的同时大幅度提高了计算效率。通过与随机森林（RF）、极限梯度提升（XGBoost）算法相比,其准确度和计算效率的表现更好。最后提出了贝叶斯参数优化的LightGBM算法,准确度进一步提高,为提前采取措施保障拖曳系统的作业安全提供了一条高效的技术途径,同时为建立拖曳系统数字孪生体提供了有力的技术支撑。     

Computationally efficient approaches to fatigue analysis of deepwater risers

Riser long-term fatigue performance is an important design consideration. Although extensive application of irregular sea analysis in time domain with the rainflow counting technique for post-processing is regarded as the most accurate of the approaches for fatigue analysis, it does suffer from some limitations, such as the computational effort. For this reason, two computationally efficient approaches are employed to perform the fatigue analysis of a deepwater top-tensioned riser, based on the Longuet-Higgins distribution and time domain scaling respectively. With Longuet-Higgins distribution irregular wave sea states are expanded into their individual wave bins. These regular wave simulations are of short duration and consequently run quickly. Using the time domain scaling technique, the number of irregular wave runs can be performed for a comparatively small number of load cases and hence reduces the calculation time. The results showed a reasonable accuracy and significant efficiency for both approaches, compared with those from the equivalent rainflow analysis. With much less computational effort and disk storage requirement, the approaches outlined in this paper can therefore be used for the fatigue assessment of deepwater risers in industry practice.     

深水立管波致疲劳分析的确定性方法

采用Longuet-Higgins法将表征长期海况的波浪散布图分解为单个规则波的散布图,以此进行深水立管的疲劳分析,并与雨流法的计算结果进行了比较。结果表明,该方法可以准确捕捉最大疲劳损伤发生的位置,计算得到的疲劳寿命和雨流法的结果较为接近;且随着离散的精细,其结果和雨流法的结果越加接近。该方法计算速度快,占用硬盘空间少,可用于初始设计阶段立管的疲劳分析及疲劳参数敏感性分析。在疲劳分析的校核阶段,该方法因计算结果是保守的,依然可以应用。