Datas for optimization

To resolve the long-standing conflict between accuracy and efficiency in designing mooring systems for deep-water cylindrical FPSO (Floating Production Storage and Offloading), an intelligent design framework that integrates high-fidelity numerical simulation, a BP neural network surrogate model, and a global optimization algorithm was proposed and validated. The method established and verified a high-precision coupled dynamic model using OrcaFlex and analyzed the impact of key parameters through 720 sets of simulations. Based on these simulation data, a BP neural network surrogate model was further designed and trained, with design variables (such as segmented line length, number of cables, mooring radius, etc.) as inputs and key responses (such as platform displacement, line tension, etc.) as outputs. With the goal of minimizing the total material usage and in accordance with relevant industry standards as constraints, global optimization was carried out using the Levenberg-Marquardt algorithm. The results show that the neural network model has a high prediction accuracy (coefficient of determination R2>0.98). The optimal solution obtained is a 3×3 mooring layout with a mooring radius of 1670.50 m. Under the premise of satisfying the API specification through calculation verification, the maximum platform displacements in the intact and single-cable failure conditions are 80.61 m and 85.23 m respectively, and the maximum mooring tensions are 13507 kN and 15455 kN respectively, all of which meet the safety thresholds. This framework ultimately achieves a 49% reduction in material usage and cost savings of $1,095,238, while improving computational efficiency by a factor of 124.

为解决深水圆柱形浮式生产储卸油装置（FPSO，Floating Production Storage and Offloading）系泊系统设计中长期存在的精度与效率冲突，本研究提出并验证了一套融合高保真数值模拟、BP神经网络代理模型与全局优化算法的智能设计框架。该方法借助OrcaFlex软件建立并验证了高精度耦合动力学模型，并通过720组仿真分析了关键参数的影响规律。基于上述仿真数据，进一步设计并训练了BP神经网络代理模型，以设计变量（如分段缆绳长度、缆绳数量、系泊半径等）作为输入，以平台位移、缆绳张力等关键响应作为输出。以总材料用量最小化为优化目标，并以相关行业标准作为约束条件，采用列文伯格-马夸尔特（Levenberg-Marquardt）算法开展全局优化。结果表明，该神经网络模型具备极高的预测精度，决定系数R²大于0.98。所得到的最优解为系泊半径1670.50米的3×3系泊布局。经计算验证满足API规范的前提下，完整工况与单缆失效工况下的平台最大位移分别为80.61米与85.23米，最大系泊张力分别为13507千牛与15455千牛，各项指标均满足安全阈值要求。该框架最终实现了49%的材料用量削减，节省成本1095238美元，同时将计算效率提升124倍。