乏信息条件下面向复杂海洋环境的舰船效能智能评估方法研究

In the early design and demonstration stage of a warship, an effective evaluation of ship effectiveness is of great significance for decision-making. The ship effectiveness is a non-linear combination of its performance, which is affected by a complex naval battlefield. In the demonstration stage, due to the lack of information of system performance and effectiveness, there are quite a few problems, such as the strong subjectivity of expert rating method, insufficient consideration on the impact of complex marine environment and the difficulty of describing the coupling relations among the sub-systems. Based on the previous research on ship reliability and taking the lack of information and the complex marine environment into account, this research will explore the use of intelligent technologies such as support vector regression and combat model, and break through the three difficult problems: the prediction of warship performance by machine learning under the lack of information, the quantification of warship effectiveness considering the impact of complex marine environment and the construction of a non-linear coupling model of overall effectiveness. The warship effectiveness evaluation method proposed by this research has the characteristics of fusing test and simulation data, loading complex marine environmental impact and reflecting complex coupling relation among factors. The results will found a theoretical basis for the design and evaluation of deep sea warships.

在舰船的早期设计论证阶段，有效评估舰船的效能，对于舰船装备开发决策，具有重要意义。舰船的效能是其单项性能的非线性组合，同时受到复杂多变的海战环境的影响。在论证阶段，由于缺乏系统层级的性能、效能信息可循，长期以来，其效能评估存在专家评分主观性强，对复杂海洋环境的影响考虑不足，子系统间复杂耦合关系难以反映等问题。本项目拟在前期舰船可靠性研究基础上，针对乏信息特点，面向复杂海洋环境，探索利用支持向量机、对抗模型等机器学习技术，重点突破乏信息条件下舰船性能指标的学习预测、考虑复杂海洋环境影响的效能量化、考虑非线性作用的综合效能耦合模型等三个难点问题。研究所提出的舰船效能评估方法具备可融合试验与仿真数据、可加载复杂海洋环境影响、可反映因素间复杂耦合关系等特点。研究成果将为深远海舰船设计评估工作提供理论基础支撑。

构建一套可融合试验与仿真数据、可加载复杂海洋环境影响、可反映因素间复杂耦合关系的舰船效能智能评估模型，对于舰船装备研制、测试、改进以及军事行动决策，具有重要意义。本项目基于先进机器学习技术和效能评估模型，研究复杂海洋环境作用下的舰船性能预报方法，探究舰船性能非线性变化与耦合对装备效能的影响机理。项目提出了虚（仿真模型）实（实船测试）融合的舰船性能测试验证平台架构及其指标体系，建立了小子样乏信息条件下多源数据融合机理与算法；基于深度神经网络，开展了舰船实海域操纵响应、能耗等航行性能预报研究，提出了基于环境激励时频特征的预报模型优化方法；对基于贝叶斯网络的传统效能评估模型进行时间域离散，引入了储备因子和共因失效因子，建立了考虑动态相关性的系统效能耦合分析方法。研究成果应用于JARI-USV等型号装备测试评估，并形成测评软件系统，被中国船级社、中船集团第701研究所、702研究所等单位采购应用，支撑我国智能舰船装备发展。.在项目执行期间，发表了高水平学术论文15篇，其中SCI收录9篇；授权国家发明专利3项，软件著作权3项；出版英文专著1部；制定国家标准1项（排3）；获得中国造船工程学会科技进步奖一等奖1项（排1）；培养硕士研究生2名，协助培养硕士研究生1名、博士研究生2名。