项目名称: 面向复杂流体模拟数据的特征提取技术研究
项目编号: No.10876036
项目类型: 面上项目
立项/批准年度: 2009
项目学科: 金属学与金属工艺
项目作者: 解利军
作者单位: 浙江大学
项目金额: 30万元
中文摘要: 复杂流场的特征抽取和智能分析在现代科学计算中可以发挥重要作用,以往研究主要集中在基于 数值方法的2D 和3D 流场拓扑结构抽取上。本申请拟针对多介质辐射流体力学问题的时序数据,基于流场的局部特征(包括涡、激波、分离和粘附线、回流区、边界层)和时序事件(生成、持续、分裂、融合和耗散),采用数据挖掘方法,实现高层特征抽取(如涡的分布规律、激波的生成预测等)。具体包括:(1)研究数据挖掘方法处理流场数据的基本范式,构建流场数据智能分析的框架;(2)研究三类挖掘算法(聚类算法、关联规则算法和预测模型类算法)在复杂流场上的应用,挖掘流场中对应的三类主要知识;(3)研究抽象特征的映射方法,将挖掘到的特征进行可视化显示;(4)结合中物院的需求,将发展的方法集成为软件开发包,并应用于多介质辐射流体力学问题,验证挖掘结果。
中文关键词: 流场特征抽取;数据挖掘;特征可视化
英文摘要: Intellective feature extraction and analysis of complex flow plays an important role in modern science. Currently, research mainly focuses on flow field topology features extraction based on numerical methods. This application plans to study high-level features (such as distribution of vortexes, prediction of shock wave, etc.) extracting methods for the radiation hydrodynamics of the time-series data based on data mining methods. We will try to utilize data mining methods on the flow of local features (including vortex, shock, isolation and adhesion line, circumfluence, boundary layer) and the timing events (generation, sustained, separatism, integration and dissipation). The research will includes (1) Study the basic paradigm to use data mining approaches to CFD simulating results (2) Utilize three types of mining algorithms (clustering, association rules and prediction models) to analyze complex flow, (3) Study visualization methods to map the extracted features into graphics, (4) Develop a software tool according the requirement of ZhongWu Yuan and collaborate with them to mine their own data.
英文关键词: Fluid feature extraction; data mining; feature visualization