面向高维动态大数据的多粒度知识协同及推理

当前，领域数据呈现大规模、高维、动态变化等特点。粒计算能有效发掘样本共性特征，已成为领域大数据分析的一种新的范式。现有粒计算模型常侧重静态数据或基于增量的动态数据分析，较少从知识协同和推理角度实现跨特征空间和跨时间维度的群体知识发现。本项目拟采用粒计算问题求解思路，系统研究高维动态大数据特征提取、知识表示以及知识推理，为发掘数据中蕴含的群体共性知识提供新的理论和方法。研究内容包括：（1）流形保持下的模糊无监督学习特征选择，实现问题空间由高维向低维转换；（2）基于阴影集的多粒度信息粒构建，实现研究对象由样本个体向信息粒转变；（3）基于协同聚类的知识协同及基于粒模糊模型的知识推理，实现分析模型由单模静态向多模动态迁移；（4）理论成果应用于废旧织物材质分类行业难点问题。本项目涉及多学科领域交叉，旨在建立一套完善的高维动态大数据多粒度分析方法，丰富粒计算理论体系，助力其在大数据挖掘中的实际应用。

Currently, domain data is characterized by large-scale, high-dimensional, and dynamic changes with the rapid development of information technology. Granular computing can effectively discover common characteristics of samples and has become a new paradigm for analyzing domain big data. Existing granular computing models often focus on static data or incremental-based dynamic data analysis, and seldom implement group knowledge discovery across feature spaces and across time dimensions from the perspective of knowledge collaboration and reasoning. This project intends to use the granular computing problem solving ideas to systematically study high-dimensional dynamic big data feature extraction, knowledge representation, and knowledge reasoning, and provide new theories and methods for discovering the group knowledge contained in data. The main research contents include: (1) Feature extraction based on fuzzy unsupervised learning with spatial geometric structure preservation, so that the problem space is transformed from high-dimensional to low-dimensional; (2) Construction of multi-granules based on shadowed sets to realize the transformation of research objects from sample individuals to information granules; (3) Knowledge collaboration based on collaborative clustering and knowledge reasoning based on granular fuzzy model to realize the transformation of analysis models from single-mode static scenarios to multi-mode dynamic scenarios; (4) The theoretical results are applied for classifying waste fabric materials which is a difficult problem in the factory. This project involves cross-disciplinary fields, and aims to establish a complete set of multigranular analysis methods for high-dimensional dynamic big data. The results will enrich the theoretical system of granular computing, as well promote its practical applications in big data mining.