动态非平稳环境下的数据描述与系统建模：粒计算的框架

The amount of data that is being created by various systems keeps growing at a high pace, yet only a small percentage of data is actually analyzed. Deciding how to effectively analyze big data to support better decision making and facilitate strategic business moves has become a challenging and open problem, which calls for innovative technologies of information processing. Granular computing inherently oriented on processing information granules emerges here as a highly promising alternative. As the name stipulates, granular computing is focused on forming, processing, and communicating meaningful chunks of information – information granules being composed of entities linked together because of their similarity, coherency, and functionality. Recognizing data abstraction and processing as an ever-present and acute need in perceiving and modeling complex systems, we establish a coherent and comprehensive conceptual platform of data analysis and granular computing. The proposed project is concerned with a methodology and an algorithmic framework supporting a comprehensive linguistic data description and their analysis in non-stationary and dynamic environment. The ultimate objective of the project is to realize data analysis by striking a sound tradeoff between precision (granularity) and interpretability of the results – the two crucial requirements present here. The non-stationary and dynamic environment (which is predominant in data stream analysis) makes the analysis of data a complex undertaking. We develop a concept of so-called referential information granules which help establish a unified view at the analysis of the data that come in the form of data stream changing over time (non-stationary aspect) with a varying data space (resulting in the dynamic character of data). The proposed investigation of novel and promising granular models will develop, conceptualize, and produce algorithms for constructing intelligent systems to benefit environmental, agricultural, energy and human health sectors with expected ancillary advantages to machine learning, data mining and high-performance computing.

随着智能系统所产生的数据量持续增长，如何分析和处理庞大的数据以更好地支持决策制定和商业战略实施成为一个很有挑战性的问题。在这个问题上，面向处理信息粒的粒计算成为一个非常极具发展前景的途径。粒计算旨在处理通过相似性，一致性和功能相关性关联在一起的、以信息粒形式出现的复杂信息集合,我们需要建立针对数据分析和粒计算的一个统一的综合性概念“平台”。本项目同时致力于建立一种在非平稳动态环境下能够支持全面语言性数据描述和分析的方法论和计算框架。这两个关键问题将通过聚焦解决分析结果的准确性和可解释性这两个所有的数据分析框架都面临的关键需求来完成。在非平稳动态环境下，数据的分析会变得更加复杂。通过信息粒建立一种对于在变化的数据空间中随时间改变的以数据流的形式产生的数据的统一性视图。通过这种全新的、具有良好应用前景的粒模型, 本项目将发展和建立一系列智能系统概念框架和算法基础。

随着智能系统所产生的数据量持续增长，如何分析和处理庞大的数据以更好地支持决策制定和商业战略实施成为一个很有挑战性的问题。在这个问题上，面向处理信息粒的粒计算成为一个极具发展前景的途径。粒计算旨在处理通过相似性，一致性和功能相关性关联在一起的、以信息粒形式出现的复杂信息集合,我们需要建立针对数据分析和粒计算的一个统一的综合性概念“平台”。本项目同时致力于建立一种在非平稳动态环境下能够支持全面语言性数据描述和分析的方法论和计算框架。.为了达到上述目标，本项目研究解决了以下基本问题：（1）基于合理粒度原则创建信息粒描述符与信息粒的性能评估；（2）信息粒的编码与解码；（3）构建层次状的粒模型；（4）高效的粒度数据描述策略；（5）通过缺失值插补提高数据的整体质量；（6）开发了基于粒计算的数据处理原型框架。.本项目的原创性和研究成果主要体现在以下方面。第一，本项目从一个崭新的视角来描述信号数据，对数据进行处理和解释； 第二，解决了粒计算领域的一系列基础性问题，相关的研究成果发表在IEEE Transactions等高水平期刊，它将极大地推动计算智能和粒计算领域的发展和前进；第三，成功地将研究成果应用于一些预测系统，取得了良好的效果。