考虑噪声分布的鲁棒模糊粗糙集模型及算法研究

项目名称： 考虑噪声分布的鲁棒模糊粗糙集模型及算法研究

项目编号： No.61202259

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 安爽

作者单位： 东北大学

项目金额： 24万元

中文摘要： 在工程应用中，采集的数据往往被叠加上一定程度的且服从一定概率分布的数据噪声。因此，设计鲁棒的数据分析模型是十分必要的。模糊粗糙集理论为处理数据中的不确定性提供了一种有效的数学工具，该理论对噪声的敏感性引起了各国学者的广泛重视。当前未考虑噪声的分布信息是鲁棒模糊粗糙集建模研究中仍然存在的一个问题。本项目尝试解决这一问题，建立考虑噪声分布的鲁棒模糊粗糙集模型。 本项目突破当前鲁棒模糊粗糙集模型研究中"不考虑噪声分布"的建模方式，通过分析服从不同分布的噪声数据给模糊粗糙集的近似计算带来的干扰特性，总结噪声分布模型与鲁棒模糊粗糙集建模之间的关系；以经验风险最小化原则为指导建立考虑噪声分布的鲁棒模糊粗糙集模型。然后，基于考虑噪声分布的模糊粗糙集模型建立鲁棒的分类学习算法。此外，本项目将所提出的模型及算法应用于短期风电功率预报，验证考虑噪声分布的模糊粗糙集模型在实际应用中的有效性和鲁棒性。

中文关键词： 模糊粗糙集；数据分布；鲁棒性；分类与预测；风电预报

英文摘要： In project, data collected usually corrupted by noise which follows a certain probability distribution. It is necessary to design robust models in data analysis. Fuzzy rough set theory presents an effective mathematical tool for handling uncertainty in data, and the sensitivity to noise of the theory causes extensive attention of scholars from various countries. Currently, unconsidering distribution models of noise is still a problem in modeling robust fuzzy rough sets. This item attempts to resolve the problem, and establishes robust fuzzy rough set models considering distribution information of noise in data sets. In this work, the existent modeling techniques that do not unconsider distribution models of noise are changed.The relation between distribution models of noise and establishing robust models is summarized by analyzing the disturbing characteristic of fuzzy approximations produced by noise with different distribution models. And then the robust fuzzy rough set models considering noise distribution are built inspired by empirical risk mininization principle. Based on the proposed robust models, robust classification algorithms are introduced. The new models and algorithms introduced in this item are applied in short-term wind power prediction to test the effectiveness and robustness of the prop

英文关键词： Fuzzy rough sets；data distribution；robustness；classification and prediction；wind power forecast