面向决策知识集成创新的多源知识融合方法研究

项目名称： 面向决策知识集成创新的多源知识融合方法研究

项目编号： No.71201020

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 管理科学与工程

项目作者： 王晓欢

作者单位： 东北大学

项目金额： 19万元

中文摘要： 面向决策知识集成创新的多源知识融合方法研究，是根据目前国内外解决复杂决策问题相关理论与方法的研究发展动态，及复杂决策问题求解中对决策知识推送的实际需求而提炼出来的、具有大量实际背景的、新的研究课题。本项目的主要研究内容是：研究面向决策知识集成创新的多源知识融合的理论框架，提出多源决策知识预处理方法、多源决策知识融合的强化学习模型及求解算法；进一步地，开发多源决策知识融合子系统的原型系统，并以现实中若干典型复杂决策问题为背景，针对求解过程中对决策知识推送的需求，进行有针对性的理论框架、方法与应用研究。本项目的研究目的是：基于多源决策知识融合的视角，立足解决复杂决策问题求解时需要多源决策知识推送的问题，提出面向多源决策知识集成创新的基于强化学习的多源知识融合方法与技术，拟取得高水平的、创新性的研究成果，弥补目前已有的决策模型和决策方法在决策知识推送方面的不足。

中文关键词： 复杂决策；决策知识集成创新；决策知识推送；决策知识融合；强化学习

英文摘要： Multi-source knowledge fusion for decision knowledge integration and innovation is a new and practical research topic, which is proposed considering the research trends of theories and methodologies to resolve the complex decision-making problems and the practical requirements of decision knowledge push of decision makers. The research is structured as follows: firstly, a theoretical architecture of multi-source knowledge fusion for decision knowledge integration and innovation is developed; then, the knowledge pretreatment method is proposed, preparing knowledge for multi-source decision knowledge fusion; next, the reinforcement learning-based model and methods are proposed to realize decision knowledge integration and innovation; lastly, a multi-source decision knowledge fusion sub-system is developed, and the proposed theoretical architecture and methodologies are realized and evaluated in a typical and practical case study. The main objective of this research project is to propose the methodologies and technologies of reinforcement learning-based multi-source decision knowledge fusion to realize decision knowledge push in complex decision-making problems, from the perspective of decision knowledge fusion. Creative and original research results are supposed to be achieved. It is also to cover the shortages of

英文关键词： complex decision-making；Decision knowledge integration and innovation；Decision knowledge push；Decision knowledge fusion； Reinforcement learning