基于多模态信息集成的组合预测模型及其应用研究

项目名称： 基于多模态信息集成的组合预测模型及其应用研究

项目编号： No.71501002

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 管理科学

项目作者： 刘金培

作者单位： 安徽大学

项目金额： 17.4万元

中文摘要： 信息技术的发展和现实生活中传感器的广泛应用为我们提供了大量的多模态信息。本项目基于多模态信息的多源性、跨媒体性、交互性、异构性等特点，研究不确定环境下基于多模态信息集成的组合预测模型。利用模式识别技术对多模态信息进行特征提取和加权属性图建模，获取多模态信息的聚类结果。在对多模态信息一致化处理和质量评价的基础上，考虑到信息之间的交互影响，提出一系列广义多模态信息集成算子，构建若干基于多模态信息集成的单项预测模型。对单项模型的预测结果进行更广义的组合，建立新的基于多模态信息集成的多目标组合预测优化模型。提出组合预测模型的有效解、满意解等概念，研究非劣性以及优性组合预测存在的有效性条件，同时设计组合预测模型的智能求解算法,并探讨在水污染预测中的应用。本项目通过数理论证和实证分析，不仅能完善基于多模态信息集成组合预测理论和应用的研究框架，而且可为相关部门的预测和决策分析提供理论依据和技术支持。

中文关键词： 组合预测；集成算子；多模态信息；多目标优化；有效性

英文摘要： With the development of information technology and the widely used sensors in our modern life, large amount of multimodal information is generated. This project develops combination forecasting model based on multimodal information aggregation in the uncertain environment according to the multi-resource, cross-media, interactivity and isomerism of the multimodal information. This project considers extracting the feature of multimodal information using the pattern recognition method and then modeling it by weighted attributed graph in order to cluster the multimodal information. Based on the uniformization and quality evaluation of the multimodal information and interactions among the information, several generalized multimodal information aggregation operators are proposed and single forecast models of multimodal information aggregation are constructed. The single forecast results are generalized to build novel combination forecast optimal model for multi objectives based on the multimodal information. Furthermore, the concepts of efficient solutions and satisfactory solutions are presented, and the corresponding optimality conditions for the non-inferior combination forecast and the superior combination forecast existing are given. Meanwhile, the application of the developed combination forecast model in water pollution forecast is discussed. This project not only improves the theory and application of the combination forecasting methods based on the multimodal information, but also provides the theory basis and technology support for the relevant forecasting and decision analysis problems.

英文关键词： combination forecasting;aggregation operator;multimodal information ;multi-objective optimization;effectiveness