CAREER: Theory and Application of Hybrid Estimation

从故障检测和隔离到目标跟踪和识别，从生物医学信号处理到复杂系统分解，混合估计作为一种强大的鲁棒自适应方法，近年来在解决结构和/或参数不确定性问题方面取得了巨大成功。混合估计目前存在的瓶颈问题是，现有的技术对于大多数现实世界中涉及大量模式的混合估计问题来说，成本效益不够。进一步发展混合评估的最严重障碍与现有技术局限于固定结构这一事实有关。换句话说，现有的混合估计有一个基本的限制，固有于固定结构，因为它不能有效地处理涉及多个模式的问题，不幸的是，这是大多数现实世界问题的情况。这种限制源于现有技术的基本假设，即任何时候的系统模式都可以由一组可以先验地确定的小而固定的模型中的一个来精确地表示。因此，现有的固定结构的混合估计已经到了这样一个阶段，即在其固定结构内无法再期望有很大的改进。然而，现有的研究工作几乎完全局限于这种固定结构。本文对混合估计技术的理论和应用进行了研究。提出的理论研究包括开发几种有前途的互补变结构算法，这些算法具有很高的成本效益，简单且足够通用，可以适用于大多数混合估计问题，并且一些工作朝着混合估计的统一理论发展。所提出的研究将采用的技术方法是在混合估计中使用可变结构：模型集，而不是模型本身，可以根据测量序列进行自适应，其中包含有关当前有效的系统模式的有价值信息。这是PI最近发起的对现有技术进行重大改进的一个有希望的新方向。PI最近开发的这种算法证明了它的巨大潜力，该算法简单，普遍适用，并且比最先进的固定结构混合估计器更具成本效益。预计该项目将对估算、决策和识别及其应用领域的理论和应用研究产生重大影响。从短期来看，它将显著推进最先进的混合估计技术，从长远来看，它将为随机系统理论、随机信号处理和统计推断的未来研究提供洞察力和推动力。教育和人力资源发展是拟议的职业发展计划的组成部分。该计划的主要教育目标是培养学生发展科学探究和技术发现与发明的能力、知识和技能，从而为他们从事工程职业做好准备。这是通过以下方式实现的：1)在PI的创新和密切指导下，为学生提供参与高级研究项目的机会；2)开设新的研究生和本科课程，更好地教育和培养学生；3)通过编写教材的方式，将研究成果传播给更多的学生

9734285LiHybrid estimation, as powerful robust adaptive approach, has found great success recently in solving many problems compounded with structural and/or parametric uncertainty, ranging from fault detection and isolation to target tracking and recognition, and from biomedical signal processing to complex system decomposition. The bottle-neck problem currently exists with hybrid estimation is that existing techniques are not cost-effective enough for most real-world hybrid estimation problems, which involve a large set of modes. The most serious barrier for the further development of hybrid estimation is associated with the fact that existing techniques are confined to a fixed structure. In other words, existing hybrid estimation has a fundamental limitation, inherent to the fixed structure, in that it cannot handle effectively problems involving more than just a few modes, which is, unfortunately, the case for most real-world problems. This limitation stems from the fundamental assumption of the existing techniques that the system mode at any time can be represented accurately by one of a small and fixed set of models that can be determined a priori. As a result, existing hybrid estimation with a fixed structure has arrived at such a stage that great improvement can no longer be expected within its fixed structure. Nevertheless, existing research effort is confined almost entirely to this fixed structure. In this proposal, an investigation of the theory and application of hybrid estimation techniques is proposed. The theoretical research proposed includes the development of several promising and complementary variable structure algorithms that are highly cost-effective, uncomplicated and general enough to be applicable to most hybrid estimation problems, and some work toward the development of a unified theory of hybrid estimation.The technical approach that will be undertaken for the proposed research is to use variable structure in hybrid estimation: The set of models, rather than the models themselves, may be made adaptive based on measurement sequences, which carries valuable information about the system mode currently in effect. This is a promising new direction for improving significantly on the existing techniques, which was initiated by the PI recently. Its great potential has been demonstrated by the recent development of such an algorithm by the PI that is uncomplicated, generally applicable, and significantly more cost-effective than the state-of-the-art fixed structure hybrid estimators. This project is expected to have a significant impact on the theoretical and applied research in estimation, decision and identification, and their application areas. In the short term it will advance significantly the state-of-the-art hybrid estimation techniques, and in the long term, it will provide insight and thrust to future investigation in stochastic systems theory, random signal processing and statistical inference.Education and human resource development are integrated components of the proposed career development plan. The primary education objective of this plan is to train students to develop ability, knowledge, and skills for scientific inquiry and technological discovery and invention, hence preparing them well for their pursuit of an engineering career. This is achieved by the following: 1) Provide students opportunities to participate in advnced research projects under innovative and close guidance of the PI; 2) Develop new graduate and undergraduate courses for better educating and training students; 3) Disseminate results from the proposed reseach by textbook writing to reach out to more students.***