Anytime algorithms for estimation-based model predictive control

在数字化不断推进的过程中，对大量测量数据进行系统、高效的分析，特别是利用基于优化的估算程序，是未来技术过程自动化与控制领域的重大挑战之一。特别是，这适用于现代基于模型的预测控制方法，在许多情况下，需要估计当前系统状态和/或其他过程参数的测量，以预测和优化系统行为。然而，尽管具有巨大的概念和实际意义，但到目前为止，还没有任何方法可以将动态在线测量在理论上合理且在数值上有效地集成到基于模型的预测控制方案中。因此，本课题的研究动机和基本思想是设计和开发一类新的基于估计的模型预测控制程序，该程序将相关状态信息和过程参数的估计在一个基于集成和优化的框架中与预测控制方案相结合。特别是，总体目标是为所谓的任何时间算法的设计提供创新和系统理论上健全的方法，即使在底层优化算法仅执行有限次数迭代的情况下，例如由于大量测量或有限的计算资源，也可以确保重要的稳定性和实时保证估计和控制。

In the course of the continuously advancing digitization, the systematic and efficient analysis of a big amount of measurement data, particularly by means of optimization-based estimation procedures, is one of the big future challenges in the field of automation and control of technical processes. In particular, this applies to modern model-based predictive control approaches, which in many cases require to estimate the current system state and/or other process parameters from measurements in order to predict and optimize the system behavior. However, despite the enormous conceptual and practical relevance, there exist up to now no approaches that allow for a systems theoretically sound and numerically efficient integration of dynamical on-line measurements into model-based predictive control schemes. Motivation and basic idea of this research project is therefore the design and development of a novel class of estimation-based model predictive control procedures, which combine the estimation of relevant state information and process parameters in an integrative and optimization-based framework with a predictive control scheme. In particular, the overall goal is an innovative and systems theoretically sound methodology for the design of so-called anytime algorithms, which allow to ensure important stability and real-time guarantees of estimation and control even in cases where the underlying optimization algorithms perform only a limited number of iterations, for example due to big amounts of measurements or limited computational resources.