经济模型预测控制的嵌入式优化算法

Engineers routinely use optimal control algorithms for designing and operating dynamic processes such as industrial production processes, energy generating devices, or state of the art transportation technologies. Here, the objective is to improve performance from an economic standpoint, in terms of its environmental impact, or its safety. State of the art algorithms are naturally challenged by new technologies, in particular in the field of renewable energy systems, whose dynamics become faster and faster. Due to their complexity, many of these advanced technology based processes cannot be operated or even tuned anymore by human beings and new types of intelligent computer based optimization and decision making codes are required. In particular, economic model predicitve control algorithms are becoming more and more important to support the online design and control of dynamic processes but also as an important tool for enforcing robustness and safety. The proposed project is about economic nonlinear model preditive control, a modern receding horizon based control technique that minimizes at each sampling instance an economic objective on a moving horizon subject to the model dynamics and, optionally, control and state constraints. We focus on the theory, numerical algorithms and software implementation analyzing (i) the interplay between design, performance and stability of economic model predicitive controllers and (ii) how a safe operation can be ensured by taking uncertain inputs into account. Moreover, a major goal of this project is to develop and implement open-source software for implementing real-time algorithms for robust and economic model predicitive control on embedded hardware. While the theory for standard tracking model predicitve control problems is well developed and there exist a wide variety of local optimization algorithms, there are at the current status only a few initial attempts to establish stability results for economic model predicitive control and there is basically no structure exploiting software available. The proposed project will be among the first attempts to establish a rigorous stability theory for economic model predictive control methods coming along with a mature software implementation.

工程师们经常使用最优控制算法来设计和操作动态过程。在这里，从经济角度来看，目标是从其环境影响或安全性方面来提升系统性能。由于它们的复杂性，许多基于这些先进技术的过程很难被操作或者甚至很难再由人为进行调整，所以产生了对新型的基于智能计算机的最优算法和决策代码的需求。特别的，为了支持对动态过程的在线设计和控制，经济模型预测控制算法正变得越来越重要。本项目主要研究经济非线性模型预测控制，旨在在一个受控于动力学模型和控制及状态约束的移动时域，在每个采样时刻最小化经济目标。我们将专注于理论、数值算法和软件实现来分析经济模型预测控制器的设计、性能和稳定性之间的相互作用，以及当考虑不确定输入时，如何保证一个安全的操作。此外，该项目的一个主要目标是开发和实现对在嵌入式软件上实现鲁棒和经济模型预测控制实时算法的开源软件。该拟建项目将对经济模型预测控制建立一个严格的稳定性理论，同时实现一套相关的成熟软件。

此次国家自然科学基金项目分别在实时优化算法、经济模型预测控制等领域取得了长足的进展与突破。在过去四年内，Boris Houska教授和他日益壮大的课题组一起，完成了一系列突破性贡献——建立了经济模型预测控制的稳定性、对实时控制系统的不确定性进行了分析。这些成果都已经在顶级国际期刊上发表，如《Automatica》和《IEEE Transaction on Automatic Control》。这份报告将侧重于总结这些研究成果，但与此同时也希望能强调因国家自然科学基金项目而得以开展的一系列研究活动所产生的影响力。尤其是在报告中展现Boris Houska教授和他在上海科技大学的课题组在组织国际会议和研讨会方面，以及在开展国际合作方面的成果。