Optimization and Control in Large-scale Uncertain Processes

点击翻译按钮获取中文摘要

Technological advances on sensing equipment, communication networks and computational hardware have contributed to a dramatic increase in the availability and storage of data and the transmission of information both within and between the social, commercial, financial and industrial sectors. The interaction between information management systems, data transmission networks, and systems in the physical world is indeed becoming more and more prevalent in various application domains.          Data is not only increasingly available, there is also an urgent need for its careful exploitation. There is a growing need for the development of data-based methodologies for system automation and optimization. With the advent of recent developments in learning-based controller design techniques, such as reinforcement learning (RL), iterative learning control (ILC) and extremum-seeking control (ESC), control experts are now equipped with a growing set of model-free controller design techniques that can solve complex control problems in the absence of exact knowledge of process dynamics. Despite their obvious commonalities, leading model-free learning control and optimization techniques have evolved almost independently of each other.          The proposed research program tackles three themes focussed on the application of model-free techniques. The first theme considers the development of ESC techniques for the solution of real-time optimization problems in complex systems. The second theme considers the design of distributed optimization systems operated over unreliable peer-to-peer communication networks subject to unknown dynamics and uncertain communication network structure. The last theme considers the synergy of learning techniques and model-free control techniques. The goal is to expand the scope of application of these data-driven techniques by exploiting the generality of learning approaches and the stability and convergence certificates of leading model-free techniques such as ESC.           ESC is now widely used in industry and enjoys a growing list of applications in diverse challenging fields from electron microscopy to remote sensing in space exploration. The research proposed in this program further extends the applicability of ESC techniques to tackle large-scale systems with complex dynamics. The use of learning technology further expands the application of these techniques to achieve optimal performance in systems operating in uncertain environments. Of particular interests are applications in the area of energy efficiency, energy utilization and storage in large-scale systems.          The research program proposes significant contributions in control engineering and learning technology that are directly relevant in industrial applications. Three PhD and two MSc are to be trained in this program. It provides a unique opportunity for the training of HQP in the fields of control and optimization that are strategic to industry.