Search

Data-driven control offers an alternative to traditional model-based. Most present data-driven control strategies either involve model identification or need to assume availability of state information. In this work, we develop an input-output data-driven control method through dissipativity learning. Specifically, the learning of the subsystems' dissipativity property using one-class support vector machine (OC-SVM) is combined with the controller design to minimize an upper bound of the $L_{2}$ -gain. The data-driven controller synthesis problem is then formulated as quadratic-semidefinite programming with linear and multilinear constraints, solved via the alternating direction method of multipliers (ADMM). The proposed method is illustrated with a polymerization reactor.

Alternating direction method of multipliers (ADMM), as a powerful distributed optimization algorithm, provides a framework for distributed model predictive control (MPC) for nonlinear process systems based on local subsystem model information. However, the practical application of classical ADMM is largely limited by the high computational cost caused by its slow (linear) rate of convergence and non-parallelizability. In this work, we combine a recently developed multi-block parallel ADMM algorithm with a Nesterov acceleration technique into a fast ADMM scheme, and apply it to the solution of optimal control problems associated with distributed nonlinear MPC. A benchmark chemical process is considered for a case study, which demonstrates a significant reduction of computational time and communication effort compared to non-parallel and non-accelerated ADMM counterparts.

The effect of network topological features, specifically the modularity or the existence of communities in the network, on the control performance, remains an unexplored problem. In this work, we investigate the role of communities in feedback control in the framework of sparsity-promoting control and structured optimal control of a Laplacian dynamics defined on a network. We find that for modular networks, the community structures in the network correspond to feedback channels necessary for maintaining control performance, and result in an advantage in total control cost over non-modular networks as the cost of feedback channels becomes significant. For decentralized control, the best control performance is approached with a decomposition according to the communities. Thus, the communities act as the “core” of sparse feedback control. We then discuss the significance of this conclusion in understanding the behavior and evolution of networks as well as guiding the design of control strategies for large-scale and complex systems.

Subsea production system is a key technique in the development of oceanic oil/gas resources, which has experienced a long-time evolution. Control system is the core technique of subsea production system, which is indispensable in subsea production. We introduce eight typical control architectures of different structures, including direct hydraulic control system, piloted hydraulic control system, sequenced hydraulic control system, direct electro-hydraulic control system, multiplexed electro-hydraulic control system, all-electric control system, powered autonomous remote control system and integrated control buoy. We make a comparison of the above eight control techniques in terms of emerging background, constituents, application ranges, characteristics, advantages and disadvantages. Finally, the development trends of the subsea production system are presented.

Big Data is changing this world at a surprising speed, and MapReduce plays a critical role in finding insights in Big Data. However, to efficiently extract insights from Big Data, performance optimization of MapReduce applications is a challenging task. To facilitate the full-system optimization of MapReduce applications, we propose a correlation-based performance analysis approach to efficiently identify critical outliers. The basic intuition is that critical outliers are key to the overall performance and they can only be accurately identified by correlating different phases, tasks and resources. Based on the proposed approach, we further implement a correlation-based performance analysis tool, called Sonata. It can efficiently identify critical outliers, and then, recommend optimization suggestions for practitioners based on embedded rules. Since the performance overhead is key to the applicability of a performance tool, we conduct experiments to demonstrate that Sonata is a practical tool with less than 5% overhead and good scalability. To demonstrate the effectiveness of Sonata, we share several cases during the performance tuning of IBM Platform SymphonyTM with the help of Sonata.