Smart Power and CPS WorkshopSmart Power and Cyber-physical Systems Theme: With fast-growing adaptability and integration with computing technologies, planning and operation of cyber-physical systems (CPSs) are facing new challenges of uncertainty, security, and scalability. Cutting-edge techniques in optimization, control, and big data analytics are utilized to unlock the considerable potential of those CPSs with improved system efficiency and resilience. In this context, prospective speakers of the workshop session are invited to present recent advancements that leverage modern tools such as decentralized control, online optimization, deep learning, and Internet of things (IoT) for the modeling and analysis of CPSs including e.g., smart power grids, intelligent transportation, geo-distributed data centers, communication networks, and networked robotic systems. Aiming at promoting cross-fertilization among researchers, the topics of interests include, but are not limited to, the following aspects:
Agenda (Introdution slides) 1:30-2:45pm — Session I Ming Jin, Postdoc Scholar, IEOR UC Berkeley Title: Stability-certified Smooth Reinforcement Learning: A Control-theoretic Perspective Abstract: It is critical to obtain stability certificate before deploying reinforcement learning in real-world mission-critical systems. In this talk, I will justify the intuition that smoothness (i.e., small changes in inputs lead to small changes in outputs) is an important property for stability-certified reinforcement learning from a control-theoretic perspective. The smoothness margin can be obtained by solving a feasibility problem based on semi-definite programming for both linear and nonlinear dynamical systems, and it does not need to access the exact parameters of the learned controllers. Numerical evaluation on nonlinear and decentralized frequency control for large scale power grids demonstrates that the smoothness margin can certify stability during both exploration and deployment for (deep) neural-network policies, which substantially surpass nominal controllers in performance. The study opens up new opportunities for robust Lipschitz continuous policy learning.
Yihsu Chen, Associate Professor, UCSC Title: A Power Market Model in Presence of Strategic Prosumers (joint work with Sepehr Raymar) Abstract: Prosumers, with ability to act both as a supplier and a consumer in a power market, have received considerable attention recently. Possessing with distributed energy resources, their capability to operate in an isolated mode, shielding from the main grid, has also been promoted as a vital option to enhance the power system's resilience. One emerging concern is the prosumer's ability to manipulate the power market as a buyer or as a seller. This talk discusses the preliminary results of a study that vets the outcomes of a power market in presence of strategic prosumers. We posit a situation where a strategic prosumer owns a renewable unit with variant output and a dispatchable backup unit and participate in a competitive market. The prosumer is assumed to maximize its benefit by deciding amount of power to buy from or sell into the main grid, amount of renewable power to forego consumption, and amount of power to produce from backup unit. The model is applied to a simple case study as an illustrative example.
Ram Rajagopal, Associate Professor, Stanford Title: Online Learning and Optimization in Distributed Energy Systems Abstract: Distributed Energy Resources are being progressively adopted by consumers and are significantly changing how we monitor and optimize power distribution networks. Some major challenges are learning accurate models of distribution networks to incorporate in system management, optimizing resource coordination under realistic data ownership and communication constraints or in a model-free manner and deciding where to place resources under realistic system models. In this talk, we introduce various formulations inspired by practical applications and real data and review algorithms that address them. We list various open algorithm design and performance analysis questions related to these problems.
2:45-3:15pm — BREAK, Refreshments and Discussion 3:15-4:30pm — Session II Ruoxi Jia, Postdoc, EECS UC Berkeley Title: Accountable Data Analysis for Cyber-Physical Systems Abstract: With the deployment of large sensor-actuator networks, Cyber-Physical Systems (CPSs), such as smart buildings, smart grids, and transportation systems, are producing massive amounts of data often in different forms and quality. These data are in turn being used collectively to inform decision-making of the entities that engage with the CPSs. The impact of these systems on people's lives has led to a strong call for accountability of system decisions made based upon various data sources. In this talk, I will discuss a principled way to characterize the value of different data sources for any given data-enabled decisions or services, and provide efficient algorithms for data valuation. This not only enables us to better understand black-box predictions through the lens of training data but allows for the fair allocation of the profit generated from a model that is built with data from cooperative entities. We use the proposed data value notion to develop an effective data sanitization mechanism, which can effectively screen off low-quality or even adversarial data instances from the training set.
Atif Maqsood, PhD Student, ECE UCSC Title: Optimal Rotational Load Shedding via Bilinear Integer Programming Abstract: The focus of this talk will be to address the problem of managing rotational load shedding schedules for a power distribution network with multiple load zones. An integer optimization problem is formulated to find the optimal number and duration of planned power outages. Various types of damage costs are proposed to capture the heterogeneous load shedding preferences of different zones. The McCormick relaxation along with a simple feasibility recovery procedure is developed to solve the complicated bilinear integer program, which yields a high-quality suboptimal solution. Extensive simulation results corroborate the merit of the proposed approach, which has a substantial edge over some existing load shedding schemes. During introduction I will go over the concept of load shedding and why it is implemented in some parts of the developing world. Through the specific example of LESCO (Lahore electric supply company) I will demonstrate the expanse of this problem and how the solution presented in this talk can help managing the distribution of power in a most cost-effective way.
Hongcai Zhang, Postdoc Scholar, Department of Civil & Environmental Engineering, UC Berkeley Title: Data-driven Approach to Harnessing Flexibility of Large-scale Distributed Energy Resources for Power Grids Abstract: Distributed energy resources (DERs) such as battery storage systems, electric vehicles, and shapeable loads, promise the indispensable ability of providing flexibility services to power systems. In order to access some lucrative grid service markets, e.g., regulation market, a large population of small-capacity DERs usually have to jointly participate the market under the coordination of an aggregator. Because of DERs’ large-population, stochasticity and heterogeneity, the aggregator is faced with three difficult challenges to construct a profit-maximizing bid: 1) Accurately evaluate DERs’ flexibility ahead-of-time; 2) Optimize large-scale DERs’ power schedules with affordable computational capability; and 3) Realize accurate & efficient online control. In this talk, I will first propose a data-driven approach for aggregate modeling of large-scale DERs. Then, I will discuss how we can effectively tackle the aforementioned challenges based on the proposed aggregate model combined with distributionally robust optimization and heuristic laxity-based controller.
|