ABSTRACT

The grid is undergoing a profound transformation with increased penetration of renewables, the adoption of storage and electrification of transportation, and the rise of connected consumers that produce electricity and shape their consumption and the fast adoption of novel power control and monitoring technologies, among various other major trends. These trends are dramatically increasing the complexity of managing the grid by interconnecting networks (e.g., transportation, power, water), increasing variability and uncertainty in production and consumption, and creating more granular and real-time feedback loops in the system. At the same time, systems of the future will generate massive amounts of different types of data, and offer opportunities to support decision making at multiple time and spatial scales. In this talk we address how to utilize this data to map and reconstruct the electric power grid with high resolution. We combine a variety of approaches, including computer vision applied to satellite imagery, graphical models applied to power measurements, and natural language processing applied to building information to obtain detailed and dynamic resource maps that include significantly more assets than existing data collection efforts. We conclude the presentation demonstrating how such high resolution data can reveal critical system trends and highlight the disparity of the impact of new technologies in the grid.

SPEAKER

Ram Rajagopal is an Associate Professor of Civil and Environmental Engineering at Stanford University, where he directs the Stanford Sustainable Systems Lab (S3L), which is focused on large-scale monitoring, data analytics, and stochastic control for infrastructure networks, in particular, power networks. His current research interests in power systems are in the integration of renewables, smart distribution systems, and demand-side data analytics. He holds a Ph.D. in Electrical Engineering and Computer Sciences and an M.A. in Statistics, both from the University of California, Berkeley, a Masters in Electrical and Computer Engineering from University of Texas, Austin and a Bachelors in Electrical Engineering from the Federal University of Rio de Janeiro. He is a recipient of the NSF CAREER Award, a Powell Foundation Fellowship, the Berkeley Regents Fellowship, and the Makhoul Conjecture Challenge award. He holds more than 30 patents and several best paper awards from his work and has advised or founded various companies in the fields of sensor networks, power systems, and data analytics.

ABSTRACT

Inverter-based resources such as solar and storage provide us with more flexibility in the control of power systems. Through their power electronic interfaces, complex control functions can be implemented to quickly respond to changes in the system. Recently, reinforcement learning has emerged as a popular method to find these nonlinear controllers. The key challenge with a learning-based approach is that stability and safety constraints are difficult to enforce on the learned controllers. Using a Lyapunov theory-based approach, we show how to explicitly engineer the structure of neural network controllers such that they guarantee system stability. The resulting controllers only use local information and outperform conventional droop as well as strategies learned purely by using reinforcement learning.

SPEAKER

Baosen Zhang is the Keith & Nancy Rattie Endowed Career Development Professor in the Department of Electrical and Computer Engineering at the University of Washington. He received his undergraduate degree in engineering science from the University of Toronto in 2008; and the Ph.D. degree in Electrical Engineering and Computer Science from the University of California at Berkeley in 2013. Before joining UW, he was a postdoctoral scholar at Stanford University. He has received the NSF CAREER Award, as well as a number of best paper awards.

ABSTRACT

In this talk, we introduce methods that remove the barrier for applying neural networks in real-life power systems, and unlock a series of new applications. First, we introduce a framework for (i) verifying neural network behavior in power systems and (ii) obtain provable worst-case guarantees of their performance. Up to this moment, neural networks have been applied in power systems as a black-box; this has presented a major barrier for their adoption in practice. Using a rigorous framework based on mixed integer linear programming, our methods can determine the range of inputs that neural networks classify as safe or unsafe; and, when it comes to regression neural networks, our methods allow to obtain provable worst-case guarantees of the neural network performance. Such methods have the potential to build the missing trust of power system operators on neural networks, and unlock a series of new applications in power systems and other safety-critical systems. Second, we present a framework for physics-informed neural networks in power system applications. Exploiting the underlying physical laws governing power systems, and inspired by recent developments in the field of machine learning, we propose a neural network training procedure that can make use of the wide range of mathematical models describing power system behavior, both in steady-state and in dynamics. Methods such as the ones we will discuss in this talk unlock the potential of neural networks to perform power system tasks at extremely fast computing times while maintaining verified accuracy.

SPEAKER

Spyros Chatzivasileiadis is an Associate Professor at the Technical University of Denmark (DTU) and the Group Leader of the Electric Power Systems Group at the Center for Electric Power and Energy at DTU. Before that he was a postdoctoral researcher at the Massachusetts Institute of Technology (MIT), USA and at Lawrence Berkeley National Laboratory, USA. Spyros holds a PhD from ETH Zurich, Switzerland (2013) and a Diploma in Electrical and Computer Engineering from the National Technical University of Athens (NTUA), Greece (2007). He is currently working on machine learning applications for power systems, and on power system optimization, dynamics, and control of AC and HVDC grids. Spyros is the recipient of an ERC Starting Grant in 2020.

ABSTRACT

Deep learning systems have achieved state of the art performance in many challenging, nonlinear control problems, yet has found relatively limited application in safety critical systems. In this talk, I will present a number of paradigms for incorporating hard constraints within deep learning systems, using the frameworks of differentiable optimization and projection layers. In particular, I will highlight how we can use these techniques to apply “standard” deep learning methods to inverter and microgrid control systems, improving upon the performance of classical control methods, while still obeying the typical constraints of robust control.

SPEAKER

Zico Kolter is an Associate Professor in the Computer Science Department at Carnegie Mellon University, and also serves as chief scientist of AI research for the Bosch Center for Artificial Intelligence. His work spans the intersection of machine learning and optimization, with a large focus on developing more robust and rigorous methods in deep learning. In addition, he has worked in a number of application areas, highlighted by work on sustainability and smart energy systems. He is a recipient of the DARPA Young Faculty Award, a Sloan Fellowship, and best paper awards at NeurIPS, ICML (honorable mention), IJCAI, KDD, and PESGM.

ABSTRACT

Policy optimization is a key ingredient of modern reinforcement learning, and can be used for efficient design of optimal architectures and feedback schemes in engineering systems, with implicit on-line learning. In the design of feedback schemes, certain constraints are generally enforced on the policies to be implemented, such as stability, robustness, and/or safety concerns on the closed-loop system (that is, after the feedback loop is closed). Hence, policy optimization entails, by its nature, a constrained optimization in most cases, which is also nonconvex, and partly because of that, analysis of its global convergence is generally very challenging. Further, another element that compounds the difficulty is that some of the constraints that are safety-critical can be difficult to enforce on the system while being learned as the policy optimization methods proceed. We have recently overcome this difficulty for a special class of such problems, by introducing two policy optimization algorithms featuring what we call the “implicit regularization” property, which I will discuss in this presentation, while also placing this in a broader multi-agent context, including coping with adversarial interventions. I will also discuss extensions to the model-free framework and associated sample complexity analyses, and potential applications in power systems and networks, where cyber-secure, robust, and resilient operations are called for.

ABSTRACT

This talk reviews a number of applications of differential privacy (DP) for energy systems. It reviews DP mechanisms for the obfuscation of power systems to release of realistic test cases, the optimal power flow of distribution systems, and the market clearing of sequential energy markets. The talk also reviews the potential of deep learning in speeding these applications.

ABSTRACT

Signals from physical infrastructures, such as the grid, have structural properties that are naturally governed by and can be explained using the laws of physics. This talk suggests how to view those physical laws as instances of a Graph Signal Processing (GSP) signal, in a way that unveils a series of statistical and algebraic properties of the signal and that opens the door for applying GSP to develop machine learning algorithms using a parametric Bayesian framework. In this talk we introduce the algorithmic foundations of GSP modeling for the analysis of signals and multivariate-time series whose support is a graph and unveil how power systems modeling supports the notion that the grid voltage phasors are a low-pass graph signal process. Having established this connection we explore how to use GSP modeling and Graph Fourier Transforms (GFTs) allows directly to derive sampling schemes, denoising, interpolation, general classification and system identification problems with application to data injection attacks, as well as compression of grid data.

SPEAKER

Anna Scaglione (M.Sc.’95, Ph.D. ’99) is currently a Professor of Electrical, Computer and Energy Engineering at Arizona State University. Scaglione’s expertise is in the broad area of statistical signal processing with application to communication networks, electric power systems/intelligent infrastructure and network science. Dr. Scaglione was elected an IEEE fellow in 2011. She is the recipient of the 2000 IEEE Signal Processing Transactions Best Paper Award, the 2013, IEEE Donald G. Fink Prize Paper Award for the best review paper in that year among all IEEE publications. Also, her work with her student earned the 2013 IEEE Signal Processing Society Young Author Best Paper Award (Lin Li) and several best conference paper awards. She was SPS Distinguished Lecturer for the years 2019-2020 and is the recipient of the 2020 Technical Achievement Award from the IEEE Communication Society Technical Committee on Smart Grid Communications

ABSTRACT

Residential loads have great potential to enhance the efficiency and reliability of electricity systems via demand response (DR) programs. One major challenge in residential DR is to handle the unknown and uncertain customer behaviors, which is further influenced by time-varying environmental factors. In this talk, we study automated control method for regulating air conditioner (AC) loads in residential demand response (DR) by modeling it as multi-period stochastic optimization and learning problem. Machine learning techniques including Gaussian process and logistic regression are employed to learn the unknown thermal dynamics model and customer opt-out behavior model, respectively. Based on the Thompson sampling framework, we propose an online DR control algorithm to learn customer behaviors and make real-time AC control schemes. This algorithm considers the influence of various environmental factors on customer behaviors and is implemented in a distributed fashion to preserve the privacy of customers. Numerical simulations demonstrate the control optimality and learning efficiency of the proposed algorithm.

SPEAKER

Na Li is a Gordon McKay professor in Electrical Engineering and Applied Mathematics at Harvard University. She received her Bachelor degree in Mathematics from Zhejiang University in 2007 and Ph.D. degree in Control and Dynamical systems from California Institute of Technology in 2013. She was a postdoctoral associate at Massachussetts Institute of Technology 2013-2014. Her research lies in control, learning, and optimization of networked systems, including theory development, algorithm design, and applications to real-world cyber-physical societal system. She received NSF career award (2016), AFSOR Young Investigator Award (2017), ONR Young Investigator Award(2019), Donald P. Eckman Award (2019), McDonald Mentoring Award (2020), along with some other awards.

ABSTRACT

In the power grid optimization literature, one often finds clean problem formulations with continuous decision variables and deterministic data. Reality is different. Specifically, this talk focuses on two tough problems my research lab has recently faced: (i) large-scale mixed integer programs, and (ii) power pricing and scheduling in the context of human choices. Specifically, large-scale mixed integer programs arise when managing large-populations of distributed energy resources with binary (on/off) control. We present a novel (yet historic) heuristic solution known as Hopfield methods. The problem of human choices in-the-loop is fundamental to our current Smart LeaRning Pilot for EV charging stations (SlrpEV). Specifically, we present a menu of differentiated charging service options to EV drivers, and optimize the pricing and charge scheduled based on their preferences, to maximize the operator’s net profit. I close the talk with perspectives on tough problems that deserve increased attention for realizing sustainable and resilient power grids of the future.

SPEAKER

Scott Moura is the Clare and Hsieh Wen Shen Endowed Distinguished Professor in Civil & Environmental Engineering and Director of the Energy, Controls, & Applications Lab (eCAL) at the University of California, Berkeley. He received the B.S. degree from UC Berkeley and the M.S. and Ph.D. degrees from the University of Michigan, Ann Arbor, in 2006, 2008, and 2011, respectively. From 2011 to 2013, he was a Post-Doctoral Fellow at the Cymer Center for Control Systems and Dynamics, University of California, San Diego and a Visiting Researcher at the Centre Automatique et Systèmes, MINES ParisTech in 2013. His research interests include control, optimization, and machine learning for batteries, electrified vehicles, and distributed energy resources.

ABSTRACT

We do research on leveraging machine learning to improve assessment tools and optimization methods to support decision making in the context of bulk power systems reliability management. One of the challenges is the large-scale nature of these problems, and to solve them there are many opportunities to develop more effective approaches that would combine prior physical knowledge with state-of-the-art optimization and machine learning methodologies. We present some research carried out along this topic in our group at the Montefiore Institute.

SPEAKER

Louis Wehenkel, Professor of EECS at the University of Liège graduated in Electrical Engineering (Electronics) in 1986 and received the Ph.D. degree in 1990, both from the University of Liège (Belgium), where he is full Professor of Electrical Engineering and Computer Science. His research interests lie in the fields of stochastic methods for modeling, optimization, machine learning and data mining, with applications in complex systems, in particular large scale power systems planning, operation and control, industrial process control, bioinformatics and computer vision.

ABSTRACT

A demand response program motivates residential customers to shift electricity consumption to off-peak hours, thereby reducing overall peak load. In this talk, I present a demand response algorithm that accounts for uncertainty in load demands and electricity price, addresses customer privacy concerns, and satisfies distribution network and operational constraints. The algorithm casts the load scheduling problem as a Markov decision process and solves it using deep reinforcement learning along with the actor-critic method. Additionally, federated learning enables customers to determine neural network parameters in a decentralized manner without revealing private information like their load demands and discomfort costs. Even so, the resulting scheduled loads may not yield a power flow solution that satisfies all network and operational constraints. To overcome this challenge, we formulate and solve an optimization problem that projects the potentially constraint violating scheduled loads onto the space of feasible ones. Performance evaluation via numerical simulations demonstrates the following benefits of the algorithm: (i) it converges to the same solution as the corresponding centralized load control problem with full information, (ii) it is scalable to large test distribution systems with real-time pricing, and (iii) the load scheduling policy updates to satisfy operational constraints.

SPEAKER

Christine Chen is an Assistant Professor with the Department of Electrical and Computer Engineering at The University of British Columbia. Her research interests include power system analysis, monitoring, and control. She received the B.A.Sc. degree in engineering science from the University of Toronto in 2009, and the M.S. and Ph.D. degrees in electrical engineering from the University of Illinois at Urbana-Champaign in 2011 and 2014, respectively. Christine is a recipient of the 2017–2018 Best Paper Award from the IEEE Transactions on Energy Conversion. She serves on the editorial boards of the IEEE Transactions on Power Systems and the IEEE Transactions on Energy Conversion.