Los Alamos National Labs with logo 2021

Robust Real-Time Control, Monitoring, and Protection of Large-Scale Power Grids in Response to Extreme Events

The protection and emergency control system for the U.S. transmission grid is one of the most crucial points of U.S. national security.

Contact us  

  • Phillip Top
  • Lawrence Livermore National Laboratory
  • (925) 422-1100
Fig. 1: Robust emergency control scheme

Fig. 1: Robust emergency control scheme

Project purpose

The protection and emergency control system for the U.S. transmission grid is one of the most crucial points of U.S. national security. An emergency occurs when the power system is close to or beyond secure operation limits, which are quantified in terms of feasibility/security of voltage magnitudes, frequency, active and reactive power transfer. Emergency control actions, in principle, are designed to move the power system from an insecure operating point to a secure operating point within desired time-frames and protect the system from cascading outages and blackouts. Current operational paradigms for emergency control primarily rely on relay protection devices and automated schemes such as under-voltage load shedding that are effective in restricting emergencies to local regions and preventing cascading outages.

To further improve system reliability and performance during maintenance scenarios or in extreme events, many utilities implement Remedial Action Schemes (RAS) or Special Protection Schemes (SPS). These schemes may significantly improve the response of the system to failures and are believed to enable better integration of renewable energy sources. However, while the benefits of SPS have led to widespread adoption, the potentially deteriorating effects of misoperation and unintended interference between different schemes can pose a serious risk to system operations.

This project aims at designing real-time extreme event monitoring and identification methods; and providing the computation and implementation of fast control actions in a similarly short time-frame as the traditional special protection schemes, but revisit the way those systems are designed. Specifically, we will consider a context where increased variability in the system state (due to, e.g., increased levels of renewable energy) and more frequent extreme events (such as snowstorms and tornadoes) lead to greater demand and diversity in the required control actions. To develop advanced automatic control schemes that are necessary to address this situation, we will leverage a limited set of non-local measurements provided by high-fidelity sensors such as Phasor Measurement Units (PMUs) and draw on recent advances in robust, stochastic and data-driven power system optimization and control.

Technical approach

First, we will focus on emergency control actions for situations where the system stabilizes at a short-term stable state after a big initiating event. In this situation, short-term stability implies that the system reaches a steady state, but is vulnerable to cascading overloads and long-term voltage instability. Hence, it is necessary to immediately identify and implement control actions that mitigate the risk and avoid further deterioration. In particular, we are interested in developing emergency control schemes that account for variability in the pre-event system state, as well as limited information about which parts of the system were impacted by the event. To tie our activities to the current state-of-the-art in the electric power industry, we will review existing methods to identify and implement emergency control actions. This will include a review of centralized security assessment methods, as well as the design of special protection and remedial action schemes. Our goal is to develop algorithms that inherit the flexibility of centralized algorithms while requiring fewer measurements and acting as fast as special protection schemes. We will do this by utilizing recent developments in robust optimization and robust parameter estimation. A key aspect of our work will be to incorporate uncertainty about system topology, and where to invest in measurement capabilities such as pastor measurement units (PMUs) to reduce it.

Second, we focus on augmenting such control polices with the knowledge of load and generator dynamics. Knowledge of grid dynamics, possibly non-linear, can ensure that the prescribed control policies are achievable in the post-event scenario and do not lead to further system deterioration. However, information on individual and aggregated dynamics of grid equipment is seldom accurately available, more so for load equipment. Hence, we will strategize the optimal use of streaming PMU data of high fidelity to learn approximate dynamic models and use them for augmenting control design. On this way, we investigate data-driven reduced-order models for load aggregations and system dynamics (using advanced kernel methods and neural networks), and non-linear controls (using Kernel methods and Koopman operator theory). We plan to design data-driven grid monitoring and failure prediction methods based on adaptive importance sampling and neural networks trained over PMU data. 

To validate, calibrate, and test our emergency control propositions, we will utilize and investigate an open-source DOE supported software suite GridDyn. The modular, flexible, and open nature of GridDyn allows tighter integration of the control schemes and much more flexible modeling than could be achieved with commercial tools. Integration with HELICS allows testing and integration with other components such as communication systems and other types of models. 

Plans and results

In 2020 we plan to develop a formulation and initial implementation of optimal power flow-based emergency control with the system model uncertainty. We also plan to design a reduced (machine-learning) based model to describe the non-linear load/generator behavior based on the PMU data and determine feasible operating conditions.  We suggest to merge models from Modelica into the dynamic module in GridDyn and demonstrate dynamic and steady-state functionality with control modules on appropriate test cases.

 In 2021 we plan to formulate augmented emergency control algorithms that utilize partial information about system state and parameters (as obtained through system estimation and PMU measurements) to reduce uncertainty and conservativeness. We propose to design non-linear emergency control methods with the incorporation of equipment and grid dynamics, using data-driven Kernel and Koopman operators. To support our theoretical investigations, we plan to develop a stable and scalable GridDyn test environment with verifiable performance of linear emergency control policies and measurement collection. In 2022 we will integrate algorithms developed earlier with GridDyn. And will use simulations and data collection to identify possible improvements/augment control policies designed earlier.

Team and Biographies

Yury Maximov (PI) is a computer scientist working on power systems, high-dimensional statistics, empirical process theory, machine learning and optimization methods behind them. Yury obtained his Ph.D. in Math from Moscow Institute of Physics and Technology in 2013. During his study, Yury spent several years in the industry as a software engineer and project manager working on industry-grade machine learning, information retrieval and business intelligence solutions. He joined LANL in October 2016 as a postdoc in T-4 and was converted to a scientist in T-5 in March~2018. At LANL he is a PI of the DOE AGM “Robust Real-Time Control, Monitoring, and Protection of Large-Scale Power Grids in Response to Extreme Events” and DOE GMLC “Emergency Monitoring and Control Through New Technologies and Analytics” project. He is an author of more than 20 peer reviewed publications including the top ML/statistical venues such as NIPS, ICML, JAIR, EJOS, IJCAI etc.

 Deepjoyti Deka is a staff scientist at Los Alamos National Laboratory, where he was previously a postdoctoral research associate at the Center for Nonlinear Studies. His research interests include the design and analysis of power grid structure, operations and data security, and modeling and optimization in social and physical networks. At LANL, Dr. Deka serves as a co-principal investigator for DOE-GMLC project on machine learning in distribution systems and an LDRD project in cyber-physical security. Before joining the laboratory, he received the M.S. and Ph.D. degrees in Electrical Engineering from the University of Texas, Austin, TX, USA, in 2011 and 2015, respectively.

 Wenting Li is a postdoctoral researcher at Los Alamos National Laboratory, where she joined Applied Mathematica and Plasma Physics group (T-5) and the Center of Nonlinear Studies. She received her Ph.D. in electric engineering in 2019 at Rensselaer Polytechnic Institute, Troy, NY, and the B.E. degree from Harbin Institute of Technology, Heilongjiang, China, in 2013. Her research interests include high-dimensional data analytics, feature extraction, application of machine/deep learning methods to identify and locate events in power grids.

 Additional contributors: Philip Top (LLNL, Lab+1 PI), Prof. Line Roald (University of Wisconsin, Madison), Prof. Audun Botterud (MIT, Massachusetts Institute of Technology)

  1. Li, W., Deka, D., Chertkov, M., & Wang, M. (2019). Real-Time Faulted Line Localization and PMU Placement in Power Systems Through Convolutional Neural Networks. IEEE Transactions on Power Systems, 34(6), 4640-4651
  2. Owen, A.B., Maximov, Y. and Chertkov, M., 2019. Importance sampling the union of rare events with an application to power systems analysis. Electronic Journal of Statistics, 13(1), pp.231-254.
  3.  Artem Mikhalev, Alexander Emchinov, Samuel Chevalier, Yury Maximov, Petr Vorobev. A Bayesian Framework for Power System Components Identification. IEEE PES General Meeting, 2020.
  4. Nikolay Stulov, Dejan J Sobajic, Yury Maximov, Deepjyoti Deka, Michael Chertkov. Learning a Generator Model from Terminal Bus Data. Power Systems Computation Conference 2020.
  5. I Mezghani, S Misra, D Deka. Stochastic AC optimal power flow: A data-driven approach. Power Systems Computation Conference, 2020.
  6. N Stulov, D J Sobajic, Y Maximov, D Deka, M Chertkov. Learning a Generator Model from Terminal Bus Data. Power Systems Computation Conference, 2020.
  7.  Artem Mikhalev, Alexander Emchinov, Samuel Chevalier, Yury Maximov, Petr Vorobev. A Bayesian Framework for Power System Components Identification. IEEE PES General Meeting, 2020.