High Performance Adaptive Deep-Reinforcement-Learning-based Real-time Emergency Control (HADREC) to Enhance Power Grid Resilience in Stochastic Environment

Default ARPA-E Project Image


Program:
OPEN 2018
Award:
$3,500,000
Location:
Richland, Washington
Status:
ALUMNI
Project Term:
08/16/2019 - 11/15/2022
Website:

Critical Need:

Significant investment and efforts are devoted to hardening grid infrastructures in the U.S. Preventive control measures, such as out-of-merit generation dispatch, have been widely adopted to ensure adequate security margins. However, on any given day, about 500,000 customers are without power for 2 hours or more. Several large blackouts have occurred in the U.S. in the last 20 years. A recent study estimated that annual economic loss from all U.S. outages is between $30–50 billion. These statistics suggest that current preventive measures are inadequate. Emergency control is imperative in real-time operation to minimize the occurrence and impact of power outages and widespread blackouts.The challenge is that existing analytical-based technologies cannot provide effective emergency control actions to operators in real time based on actual system conditions. Due to time constraints, most control actions are predefined at least one day ahead and depend heavily on operators to identify the causes and take proper action. Consequently, these control actions tend to be overly conservative and inefficient.

Project Innovation + Advantages:

Pacific Northwest National Laboratory (PNNL) will construct an intelligent, real-time emergency control system to help safeguard the U.S. electric grid by providing effective and fast control actions to system operators in response to large contingencies or extreme events. PNNL’s scalable platform will utilize advanced machine learning techniques (deep-meta-reinforcement learning) as well as high-performance computing to automatically provide effective emergency control strategies seconds after disturbances or attacks. Platform development will focus on the determination, timing, coordination, and automation of control actions, including adaptation under uncertainty. The technology will diminish the need for costly preventive security measures as well as reduce action time sixtyfold and system recovery time by at least 10%, enabling more efficient and resilient grid operation.

Potential Impact:

If successful, PNNL's HADREC will displace existing technologies for grid centralized emergency control by solving the key technical hurdles for determining and coordinating emergency control actions based on real-time system states.

Security:

HADREC can ignite and facilitate development and automation of fast and innovative grid control technologies. This will prevent power outages and blackouts while maintaining U.S. leadership in advanced grid control and operation.

Environment:

HADREC’s advanced control capabilities will allow for a more secure and responsive grid, which facilitates the efficient use of clean-energy generation resources and network assets.

Economy:

Using HADREC, grids in the U.S. could potentially avoid billions of dollars from outage costs each year.

Contact

ARPA-E Program Director:
Dr. Richard O'Neill
Project Contact:
Yousu Chen
Press and General Inquiries Email:
ARPA-E-Comms@hq.doe.gov
Project Contact Email:
yousu.chen@pnnl.gov

Partners

Google
V&R Energy

Related Projects


Release Date:
11/15/2018