- Aug 18, 2021 7:24 pm GMT
Power grid operators monitor and control the electricity network 24/7/365. As increasing numbers of low capacity and variable distributed energy resources, such as renewables (wind and solar), and advanced network components come onto the grid - combined with significant increases in demand for energy and variability of demand from electric vehicle penetration and electrification efforts, control room operators are being stretched to their limits. On top of the list of demands during normal system operations, operators are up against a barrage of extreme weather events that cause major disruptions to bulk power systems around the world.
To accelerate solutions that could alleviate these increasingly complicated challenges, EPRI co-sponsored the Learning to Run a Power Network (L2RPN) Challenge with Réseau de Transport d'Électricité (RTE), Google Research, ChaLearn, IQT Labs, University College London, V&R Energy, Pacific Northwest National Lab and several European- and US-based universities. The challenge, developed by RTE, is aimed at developing artificial intelligence tools to support operators’ decision-making to manage the increasingly complex system of the future.
Challenge participants were asked to create a reinforcement learning (RL) based agent that would be trained and evaluated on a virtual power network. The aim was to create an agent that would monitor and control the synthetic power network as it changed over a year in five-minute intervals. The agents had to respond to disturbances on the network and ensure supply was maintained at all times in a safe and cost-efficient manner. In grid operations, RL has the potential to aid operators by rapidly assessing conditions in real time and looking ahead and providing recommendations, boosting grid stability, and ultimately improving reliability to customers. The agents would act as an “autopilot” in real time, guiding the decisions of the operator. In 2021 the L2RPN competition was expanded to develop “trust in AI,” whereby agents are designed to flag confidence in their decision making, identifying future scenarios where unstable states would occur, and identifying the location on the network where the issues will arise, if not corrected. This trust building is essential for human-machine coordination in the control center of the future.
The solutions created by the challenge participants are driving the development of RL methods for power network operation in the real world. Ultimately, it is hoped that the best RL agents will optimize network operation, identify issues and associated actions, and decrease grid congestion, resulting in fewer brownouts and blackouts. Additionally, the agent will allow operators to enhance future-looking decisions as the model becomes more predictive. Long-term, models can be applied beyond electricity flow simulations, and used for grid protection studies, grid market studies, energy system planning and dynamic simulation.
Learn more about how EPRI is convening the AI community to produce transformative solutions for the grid of the future at EPRI’s AI and Electric Power Summit on September 28-29, Register here.
Get Published - Build a Following
The Energy Central Power Industry Network is based on one core idea - power industry professionals helping each other and advancing the industry by sharing and learning from each other.
If you have an experience or insight to share or have learned something from a conference or seminar, your peers and colleagues on Energy Central want to hear about it. It's also easy to share a link to an article you've liked or an industry resource that you think would be helpful.