The AI Grid of the Future Featured Pattern: P1401 September 2019

Author: David Strachan-Olson (Send us feedback.)

AI can help integrate renewable energy into electric grids and help make electric grids more robust against disturbances.

Abstracts in this Pattern:

Researchers are developing new techniques to help make electric grids smarter and more resilient, and many of these new techniques leverage machine learning. The German Federal Ministry for Economic Affairs and Energy (Berlin, Germany) recently funded InnoSys 2030—a new project that will research how digitalization can help to improve the efficiency and stability of electric grids. Participants in the project include numerous German research institutes and universities, four German transmission-system operators, multiple distribution-network operators, and two industrial companies. Ideally, the project will lead to the creation of grid-management systems that automatically detect faults in real time and improve efficiency to limit the need for grid-extension measures. Similarly, researchers from the Fraunhofer Institute for Optronics, System Technologies and Image Exploitation (Fraunhofer Society for the Advancement of Applied Research; Munich, Germany) have developed artificial intelligence to process data from phasor-measurement units, which measure the amplitude and phase of current and voltage on the gird many times per second. During the first stage of research, the Fraunhofer researchers developed a compression technique that reduces the size of the data that phasor-measurement units generate by 80%. During the second stage, researchers used historical system outages to train neural networks. The trained neural networks were then able to detect the type and location of disturbances in real time. Researchers and grid operators still need to develop control policies and systems that react automatically once the neural networks detect disturbances.

Electric-grid operators are also facing new challenges as they integrate more renewable energy into the electric grid, because renewable-energy sources often generate energy intermittently. Researchers from DeepMind Technologies (Alphabet; Mountain View, California) have developed an AI model that can predict the energy output of wind farms 36 hours in advance. The researchers applied the model to the central-US wind farms of Alphabet subsidiary Google to make day-ahead power-delivery commitments to grid operators. By implementing this system, DeepMind researchers were able to increase the value of Google's wind energy by 20%.