Model-based Reinforcement Learning with Active Learning for Efficient Electrical Power Converter Design
Technology Description:
IBM Research will develop a reinforcement learning (RL)-based electrical power converter design tool. Such converters are widely used and critically important in many applications. Designing a specific converter is a lengthy and expensive process that involves multiple manual steps—selecting and configuring the correct components and topologies; evaluating the design performance via simulations; and iteratively optimizing the design while satisfying resource, technology, and cost constraints. In this project, the design problem will be formulated as mixed integer optimization to be intelligently and automatically solved using an RL-based optimizer. Because the physics-based simulation models used for design are computationally expensive and time-consuming, IBM’s framework will use surrogate models with similar fidelity but lower computation cost and query the physics-based models only infrequently via intelligent strategies. The proposed RL-enhanced automated design can explore the design space much more effectively, decreasing development time and cost without compromising performance.
Potential Impact:
DIFFERENTIATE aims to enhance the productivity of energy engineers in helping them to develop next-generation energy technologies. If successful, DIFFERENTIATE will yield the following benefits in ARPA-E mission areas:
Security:
Seek U.S. technological competitive advantage by leading the development of machine-learning enhanced engineering design tools.
Environment:
Use these tools to solve our most challenging energy and environmental problems by facilitating an economically-attractive transition to lower carbon-footprint energy sources and systems.
Economy:
Reap the economic productivity benefits associated with the commercial adoption of the resulting higher-value energy technologies and associated products.