Machine learning for solar power is a supercomputer killer

Credit: CC0 Public domain

Supercomputers could end up jobless thanks to a suite of new machine learning models that produce fast and accurate results using a normal laptop.

Researchers at the ARC Center of Excellence in Exciton Science, based at RMIT University, wrote a program that predicts the bandgap of materials, including for solar power applications, via free, easy-to-use software . Bandgap is a crucial indication of a material’s efficiency when designing new solar cells.

Bandgap predictions involve chemical calculations at the quantum and atomic scales and are often made using density functional theory. Until now, this process required hundreds of hours of expensive processing on a supercomputer, as well as complicated and expensive software.

To solve this problem, the researchers trained a machine learning model using data generated from 250,000 previous supercomputer calculations. The results were published in Cheminformatie Journal.

Significantly, while the program is able to include multiple variables, it has been found that only one factor, stoichiometry, contains, in almost all cases, enough information to accurately predict the bandgap. Stoichiometry is the numerical relationship between chemical reagents and products, like the volume of ingredients in a recipe for baking a cake.

Further work is needed to fully understand why stoichiometry alone has proven so useful. But it raises the exciting prospect that lengthy supercomputer calculations may no longer be required for some applications. The artificial neural network that powers machine learning programs may one day be replaced by software that performs a function similar to density functional theory, but with much more simplicity.

Lead author Carl Belle says that “if you want to do simulations but need to have millions of dollars of supercomputing infrastructure behind you, you can’t do it. If we can dig into why the stoichiometric setup is so powerful, then it could mean that supercomputers aren’t needed to select candidate materials, nor for precise simulations. It could really open things up for a whole new group of scientists to use. “

The machine learning program is not limited to the forbidden band. It can be used to predict the properties of many other materials for other contexts and has been developed by a professional programmer, which makes it useful not only for scientists and academics, but also for companies and applications of business.

“It’s built to industry standards and it’s designed to be collaborative,” Belle said.

“The website has a fully relational database. It has millions of records. It’s all there and available for free. We’re good to go.”


Machine learning speeds up material science simulations


More information:
Carl E. Belle et al, A Machine Learning Platform for Material Discovery, Cheminformatie Journal (2021). DOI: 10.1186 / s13321-021-00518-y

Provided by the ARC Center of Excellence for Exciton Science

Quote: Machine Learning for Solar Power is Supercomputer Killer (2021, June 23) retrieved June 23, 2021 from https://phys.org/news/2021-06-machine-solar-energy-supercomputer -killer.html

This document is subject to copyright. Other than fair use for private study or research purposes, no part may be reproduced without written permission. The content is provided for information only.


Source link

About Mariel Baker

Check Also

Getting up to date on the proton – sciencedaily

Scientists are developing a revolutionary theory to calculate what happens inside a proton traveling at …