(Photo courtesy AES)
At the newly completed Baldy Mesa solar+storage farm in Southern California’s Mojave Desert, Amazon is using machine learning (ML) models to help predict when and how its batteries should charge and discharge energy back to the grid.
Digitalization and the cloud have enabled a “surge” of data, Amazon says, allowing clean energy owners and operators to access real-time weather data, historical grid data, and more.
“AI is an important tool that’s already helping our society make the transition to carbon-free energy and address climate change at scale,” said Kara Hurst, Amazon’s vice president of worldwide sustainability. “Pairing solar projects enabled by Amazon with AI technologies powered by AWS helps to ensure the grid and the customers it serves receive a steady supply of carbon-free energy for more hours each day, while also helping Amazon make progress toward our commitment to be a more sustainable company.”
For the Baldy Mesa project, software built using the Amazon Web Service (AWS) product SageMaker is expected to analyze up to 33 billion data points per year, according to Fluence, the solutions provider. The software is designed to optimize when to buy, store, and sell energy based on grid solutions. Fluence says the same ML solution was deployed at a similar site in California, and was able to help anticipate 2023’s state-wide heatwave.
Amazon is also hoping to leverage ML to optimize energy usage at its facilities. At its San Bernardino Air Hub, Amazon’s 5.8 MW rooftop solar array is paired with a 2.5 MW battery. Its teams are in the process of developing an AI model that would leverage ML capabilities along with performance data from Amazon rooftop solar arrays to help the Air Hub and other facilities minimize their energy usage.
After aggregating performance information from its rooftop solar+storage installations, Amazon will combine it with local weather and building data at a central location. (AWS Data Lake). Once launched, Amazon anticipates the model can be used for predictive insights on site performance and energy generation – a task it says humans cannot perform, as it requires manual analysis and monitoring of each system and building.
