Can sustainable AI practices help reduce electricity consumption?

Machine Learning

So why can't AI help us expand the use of wind, solar, and hydroelectric power? Last year, the UK generated enough clean energy to power every household. The problem comes when you try to use green electrons for demand response for utilities. In states where renewable capacity is growing rapidly, such as California and Texas, Mousavi says, “the grid is not optimized to get renewable energy to where it's needed. When there isn't enough transmission capacity for renewable generation, especially at midday, when solar generation peaks, we have to resort to curtailment, or simply taking renewable capacity off the grid.”

He cites California's Central Valley's massive investment in solar power plants as a key example. “The electric grid was not designed, and has not yet been upgraded, to transmit the majority of the electricity needed in the San Francisco Bay Area from areas in the Central Valley, such as suburbs Davis and Merced, where solar capacity is rapidly increasing,” he explains. “As a result, the Bay Area has high consumption demand, but does not have the appropriate electric grid infrastructure to supply all the renewable energy needed there. So, to maintain grid stability, they are forced to disconnect from installed and readily available renewable energy sources, such as solar, during peak demand periods and instead use local, traditional power plants (usually called 'peaker plants') during high demand hours.”

Infrastructure investments will help, but building out power systems takes years, or even decades. What can be done about surging AI power demands today? One option is to ration computing power, as AWS has reportedly been forced to do in Ireland, or to incentivize the relocation and expansion of computing power to renewable energy-rich areas, relieving the additional pressure on the power grid to match demand with renewable energy supply. Another is to keep burning fossil fuels. Despite their problems, oil and gas power plants are abundant, reliable, and evenly distributed near population centers.

But the more these sources of energy are used, the less urgency there is to bring new renewable energy projects online. There are many clean energy projects in the U.S. that are waiting for regulatory approval. But as AI expands its influence from businesses to homes and personal devices, the “additionality” of new renewable energy generation could be delayed.

To overcome the lack of access to clean energy, data centers have adopted workarounds such as power purchase agreements (PPAs) and renewable energy certificates (RECs) to offset or deduct their carbon emissions. “PPAs and RECs are two of the drivers of increased power generation and have been a boon for the renewable energy industry, but they are not a panacea,” Mousavi explains.

“They do not do much to improve transmission, they do not enable functional accountability mechanisms by properly allocating emissions responsibility, i.e. e-responsibility, to the end-consumers of electricity, and trading emissions attributes through mechanisms like RECs makes it virtually impossible for anyone to know the true emissions impact.”

“As discussed in the paper 'What is Scope 2 Good for?', without Scope 3 Category 3 (S3C3) we don't know our true emissions impacts, and market-based Scope 2 (MBS2) emissions don't,” Mousavi continued. “We need to improve the balance between renewable energy production and demand. We've rapidly increased clean electricity generation over the last few years, but we haven't proportionately improved the infrastructure to get that increased capacity to where it's needed.”

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *