Towards better solar forecasting

The intermittency of renewable energy historically has been its primary barrier to adoption, aside from capital cost which is on a decades long decline. Solar power continues to decline in price as 39 gigawatts of solar were installed last year at a lower investment cost than the 31 gigawatts installed in 2012. Still, intermittency leads to energy storage requirements, which often means batteries or pumped hydro or compressed air. Which means additional cost.

A great deal of VC has poured into energy storage startups over the past decade, leading to a slew of novel technologies trying to cross the valley of death toward commercialization. But as some recent Gigaom reporting points out, another way to help stabilize the grid and address intermittency is through improved solar forecasting.

To address the issue of solar forecasting, IBM is actually working on a machine learning project appropriately named Watt-sun. The platform crunches data from multiple solar forecasting models and adds in localized environmental data. Reports are that the technology is already 35 percent more accurate than the next best solar forecasting models and it’s improving.

For utilities, the ability to predict solar power output up to 24 hours ahead helps them balance the grid, particularly when it’s combined with load forecasting analytics about what sort of demand will be coming off the grid. I imagine as well that there could be some value for utilities in having this data as they participate in day-ahead energy markets.

Integrating solar onto the grid will always require a degree of variability that makes it more complex to handle than fossil fuels. But it would appear that at least one big IT player sees an opportunity to use software to tackle the problem.