September 27, 2019
Share this post:
Author: Dr. Julian De Hoog, Technical Lead – Energy & IoT, IBM Research Australia
Staring up at the roof of his house, my neighbour Oscar asks me, “So, do you think it would be worth it?” He is referring of course to the possibility of installing solar PV panels on his roof. Across Australia, many others like him have already taken that step: more than 2 million such systems have been installed across the country, and with more than 25% of homes having solar now, Australia is leading the world. Many of these home owners are now also looking at installing energy storage to get more value out of their solar, and to further reduce their dependence on the grid. And it’s not just home owners – businesses of all types and sizes are now becoming more and more interested in understanding their energy use and taking some control into their own hands.
Impacting not just your bill, but also the environment
For most people like Oscar, the main motivation is to reduce their energy bills. Solar PV has come down in cost so far that the typical payback for many owners is now 5 years or less. Given that solar PV systems often come with warranties lasting 10 years (inverters) or 20 years (solar panels), the economics speak for themselves. At the same time, it feels good to be powering your home from the sun, rather than from the fossil fuel dominated generation mix we have today (at least in Victoria).
However, the situation can change from one house to another. In Oscar’s case, the neighbourhood is leafy and much of his roof is shaded by nearby trees. The pitch of the roof is steep, and the sunniest parts are West-facing, meaning that they wouldn’t generate much electricity until afternoon.
The case for AI
Our team at IBM Research Australia is familiar with these challenges. Over the past five years we have analysed data from dozens of homes and recognised how unique a home’s profile can be. In a project with two industry partners, Selectronic (an inverter manufacturer) and Relectrify (a provider of energy storage), funded by the State of Victoria, we learned how batteries, solar panels, inverters, and cloud-based services can best be orchestrated to get the maximum possible value out of these kinds of systems. Unlike solar panels, batteries are still fairly expensive (at least for now), so it is essential to operate them in the best possible way to ensure they provide an appropriate economic return.
This is where forecasting becomes essential. For example, on a sunny day, someone like Oscar would want to start the day with an empty battery, so that the full battery capacity can be used to soak up any excess solar generation. But on a cloudy day, Oscar is better off with a full battery in the morning, so that he can take advantage of lower power prices overnight. These kinds of decisions shouldn’t be left to the owner; they should be made automatically, taking into account anticipated solar generation, predicted energy demand, and an understanding of the electricity tariff structure. Some retailers are even offering attractive rewards to battery owners who discharge to the grid during peak price times – such events must be forecast as well.
The amount of energy generated by a rooftop solar PV system depends not just on sunshine and weather, but also on local impacts. Throughout most of the day, this system is affected primarily by cloud movement, but in the late afternoon, a nearby tree casts a shadow over part of the system. These impacts needs to be taken into account individually for every site.
Forecasting one system at a time
At IBM Research Australia, we’ve developed an optimal control system that can do exactly this, and we’re now testing it at real sites together with our industry partners. We’ve learned that to effectively forecast solar PV generation, you really need to generate a model that is custom-tailored to each site. Forecasting solar irradiation (the amount of solar energy that reaches your panels) is already difficult on its own, and we are fortunate to have access to such forecasts via The Weather Company.
IBM Research Australia’s energy storage optimisation project team. L-R: Rod Scott, CEO, Selectronic; Valentin Muenzel, Cofounder and CEO, Relectrify; Julian de Hoog, Research Scientist, IBM Research; Dan Crowley, Cofounder and CTO, Relectrify; Paras Karki, Electronics Engineer, Selectronic.
But that’s only half the picture: even if you know how much solar energy is hitting your roof, there are still many other impacts that affect how much of that solar energy is turned into electricity. For example, there can be local shading impacts (trees or roofline features), multiple sets of panels facing different directions, and losses in the wiring and inverter that reduce the amount of usable energy. We have been careful to design our methods so that they scale: by using a data-driven approach that requires only two weeks of recent generation data from the site, we’ve been able to significantly improve our solar PV forecasts without having to build complex models. In a collaboration with our Research lab in Ireland, we are now further testing these forecasting approaches on 49 sites across Europe.
What about businesses?
All of this hold true for many businesses as well; they may have different incentives, but ultimately need to solve the same problems. For example, in the work our group has done managing energy consumption for commercial building owners, we’ve learned about the cost of exceeding peak demand thresholds. In our discussions with several clients, we’ve learned about the value of being able to forecast rooftop solar generation across large parts of the network. Many businesses are now actively seeking to reduce their carbon footprint, either in response to regulation or of their own accord. Increasingly, businesses are becoming conscious of the risks of extreme climate events and the need to decrease their exposure to these risks.
Bring on the renewables revolution
So should Oscar install solar panels on his roof? “Absolutely!” I tell him. With modern solar panel technology and the possibility of further rises in electricity prices, the chance that it will pay for itself is high, even if his roof isn’t optimally angled and exposed. And what about energy storage? That question is harder to answer, and depends quite a bit on the usage patterns from one home to another … but with our forecasts and optimal control systems, we hope that batteries will become affordable sooner, for more people, and allow us to enable a faster transition to renewables.