- Solar forecasting has improved over time
- Applying artificial intelligence in solar energy has been a driver of this improvement
- Looking ahead, Manildra will harness the latest predictive technologies to help improve functionality.
The application of artificial intelligence-driven solar forecasting to the operation of solar power plants could improve the efficiency of solar power by better managing one of the most significant impediments to its constancy: weather variability.
Despite the increasing use of renewables, the weather remains a significant consideration in the wider adoption and implementation of solar and wind power resources. The output from wind and solar plants varies based on the weather’s intermittent nature and predicting the scale and timing of this variation is important for demand and supply planning for electricity systems the world over.
Developments in artificial intelligence (AI) have become important in more accurately forecasting the weather and, in turn, improving renewable energy efficiency and accessibility.
"New Energy Solar’s Manildra Solar Plant is set to harness the latest predictive technologies..."
Understanding AI and forecasting techniques
AI encompasses a broad field of science that includes computer science, psychology, philosophy and linguistics and aims to enable computers to perform tasks that would normally require human intelligence.1 It emerged as a mainstream computer science discipline in the mid-1950s and has since produced a number of powerful tools used to solve complicated practical problems in multiple areas of engineering and technology2 across a variety of industries.
Historically, weather forecasts have been generated by powerful computers processing large amounts of atmospheric and oceanic data, often coupled with data from other sources such as ocean buoys and independent weather trackers, using models that simulate the physics associated with weather.3 The challenges associated with renewable energy’s variability are thought to be exactly the kind of issues for which AI is most applicable.4 AI can play an important role in the modelling, analysis and prediction of both the weather and the performance of renewable energy.5
Currently for the modelling – that is, prediction of performance and control of renewable energy processes – analytic computer codes use complex algorithms solving differential equations. This method requires substantial computing power and considerable time to develop accurate predictions.6 In contrast, AI systems can learn key information patterns, obviating the need for complex rules and mathematical routines. AI offers the potential to make better, quicker and more practical predictions than any traditional methods.7
Solar forecasting in Australia
In Australia, the Australian Energy Market Operator (AEMO) announced in early 2018 that solar plants could self-forecast the amount of electricity generated as opposed to the previous system where AEMO was solely responsible for forecasting the output of generators. Self-forecasting, Federal Energy Minister at the time Josh Frydenberg said, “would allow wind and solar farms to not only maximise the amount of renewable energy dispatched into the grid but also to avoid the need to pay for frequency control services”.8 This has prompted Australian researchers to work toward developing technologies to improve short-term weather forecasting and thereby improve the ability of solar farms to predict output.9
New Energy Solar’s Manildra Solar Plant is set to harness the latest predictive technologies10 through a partnership with Solcast, a solar forecasting vendor based in Canberra, Australia. A better understanding of Manildra’s actual and expected output through the use of Solcast’s global solar forecasting application programming interface is expected to reduce the need for frequency control services and to better manage supply and demand across the National Electricity Market. In time, this technology is expected to improve system reliability and bring down electricity prices.11
Solcast’s technology employs a third-generation satellite “nowcasting” system that can detect and predict cloud characteristics, track aerosols and utilise numerical weather model data, as well as model solar radiation and PV power output using proprietary algorithms.12 With five weather satellites contributing to 18 model predictions during each update, Solcast estimates that it calculates more than 600 million forecasts per hour and can forecast up to seven days ahead.13 According to Utility Magazine, Solcast’s 2017 pilot project yielded a 10–15% improvement over AEMO’s own models.14
In future, NEW anticipates that the Solcast technology will improve Manildra’s operations, and that the further development of this intelligent technology will be a powerful tool to improve accessibility to solar resources.