Machine-learning-based forecasting of distributed solar energy production
The Australian National University (ANU) lead a project to develop techniques for forecasting the generation of electricity from distributed solar photovoltaic (PV) systems, like those installed on household rooftops, with the objective of developing a commercial product that can be sold to utilities and other companies to accurately predict the output of of PV systems and therefore demand on the grid for improved efficiency and stability.
Lead organisation:
Australian National University
Project partners:
Telstra, SolarHub, ActewAGL, AEMO, University of California San Diego, University of Central Florida, LAROS Technologies, Armada Solar
Location:
Canberra, ACT
Technology:
Solar energy
ARENA program:
Former Australian Solar Institute initiatives and programs
Start date:
1 Jan 2013
Finish date:
30 Sep 2016
Need
The growing uptake of residential PV systems creates challenges for electricity grid management and instability in the electricity grid due to the fundamental intermittency of sunlight. If there exists no reliable method to estimate the contribution of distributed residential PV systems to the power grid, this leaves grid operators vulnerable to power quality issues. By developing this modelling system, the researchers are aiming to reduce the impact of weather events on the reliable supply of electricity.
Project innovation
Using sky images to significantly reduce errors in solar prediction is highly innovative and is a breakthrough which can be applied to a broader range of forecasting tasks where useful information can be retrieved from images. Accurate forecasting, paired with energy storage and/or fast response alternative energy generation provides for cost-effective management of the electricity network. .
Benefit
This modelling will assist with predictability for energy providers and improve the stability of the grid by observing the interaction of PV systems with the electricity network, and increasing the amount of electricity that can be provided by PV systems through forecasting when and where clouds will affect electricity generation by PV systems.
Lessons learned
A key enabler for accurate forecasts of output is accurate estimation of the real-time contribution of distributed PV systems. Through the use of clear-sky radiation modeling, the theoretical power output for a given PV system can be calculated and given a PV clear-sky index which can be used to estimate the output of nearby PV systems. Power output during weather events can then be estimated to inform energy market operators and therefore improve reliability of the grid.
