Machine-learning-based forecasting of distributed solar energy production

australian national university (anu), nicta (australia's ict research centre of excellence), university of california san diego, university of central florida, laros technologies, armada solar

Solar energy is expected to become the best, low-cost solution for generating the bulk of Australia’s future electricity. However the availability of sunlight is inherently intermittent, leaving potential grid operations vulnerable to power quality issues.Accurate solar forecasting is essential to mitigate this intermittency and achieve high penetration of solar energy in the Australian grid, decrease its total operational costs, reduce risk, and enhance its operational efficiency.

This project will explore real-time data mining of some of the 650,000 widely distributed residential PV systems in Australia. The output of these systems, when mapped in real time, is expected to enable accurate estimation  of cloud location, motion and opacity, inherently matched to the characteristics of PV systems. It will also enable the development and deployment of an experimental network of low-cost all-sky cameras to diversify the cloud detection methods available for synthesis.

The data and technologies developed will provide power utilities with their first tools for managing the technical issues associated with high penetration of rooftop PV, ensuring that these tools are low-cost will further increase their commercial viability. Consequently, the results produced by this project will simplify the widespread uptake of PV, expand the potential PV market, lower the costs of our finance for solar power, and increase the market value of solar energy generation capacity.

This project will help utilities, generators and market operators prepare for the next generation of solar energy.  It will also develop quantitative measures of performance and aim to determine fundamental limits on how accurately the output from distributed PV systems can be forecast on operationally-relevant temporal and spatial scales.

The project team comprises world class experts in all areas needed to develop commercially relevant solar forecasting. They will collaboratively collect, curate and analyse high-quality field data from sites that span a diverse range of environmental conditions. The industry partners involved represent a cross section of the sectors interested in solar energy forecasting and bring critical understanding of commercial operating requirements, access to unique operational data, and intimate knowledge of multiple Australian market sectors.

Fact Sheet: Machine learning based forecasting of distributed solar energy production (PDF 362KB)