This report details techniques for forecasting the generation of electricity from distributed solar photovoltaic (PV) systems.
A high uptake of photovoltaic (PV) systems creates instability in the electricity grid due to the intermittency of sun light. 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.
This project explores real-time data mining from a few hundred widely distributed residential PV systems combined with image data from inexpensive sky cameras. The forecast horizon was chosen to lie between 5 minutes (the shortest time to place a bid on the energy market) and 60 minutes (lead time for diesel or gas turbines to provide additional power if necessary).
The number of all PV systems is often much larger than the number of systems from which data can be collected. Statistical modelling can return averages which closely describe the behaviour of the whole system without collecting data for each single part of the system. A spatially distributed system of PV generators changes its power output rapidly when clouds move in or out. It therefore has to be exploring how many data sources need to be sampled in which frequency in order for the resulting model to represent the behaviour of all PV systems.
The quality of publicly available real-time PV data varies largely due to the fact that most of the data are created and recorded by individuals. Therefore, this project also chose to create a network of 100 low-cost data loggers installed in residential homes which measure the power output of the PV system in short intervals. The data loggers are distributed over the whole ACT providing high quality measurements of the PV system power.