8.6 Solar forecasting
Forecasts of solar output are required for periods ranging from days ahead down to hours and tens-of minutes ahead. In Australia, forecasts of solar generation would be required up to two years out, as is currently required for wind generation. Forecasts should include information about the expected output and the degree of uncertainty in the expected output to indicate particularly volatile periods. Short-term forecasts are aided by the fact that clouds can be observed. Forecasts are an important method for managing both the intermittency and uncertainty of solar generation and should be incorporated into system planning and operations.
The Australian Energy Market Operator (AEMO) has critical forecasting needs: both short-term to long-term energy growth (20 years). AEMO requires five-minute pre-dispatch data and the Australian Wind Energy Forecasting System (AWEFS) was developed in response to the growth in intermittent generation in the National Energy Market (NEM). AWEFS was developed to provide AEMO with more accurate wind generation forecasts to facilitate the operation of the market. Questions were raised by industry experts as to whether AWEFS can be expanded for solar forecasting and what information is needed to be able to piggyback off wind forecasting systems. The size of wind and solar systems need to be noted in this discussion. Wind is mostly large-scale (mostly MW) while solar is both small and large-scale systems (both kW and MW). There were also questions about the possibility of applying the European ANEMOS short-term wind forecasting system to solar forecasting, and AEMO is currently in discussion with the federal government regarding this. Wind energy output can be controlled using pitch control and discussions arose during the workshop about possible methods to control solar output. The ability to curtail output does not exist for current rooftop solar systems. Solar appears more predictable than wind, however wind forecasting has taken five to seven years to develop and many issues are yet to be resolved. There are likely to be a lot of unknown areas which need to be investigated and resolved when developing a solar forecasting system.
Solar intermittency is believed to be more predictable than wind intermittency as it is affected by fewer climatic events, such as cloud cover and temperature. The ramp rates for solar (mainly PV), however, are potentially significantly higher than wind due to the lack of inertia: wind turbines have rotary components which provide some inertia for wind systems. Solar output is more predictable in the very short term because the movement of clouds is visible, but accurate prediction for long-term solar output is needed to determine ways to effectively compensate for solar intermittency.
Some key requirements for data collection and forecasting of solar believed to be important are:
- temperature, irradiance, location
- correlation between weather patterns in nearby geographical locations
- ability to obtain samples at all required time intervals (i.e. seconds, minutes, hours, days)
- cross-checking of data with satellite data
- seasonal data to cater for seasonal variations
- the requirement of a large network of ground based monitoring stations to correlate satellite data with their output, and this should be established and maintained for several years
- accuracy of data
- sky-view cameras for large-scale solar farm to monitor the movement of clouds.
Data collection and forecasting differ between small- and large-scale solar systems in the following ways:
- large systems are site specific, whereas small systems over a large geographic area need more measurement points
- small systems are affected by environment and orientation more greatly, and these factors need to be measured as well at the solar resource
- the accuracy of data and forecasting might have an impact on power quality aspects for small scale systems, while it affects energy output for large-scale solar systems.
The time-scales for data collection and forecasting of solar power systems, believed to be appropriate by industry experts, varied between sub-second and ten minutes. It was also mentioned that the time-scale can be larger for large-scale systems as variations in power output are likely to be significantly lower due to larger area coverage. A number of industry experts indicated the need for further research to determine the appropriate time-scales for data collection and forecasting for solar energy systems of different sizes. It would also be of interest to the Australian power industry to know how these forecasting issues are being dealt with in other parts of the world.