Solar and wind generators stand to gain if investment in short-term forecasting reaps better results than AEMO can deliver, writes Dr Colin Bonner.

In March the Australian Renewable Energy Agency (ARENA) in partnership with the Australian Energy Market Operator (AEMO) announced $9.41 million in funding for 11 projects to trial short-term forecasting at large-scale wind and solar farms on the National Electricity Market (NEM). Under the market changes facilitated by AEMO, wind and solar farms will be able to provide their own 5-minute dispatch forecasts directly to AEMO, as opposed to AEMO using its internal forecast for each generator.

The main goal is to reduce frequency control ancillary services (FCAS) charges to generators, which can exceed $500,000 a year for a 100MW solar farm. The industry is more than ready for change but can self-forecasting outperform AEMO’s current forecasting system?

AEMO performs demand and generation capacity forecasts for the entire NEM. While fossil fuel, hydro and battery generators have excellent control of what they can generate and when, the future generation capacity of wind and solar farms is dependent on local weather variations. AEMO forecasts these variations with the Australian Wind Energy Forecasting System (AWEFS) and the Australian Solar Energy Forecasting System (ASEFS) based on SCADA data from the generators, satellite data and weather models. These forecasts – along with current generation – are used for 5-minute ahead dispatch forecasts which each generator must meet.

Due to the imperfect nature of loads and generators on the NEM, AEMO routinely procures ancillary services to ensure the stability and quality of the NEM. These services are typically frequency “raise” and “lower” requests that add or remove energy from the grid to keep the synchronous machines spinning at 50.00Hz.

Overestimates by AWEFS and ASEFS provided by AEMO potentially mean wind and solar generators are unable to perform as instructed by AEMO. As a result, AEMO has to procure FCAS to make up for the energy shortfall, resulting in “causer pays” penalties for the generator. Generators are effectively paying AEMO to procure FCAS due to poor AEMO forecasts.

Self-forecasting future

AWEFS and ASEFS have been operating for many years now. The question is can the industry outperform AEMO’s current forecasting system? The answer is yes. AWEFS and ASEFS forecasts rely on low-dimensional models with limited data from the generators. Forecasts for both systems have significant components of persistence where the forecast models assume the generator’s capacity in 5 minutes will be largely what it is generating now. Inputs to AWEFS and ASEFS are averages per “cluster”, which for solar is generally in the order of several hundred meters and several kilometres for wind farms.

Short-term forecasting at 5-minute dispatch intervals is trying to forecast micro-β meteorological time scales. The chart on this page shows the relationship between the meteorological time scales and horizontal length scales for various weather events. However, the current inputs to AWEFS and ASEFS are on the micro-α and meso-g length scales, which really only allows modelling and forecasting of atmospheric variations on the micro-α and meso-g time scales.

The real power of self-forecasting is that it allows integration of a lot more data directly from the systems’ SCADA, onsite cameras and additional sensors around the generator. The additional data allows sampling and modelling of the atmosphere on the micro-β scales, which enables accurate forecasting at the 5-minute dispatch interval.

Improved forecasting will allow better integration of renewable energy into the NEM in a smarter way. In addition, with 45% of the current wind and solar capacity on the NEM participating in the trial it is an incredibly strong signal from the industry that a change is needed.

Time and spatial scales of weather events
Source: Mesoscale Meteorology in Midlatitudes, Markowski & Richardson, 2010.

Dr Colin Bonner is technical director and co-founder of Fulcrum3D.