Some generation groups can be inaccurate

A summary of the accuracy of generation groups follows.

NSW1 (DAEG AND WAEG)

NSW1 tends to be very accurate with the following exceptions:

Bayswater(1,2) and Bayswater(3,4) however the aggregate value is very accurate.

Smithfield groups can overpredict generation output however overall the forecasts are pretty good.

VIC1 (DAEG AND WAEG)

Newport, Somerton and Laverton example

The generation groups of Newport and (Somerton + CherryTreeWF) and Laverton are often impacted by numerical instability. These three groups are tightly connected therefore the aggregate of these three groups is pretty accurate.

For this reason you will often see these generation groups forecasting a small or uneven generation profile. Nonetheless for the most part you should see the step change profile indicating a unit commitment. Similarly if a unit is committed to run a full capacity, DAEG can on occasions predict an output greater than capacity. We deliberately don’t cap DAEG output to help highlight the potential inaccuracy.

Mortlake is very accurate due to AEMO adding a new constraint about a year ago.

VIC1 WAEG

Currently Vic is mostly a lost cause for WAEG. However there are still units that solve early which are very accurate. The accurate generation groups include:

• The two Yallourn groups

• Mortlake

• All the Hydros including Murray

• Most of the renewables

• BDUs are ok except Rangebank which is unreliable.

• Unfortunately Loy Yang, Jeeralang, Somerton, Laverton, Newport, Valley Power and duids connected to these simply can’t resolve a solution because of insufficient inputs and near parallel planes in the feasible solution space which causes numerical instability.

QLD1

QLD1 is getting progressively more difficult to solve due to additional duids, most recently ClarkeCreekWF and GreenBankBDU.

Groups in QLD are resolved at various stages in the solve process. Generation groups that are resolved early in the solve cycle and are accurate include:

• Callide C

• Gladstone(3,4)

• Kogan

• Millmerran

• Tarong North

• Yabulu

• Yarwun

Generation groups that can be somewhat inaccurate, mainly during daylight hours include:

• Barron + Kareeya

• CallideB - 0.06*Wivenhoe

• DulaccaWF + Roma

• Gladstone(1,2,5,6) - 0.06*Wivenhoe

• MtStuart

• Oakey (Oakey + OakeySF is very accurate)

• Stanwell - 0.06*Wivenhoe

• Swanbank + Wivenhoe

• WivenhoePump (can be quite inaccurate but the overall daytime operating profile is good)

• ChinchillaBDU + WandoanBDU + WesternDownsBDU (just ok)

Once the sun goes down we set the solar farms to zero which improves overall accuracy.

Groups that are no longer reliable and provide little information either due to the size or the accuracy of the group include the Braemar/DarlingDowns groups and the Tarong groups.

SA1

Overall very accurate although there are minor errors for most aggregate groups.

Reasons for inaccuracy

Number of iterations to resolve groups

Some generation groups are resolved after many tens of thousands of computations and iterations. This occurs for a number of reasons, for example when a set of constraints are very near parallel which means the intersection of two constraints is difficult to accurately solve and can cause numerical instability.

Why accuracy can change

Constraints are updated from time to time, new constraints are added and old constraints are removed. For example Mortlake was once difficult to solve however AEMO invoked a new constraint which means that Mortlake is now very accurate.