Wind energyProject Lake Bonney Stages 2/3
Report: Vestas Short Term forecasting of Wind Farms Lessons Learnt (PDF 96KB)
This Vestas Short Term Forecasting Lessons Learnt Report outlines the key learnings Vestas has learnt in the six month period of April 2019 – October 2019 in regard to their involvement in the ARENA Short Term Forecasting trial.
The initial project plan anticipated a shorter duration for establishing the infrastructure connectivity using the API and get to a stable state for the forecast submittal via this API. Future deployment should allow for additional time to test and stabilize the self-forecast pipeline. Additionally, allowing for longer operational runtime allowed the system to collect critical information regarding both the reliability of the forecast submittal as well as different operating conditions of the farms that may impact the forecast accuracy. A good understanding of the potential operational constraints for instance can help further refine the forecast model accuracy. The team will continue to leverage the results observed to improve subsequent releases of the forecast model. Similarly, while the team focused on making the infrastructure resilient to ensure high availability of the self-forecast, the need for additional monitoring of the infrastructure was identified and deployed to strengthen the 24/7 operational support.
The assessment condition stating that any AEMO constraint on a park that is equal to the park’s forecasted AWEFS availability will not be excluded from the MAE/RMSE calculation should be revised. This condition will favour the AWEFS when the actual availability and self-forecast availability is higher than the AWEFS availability, as the actual MW will directly follow the AWEFS availability due to the constraint limiting the park. The assessment would be more reflective of self forecasting reliability over AWEFS reliability if all periods that an AEMO constraint is applied to the park are excluded from the assessment.
In conjunction to the above, the assessment condition that 80% of periods must be available for the forecast should also be revised. As, over the last 20 days, AEMO constraints (where the constraint was both less than or equal to the AWEFS availability), were binding at Lake Bonney 3 for over 24% of the periods. Therefore, if Lake Bonney were to suppress the forecast where AEMO is constraining the park to its AWEFS maximum availability, then the trial might not pass the condition that forecast data must cover 80% of the trial period.
An additional assessment condition that should be revised is that self-forecast values must be positive. While AWEFS also provides only positive values, and therefore should not affect the MAE/RMSE comparison between the two values, it is still believed that accurate, negative value forecasts can be produced for the parks.