This Lessons Learnt Report outlines the key learnings Fulcrum3D has learnt in the six month period of April 2019 – October 2019 in regard to their involvement in the ARENA Short Term Forecasting trial.
Report extract
Key Lessons
- The hardware on the CloudCAM systems (cameras, pyranometers, logger, power supplies, network hardware etc.) is performing reliably. We are very pleased with this outcome. Other CloudCAM systems have operated continuously in harsh environments for over three years and this is another affirmation that the hardware is robust and well designed. This is particularly encouraging given the heavy rains during the February floods and summer temperatures consistently over 40C.
- Our remote communications are working well (via our dedicated NBN connection that we established last December). This enables us to perform system maintenance remotely.
- Connection to the AEMO MP5F API took longer than expected because:
- There is complexity around the arrangement for when more than one forecaster is
on the same site. - Formal communications with AEMO must be via the DUID participant and not with independent forecasters. The participants have not necessarily followed the MP5F development as closely as the forecasters and are still learning how it works.
- There is complexity around the arrangement for when more than one forecaster is
- Power modelling is non-trivial but providing bonus insights. It is confirming our original hypothesis that a detailed power model is essential for accurate forecasting.
- We have growing confidence that onsite high dimensional forecasting models can significantly outperform ASEFS.
- Insights around the power modelling:
- Inter-row shading is more complex than initially thought – we have been able to model that complexity.
- The behaviour of all inverters on the solar farms did not necessarily match when they were not clamped by set point. This was not expected.
- DC current sensors are not always accurate. Better DC current sensor accuracy would assist better forecasting.
- Power point controller and inverter controller algorithms are obscure (they are not well documented by the suppliers). This is another challenge for forecasting – we have built work arounds for this.