Wind energyProject Lake Bonney Stages 2/3
This report details how Vestas is utilising four separate forecasting model approaches to assess the incremental improvements that meteorological masts and weather models have on the accuracy of wind production forecasts.
The Lake Bonney Stages 2/3 short-term forecasting project for Infigen Energy utilised four separate forecasting model approaches (implemented at different stages) to assess the incremental improvements that meteorological masts and weather models have on the accuracy of wind production forecasts when combined with high-resolution wind turbine Supervisory Control and Data Acquisition (SCADA) data, and machine learning algorithms.
The first forecast model developed was the “SCADA data only” model, which was most critical because it acted as the baseline for the other models, and was the first to be submitted into the Australian Energy Market Operator (AEMO) production environment on September 10, 2019 for both Lake Bonney 2 (LB2) and Lake Bonney 3 (LB3).
Prior to submitting forecasts to AEMO’s application programming interface (API), Infigen developed its own API layer that was configured as an intermediary interface between Utopus Insights’ forecasting system and that of AEMO. This layer allowed Infigen to implement additional cyber security controls on the automated submission system. A Self-Forecast Suppression system and user interface was also designed by Utopus Insights to allow Infigen’s operators to supress self-forecasted values being submitted to AEMO. During these times, the Australian Wind Energy Forecasting System (AWEFS) forecast would instead be used for AEMO calculation of the dispatch target value. This tool allowed operators to minimise the times when the self forecast was submitting unfavourable values that would negatively impact the average performance values. During the course of the project, voluntary participant suppression was used less than 0.1% of the time for LB3 and never for LB2, a reflection of the stability and reliability of the forecast models deployed.