Wind energyProject Ararat Wind Farm
Report: DNV GL – Multi-Model and Machine Learning Wind Forecast Project Lessons Learnt 3 (PDF 142KB)
This DNV GL Wind Forecasting Lessons Learnt Report outlines the key learnings DNV GL has learnt in the period from May 2020 – September 2020.
DNV GL have entered a funding agreement (the Project) with Australian Renewable Energy Agency (ARENA) to provide short term wind power forecasts for the Ararat Wind Farm. The purpose of the Project is to explore the potential for wind farms to provide their own, more accurate, forecasts as inputs into AEMO’s central dispatch system.
DNV GL has partnered with the Ararat Wind Farm Pty Ltd (AWF) and RES Australia Pty Ltd (RES). AWF represent the wind farm for which forecasts are to be provided, and are providing access to relevant data and allowing forecasts to be submitted on behalf of the Project. RES are providing support with facilitation of the Project, and evaluation of the value that accurate forecasts can provide to an operational project.
Two of the main challenges associated with the Project to date have been successful recovery of real-time data from the site, and communication via the AEMO MP5F API. These challenges were also present earlier in the Project [DNV GL Lessons Learnt Report 1] and [DNV GL Lessons Learnt Report 2] and whilst they have largely been overcome, they are still persisting to a degree, and have the potential to impact upon the reliability of forecast delivery. The key lessons learnt are that it is necessary to allocate adequate resources to both aspects of the Project, across all stakeholders involved in the Project, and that it is important to establish processes to pre-emptively prevent outages or resolve them when they occur.
Further challenges have included developing and implementing a machine learning model to generate accurate short horizon forecasts. Considerable effort has been made to fine tune the machine learning parameters while also reducing latency in live site data and model run time.