This is the final project summary from the Short-term Forecasting project, Solar Power Ensemble Forecaster, undertaken by Industrial Monitoring & Control (IMC).
Report extract
Fourteen real-time solar power forecasting models were developed, deployed and continuously improved on five solar farms over nine months. Individual models’ forecasts were combined into Ensemble models, improving the overall prediction accuracy. Finally, a reimplementation of the Causer Pays procedure was developed to allow financial comparison of the fee reduction resulting from the individual models’ forecasts.
Challenges around site integration, project delays, and high solar farm staff turnover in this young industry created deployment difficulties. Technical issues with the varying data quality and precision from the different sites initially caused problems for forecast accuracy and stability, until technical solutions were developed. Significant delays in obtaining data and information about the Causer Pays procedure delayed the financial analysis and limited the development of the planned financial model optimisation algorithms.
Despite these challenges, a high performing ensemble of models is now producing live forecasts used in real- time energy market dispatch and is shown to be generating large fee reductions for the participating solar generators, helping to enable a more stable grid with higher renewable energy penetration, lower energy prices and ultimately lower carbon emissions.
