Scipher.Fx 2.0 – Ensemble Forecasting  

Summary

At Utopus Insights, we understand the values associated with accurate forecasts for our customers – from minimizing imbalance costs, complying with grid regulations, optimizing O&M activities, to optimizing energy bidding strategies. 

To bring more value to our customers, we have evolved from the previous methodology we launched 2 years ago to an improved new methodology. We are pleased to announce the release of Scipher.Fx 2.0, Utopus Insights’ ensemble wind power forecasting model, developed to improve our forecast accuracy and enhance our system reliability through optimizing and combining multiple global and regional weather forecast models as input. 

We chatted with our Product Owner, Ayumu Suzuki, to learn more about the enhanced forecasting capabilities. 

What has changed in the way power forecasting will be done now, as compared to previously? 

AS: We have been training and deploying the new ensemble forecasting models in stages to our customers throughout November, and you should see the benefits of the new models over the coming months. 

It should be noted that winter months with higher windiness are typically harder to forecast more accurately than the calmer periods in the summer. We expect month to month error fluctuations due to windiness to persevere, but we are confident that the new ensemble model will outperform the first methodology in terms of relative accuracy gains. 

Do you have any reference examples where improvements have been seen already? 

AS: To give some context to the expected accuracy gains, below are some results for a wind farm in Europe, where forecasting models were trained using data over March 2020 – August 2021, and accuracies of those models were evaluated over a one month period in September 2021. 

The chart below shows the normalized Mean Absolute Error (nMAE, as % of installed capacity) for each forecast source, where the nMAE values have been averaged over 0-48 hour horizons. The yellow bars represent the individual Numerical Weather Prediction (NWP) based power forecasts which are combined to produce the ensemble forecasts.