Games and other news from IEA wind task 36 forecasting for wind energy

The International Energy Agency (IEA) Wind Task 36 on Forecasting for Wind Energy organises international collaboration, among national weather centres with an interest and/or large projects on wind forecast improvements (NOAA, DWD, ...), forecast vendors and forecast users to facilitate scientific exchange to be prepared for future challenges.

The talk discusses the general setup of the Task, and the latest developments. Among those are decision making under uncertainty. To this aim, actual forecasting situations were gamified and was tested by a wide audience. During the game, the participants could experience the benefit of probabilistic information on their decisions to trade.

A major effort of the Task is the IEA Recommended Practice for Forecast Solution Selection which is divided into 3 parts: (1) “Forecast Solution Selection Process”, (2) “Designing and Executing Forecasting Benchmarks and Trials”, and (3) “Evaluation of Forecasts and Forecast Solutions”. The group initially identified three key contributing factors for failing to get the optimal value from the forecasts: (1) the specification of the wrong forecast performance objectives in the forecast solution selection process, (2) the use of poorly designed benchmarks or trials to select a forecast solution for a user's application and (3) the use of non-optimal evaluation metrics to assess the performance of candidates or existing forecast solutions.

For this year, we intend to update the guideline in the light of the experiences throughout the industry in its initial application, and after collecting this experience at 3 Open Space workshops. A fourth part will be added, extending the guideline to Measurements for Real-time Forecasting Applications.

Another current activity of the task is the Numerical Weather Prediction (NWP) benchmark. NWP model providers have been asked to run two well-measured sites. One is from the Wind Forecast Improvement Project 2 and represents a difficult case of mountain waves in the US, the other one is based on a week of data from an offshore wind farm in the Baltic.

A common validation framework was developed for that case and is available from Github. Finally, we submitted a paper on the uncertainty propagation throughout the entire modelling chain, where we investigated the part of the uncertainty that is coming from the model, that of the model inputs, and the part that is weather related.

The analysis is done separately for the planning phase, operational phase and market phase of the plant and forecast system.