Automated analysis workflows implemented in our DataModeler Pro helped to find a minimally sufficient set of weather variables impacting wind energy output of a wind farm and reliably predict the production from easy to measure weather data
Industry Example: Renewable Energy
Who Cares: Investors, Site owners, Energy Suppliers
What Enabled: Prediction, Forecasting
Collaborators: Prof. Frank Neumann, and Markus Wagner (The University of Adelaide), Prof. Tobias Friedrich (Friedrich-Schiller University of Jena)
Wind energy is currently a key player in the area of renewable energy. An accurate and timely prediction of it's output is critical at the stage of validating an investment decision to build a new wind farm, as well as in the energy load balancing for coordinating production of traditional power plants and weather-dependent sites.
Weather forecasts have become extremely effective and precise thanks to development of more and more sophisticated weather models. Avaliability of precise weather data and forecasts has motivated us to look for patterns and dependencies of weather features to wind energy outputs. In this case study we analyzed several months of data from an Australian wind farm in Tansania together with the publicly available weather data measured on the same location. Robust solution workflows implemented in DataModeler Pro helped us iteratively identify the minimal list of weather features impacting the energy output and construct concise and transparent predictive model ensembles, which only contained these driving features.
Initially 16 features were used to predict energy output, out of which only two features - a wind gust and a dew point were identified as drivers with the quantified contribution of each factor to the accuracy of predictions.
...the presented framework is so simple that it can be used literally by everybody for predicting wind energy production on a smaller scale—for individual wind mills on private farms or urban buildings, or small wind farms.
Models were developed using the data collected from October 2010 to June 2011. Their predictions however were tested on the weather data collected in July 2011 (an entirely unseen season). The results showed a very good prediction accuracy (of 12% RMSE).
Related Links/ Publications/ Resources:
Ekaterina Vladislavleva, Tobias Friedrich, Frank Neumann, Markus Wagner – Predicting the energy output of wind farms based on weather data: Important variables and their correlation. Renewable Energy Journal, 2013, vol 50, p.236-243
Key Words: Wind energy, Prediction, Genetic programming, Symbolic Regression, DataModeler Pro