New Model Reduces Forecasting Errors in Predicting Wind Energy
As the number of wind energy installations continues to grow, so does the recognition that energy from this source will be a key part of a significantly greener grid in the coming decades. But wind energy sources are inherently variable and have limited predictability, which has led to a significant rise in forecasting errors. In a new study, researchers developed a model for predicting wind speed that significantly reduces forecasting errors in predicting wind energy. The model, which was evaluated on data from more than 500 wind farms in the Midwest United States, outperformed traditional baselines.
The study, by researchers at Carnegie Mellon University (CMU) and the Georgia Institute of Technology, is a finalist for both the 2023 Quality Statistics and Reliability (QSR) Best Student Paper Competition and the Best Applied Paper Competition at 2023 Institute for Operations Research and the Management Sciences (INFORMS) workshop on data mining and decision analytics, which will be presented at INFORMS later this year.
Predicting problems associated with wind energy sources have led to significant increases in forecasting errors for some independent system operators. This, in turn, has introduced additional challenges for market-clearing algorithms that are the foundation of grid operations.
“There is a strong need for new forecasting methods that can quantify and reduce the uncertainty of wind power predictions and enable grid operators to act effectively,” says Shixiang Zhu, assistant professor of data analytics at CMU’s Heinz College, who led the study.
Using data from more than 500 wind farms operated or planned by the Midcontinent Independent System Operator, which delivers electric power to 42 million customers in 15 U.S. states and the Canadian province of Manitoba, researchers developed a spatio-temporal model to address these problems. Their model assumes that the wind prediction of a cluster is correlated to its upstream influences in recent history and the correlation between clusters is represented by a directed dynamic graph.
The researchers describe a Bayesian approach in which prior beliefs about the predictive errors at different data resolutions are represented in the form of Gaussian processes. The joint framework enhances predictive performance by combining results from predictions at different data resolution points; in doing so, it significantly reduces forecasting errors and provides reasonable quantification of uncertainty.
“Wind energy is predicted to surpass other sources of renewable power generation this decade,” notes Zhu. “Our model can inform efforts to plan for and optimize the allocation of wind energy.”
Summarized from a working paper, Multi-Resolution Spatio-Temporal Prediction with Application to Wind Power Generation, by Zhu, S (Carnegie Mellon University), Zhang, H (Carnegie Mellon University), Xie, Y (Carnegie Mellon University), and Van Hentenryck, P (Georgia Institute of Technology). Copyright 2023. All rights reserved.
About Heinz College of Information Systems and Public Policy
The Heinz College of Information Systems and Public Policy is home to two internationally recognized graduate-level institutions at Carnegie Mellon University: the School of Information Systems and Management and the School of Public Policy and Management. This unique colocation combined with its expertise in analytics set Heinz College apart in the areas of cybersecurity, health care, the future of work, smart cities, and arts & entertainment. INFORMS named Heinz College the #1 academic program for Analytics Education. For more information, please visit www.heinz.cmu.edu.