This is an excerpt from the Second Quarter 2012 edition of the Wind Program R&D Newsletter.
Since 2008, Argonne National Laboratory and INESC TEC (formerly INESC Porto) have conducted a research project to improve wind power forecasting and better use of forecasting in electricity markets. One of the main results from the project is ARGUS PRIMA (PRediction Intelligent MAchine), a software platform for testing advanced statistical algorithms for short-term wind power forecasting. The software platform contains a set of novel statistical algorithms, also developed during the project, to generate more accurate wind power point and uncertainty forecasts.
Improved wind power forecasting is key to achieving more efficient operation of power systems with large shares of wind power and other renewable energy resources. The improved statistical algorithms in ARGUS PRIMA can help lower the cost of integrating wind power into the electric power grid.
"We are excited about the results from testing the new algorithms. Our test results show significant improvements in forecast performance compared to more traditional approaches," said Audun Botterud, principal investigator for the project.
For wind power point forecasting, ARGUS PRIMA trains a neural network using data from weather forecasts, observations, and actual wind power generation, to produce the most accurate power prediction. The key novelty is the training criteria used in the neural network, which is based on concepts from information theoretic learning (ITL). Results from tests on real-world data from two large-scale wind farms in the Midwestern United States, using the new ITL training criteria as compared to the traditional minimum square error criterion (see Relative comparison figure), show distinct reductions in forecasting errors.
However, despite reduced forecasting errors, there will always be a degree of uncertainty in wind power forecasts. It is therefore important to estimate the level of uncertainty to better account for it when making operational decisions. ARGUS PRIMA contains two new computational learning methods for estimating the forecast uncertainty. The new algorithms have been tested on datasets from the Eastern Wind Integration and Transmission Study, as well as on two wind farms located in the Midwestern United States. Testing shows that the new algorithms better match the observed wind power probability distribution than results obtained through more traditional statistical approaches.
ARGUS PRIMA is made available to users under a licensing agreement.
"We are seeing a lot of interest in our research. The licensing arrangement helps to facilitate transfer of the statistical learning algorithms developed in the project to industry use. A leading forecast provider in the United States has just licensed our software," said Botterud.
The research has been sponsored by the U.S. Department of Energy's Wind Program. For more on this project, visit Argonne's Wind Power Forecasting and Electricity Markets Web page.
The Argonne National Laboratory outside Chicago, Illinois, supports the Energy Department's Wind Program by researching improved methodologies for wind power forecasting and analyzing the use of wind power forecasting in power systems operations. The lab is also addressing wind turbine drivetrain reliability issues, developing advanced superconducting drivetrain designs, assessing wind energy developments in critical wildlife habitats, and developing a GIS-based visual impact risk analysis and mitigation system.