Dynamic Linear Model Forecasts for Wind Direction and Speed

• Activity Number: 522 - Contributed Poster Presentations: Section on Physical and Engineering Sciences
• Type: Contributed
• Date/Time: Wednesday, August 2, 2017 : 10:30 AM to 12:20 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Patrick Edmonds* and Cindy L Yu
• Companies: Iowa State University and Iowa State University
• Keywords: Forecasting ; Meteorology ; DLM ; Numerical Weather Prediction

Abstract

Wind energy continues its long set path of industrial and market expansion. Numerous private and public entities the world over reap the economic benefits of the industrial seeds sown over the past twenty years of wind energy development. However, growth of the wind energy sector remains inhibited by the problem of uncertainty in atmospheric forecasts. Both wind speed and wind direction are of critical importance to power system operators, yet few quality methods exist for wind direction prediction. Forecasting methods are classified as numerical weather prediction (NWP) models, statistical, or hybrids of both. The current study begins under the purely data-driven approach and moves into the hybrid class. A series of dynamic linear models is proposed for simultaneous bi-variate wind speed and direction prediction. Every member of the model series is tested against the method of persistence prediction and the most viable autoregressive-moving-average model. The process is repeated on NWP output to assess long-term forecast bias correction. All models are assessed for forecasting ability through the measure of mean absolute error over a given time horizon.