Spatial Analysis of Analog Ensemble Forecasts for Wind Forecasting

Leveraging renewable energies, as a power source in an electrical grid, is a difficult problem that presents engineering and software challenges. Optimization algorithms based on artificial intelligence and data-enabled science can be employed to address several of these challenges. Improving short-term wind forecasts is paramount to aiding utility companies in estimating the amount of power available at specific times, and in turn to cope with over- or under-generation of power using wind turbines. It also helps utility companies plan and implement strategies to sell the electricity when in excess or generate and buy it when in deficit. The integration of data mining and machine learning algorithms with meteorological data from the environmental sciences can lead to new advancements.

This research presents an application of an analog-based method for short-term wind speed prediction. The long-term goal is to use this methodology operationally in the management of wind farms. Further investigation of the application of analog models to improve the short-term prediction of wind speed and direction is needed. Utilizing past weather observations and forecast model output, Monache et al. (2013) worked to discover the best analog members for a specific site. The current research investigates how the addition of spatial analytics impacts the selection of ensemble members. Supervised machine learning and spatial analytics will be used to categorize regional subsets of stations to determine the impact on the analog ensemble output when spatial relationships are integrated into the effort. The hypothesis is that observations from neighbor stations will improve the short-term forecast predictions.