Gaussian Processes (GPs) for spatial-temporal power curve estimation

In this study, we address the challenge of modeling wind power curves by incorporating the effects of terrain. Unlike environmental variables such as wind speed and temperature, which vary both spatially and temporally, terrain features such as ruggedness, slope, and ridges vary between turbine locations but remain constant over time. To effectively handle these two modes of variation, we employ a Kronecker product Gaussian Process (GP) model. By jointly modeling data from multiple turbines, this approach integrates terrain information into the power curve estimation process, leading to more accurate, site-specific predictions. Moreover, it enhances the model’s generalizability to turbines located in different terrains, underscoring the value of incorporating spatially-varying terrain data in wind energy forecasting.