Geospatial Data Mining for Turbine Layout and Energy Forecasting in Offshore Wind Farms

This research leverages geospatial data mining to support decision-making in the realm of renewable energy, more specifically offshore wind energy farms. It focuses on wind pattern recognition via wind vectors (speed and direction at a given time), wind roses (circular plots display wind speed and direction distributions over a time period) and analysis of a phenomenon called wake effect (where relative positions of turbines can decrease wind speed and increase turbulence downwind). Using geospatial wind data and real turbine observations, this research explores how wind vectors and turbine layout impact total energy output. Sentinel-1 satellite data offer high spatial wind vector resolution along the northeastern U.S. coast, while operational data from the Jersey-Atlantic Wind Farm deliver high temporal power output resolution. K-means clustering is deployed to uncover geospatial wind patterns, while PCA is used to interpret the relationships between wind features and energy output trends. Supervised learning models, including a feedforward ANN and a CNN, are used to predict turbine-level energy output. The CNN outperforms other models at capturing spatiotemporal relationships, achieving R² scores over 0.93. Wake effects are inferred from prediction discrepancies related to turbine positioning. This geospatial and temporal analysis offers domain-specific guidance on enhancing relative positioning of turbines with respect to the wind, to reduce wake effects and maximize energy generation in offshore wind farms. This work exemplifies spatiotemporal data mining empowering offshore wind development with data-driven tools for wind farm turbine layout and energy forecasting. Lastly, this study highlights the power of spatial data mining in domain-specific decision support.