Evaluation of Offshore Wind Power Potential Using Large-Scale Spatiotemporal Data Mining

Offshore wind power plays a vital role in the global energy transition. The escalating demand for green energy has necessitated the development of computationally efficient and accurate wind farm assessment systems. Existing assessment methods, based on numerical simulations or in-situ observations, are often constrained by high costs and limited spatiotemporal resolution when applied to large-scale studies. Thus, by integrating historical meteorological data with machine learning algorithms, this research aims to establish a framework for assessing wind farm potential and develop a corresponding predictive model.

This study utilizes ERA5 global atmospheric reanalysis data and GEBCO bathymetric datasets. First, K-means cluster analysis is employed to identify high-potential development areas in the offshore waters of Taiwan, considering both wind resource potential and bathymetric constraints. Subsequently, this research combines wavelet analysis and principal component analysis for feature extraction to build optimized machine learning models. Furthermore, the predictive performance of various models is evaluated, and the correlations among key variables are examined. 

Results indicate that the proposed assessment framework effectively identifies optimal locations for offshore wind farms and enables precise forecasting of future wind energy potential. Additionally, the analysis reveals a weakening temporal correlation between the Southern Oscillation Index and local wind speeds—a phenomenon that may be attributed to global climate change. These findings offer significant practical value for engineering; not only do they provide decision-making support for offshore wind farm site selection, but they also serve as a scientific basis for optimizing power generation strategies and grid dispatching.