Multi-relational clustering of spatiotemporal wind velocity in South Africa

As the global shift towards sustainable energy sources gains momentum, harnessing wind energy has become a significant focus. Effective utilization of wind resources requires a comprehensive assessment of resource characteristics and energy potential. Classifying locations with similar wind velocities using distance metrics poses a challenge, particularly when dealing with mixed data types. This paper investigates two methodologies for distance-based cluster analysis of mixed wind velocity data. The first method involves transforming continuous and circular variables into a common type, followed by clustering using the Euclidean distance metric. However, this approach can distort the original circular space and lead to increased feature dimensionality. In contrast, Gower’s distance metric is explored as a solution. Gower’s metric facilitates clustering while preserving original data types, accommodating various variable combinations without the need for explicit transformations.This study focuses on wind velocity in the context of South Africa’s drive to enhance renewable energy, particularly wind power, in its energy mix. The research compares two clustering methods: one utilizing the Euclidean distance metric for spatiotemporal wind velocity data, and the other using Gower’s distance metric for Relational Data analysis. These methods are applied to the South African Renewable Energy Development Zones, aiming to identify similar clusters of locations based on the wind characteristics. The investigation considers two cases, only wind direction, as well as the wind speed and wind direction combined. The K-medoids algorithm is used for clustering, and the results are compared using a variety of validation methods. These methods include elbow plots, silhouette scores, within-class to between-class scatter ratios, and probability density functions, which are employed to assess the quality of the cluster outputs.