@article{4814,
  author = {Olaniyi Olawale Omoyajowo, Ezendu Ariwa and Yue Yong},
  title = {Spatial Clustering of the NREL PVDAQ Photovoltaic Systems: Geographic Structure of a National Performance Dataset},
  journal = {Journal of Information Organization},
  year = {2026},
  volume = {16},
  number = {3},
  doi = {https://doi.org/10.6025/jio/2026/16/3/99-121},
  url = {https://www.dline.info/jio/fulltext/v16n3/jiov16n3_1.pdf},
  abstract = {The increasing deployment of photovoltaic (PV) systems has created a growing need for reliable, geographically
representative datasets for performance assessment, degradation analysis, and power forecasting.
However, the spatial organization of large PV monitoring archives is often not well understood, which can
introduce sampling bias and limit the generalizability of predictive models. This study presents a spatial
clustering analysis of the National Renewable Energy Laboratory (NREL) Photovoltaic Data Acquisition
(PVDAQ) public dataset, examining the geographic structure of 1,862 PV systems distributed across the
United States. Using latitude and longitude as clustering variables, three complementary unsupervised
methods are applied: K-Means, DBSCAN, and hierarchical Ward clustering. Internal validation using silhouette
scores indicates strong and consistent spatial structure. K-Means identifies 16 macro regions with a silhouette
score of 0.701, hierarchical Ward clustering identifies 17 similar macro regions with a silhouette score of
0.697, and DBSCAN identifies 34 denser local clusters while classifying 5.16% of systems as spatial outliers.
The resulting clusters correspond to recognizable deployment and climatic regions, with pronounced
concentration in California and the Northeast corridor. The findings demonstrate that the PVDAQ network is
not randomly distributed but exhibits a clear geographic architecture. These spatial strata provide a
foundation for stratified sampling, regional performance analysis, outlier detection, and future spatially
aware PV forecasting models, including deep learning and physics informed machine learning approaches.},
}