Comparative Analysis of Dimensionality Reduction Methods for Optimizing Load Profiling in Smart Meter Data

Data metering is a major component of the smart grid, which is used to monitor and measure energy consumption across seasons, zones, and weather conditions. The emergence of smart meters poses a challenge for utilities in handling the substantial amounts of data, commonly known as Big Data. It is crucial to extract important reduced-dimensional features from the original measurements for effectively analysis of this large dataset. This research evaluates the effectiveness of three dimensionality reduction methods—Principal Component Analysis (PCA), t-Distributed Stochastic Neighbor Embedding (t-SNE), and Random Projection—using a dataset from the UMassTraceRepository with Python’s Scikit-learn. Metrics like silhouette score and root mean square error (RMSE) are used for comparison. Findings reveal that PCA outperforms Random Projection in identifying distinct clusters, while t-SNE excels in clustering load profiles. These results highlight the critical role of choosing appropriate dimensionality reduction techniques to enhance clustering accuracy in utility load profiling, thereby improving demand-side management.