A Comparative Study of Clustering Algorithms on Power Consumption Data: The Role of Cyclic Feature Extraction

Power consumption data, as an increasingly important type of data in the digital era, plays a key role in power forecasting and consumption clustering analysis. In this study, we hypothesize that meaningful features can be extracted from the cyclic nature inherent in energy consumption data. The ability to interpret and classify this data from new perspectives provides valuable contributions to the field of data science. To test this hypothesis, we examine clustering performance using hybrid methodologies, focusing on extracting non-cyclic features from various data types and cyclic features from wavelet-transformed data. By conducting separate analyses for each, we assess the outputs of two well-known algorithms, namely, OPTICS and DBSCAN, both of which depend heavily on precise parameter selection and optimization for accurate results. After obtaining the results, we observe a clear and decisive outcome in OPTICS clustering. On the other hand, in the DBSCAN clustering algorithm, effective clustering is observed regardless of whether the data is non-cyclic or cyclic.