Disaggregation of household energy consumption with principal component analysis and sparse filtering

A detailed understanding of household electricity consumption across heating, cooling, and baseline electrical loads is critical for effective demand-side energy management and the development of targeted energy efficiency strategies. This paper presents two methodologies—Principal Component Analysis (PCA) and Sparse Filtering—for disaggregating total daily electricity consumption into baseline electricity, heating, and cooling loads, using data collected from 68 households in Cyprus over one year. PCA, a well-established technique for dimensionality reduction, identifies two dominant components explaining 52.27% and 14.83% of the total variance, attributed primarily to cooling and heating, respectively. Sparse Filtering, in contrast, produces two distinct temporally localized features: one active in winter, the other in summer. This suggests a disentangled, seasonally aligned representation of energy use. While PCA emphasizes global variance, Sparse Filtering captures household-specific patterns – even when PCA components are near zero or negative – highlighting periods of low consumption potentially attributable to base electrical loads. These findings demonstrate the potential of Sparse Filtering to offer more interpretable and granular insights into occupant energy consumption than traditional PCA-based approaches. The complementary strengths of the two approaches improve the interpretability of disaggregated household load profiles and increase the robustness of unsupervised energy disaggregation methods. • Energy disaggregation into heating, cooling, and base loads from total consumption. • Applying PCA and Sparse Filtering for disaggregation using only power data. • Enhancing NILM by reducing data size while preserving usage patterns. • Comparing PCA vs. Sparse filtering for separating seasonal energy loads.