Canonical Correlation Analysis and Visualization for Big Data in Smart Grid

Abstract

Electricity consumption behaviors are influenced by various external and internal factors such as climate, location, building type, consumer characteristics and even other energy consumption. In order to investigate the electricity consumption behaviors of diverse consumers, we propose a methodology based on canonical correlation analysis to explore the correlation among electricity consumption, gas consumption and climate change under different circumstances. We first preprocess three multivariable datasets that contain 24-value daily data in a one-year period, and conduct consumer segmentation based on climate zones, locations and building types. Then an optimized canonical correlation analysis model with an optimal result selection mechanism is adopted to calculate the canonical correlations and weights of every set of daily data. Finally, we propose a post-processing analysis for further comparison on the calculated results. We investigate three research questions to present and discuss the analysis results, including canonical correlation and weights overview, typical patterns analysis, and comparison on climate zones and locations.