Fractal analysis of complex power load variations through adaptive multiscale filtering

Power load analysis is important for optimizing resource allocation, planning the production of electricity, and predicting power markets. Yet, it is challenging, since load data exhibit both periodic and stochastic features, and is affected by a multitude of factors including social, economic, political, and climatic factors, as well as industrial structure, living standards, and user behaviors. In this paper, we employ a multiscale framework to systematically analyze load data from two electric utilities in two cities of different size in China. The low frequency trend signals in both load data sets are quite irregular. The detrended data of the load time series are further denoised to remove high frequency noise. Fourier spectral analysis of the original and filtered data shows that the load time series has very strong spectral peaks corresponding to a period of one day. Using adaptive fractal analysis, which can best extract fractal behaviors from signals with strong oscillatory trends, we further show that load time series has long-range correlations. Amazingly, maxima of the temporal variations of the long-range correlations correspond well with temperature minima, highlighting that long-range correlations are stronger in winter than in summer.