Abstract: We focus on the analysis of electricity power load in Czech Republic which exhibits seasonality as well as other periodic trends typical for European countries. The presented approach uses Multi-fractal Detrended Fluctuation Analysis method (MF-DFA) from statistical physics to analyze extremely large power load datasets with one minute resolution. Extraction of stochastic part of the signal using Fourier transform allows us to apply this method and to estimate Hurst exponent. The resulting fluctuation function of the dataset is characterized by heavy-tail distribution with no finite moment. Generalized Euclidean metric with variable exponent q is used to analyze this fluctuation function. We found that the autocorrelation function exhibits persistent behavior for q<1 and it is anti-persistent for q>1. Knowledge of properties of autocorrelation function allows better prediction of electricity load as well as measure of risk.