Energy consumption prediction using people dynamics derived from cellular network data

Bogomolov, Andrey; Lepri, Bruno; Larcher, Roberto; Antonelli, Fabrizio; Pianesi, Fabio; Pentland, Alex

doi:10.1140/epjds/s13688-016-0075-3

EPJ Data Science

Table 4 Peak daily energy consumption features. Mean daily energy consumption features

From: Energy consumption prediction using people dynamics derived from cellular network data

Rank	Feature	Decrease in MSE	Decrease in node impurity
1	Number of consumers per grid powerline	21.22	132,035.53
2	Entropy of temporal sum of outgoing calls	5.86	24,013.99
3	Kurtosis of temporal entropy of Internet activity	4.33	15,479.71
4	Entropy outgoing calls fundamental frequency	4.10	14,650.13
5	Sum of 4 harmonic of incoming calls	3.96	11,449.25
6	Skewness of 5 harmonic of incoming calls	3.66	9,595.91
7	Skewness of temporal entropy of Internet activity	3.51	9,498.52
8	Sum of spectral distribution skewness of outgoing calls	3.47	2,108.19
9	Sum of temporal distribution kurtosis of Internet activity	3.35	2,267.17
10	Spatial variance of spectral variance of incoming SMS	3.30	3,971.29
11	Sum of 5 harmonic of incoming calls	3.27	9,468.86
12	Sum of calling direction area codes	3.25	6,815.90
13	Spectral variance of outgoing SMS activity total in time	3.20	13,899.48
14	Spatial skewness spectral skewness distribution of Internet activity	3.19	3,512.42
15	Kurtosis of temporal mean distribution of outgoing SMS	3.17	27,704.10
16	Spectral variance of temporal median distribution of outgoing SMS	3.02	15,366.03
17	Spatial median of incoming temporal sum	2.99	2,642.34
18	Spectral variance of outgoing SMS fundamental frequency	2.96	11,624.28
19	Skewness of incoming calls temporal entropy	2.92	4,174.09
20	Sum of outgoing calls 5 harmonic	2.89	10,251.36
21	Variance of Internet activity temporal entropy	2.86	5,382.14
22	Median of calling direction area codes	2.85	3,591.92
23	Sum of outgoing calls 4 harmonic	2.84	4,440.13
24	Median of incoming calls 16 harmonic	2.74	943.98
25	Spatial standard deviation of outgoing call temporal skewness	2.73	2,502.05
26	Spatial standard deviation of outgoing call temporal kurtosis	2.72	2,888.08
27	Skewness of incoming calls 4 harmonic	2.71	8,041.38
28	Spectral distribution skewness of mean Internet activity	2.69	1,759.88
29	Spatial standard deviation of temporal Internet activity entropy	2.68	4,233.89
30	Variance of outgoing calls temporal distribution kurtosis	2.67	3,283.40
31	Kurtosis of spectral distribution skewness.	2.67	5,160.36
32	Sum of Internet activity 4 harmonic	2.65	1,695.49

For each variable from CDRs we computed the mathematical functions, which characterize the distributions and measure the information theoretic and statistical properties of such variables, e.g. mean, median, standard deviation, min and max values and Shannon entropy. Moreover, the same computations were repeated on sliding windows of variable length (1-hour, 4-hour and 1 day), producing second-order features that capture temporal relationships.

Back to article page