Recovering time-series missing data - an evolutionary approach to GECCO 2015 industrial challenge

GECCO 2015 industrial challenge [1] aimed to compare procedures for recovering missing information in a heating system. In this competition, four timeseries (Figure 1) were provided, measuring water temperature or heating power at minutely intervals. To simulate missing data, parts of the time-series were dropped, with size and frequency of the gaps sampled from an exponential distribution function. The data was divided to 7 periods: The first and the last periods had no missing data, four other periods only had data from one of the time-series missing, and finally, in one of the periods, all of the series were missing at random intervals at the same time [2]