Anticipation of minutes-ahead household active power consumption using Gaussian processes

In price directed electricity markets, participants continuously monitor the cleared electricity prices and respond to them with the amount of energy they would like to purchase. Thus, electricity purchase decisions are significantly facilitated by anticipating future consumption. In this paper, a data-driven method for anticipating the active power consumption in households is presented. In particular, Gaussian processes (GP) from the machine-learning realm are used for anticipation of electrical consumption at the level of individual households. Additionally, the performance of Gaussian processes equipped with various kernel functions is benchmarked against the approach of autoregressive moving average (ARMA) for anticipation of ten minute-ahead household consumption. The results indicate that GP outperforms ARMA in minute-ahead consumption anticipation, while there is not a dominant kernel that outperforms the rest within the GPR models.