Probabilistic baseline estimation via Gaussian process

Demand response aims at utilizing flexible loads to operate power systems in an economically efficient way. A fundamental question in demand response is how to conduct a baseline estimation to deal with increasing uncertainties in power systems. Unfortunately, traditional baseline estimation lacks the ability to characterize uncertainties due to their deterministic modeling. This deficiency often results in erroneous system operations and miscalculated payments that discourage participating customers. In this paper, we propose a Gaussian process-based approach to mitigate the problem. It features the ability to use all historical data as a prior knowledge, and adjust the estimation according to similar daily patterns in the past. To characterize the uncertainties, this method provides a probabilistic estimate that can be used to not only increase estimation confidence for system operators but also to fairer treatment to customers. Finally, simulation results from Pacific Gas and Electric Company data show that this new method can produce a highly accurate estimate, which dramatically reduces the uncertainties inherent in the distribution power grid. Such a work opens the door for power system operation based on probabilistic estimate.