Probabilistic baseline estimation based on load patterns for better residential customer rewards

Abstract

Residential customers are increasingly participating in demand response program for both economic savings and environmental benefits. For example, baseline estimation-based rewarding mechanism is currently being deployed to encourage customer participation. However, the deterministic baseline estimation method good for commercial users was found to create erroneous rewards for residential consumers. This is due to larger uncertainty associated with residential customers and the inability of a deterministic approach to capturing such uncertainty. Different than the deterministic approach, we propose to conduct probabilistic baseline estimation and pay a customer over a period of time when the customer’s predicted error decreases due to reward aggregation. To achieve this goal, we analyze 12,000 residential customers’ data from PG&E and propose a Gaussian Process-based rewarding mechanism. Real data from PG&E and OhmConnect are used in validating the algorithm and showing fairer payment to residential customers. Finally, we provide a theoretical foundation that the proposed method is always better than the currently used industrial approaches.