Self-supervised learning method for consumer-level behind-the-meter PV estimation

Driven by cost reduction and sustainable policies, the penetration of distributed photovoltaic (PV) systems has deepened in recent years. Most of these PV systems are installed behind the meter (BTM), where utilities cannot monitor their output levels directly. Some supervised methods have been studied to estimate BTM PV generation. These methods, however, cannot achieve accurate estimation without the dependency on training data labeled by additional measurements. As an alternative, a self-supervised learning method is proposed in this paper to train supervised estimation models from unlabeled data. Specifically, our proposed method synthesizes pseudo labels for unlabeled net load measurements using PV generation measurements of a small group of PV sites. Moreover, an end-to-end network architecture is proposed as the base estimation model. Based on a linear embedding of PV generation, the proposed end-to-end architecture can be directly trained with PV generation labels, which leads to a simplified training process and improved estimation performance. Extensive numerical simulations on two datasets from different hemispheres are carried out to verify the effectiveness of the proposed methodology.

Automated summary

Driven by cost reduction and sustainable policies, the penetration of distributed photovoltaic (PV) systems has deepened in recent years. This paper proposes a self-supervised learning method to train supervised estimation models from unlabeled data. The method synthesizes pseudo labels for unlabeled net load measurements using PV generation measurements and utilizes an end-to-end network architecture for improved estimation performance.