Phaser measurements estimation on distribution networks using machine learning

Abstract

The uptake of distribution generation on electricity distribution networks imposes the operators to install new measurement devices such as phasor measurement units to achieve network observability. In this paper, we propose a framework for estimating synchronized phasor measurements for a virtual node using the measurements from the other nodes in the network. This system uses a machine learning method, in particular supervised regression models, to provide estimates. We show the performance of the proposed framework comparing two widely used regression methods i.e., Generalized Linear Models and Artificial Neural Networks. We extensively evaluate the proposed approach utilizing a real-world dataset collected from a medium voltage ring feeder. Our results indicate very low error rates; the average error for voltage magnitude was approx. 0.2V while for phase angle was 0.7mrad. Such low errors indicate the potential for reducing the scale of the measuring infrastructure required on distribution networks and increasing their reliability.