Robust Data-driven Linearization for Distribution Three-phase Power Flow

Abstract

It is an extra heavy burden to maintain the distribution network model since of its frequent changing and large scale. In reality, the network model may involve significant parameter errors, which makes the model not applicable. This paper presents a robust data-driven approach to linearize the three-phase power flow. A three-phase linear power flow (LPF) model for active distribution networks (ADNs) is proposed. To address the issues of measurement outliers and data collinearity, the support vector regression (SVR) algorithm is applied to the linearization regression to obtain the proposed three-phase LPF model. The training process is offline and the obtained LPF model can be used to calculate the power flow online. Numerical tests demonstrate that the proposed approach can achieve satisfactory calculation accuracy with high computational efficiency, even under data collinearity and measurement outliers. Since the proposed LPF has very high accuracy, this model can be incorporated into distribution networks optimization problems.