Identifying distribution network line parameters and voltage angles by utilizing physical knowledge: A neural network approach

Abstract

The energy transition is pushing the Distribution System Operators (DSOs) to find solutions for better management of the network operating conditions. These requirements involve the installation of financially intensive metering devices or the application of synthetic datasets, due to the unobservability of distribution networks (especially low-voltage (LV) networks). The challenge is even more complex as LV networks are rarely accurately mapped in terms of their technical characteristics with non-existence of operationally recorded information such as switching states. This makes even basic network analyses, such as power flow studies, complex and often hinders further installation of low carbon technologies. To aid in the low observability challenge in LV distribution networks, this paper proposes a model based on the power flow equations for the estimation of the network admittance matrix. The neural network (NN) based approach is developed to mimic the previously estimated line switching states, while the main objective behind this model is the integration of physical information, such as power flow equations and network-operating constraints. Besides conductance and susceptance matrix, the proposed approach is applied for the estimation of voltage angle values in the network. The model behavior is examined on different network structures, as well as on the networks with a larger share of installed renewable energy systems (RES). The error rates of $\sim$1 % and $\sim$4 % in the estimation of conductance and susceptance matrix parameters indicate the importance of NN initialization parameters and demonstrate the effectiveness of the developed model in the operation of advanced distribution networks.