Design considerations for artificial neural network-based estimators in monitoring of distribution systems

Data-driven approaches based on Distributed Artificial Intelligence (DAI) such as Artificial Neural Networks (ANN) could be used to perform estimation of voltage magnitude in distribution systems for monitoring purposes. These methods may offer high accuracy and yet require relatively few measurement inputs and low computational power compared to conventional state estimation techniques. However, the number of required measurements may vary from system to system depending on several factors. Furthermore, it is important to ensure that these estimators are robust to input noise. Moreover, a factor to be considered in presence of sparse electrical measurements is that other additional inputs may be used to improve the accuracy of estimation. This paper investigates the decisive factors that affect the minimum number of input measurements for an ANN-based estimator. Furthermore, it discusses how the ANN should be designed to handle measurement noise properly in practice. Simulations are performed on benchmark networks to support the discussion.