Voltage State Estimation in a Microgrid with Temporal Convolutional Network and Various Feature Dimensions

With the rising penetration level of distributed renewable energy in the distribution network, there are new challenges related to the operation of these systems. As a subsystem within a distribution network, a microgrid is operated with a control strategy to optimally manage the customer assets. It requires to consider the voltage state and other variables. Traditional voltage state estimation relies on assumptions that do not apply to modern complex microgrids. In this paper, a machine learning approach for voltage estimation considering missing data is proposed by implementing Temporal Convolutional Networks. The missing data are imputed based on the historical data and the real online measurements. The training and evaluation are performed using the database from the network of advanced metering infrastructure in the microgrid of the University of Cyprus. The results suggest a good model fit with the mean absolute percentage errors and normalized root mean square errors below 0.1% for all studied scenarios.