Real-time prediction of grid voltage and frequency using artificial neural networks: An experimental validation

In grid-connected Distributed Generation (DG) systems, with high-penetrations of renewable and energy storage assets, the prediction of grid voltage and frequency plays an important role in enabling the power quality support, the stabilization and monitoring of distribution networks. In this paper, a method based on Artificial Neural Networks (ANNs) and Deep Recurrent Neural Networks (DRNN) has been developed for very short-term prediction of grid voltage and frequency. For different time scales (183ms, 1s, 10s, 60s), one-step and multistep ahead forecasters are developed to predict the future behavior of grid parameters. This type of predictors can be used in distributed generation systems to enhance the control performance, to prevent the occurrence of grid faults and to improve the power systems stability. The data used to establish and validate the ANNs forecasters are provided from grid connected battery storage system installed at the University of Manchester. The developed prediction models have been validated experimentally via a dSPACE real-time controller. The obtained results show that the ANNs forecasters are able to predict in real time the grid voltage and frequency with satisfactory accuracy as the largest mean absolute percent error is 0.32%.