Deep Learning Control of Transmission System with Battery Storage and Dynamic Line Rating

Abstract

Battery energy storage in utility-scale transmission grids provides the benefit of fast response, however, efficient battery control in multi-battery multi-bus systems can be challenging. We present here a battery operation strategy based on forecasted load and line ampacity. The forecasted load is serviced via conventional generators in combination with battery energy storage whose outputs are computed using non-linear programming with the objective of minimizing total battery charging and discharging. The operating strategy takes into account battery degradation, line outages, and dynamic line rating. The forecasting model is based on attention convolutional neural network architecture with bidirectional long-short term memory layers forecasting over the range calculated using the sliding windows. The strategy is tested on 24-bus reliability test system and is shown to be effective at predicting battery action.