Sizing, Operation, and Evaluation of Battery Energy Storage with Dynamic Line Rating and Deep Learning

Abstract

Integration of renewables has the potential to reduce society’s reliance on carbon-based fuels. Sizing renewable-energy installations is key to making a successful business case for the construction of new assets. Dynamic thermal line rating is the amount of current in real-time that a transmission line can safely carry, and, if utilized together with utility-scale battery installations, has the potential to reduce the capacity and power rating of batteries. Today, battery energy storage systems are operated generally with rule-based approaches where utility-scale batteries charge when they can, and discharge then they have to, or change their energy amount with the time of the day. This works aims to improve on that by introducing smart control of batteries using deep learning and deep reinforcement learning-based methods. Transmission operators today generally assume the line ampacity of the transmission lines to be stable, referred to as static line rating. Dynamic line rating, however, can be multiple times higher than static line rating, enabling the ability to sent more power across the transmission lines. The aim of this work is to size battery energy storage systems taking into account dynamic thermal line rating, transmission line outages, and battery degradation, and explore decentralized control of batteries. A combination of non-linear programming for battery action prediction is used together with deep learning-based forecasting of ampacity and load. The approach is tested on IEEE 24-bus test grid. A deep reinforcement learning-based approach is utilized to predict battery actions, and test it on IEEE 6-bus test grid. A method for evaluating battery capacity and power rating sizes based on comparing the selection criteria with the allowed tolerance in unserved energy, unserved energy duration, and the number of unserved energy events is presented. A method to generate synthetic 100-step time series of Alberta electricity pool prices and dynamic thermal line rating is demonstrated, that can be used to supplement existing data sets, and be applied in reinforcement-learning simulations.