Battery life estimation in a real-time energy management system

Batteries are complex electrochemical systems with poorly understood charge/discharge dynamics. Traditionally, batteries have been operated according to rules of thumb that require fixed charge/discharge patterns. However, these are not suitable for any sophisticated energy management system, particularly one that includes intermittent renewable generation. Such a system needs to be able to dynamically charge and discharge the battery based on the availability of renewable generation, grid tariff and load. Further, when such a system is operational, any analysis needs to be based on the observed quantities (like current and voltage). Previous research on battery models has ranged from detailed electrochemical models to simplified equivalent circuit models but none of these models simultaneously satisfy the requirements of: a) Being easy to fit (computationally and statistically) from observed quantities in real-time b) Working across various battery states and c) Being interpretable. In this paper, we propose a new model that combines a simplified equivalent circuit model with a model capturing the variation of the circuit parameters. We use a best fit algorithm to estimate model parameters. The trained model can be used both for offline diagnosis and online estimation of SOC as the battery is being used. The approach is validated using real data from lead acid batteries being operated as part of a microgrid. The results show that the model is computationally cheap, can model the I-V characteristics of the battery across discharge rates and battery states and can accurately predict how much longer the battery can be discharged for a given load profile.