Volume 10: Sustainable Energy Solutions for Changing the World: Part II

Data-driven Approach for Battery Capacity Estimation Based on In-vehicle Driving Data and Incremental Capacity Analysis Felix Heinrich, Marco Pruckner

Abstract

To ensure safety, performance and warranty of an electric vehicle, it is crucial to monitor the evolution of remaining capacity of NMC lithium-ion batteries. Estimators for the remaining capacity are often based on costly, complex and time consuming testing procedures under laboratory measurement conditions. Other methods like incremental capacity analysis require various load sequences at very low constant current rates. This is also not practical for real battery electric vehicle operation due to high and dynamic discharging rates caused by the customers individual driving behavior as well as high recharging rates.
To overcome these problems, we present a data-driven approach for battery capacity estimation in combination with incremental capacity analysis. The missing load sequences for the incremental capacity analysis are presented by the output of a recurrent neural network which describes the battery electric behavior from real in-vehicle data. Results show RMSE deviations of 1.77% to correctly estimate the remaining capacity over the whole vehicle life. This high accuracy is comparable to state of the art laboratory battery testing, but without the need of expensive experimental data. Instead only operational vehicle data can be used.

Keywords Lithium-ion battery (LIB), remaining capacity, incremental capacity analysis (ICA), electric vehicle (EV), in-vehicle data, machine learning, LSTM