A Battery Digital Twin Based on Neural Network for Testing SoC/SoH Algorithms

Abstract

Lithium ion cells are the preferred solution for the growing world of mobile applications. To avoid working conditions that can accelerate irreversible degradation reactions, two parameters must always be known, namely State of Charge (SoC) and State of Health (SoH). Since SoC and SoH cannot be measured directly with sensors on cells, they must be derived from the observation of voltage and current at the accessible connections. The literature on algorithms for SoC-SoH estimation is very rich and new advanced ones are continuously developed. However, the testing of algorithms on real batteries is very time consuming due to the need of many charge-discharge cycles in order to observe aging effects. These operations can take months in the lab. In this paper we present a Battery Digital Twin (BDT) that outputs a realistic voltage signal as a function of SoC and SoH inputs. The voltage signals produced by the BDT can later be used to feed SoC/SoH estimation algorithms. The BDT is thus a simulator for fast testing of battery parameter estimation algorithms. This paper presents a BDT developed in the Matlab-Simulink-Simscape environment. The non-linear Open Circuit Voltage (OCV) generator as a function of SoC/SoH is based on a feed forward Neural Network (NN) trained with real data from a publicly available repository. The internal complex impedance can assume fixed circuit configurations derived from typical Nyquist plots or can be dynamically adjusted by other trained NN as non-linear function of SoC/SoH.