Battery State of Charge estimator
Introduction
The invention has the function of estimating the state of charge of lithium batteries for automotive applications, using a recursive NARX artificial neural network architecture. The invention takes in input the instantaneous current, voltage and temperature signals relating to the battery pack in order to estimate the state of charge.

Technical features
The invention allows to obtain the estimated state of charge of lithium batteries in an accurate and independent way from representative models of the battery and is characterized by a reduced computational cost. The system adopts a closed-loop version of the NARX architecture for an estimation problem, using the current, voltage and temperature measurements of the battery pack. The closed-loop configuration uses in feedback the estimated state of charge, computed in the previous moments, as a further input to the network, thus allowing to enrich the set of information useful for the estimation and consequently lightening the complexity of the resulting network, as well as improving the accuracy of the estimate, which is therefore more efficient than existing solutions.
Possible Applications
-
Automotive battery packs;
-
Battery management systems (BMS).
Advantages
-
Low computational cost for implementation in on-board control units;
-
High accuracy on real charge and discharge cycles;
-
Scalability for different battery packs and usage platforms;
-
Robustness with respect to errors on the initial state of charge or on input signals.