Crisp 11

Optimization of Battery Management Systems Using
Artificial Intelligence
Iqra Mohammed Yakub, Syed Umer Ahmed, Shahd Darwish, Fahrettin Ay, Hasan Kurban
Electrical and Computer Engineering Program, Texas A&M University at Qatar, P.O. Box 23874, Doha, Qatar
ABSTRACT RESULTS
This research aimed to develop a Machine Learning (ML) To measure the goodness of fit, we used seventeen performance matrices namely, MAE (mean absolute error), RMSE (root mean
software capable of performing comprehensive regression squared error), and R-squared (coefficient of determination), etc, that are given as:
analysis through employing ensemble techniques combining σn yi )2 σn
1 n i=1(yi −ෝ i=0(yi −ෝyi )
multiple models and utilizing evaluation metrics to determine MAE = σ y − yො i , RMSE = , R2 = 1 −
n i=1 i n σni=0(yi −ഥ
y)
the most effective regression model for the given dataset. As a
case study, Electrochemical Impedance Spectroscopy (EIS)
measurements on Lithium-ion Batteries (LiBs) were used to
test the effectiveness of the software. To address this problem,
a software solution was developed through Python that
facilitates the analysis of diverse regression models and metrics
on the dataset. Our innovative approach, Comprehensive R-squared
Regression for Impedance Spectroscopy Prediction (CRISP),
leverages Artificial Intelligence (AI) to efficiently evaluate
numerous regression models across a wide range of metrics.
By focusing on enhancing EIS measurements, particularly in the
extra low frequency (ELF) region, CRISP demonstrates
MAE and RMSE Evaluation of each Regression Model
remarkable speed and accuracy, even under varying charging
states to predict real and imaginary parts of impedance
behavior Z, using a modest dataset. This pioneering software
holds significant promise for enabling online monitoring and
characterization of LiBs, thereby accelerating advancements in
LiBs technology.
INTRODUCTION
(LiBs) or lithium-ion batteries are assemblies of energy-dense
electrochemical devices (cells) whose size and effectiveness
has made their use omnipresent from simple sensors to
aircraft modules. EIS is an important technique for analytically
mapping out the characteristics and properties of LiBs. From
an electrochemical perspective, impedance can be described
as the collective opposition of the system to the flow of
electrical charges under AC excitations. Impedance Z can be
described as:
Heat Map Illustrating the Scaled Validation Score
Voting Regression Technique for Zr and Zj
This integrative, new study, combines the results of

experimental EIS measurements with a novel AI technique
called CRISP (Comprehensive Regression for Impedance
Spectroscopy Prediction) to regenerate the LiB impedance
including ELF range down to μHz region by letting a large Regenerated Nyquist Plot Visualization of the regression surface on empirical impedance data
pool of regression algorithms compete individually for
predicting both real Zr and imaginary parts Zj of Z. In this CONCLUSION & FUTURE WORKS
study the ML model is optimized by selecting the best fit In developing CRISP (Comprehensive Regression for Impedance Spectroscopy Prediction), we've successfully harnessed ensemble
function from a select possibilities to fit a given dataset. The techniques to model real-time events, specifically focusing on impedance, state of health, and charge dynamics in lithium-ion
optimization equation to find the best model for a given batteries (LiBs). Through our efforts, we have achieved our primary objective of significantly enhancing the prediction accuracy of
dataset is, F = {f |f : R → R} and a dataset ∆, determine f ∈ F, battery state of health, effectively reducing error rates typically ranging between 5-10% and hence extending the lifetime of
such that ∀g, g ∈ F → f ≤ g: LiBs. Leveraging Electrochemical Impedance Spectroscopy (EIS) data and our advanced software, we have created an application
that offers insightful analysis of battery health in real-time. Our accomplishment marks a significant step forward in revolutionizing
battery management practices, enabling the prolonged lifetimes and optimized performance of batteries across various
applications.
DESIGN AND IMPLEMENTATION NOTE The aim for the future is to deploy the CRISP application as a public web service that can be easily accessed online, along with
publishing the open-source code allowing for multiple renditions of the software. The ultimate objective is to deploy the most
Experiments were conducted to identify the most suitable pair refined version of the CRISP software onto Electric Vehicle (EV) batteries. By continuously enhancing and optimizing CRISP, we aim
of models that can fit the training data and generalize new data. to create a sophisticated battery management system tailored specifically for the unique requirements of EVs.
The protocol involved 33 regression models available in Scikit-
learn. One ensemble model was created based on the models ACKNOWLEDGEMNENT
with higher weights. The model's performance was evaluated We extend our sincere appreciation to Dr. Hasan Kurban for his outstanding leadership and invaluable guidance throughout this
using 17 different metrics, and a heat map was generated to research endeavour. Special thanks are due to Fahrettin Ay for his significant contributions to designing the machine learning
display the cross-validation scores of each model. The voting models, which greatly enhanced the depth and scope of our study. Additionally, we would like to acknowledge Dr. Selcuk Temiz for
regression model technique was used to select the top six his pivotal role in collecting the Electrochemical Impedance Spectroscopy (EIS) data, which formed the foundation of our research
models, and their weights were averaged to create an ensemble and allowed for a sophisticated testing dataset which CRISP could be executed on.
model. The data was initially split into training and testing sets.
Two experiments were conducted, one for predicting Zr and the REFERENCES
other for Zj. A machine learning software has been developed Grey, Clare P., and David S. Hall. “Prospects for Lithium-ion Batteries and Beyond—a 2030 Vision.” Nature Communications, vol. 11, no. 1, 8
to support comprehensive regression analysis on different Dec. 2020
datasets. This tool is set to be deployed soon, making it Gaberšček, Miran. “Understanding Li-based Battery Materials via Electrochemical Impedance Spectroscopy.” Nature Communications, vol.
accessible for everyone to use. 12, no. 1, 11 Nov. 2021

Crisp 11

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Crisp 11

Uploaded by

Copyright:

Available Formats

Optimization of Battery Management Systems Using

Heat Map Illustrating the Scaled Validation Score

Voting Regression Technique for Zr and Zj

This integrative, new study, combines the results of

You might also like