This presentation explains the resent methodologies utilized to estimate the Battery State of Charge with the help of a Data-Driven and Filter-Based approach for enhancing the performance parameters of a Battery Management System.
Data driven and filter based advancements in a (2) (2)
1. Battery State of Charge (SOC) Estimation Techniques
using the Data Driven and Filter Based Approach to
Enhance the Performance Parameters of a Battery
Management System (BMS).
Prepared by –
Jeet Patel, Bhaumik Joshi, Siddhi Vinayak Pandey
Department of Electrical Engineering,
Adani Institute of Infrastructure Engineering
Ahmedabad, Gujarat, India-382421.
2. Contents
Introduction
Why and How BMS?
What is SOC and SOH and why are they important?
SOC estimation using Data Driven approach
SOC estimation using Filter based approach
Conclusion
References
3. Function of Battery Management System
A BMS has the following priorities:
Protects safety of the operators from the host applications.
Detects unsafe operating conditions and responds Protects cells of battery
from damage in abuse/failure cases Prolongs life of battery (normal
operating cases)
Maintains battery in a state in which it can fulfill its functional design
requirements Informs the host-application control computer
how to make the best use of the pack right now (e.g., power limits),
control charger, etc.
4. State of Health
State of health determines the health of the battery i.e. how long the battery lasts.
There is no device to directly measure state of health of a battery.
It is measured using the values on voltage, current, energy density, power density,
temperature etc.
SOH is used for learning how much abuse of a battery is done by the user.
SOH predicts how much charge-discharge cycles are left in the battery before it
becomes futile.
5. State of Charge
State of charge determines the amount of charge present in the battery at a point
of time.
As like the SOH; SOC cannot be measured directly using a single device. There are
special techniques used to measure SOC which will be discussed later.
SOC is most important parameter of a battery. It is basically used to measure how
much power is left in the battery, which is used to estimate that how long the
battery will last before it is completely discharged.
In other words how much time it can deliver the power to a given load.
6. SOC Estimation Using Data Driven Approach
As of today, with huge computation power and lots of data available
about the battery, such as battery voltage, current value, temperature,
cycle number, etc.
Few Algorithms which will be covered in the next few slides are:
SVM (Support Vector Machine)
Neural Networks
Recurrent Neural Networks
Reinforcement Learning
7. SVM (Support Vector Machine)
This algorithm tries to find the ideal boundary in the Nth
dimensional, which easily separates your dataset as shown in
below figure.
Kernel are used to take data as input and transform it into the
required form. For example Linear, Polynomial etc.
8. Neural Networks
This is a complex algorithm, derived from the most complex thing
in the universe, “The Brain”.
It takes N-Dimensional input and tries to find the pattern in the
given input, output.
The network shown in the figure has 3 layers Input layer, Hidden
layer, and Output layer. As shown in below figure -
9. Recurrent Neural Network
An efficient method for predicting the time-series data.
It considers the relevant data taken from the past or earlier states for
estimation; which has been shown in below figure-
10. Reinforcement Learning
If there is less amount of data. This algorithm can come in handy.
In this algorithm, predictions are made in real-time and after some
iterations, the accurate predictions are made.
This algorithm decides suitable actions to be taken to maximize the
output.
11. SOC estimation using Filter Based
approach
Authentic Time Observer based series
Real time estimation of SOC and SOH
Kalman Filter (KF) for linear estimation
Extended Kalman Filter (EKF) and Double Extended Kalman Filter
(DEKF) for Nonlinear Systems.
Unscented Kalman Filter (UKF) / Sigma Point Kalman Filter (SPKF) for
highly nonlinear systems.
12. SOC estimation using Kalman Filter
Prediction of (k) based on (k-1….n).
Based on current state (k) of V,I and T; the %SOC is calculated [d].
Combination with Coulomb Counting technique, Kalman filter (KF)
estimates the accurate overvoltage and current in real time domain [e]
[f].
Estimation of overvoltage and current is done by combining the
Coulomb counting with linear KF.
It also estimates the SOC (better than the individual Coulomb counting
based estimations).
13. SOC estimation using Kalman Filter
As mentioned in below graph; the %SOC utilizing the Kalman Filter based
approach has been estimated [g] –
14. SOC estimation using Extended Kalman
Filter
The Auto Regressive Exogenous (ARX) battery model with EKF is
utilized for accurate % SOC estimation of a Lithium-ion based
battery [h].
Temperature incrementation causes the enhancement in the
nonlinearity within a battery model [h].
The EKF decreases the extreme terminal voltage error from 0.0397
to 0.0165 V. While the mean error value drops from 0.0014 to
0.0007 V [h].
15. SOC estimation using Extended Kalman
Filter
The graphical representation of ARX battery model with EKF is mentioned in below
figure [h] -
16. SOC estimation using Unscented Kalman
Filter
Sigma point is more efficient compare to mean and approximation method for the
estimation of %SOC.
The UKF filter for SOC estimation was designed using the posterior mean and
Taylor series expansion method [i].
When UKF implemented for battery cell model, the robustness for faster
convergence tracking ability of SOC is found very accurate[j].
When the LS-UPF (Least Squared Unscented Particle Filter) is compared with GA-
UPF (Genetic Algorithm and Unscented Particle Filter); performs the better %SOC
estimation utilizing the UKF technique.
17. SOC estimation using Unscented Kalman
Filter
SOC estimation through UKF/SPKF for faster convergence tracking ability is explained in
below attached figure [j] –
18. References
[a] Packtpub SVM. [Online]https://subscription.packtpub.com/book/big_data_and_
business_intelligence/9781789345070/3/ch03lvl1sec30/svm-for-churn-prediction
[b] A.A. Hussain, Kalman Filters versus Neural Networks in Battery State‐of‐Charge
Estimation: A Comparative Study, International Journal of Modern Nonlinear Theory and
Application. 3 (5) (2014) 199–209, https://dx.doi.org.104236/ijmnta.2014.35022 .
[c] Dlology SOC. [Online] https://www.dlology.com/blog/how-to-use-return_stateor-
return_sequences-in-keras/.
[d] G.L. Plett, Extended Kalman filtering for battery management systems of LiPB-based
HEV battery packs: Part 1. Background, Journal of Power Sources. 134(2) (2004) 252–261,
http://dx.doi.org/10.1016/j.jpowsr.2004.02.031 .
[e] J. Zhang, J. Lee, A review on prognostics and health monitoring of Li- ion battery,
Journal of power sources. 196(15) (2011) 6007–6014,
http://dx.doi.org/10.1016/j.jpowsour.2011.03.101
19. References
[f] W. He, N. Williard, C. Chen, M. Pecht, State of charge estimation for electric vehicle
batteries using unscented kalman filter, Microelectron
[g] S.V. Pandey, J. Patel, H. Dhiman, Battery State of Charge Modeling for Solar PV Array
using Polynomial Regression, Systems and Controls, arXiv
https://arxiv.org/abs/2008.09038
[h]Youtube SOC. [Online] https://www.youtube.com/watch?v=Gc4wjOEAoiA
[i] TOWARDSDATASCIENCE ekf, https://towardsdatascience.com/the-unscented-
kalman-filteranything-ekf-can-do-i-can-do-it-better-ce7c773cf88d/ ekf, last accessed
2020/03/26.
[j] H. He, H. Qin, Y. Shui, K. Oleksandr, Lithium ion battery SOC estimation with UKF and
RTOS µ COS – II Platform, Energy Procedia. 61 (2014) 468–471,
http://dx.doi.org/10.1016/j.egypro.2014.11.1150 .