SWDWHRICKDUJH Estimation of Lithium-ion Battery %ased on an
,mproved Kalman Filter
Hao Fang
1
,Yue Zhang
1
, Min Liu
1
*, Weiming Shen
2
1
College of Electronic and Information Engineering, Tongji University, Shanghai, China, 201804
2
The Key Laboratory of Embedded System and Service Computing, Tongji University, Shanghai, China, 201804
1531777@tongji.edu.cn,
yuezhang@tongji.edu.cn,
*
lmin@tongji.edu.cn, wshen@ieee.org
Abstract-Accurate state-of-charge (SOC) estimation is
essential to battery management system. The widely
adopted estimation methods based on Kalman Filter (KF)
fail to take the variable environmental conditions into
consideration, which may result in a poor accuracy. This
paper proposes a novel estimation model based on KF
method to estimate SOC of Lithium-ion battery. In the
proposed model, the noise variances are optimized for the
system current state at each iteration, a variable forgetting
factor is introduced to improve the algorithm’s
convergence and accuracy of estimation, and the artificial
neural network (ANN) is applied for the measurement
equation of KF. The experiments, based on Lithium-ion
Battery set of NASA, show that the proposed SOC
estimation model is valid and can improve the algorithm
performance and accuracy and robustness.
Key words-Kalman Filter (KF); state of charge (SOC);
variable forgetting factor; artificial neural network (ANN)
I. Introduction
With increasing concern about the energy shortage and the
environmental pollution, battery is being paid more attention
by batteries researchers and producers because of its high
density of energy and multi-cycle lifetime [1]. In particular,
Lithium-ion Battery attracts more interests due to its many
advantages such as high voltage, low self-discharging, long
cycle lifetime and so on. It is known that more and more this
kind of batteries are used to electric vehicles (VEs) and other
places. However batteries’ wide application is limited by
several serious deficiency of battery technology including
batteries’ misusing and crash of battery management. It leads
to short lifetime and poor security. So it is urgent to build an
effective battery management system (BMS) to make full use
of energy inside it and extend its lifetime [2]-[5]. The
functions of BMS include having knowledge of batteries’
condition and implementing suitable strategies to guarantee
batteries’ health, extend its lifetime and make the batteries
work more efficiently [6].
It is essential for BMS to estimate SOC which represents
remain amount of charge available in the batteries accurately.
Because it is a very important index to help user to determine
whether the batteries is overcharged (or over-discharged), or
whether one of the batteries is replaced because its poor
performance. However the value of can not be measured
directly by a certain sensor because of its chemical structure.
As the condition of the battery changes over time, it is
acceptable that SOC is estimated by a suitable algorithm using
relevant measurable parameters which includes terminal
voltage, charge (or discharge) current, temperature and some
other relevant factors. In recent decades, a lot of researches
have been conducted in this filed and many novel algorithms
have been proposed by these researchers, although above
studies are supposed to implement an accurate estimation,
some shortcoming still need to overcome: (1) some estimation
methods are based on an equivalent electronic circuit model,
on the one hand, it seems that the complexity of this
equivalent circuit model is closely related to the estimation
precision. But it improves estimation performance at the cost
of heavy computation burden, which will reduce algorithm’s
efficiency. On the other hand, as SOC is also related to
changing ambient temperature and aging level of battery, the
parameters inside this model are also changing over time. So
this method is suitable for rapid changed circumstances. (2)
Some estimation methods treat the battery construction as a
black-box system rather than specific physics construction.
ANN, representative of this kind of model, is capable of
approximating battery’s nonlinear relationship. But the model
error is restricted by the trained data and methods.
Nowadays, the estimation accuracy is been improved by a
certain algorithm because of its advantage, and many valid
battery data is available in some research institute. Taking
these factors into consideration, fusion model is a feasible
method to optimize estimation accuracy. Therefore, this paper
proposes an estimation model with KF and ANN. Firstly a
ANN model trained by NASA’s battery set is used as
measurement equation in KF theory, which its output is a
relatively SOC. Secondly in order to compensate the
measurement error, the recursive least squares (RLS)
algorithm with a variable forgetting factor is introduced to this
model. Finally, an accurate estimation can be obtained by
using this improved KF.
The outline of this paper is organized as follows. Related
work is presented in Section II. The battery model is
introduced in Section III. Section IV explains the proposed
SOC estimation algorithm which is utilized to combine ANN
Proceedin
s of the 2017 IEEE 21st International Conference on Computer Supported Cooperative Work in Desi
n
978-1-5090-6199-0/17/$31.00 ©2017 IEEE 515