Professional Documents
Culture Documents
in EVs
Abstract: Accurate state of charge (SOC) is critical for products. However, overcharge and over discharge will
battery energy management system in electric vehicle (EV) cause permanently and irreversibly damage to lithium
application. Overcharge and over discharge will shorten batteries, which may even lead to potentially explosive. On
batterys lifespan and induce potential safety problem, which the basis of an accurate SOC, battery management system
may even permanently damage the lithium-ion battery. Thus, a
(BMS) can help expanding lithium batterys lifespan and
data driven model is proposed for improving the accuracy of
SOC estimation in this paper. A preliminary mathematic model ensuring its safe usage. Therefore, its is critical for BMS
under constant current is established, which can match the to keep batteries within a reasonable SOC range. [6]
primary high SOC stage. Since the battery model is nonlinear, However, EVs always operate in various driving cycles,
model free adaptive control (MFAC) is used to get the dynamic which makes it difficult to obtain an accurate SOC.
linearization model and accomplish SOC estimation process on Generally, BMS should monitor the battery current, voltage,
the basis of the mathematic model. With the small sample scale temperature for safety use of EV. These data have not been
of new data updated, a data driven model based on partial least sufficiently utilized to improve the accuracy of SOC
squares (PLS) is obtained online. The accuracy of the estimation. Machine learning method is recently
mathematic model also decreases during the operating process.
developing fast and is especially suitable for data analysis.
Finally, the calculated values from the two different models are
mixed for an accurate SOC. Experimental results on lithium
Consequently, by analyzing these data, machine learning
polymer battery prove the effectiveness of the proposed method. method is utilized in this paper for an accurate SOC.
Different kinds of methods have been proposed to
KeywordsSOC estimation, PLS, MFAC, Li-ion battery. obtain an accurate SOC, among which Coulomb counting
method and OCV method are two efficiently classical SOC
I. INTRODUCTION estimation methods. As an open loop estimator, Coulomb
Since the shortage of non-renewable energy resources counting method [7] estimates SOC by the integration of
and environmental issues become increasingly prominent, current with time. Thus, current errors from sensors will
more attentions are paid to energy saving and emission accumulate, and inaccurate initial SOC will also inevitably
reduction. Diesel locomotive relies heavily on fossil fuels yield errors in the estimation process. Open circuit voltage
and produces greenhouse gases to natural environment. (OCV) has monotonic relationship with SOC. As a result,
Electric vehicle (EV) has the advantages of highly efficient, OCV method estimates SOC on the basis of OCV-SOC
zero emissions, energy-saving, and low-noise by replacing table. [8][9] Measuring OCV requires many hours even
internal combustion engine (ICE) with electric machine. [1] several days waiting for the battery inner equilibrium,
Thus, EVs attract much attentions from both academia and which is inappropriate for most online application. Extend
industry. [2][3] Since battery is the only power source in Kalman filter (EKF) is frequently appeared in multiple
pure EVs, the remaining driving distance related to EV papers for its superior performance. On the foundation of a
users experiences closely rely on SOC. Compare with closed loop structure, EKF is able to deal with unknown
other kinds of batteries, lithium batteries have advantages initial SOC in estimation process. [10][11] However,
of higher specific energy, longer cycle life and lower linearizing state space equation by Taylor series expansion
self-discharge rate. [4] [5] Hence, many manufacturers decreases its accuracy. Other similar approaches, such as
choose lithium batteries as the power source in their EV unscented Kalman filter (UKF), particle filter (PF), are too
353
are the order of system. final accuracy of the estimated SOC is improved by the
mix of the two models. Assuming that U1 is the difference
Eq.(3) are dynamic linearization as the following form
between the output of mathematic model and terminal
in MFAC:
voltage, and U2 is the difference between the output of PLS
y( + 1) = ( ) + ( ) ( ) (4) model and terminal voltage. The mix algorithm in this
paper is simply selecting the model whose output has
Where ( ) is the PPD of the control system. Input minimum difference comparing to terminal voltage. The
and output data measured from the controlled system are mix mechanism are briefly described as following: If U1 >
used to calculate the PPD online for the dynamic model. U2, we choose mathematic model in the estimation process.
On the basis of Eq. (4), the process of MFAC is as If U2 > U1, PLS model is selected for estimating. Without
follows: increasing much computational burden, this concise
( )
mechanism help making the estimation process simple and
u( ) = ( 1) + | ( )|
( + 1) ( ) (6) effective.
( 1) IV. EXPERIMENT RESULTS
( ) = ( 1) +
+ ( 1)
Lithium polymer battery manufactured by KOKAM is
( ) ( 1) ( 1) (7) selected in this paper for testing the proposed method. The
nominal capacity is 70.0 Ah and nominal voltage is 3.7 V.
( ) = (1), if | ( )| or | ( 1)| (8) The mathematic model is established firstly as the primary
Where , are the step factor, , are the model. Battery data under constant current in Fig.3 is
weighting factor, is small positive number. measured for parameters identification of the mathematic
model. In this paper, mathematic model is used as the
It can be deduced from Eq.(6)~(8) that MFAC does not following form:
rely on the mathematical model of the controlled system.
Thus, MFAC is used to estimate SOC in this paper. = + + +
Utilizing PPD and real-time data, MFAC is able to adjust ( )+ (1 ) (9)
the parameters and structure of the controlled system
adaptively. In this paper, MFAC makes full use of the Where Vmodel is the calculating voltage of mathematic
voltage difference between battery model and real battery model, Iinput is the charging or discharging current, and r is
for correcting SOC. With the benefit of MFAC, SOC the internal resistance. E0, k0, k1, k2 and k3 are the
estimation is a closed loop structure and also insensitive to parameters of mathematic model.
initial SOC.
Due to varieties in driving habits and traffic conditions,
battery pack in EVs generally works in different driving
cycles. However, measuring battery under constant current
variation is easy and feasible in most applications.
Therefore, the mathematical battery model identified under
constant current is selected as the original model in this
paper. Afterwards, on the basis of the operating data, PLS
battery model is added to the estimation structure. The
final estimated SOC are the mix of the two models. MFAC
is chosen to calculate utilizing the terminal voltage
errors. The structure of the proposed algorithm is shown in (a) Current
Fig. 3.
U
SOC* SOC Mathematica
U1
l Battery
I
Model
Complementary model
MFAC
SOC U
I U2
PLS model
T Mix mechanism
354
(c) Temperature (b) Voltage
Parameter E0 r k0
Value 3.132 0.002852 9.083e-7
Parameter k1 k2 k3
Value -0.4713 0.004097 -0.05771
(a) Current
355
The authors would like to thank the employees of
Stropower Company for their support. This work is supported
by the Key Program for International S&T Cooperation and
Exchange Projects of Shaanxi Province and the Seed
Foundation of Innovation and Creation for Graduate Students
in Northwestern Polytechnical University.
REFERENCES
[1] H. Shareef, M.M. Islam, A. Mohamed, A review of the stage-of-the-art
charging technologies, placement methodologies, and impacts of
electric vehicles, Renewable and Sustainable Energy Reviews, vol.64,
pp.403-420, 2016.
[2] P. Miller, xEV market trend and prospect, in Proc. VPPC, Seoul,
Fig.8 Errors of the proposed method
Korea, Otc. 9-12, 2012, pp.1095-1099.
[3] Y.Y. Jia, V.K. Ramachandaramurthy, K. M. Tan, N. Mithulananthan,
In Fig.6, the difference between estimated SOC and A review on the state-of-the-art technologies of electric vehicle, its
reference is increasing when the PLS battery model is not impacts and prospects, Renewable and Sustainable Energy Reviews,
vol.49, pp.365-385, 2015.
added, especially for the estimation results after 800s.
[4] M. A. Hannana, M. M. Hoquea, S. E. Penga, M. N. Uddin, Lithium-Ion
However, combining with PLS model, the SOC estimation Battery Charge Equalization Algorithm for Electric Vehicle
results closely follow the reference SOC during the same Applications, in Proc. IAS, Portland, USA, Otc. 2-6, 2016.
period. This is because the parameters of mathematic [5] A.-I. Stan, M. wierczyski, D.-I. Stroe, and R. Teodorescu, Lithium
model are just identified under constant current as ion battery chemistries from renewable energy storage to automotive
previously described, while the parameters actually change and back-up power applications-An overview, in Proc.OPTIM, Brasov,
Romania, pp. 713-720, 2014.
with temperature and SOC after operating in varying
[6] J.H. Bae, Bai Zhiguo, The development of technology on reduces the
current. In this case, mathematic model cannot guarantee SOC error rate using Hybrid Kalman Filter(HKF) and the demonstration
its accuracy in the estimated process. The proposed PLS its performance using BMS platform, in Prof. ITEC Asia-Pacific,
model is able to improve the modeling accuracy through Busan, Korea, Jun. 1-4 ,2016.
the mix mechanism in Section III. Therefore, the accuracy [7] L.W. Juang, P.J. Kollmeyer, R. Zhao, T.M. Jahns, R.D. Lorenz,
of estimation result is significantly improved by adding the Coulomb counting state-of-charge algorithm for electric vehicles with
a physics-based temperature dependent battery model, in Prof. ECCE,
complementary PLS model. Fig. 7 shows which models is Montreal, Canada, Sept. 20-24, 2015.
used in every step. It is clear shown in Fig. 7 that PLS [8] F.D. Zheng, Y.J. Xing, J.C. Jiang, B.X. Sun, J.H. Kim, M. Pecht,
model is more frequently chosen for SOC estimation after Influence of different open circuit voltage tests on state of charge
1000s. Errors in Fig.7 also prove that the estimation online estimation for lithium-ion batteries, Applied Energy, vol.183,
pp.513-525, 2016.
accuracy is improved effectively by the PLS model. The
[9] Z. Chen, Y.H. Fu, C.C. Mi, State of Charge Estimation of Lithium-Ion
errors of MFAC with PLS battery model are obviously less Batteries in Electric Drive Vehicles Using Extended Kalman Filtering,
than MFAC without data driven model in Fig.8, especially IEEE Transactions on Vehicular Technology, vol.62, no.3, pp.1020
when the parameters are changed in higher temperature 1030, 2013.
and lower SOC ranges. [10] J. Meng, G. Luo, E. Breaz, and F. Gao, A robust battery
state-of-charge estimation method for embedded hybrid energy system,
in Proc. IECON, Yokohama, Japan, pp. 001205-001210, 2015.
V. CONCLUSION
[11] M. Paschero, G.L. Storti, A. Rizzi, F. M. F. Mascioli, G. Rizzoni, A
A PLS based model for improving accuracy of the Novel Mechanical Analogy-Based Battery Model for SoC Estimation
estimated SOC is proposed in this paper. MFAC, combined Using a Multicell EKF, IEEE Transactions on Sustainable Energy,
vol.7, no.4, pp. 1695 -1702, 2016.
with the basic mathematic model, guarantees the primary
[12] J.H. Meng, G.H. Luo, F. Gao, Lithium Polymer Battery State-of-Charge
accuracy of the estimation method. PLS model is Estimation Based on Adaptive Unscented Kalman Filter and Support
established and updated online with the arrival of the new Vector Machine, IEEE Transactions on Power Electronics, vol.31, no.3,
measured data in BMS, which acts as the data driven pp. 2226 -2238, 2016.
model in estimation process. By an efficiently [13] D.M. Zhou, K. Zhang, A. Ravey, F. Gao, A. Miraoui, Online
Estimation of Lithium Polymer Batteries State-of-Charge Using Particle
computational mechanism, the proposed method mixes the Filter-Based Data Fusion With Multimodels Approach, IEEE
results of the two different models for accurate battery Transactions on Industry Applications, vol. 52, no. 3, pp. 2582 2595,
model. With a closed loop structure, the proposed method 2016.
is insensitive to initial SOC. Experimental results prove the [14] C. Dong, G.L. Wang, Estimation of power battery SOC based on
proposed method can help obtaining an accurate SOC in improved BP neural network, in Prof. ICMA, Tianjin, China, Aug. 3-6,
2014.
EV driving cycles. This paper is the initial effort for the
[15] S. De Jong, SIMPLS: an alternative approach to partial least squares
proposed method. The fusion mechanism and the test of regression, Chemometrics and intelligent laboratory systems, vol. 18,
proposed method with limited samples under various no. 3, pp. 251263, 1993.
condition still need further research in our future work. [16] Z.S. Hou, S.T. Jin, Data-driven model-free adaptive control for a class
of MIMO nonlinear discrete-time systems, IEEE Transactions on
ACKNOWLEDGMENT Neural Networks, vol.22, no.12, pp. 21732188, 2011.
356
[17] Y.M. Zhu, Z.S. Hou, Data-Driven MFAC for a Class of Discrete-Time
Nonlinear Systems With RBFNN, IEEE Transactions on Neural
Networks and learing system, vol.25, no.5, pp.1013-1020, 2014.
[18] D.Z. Xu, B. Jiang, P. Shi, A Novel Model-Free Adaptive Control
Design for Multivariable Industrial Processes, IEEE Transactions on
industrial electronics, vol.61, no. 11, pp. 6391-6398, 2014.
357