
Robust prognostics for state of health estimation of lithium-ion batteries
based on an improved PSO–SVR model
Taichun Qin
a
, Shengkui Zeng
a,b
, Jianbin Guo
a,b,
⁎
a
School of Reliability and Systems Engineering, Beihang University, Beijing 100191, China
b
Science and Technology on Reliability and Environmental Engineering Laboratory, Beijing 100191, China
abstractarticle info
Article history:
Received 25 May 2015
Received in revised form 21 June 2015
Accepted 29 June 2015
Available online 11 July 2015
Keywords:
Lithium-ion battery
Prognostic
State of health
Support vector regression
Particle swarm optimization
State of health (SOH) estimation of lithium-ion batteries is significant for safe and lifetime-optimized operation.
In this study, support vector regression (SVR) is employed in bat tery SOH prognostics, and part icle swarm
optimization (PSO) is employed in obtaining the SVR kernel parameter. Through a new validation method, the
proposed PSO–SVR model in this paper can well grasp the global degradation trend of SOH and is little affected
by local regeneration and fluctuations. The case study shows that compared with the eight published methods,
the proposed mode l can obtain more accurate SOH prediction results. Even SOH prediction starts from the
cycle near capacity regeneration, the proposed model still can grasp the global degradation trend. Furthermore,
the improved PSO–SVR model has great robustness when the training data contain no ise and measurement
outliers, which makes it possible to get satisfactory prediction performance without pre-processing the data
manually.
© 2015 Elsevier Ltd. All rights reserved.
1. Introduction
With the advantages of high energy density and lightweight,
lithium-ion batteries are widely used in various kinds of portable de-
vices, electric cars, spacecraft, etc. [1,2]. It is useful to accurately estimate
state of charge (SOC) and state of health (SOH), as they play important
roles in safe and reliable usage of lithium-ion batteries [3,4]. However,
compared to the relatively mature study on SOC issue, SOH research
still stays in the initial stage [5,6].
In general, the material properties and manufacturing assemblies
are various from batter y to battery [7]. Also, the performance of
lithium-ion batteries degrades with many factors, such as environment
temperature, and charging and discharging current, depth of discharge
[3,8]. Moreover, it is rather difficult to measure parameters of the
electrochemical reaction of lithium-ion batteries [4].Therefore,SOH
prognostics of data-driven approach seem more applicable than those
of physical-model approach. In recent years, many algorithms based
on data-driven approach have been applied to SOH prognostics, such
as extended Kalman filter (EKF) [9],particlefiltering (PF) [5],fuzzy
logic [10], neural networks [11], and regressions [12]. However, with
poor robustness, these algorithms are not good at dealing with limited
data that contain uncertain information. As prediction errors are accu-
mulating over time, these algorithms cannot be flexibly used in long-
term prediction.
The battery capacity can be easily obtained and has a clear
physical meaning [13], so capacity becomes the most common
used health indicat or of batteries. However, the uncertaint y of ca-
pacity t ime seri es, ge nerated by capacity regeneration [14],unex-
pected fluctuations, measurement error and other factors [7],isa
big challenge for battery SOH prognostics. Olivares [15] has de-
tected and isolated the effect of regenerati on phenomena within
the life-cycle model through a particle-filtering-based prognosti cs
framework. Treating re generation as a cycle process, Liu [14] has
described capacity regeneration and degradation by using combina-
tion Gaussian progress function regression. With wavelet analysis, He
[16] has separated local regeneration and fluctuation characteristics
from global degradation characteristic, which makes it possible to analyze
the two characteristics respectively. However, how to obtain accurate
global degradation prediction of SOH with raw capacity data is still a
topic that is worth studying.
In this paper, based on particle swarm optimization (PSO), a new
approach to the optimal selection of support vector regression (SVR)
parameters is proposed. This PSO–SVR model can well grasp the global
degradation trend, despite the fact that the capacity time series contain
regeneration, fluctuations, noise and outliers.
Microelectronics Reliability 55 (2015) 1280–1284
⁎ Corresponding author at: School of Reliab ility an d Systems Engineering, Beihang
University, Beijing 100191, China.
http://dx.doi.org/10.1016/j.microrel.2015.06.133
0026-2714/© 2015 Elsevier Ltd. All rights reserved.
Contents lists available at ScienceDirect
Microelectronics Reliability
journal homepage: www.elsevier.com/locate/mr