Elsevier

Applied Energy

Volume 113, January 2014, Pages 1421-1433
Applied Energy

A data-driven based adaptive state of charge estimator of lithium-ion polymer battery used in electric vehicles

https://doi.org/10.1016/j.apenergy.2013.09.006Get rights and content

Highlights

  • A lumped parameter battery model against different battery aging levels is proposed.

  • The RLS based method is used to identify the parameter of battery model in real-time.

  • A data-driven based adaptive SoC estimator is developed by RLS and AEKF algorithm.

  • The robustness of the SoC estimator against varying loading profiles is evaluated.

  • The robustness of the SoC estimator against different aging levels is evaluated.

Abstract

An accurate State of Charge (SoC) estimation method is one of the most significant and difficult techniques to promote the commercialization of electric vehicles. The paper attempts to make three contributions. (1) Through the recursive least square algorithm based identification method, the parameter of the lumped parameter battery model can be updated at each sampling interval with the real-time measurement of battery current and voltage, which is called the data-driven method. Note that the battery model has been improved with a simple electrochemical equation for describing the open circuit voltage against different aging levels and SoC. (2) Through the real-time updating technique of model parameter, a data-driven based adaptive SoC estimator is established with an adaptive extended Kalman filter. It has the potential to overcome the estimation error against battery degradation and varied operating environments. (3) The approach has been verified by different loading profiles of various health states of Lithium-ion polymer battery (LiPB) cells. The results indicate that the maximum estimation errors of voltage and SoC are less than 1% and 1.5% respectively.

Introduction

Lithium-ion battery (LiB), as an important on-board electric energy storage system, has been widely used in various electric vehicles (EVs). However, to satisfy the operation voltage and traction power requirements of electric vehicles, a battery pack has to be made with hundreds of cells connected in series or parallel to overcome the limitations of low energy density, low cell capacity and cell voltage [1], [2], [3]. Due to liquid electrolyte instability and flammability, lithium-ion cells have potential to ignite or even explode when over-charged or over-discharged. To avoid the potential safety hazard, an advanced battery management system (BMS) plays a vital role in improving battery performance, safety and reliability. For an efficient energy management of EVs, the most important key technique is to achieve an accurate State of Charge (SoC) estimate, which improves the power distribution efficiency and extends the calendar-life of the cells [4], [5]. However, since cell performance largely depends on its operation environments and aging states, it is still hard to control the battery cell accurately.

A number of studies have proposed methods for calculating battery SoC and each one has its own advantages. The most widely used method is the ampere–hour counting method. However, this open-loop calculation method can easily lead to accumulated calculation errors due to uncertain disturbances from the practice application and lack of necessary corrective resolution for the initial SoC offset [6]. Thus, to improve its performance, it is often recalibrated by other state-space based filters [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], Kalman filters being one of the most commonly used [7], [8], [9], [10], [11], [12]. The author in [7] used extended Kalman filtering (EKF) to estimate the LiPB cell SoC, the authors in [3] used multi-scale EKF to estimate the SoC of a battery pack. However, the performance of the EKF-based method has two fatal flaws. One is that this method greatly depends on the accuracy of the model; the other is that the noise information should be known in advance [9]. Authors in [9], [10], [11] used an adaptive extended Kalman filter (AEKF) algorithm to carry out the SoC estimation, which showed the prediction accuracy has been greatly improved even with an erroneous initial estimator state.

Common drawbacks with the above methods were that the cell parameter variances under different aging levels were ignored and the methods were verified under a narrow set of scenarios, without discussing different aging levels and loading profiles. In other words, the robustness of these SoC estimators against different driving cycles of varied health states of battery were not sufficiently discussed. To avoid this drawback, a data-driven parameter identification method was proposed to update the model parameters in real-time. In this method, the SoC was usually inferred by the estimated open circuit voltage (OCV) with the table of SoC vs. OCV. Authors in [11], [17] used the AEKF algorithm and recursive least squares (RLS) method to estimate OCVs and to infer SoCs. The results showed that this method had a desirable performance with an acceptable accuracy. Authors in [18] also used an online parameter identification method to estimate the OCV. To evaluate the OCV-based SoC estimation approach with the former SoC-based method, authors in [11] carried out the comparison of the SoC estimation between the AEKF-based OCV estimator and AEKF-based SoC estimator. The results showed that the latter has a better SoC estimation accuracy since its closed-loop feedback structure, but if the operation environment changed or the battery aged, the parameter variance on the battery would decrease the estimation accuracy remarkably.

A key contribution of this study is that a data-driven based adaptive SoC estimator was developed, thus the prone-error and time consuming periodic calibration of the model could be avoided. What’s more, the performance of the SoC estimator has been verified and evaluated by different loading profiles with different battery aging levels. In the proposed SoC estimator, the RLS method is employed to develop a data-driven based parameter identification method with the real-time measurement of battery current and voltage. In addition, the AEKF method is applied to establish the SoC estimator. It is noted that the AEKF algorithm can improve the prediction precision by adaptively updating the noise covariance, and the time-series based calculation process of the AEKF algorithm is shown in Table 1 [11].

The remainder of the paper is organized as follows: Section 2 describes the lumped parameter battery model and the implement flowchart of the RLS based online parameters identification method. A data-driven adaptive SoC estimator with AEKF algorithm is present in Section 3. To verify the proposed approach, different aging levels of 3.7 V/32 A h Lithium-ion polymer battery (LiPB) cells are used to execute the characteristic test and the experiment has been described Section 4. Section 5 verifies the proposed adaptive SoC estimator using the dynamic stress test (DST) and Federal Urban Driving Schedule (FUDS) test. In the final section, some conclusions and final remarks are given.

Section snippets

Battery model

Section 2.1 gives a detailed description of the lumped parameter battery model. Section 2.2 presents an implementation flowchart of the data-driven based parameter identification approach.

Adaptive extended Kalman filter-based SoC estimator

In our previous research in [6,11], the AEKF algorithm with a covariance matching technique was successfully applied to construct a reliable state estimation for a model-based battery management system application, this section focuses on building a data-driven based adaptive SoC estimator for achieving an accurate online SoC estimation against different aging levels and lading profiles. Section 3.1 gives a definition of the SoC calculation. The numerical implementation of the recursive

Data set of LiPB cells for verification

As an application case, the LiPB cells are used to verify the proposed approach. Section 4.1 gives a brief introduction for the test bench, and the test schedule is discussed in Section 4.2.

Verification and discussion

Herein we use two loading profiles to verify the proposed approach. Sections 5.1 DST loading profiles, 5.2 FUDS loading profiles verifies the proposed SoC estimator with DST and FUDS loading profiles respectively.

Conclusions

To model the dynamic voltage performance of LiPB accurately, the improved lumped parameter model has been proposed through optimizing the voltage source with an electrochemical function formed with SoC and capacity, which can improve the SoC estimation accuracy through an efficient closed-loop feedback between the SoC and open circuit voltage and achieve accurate open circuit voltage predictions against different battery aging levels.

To improve the dynamic voltage prediction precision of the

Acknowledgements

We would like to express our deep gratitude to Professor Chris Chunting Mi in the University of Michigan for many helpful discussions and Kathy McNamara for English editing. This work was supported by the National Natural Science Foundation of China (51276022) and the Higher school discipline innovation intelligence plan (“111”plan) of China.

References (24)

Cited by (207)

View all citing articles on Scopus
View full text