Elsevier

Energy

Volume 63, 15 December 2013, Pages 295-308
Energy

A data-driven adaptive state of charge and power capability joint estimator of lithium-ion polymer battery used in electric vehicles

https://doi.org/10.1016/j.energy.2013.10.027Get rights and content

Highlights

  • A data-driven parameter identification method is developed by RLS algorithm.

  • An adaptive multi-state joint estimator of the battery is developed by AEKF algorithm.

  • A data-driven SoC and SoP joint estimator is developed with the real-time measurement.

  • Robustness of the joint estimator is verified by different aging states of LiPB cells.

Abstract

An accurate SoC (state of charge) and SoP (state of power capability) joint estimator is the most significant techniques for electric vehicles. This paper makes two contributions to the existing literature. (1) A data-driven parameter identification method has been proposed for accurately capturing the real-time characteristic of the battery through the recursive least square algorithm, where the parameter of the battery model is updated with the real-time measurements of battery current and voltage at each sampling interval. (2) An adaptive extended Kalman filter algorithm based multi-state joint estimator has been developed in accordance with the relationship of the battery SoC and its power capability. Note that the SoC and SoP can be predicted accurately against the degradation and various operating environments of the battery through the data-driven parameter identification method. The robustness of the proposed data-driven joint estimator has been verified by different degradation states of lithium-ion polymer battery cells. The result indicates that the estimation errors of voltage and SoC are less than 1% even if given a large erroneous initial state of joint estimator, which makes the SoP estimate more accurate and reliable for the electric vehicles application.

Introduction

Nowadays the LiB (lithium-ion battery) is drawing a vast amount of attention as the most important onboard energy storage for electric vehicles and the rapidly growing smart grid application. LiB packs are presently the most expensive but least well understood components in electrified vehicle powertrains. To guarantee safe, efficient, and durable operations of the LiB under demanding driving conditions and interactions with an electricity grid, an effective BMS (battery management system) is necessary [1], [2], [3]. However, due to the strong time-variable and nonlinear characteristics, as well as influences by such random factors as driving loads and operating environment in its application, an accurate and reliable estimation of SoC (state of charge) and SoP (state of power capability) still remains a challenge.

Accurate battery models are the basis of state estimations. A number of studies have proposed methods for estimating battery parameters and each one has its own advantages [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], which can be generally classified into two kinds: the non-time series-iterative calculation based fitting or optimal solving (offline) method and time-series format based iterative calculation (online) method. The former method can only use archived data to solve the model parameter, but the latter method can be applied to operate with real-time measurements of battery voltage and current. Most research uses the offline method to calculate the model parameter, where the HPPC (hybrid pulse power characterization) test is the most commonly used current profile for parameter identification [5], [6], [7], [8]. However, this kind of method cannot achieve desired calculation precision when the battery aged or operating condition of the battery changed, because the parameter of the battery is very sensitive to its state. In this case, the online modeling method, which calculates the model parameter with the real-time measurement of battery current and voltage, is more suitable for the battery system and has the potential to achieve an accurate prediction performance against different battery aging levels and operating conditions. The authors [10] used the statistical-based online parameter identification method to estimate the internal resistance and OCV (open circuit voltage) of the Ni-MH battery, which achieved acceptable accuracy in its own application. However, the open loop system cannot achieve the optimal state estimate against different operating conditions, since it lacks effective feedback mechanisms. As a result, it is not suitable for accurate SoC estimation, particularly for the LiFePO4 LiB which has a flat OCV plateau.

In terms of SoC estimation, a wide variety of research methods and techniques have previously been summarized for constructing the SoC estimator, each one having its relative advantage, as reviewed by Ref. [14]. However, these SoC estimation approaches fail to achieve reliable predictions against different degradation states of LiB cells. The trajectory of the battery model parameter cannot be fully described with a limited number of experiments; however, it needs to be tested in all of its possible working conditions and aging levels. It is evident that it is not practical for the electric vehicle, which has large number of LiB cells in its battery system. Thus, a data-driven SoC prediction approach is a good choice to achieve desirable SoC estimates.

On the other hand, many energy management units in electric vehicles need the SoP information to regulate the propelling power and to coordinate the regenerative braking and friction braking [15], [16], [17]. The SoP therefore directly influences the vehicle propulsion operating strategies, performance, and comfort. Underestimates of the SoP may result in overly conservative vehicle energy management, while overestimates of the SoP may cause the battery to over-charge, over-discharge and premature failure. In addition, inaccurate SoP estimates may give rise to misleading readings on battery diagnosis and prognosis. However, the “true” power capability of the battery is difficult to determine when the battery is working. Thus the power capability estimation is primarily carried out with the design limits of the battery itself and the electric vehicle or powertrain, such as upper cutoff voltage, lower cutter off voltage, maximum continuous operating currents and powers, maximum and minimum SoC values et al. With the development of electric vehicle technology, some power capability estimation approaches are presented [12], [15], [16], [17]. In Refs. [15], [16], [17], parameters of the battery model were identified with the offline method, thus these methods cannot ensure reliable SoP estimates. In Ref. [12], although the parameter was identified with online method, it used the ampere-hour counting method for its SoC estimation; therefore the result was not reliable enough in the extensive use of electric vehicles.

A key contribution of this study is that a data-driven adaptive multi-state joint estimator for battery state of charge and power capability prediction was developed, thus the prone-error and time consuming periodic calibration experiments could be avoided. What's more, the performance of the joint estimator has been verified and evaluated by different degradation states of LiPB (lithium-ion polymer battery) cells. In the proposed multi-state joint estimator, the RLS (recursive least square) method is employed to develop a data-driven parameter identification method through the real-time measurements of battery current and voltage. In addition, the AEKF (adaptive extended Kalman filter) algorithm is applied to establish the adaptive joint estimator for battery SoC and SoP. It is noted that the AEKF algorithm can improve the prediction precision by adaptively updating the noise covariance. This approach has the potential to eliminate the drawback of parameter variation and reduce the influence from different operating conditions and aging levels.

The remainder of the paper is organized as follows: Section 2 describes the lumped parameter battery model and the implement flowchart of the data-driven parameter identification method. A review of the adaptive extended Kalman filter algorithm is presented in Section 3, and the data-driven adaptive joint estimator of SoC and SoP is developed later. The test bench and datasheets of 3.7 V/32 Ah LiPB cells are described in Section 4. Section 5 verifies the proposed data-driven joint estimator with DST (dynamic stress test) profiles. In the final section, some conclusions and final remarks are given.

Section snippets

Online modeling technology

Section 2.1 gives a detailed description of the lumped parameter battery model. Section 2.2 presents a brief review of the RLS algorithm and proposes a flowchart for the data-driven parameter identification approach.

An evolution of battery model structures

The SoC is defined as a ratio between the remaining capacity and the maximum available capacity. For applying battery model to time-series format based state estimation, the SoC should be presented as discrete-time form [22].zk+1=zkηiIL,kΔt/Cnwhere zk+1 and zk are the SoC at discrete-time index k + 1 and k respectively, and ηi is the coulomb efficiency, which is the function of current, temperature and capacity; Cn is the present maximum available capacity.

The battery model can be described

Application to LiPB cells

As an application case, three LiPB cells are chosen to verify the proposed approach from the group of tested cells [6]. Section 4.1 gives a brief introduction for the test bench, and the datasheet for our research is present in Section 4.2.

Verification and discussion

To verify whether the proposed data-driven multi-state joint estimator has a potential to achieve accurate and reliable estimation against different degradation states, three different aging levels of LiPB cells were used to carry out the verification. The verification of LiPB cell01 is presented in Section 5.1. Evaluations of other two LiPB cells are presented in Section 5.2.

Conclusions

To improve the dynamic voltage prediction precision of the lumped parameter battery model, a data-driven parameter identification approach has been proposed by the recursive least square method. With the real-time measurement of battery current and voltage, the model parameter can be adaptively updated at each sampling interval, thus the model can track the real-time characteristic of the battery.

To achieve accurate battery SoC and SoP estimation, the multi-state joint estimator has been

Acknowledgments

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 in part, the National High Technology Research and Development Program of China (2012AA111603, 2011AA11A228, 2011AA1290) in part. The authors would also like to express deep gratitude to Kathy McNamara for English editing.

Cited by (191)

View all citing articles on Scopus
View full text