Search

Battery management for plug-in electric vehicles (PEVs) has attracted extensive research attention, with most existing studies focusing on PEV operating conditions. However, battery maintenance during idling remains largely unexplored, under which electrochemical side reactions can cause battery degradation. The degradation rate depends on battery states, e.g., state of charge (SOC) and temperature. Considering that PEV idling accounts for the majority of the time, the accumulated degradation could have a major impact on battery and vehicle lifetime. In this article, battery maintenance during extended idling periods is investigated by utilizing a commonly available infrastructure, i.e., the charging unit. An optimal charging profile is designed to maintain the battery states under desirable conditions to minimize degradation over the idling period while still satisfying the charging energy requirement. Optimal charging profiles are obtained under different ambient temperatures and stages of battery life, showing different features due to respective dominating degradation mechanisms. Compared with the optimal charging profile, constant-current (CC) charging could result in up to 51.6% higher capacity loss, and that of (fast) CC-constant-voltage charging could be up to 12.3 times higher under circumstances over a 12-h overnight idling period. Integration/accommodation of user and grid demand is also addressed by augmenting the optimization framework.

Thermal management is a critical task of battery control to ensure the safe, efficient, and enduring performance of the battery system, which can be considered as an interconnected thermal network of cells. The basis of thermal management is the estimation of temperature and its gradient across the battery system, which has received extensive attention in the literature. However, existing works neglect two important constraints in practical battery systems: 1) limited number of available sensors and 2) presence of system uncertainty such as parameter error. This paper is the first to investigate robust battery system temperature estimation under sparse sensing and system uncertainty. We first propose a framework consisting of optimization problems at three different levels: 1) evaluation of the worst case estimation performance (error) under uncertainty; 2) robust observer design to minimize the worst case error; and 3) optimization of sensor locations. Two robust estimation methods are then used to solve the problem. The system uncertainty considered in this paper is the unknown resistance variability among battery cells, but the methodology can be applied to address other types of uncertainty. It is shown that the designed observers could guarantee and improve the robustness and reliability of estimation by significantly reducing the worst case estimation errors induced by uncertainty.

The commercialization of lithium-ion batteries enabled the widespread use of portable consumer electronics and serious efforts to electrify trans-portation. Managing the potent brew of lithium-ion batteries in the large quantities necessary for vehicle propulsion is still challenging. From space applications a billion miles from Earth to the daily commute of a hybrid electric automobile, these batteries require sophisticated battery management systems based on accurate estimation of battery internal states. This system is the brain of the battery and is responsible for estimating the state of charge, state of health, state of power, and temperature. The state estimation relies on accurate prediction of complex electrochemical, thermal, and mechanical phenomena, which increases the importance of model and parameter accuracy. Moreover, as the batteries age, how should the parameters of the model change to accurately represent the performance, and how can we leverage the limited sensor information from the measured terminal voltage and sparse surface temperatures available in a battery system? With a frugal sensor set, what is the optimal sensor placement? This article reviews estimation techniques and error bounds regarding sensor noise and modeling errors, and concludes with an outlook on the research that will be necessary to enable fast charging, repurposing of batteries for grid energy storage, degradation prediction, and fault detection.

This paper presents an experimental study on the parameter identification of a battery pack, which determines the relationship between identification accuracy and measurement data. Parameter identification of the lithium-ion battery is poor when the input-output data, i.e., the input current and output voltage, is not appropriate. In addition, selection/optimization of an appropriate data set for estimation needs to adapt to different applications. A first-order equivalent circuit model is adopted to model a battery pack, and the identification accuracy is analyzed for both the single-parameter and multi-parameter identification scenarios. It is found that the accuracy of different identification scenarios is influenced by the voltage noise, the current amplitude, and the current frequency. Three experiments using sine waves with different frequencies are then performed to characterize a lithium-ion battery pack. Experimental results show that the current profile with the optimal frequency content achieves the best identification performance. Therefore, it is validated that the identification accuracy can be improved when the current excitation satisfies certain criteria.

This paper provides a theoretic and systematic analysis on the battery state of charge (SOC) estimation errors caused by sensor noises. The considered noises include the bias and the variance of both current and voltage sensors. Specifically, the bias and the variance of the SOC estimation error are derived as explicit functions of sensor noises, battery parameters, and observer tuning parameters. The derivation is performed for the Kalman filter, which is the most commonly used method for SOC estimation, and the least squares method. It is found that the observer parameter tuning is subject to a tradeoff between suppressing the estimation bias and variance. Either one of these two errors becomes dominant under different observer parameter ranges. The fundamental estimation error that cannot be mitigated through observer tuning has also been identified. The results have been validated by both simulation and experiment. The obtained theoretical findings are of great practical significance as they could be used to guide sensor selection and observer design, as well as enable online uncertainty management.

This letter investigates the fundamental relationship between the estimation accuracy of the equivalent circuit dynamics and the measurement data, i.e., input current and output voltage. Specifically, the Cramer–Rao bounds of the model parameter estimation are derived analytically as explicit functions of generic input/output data. The derivation is performed in both discrete-time and Laplace representation, and for both single- and multi-variable estimation scenarios. The sinusoidal current input is then used as an example for illustration. The analytic results could benefit the estimation practice in multiple ways, including the estimation reliability/robustness evaluation, offline experiment design for system identification, and data selection for online estimation.

Data optimization, or optimal experiment design, is an effective way to improve and guarantee the accuracy of state and parameter estimation, as the quality of data has significant impact on the estimation accuracy. Such capability is especially critical for energy systems requiring high reliability. The common practice of data optimization is to design input excitation by maximizing the Fisher information, and hence minimizes the variance of the estimation error. However, such approach suffers from fundamental limitations, including negligence of estimation bias and system uncertainties in measurement, model, and parameter, which severely restrict the applicability and effectiveness of the method. This paper aims at establishing new criteria and a novel framework for data optimization and estimation error quantification to overcome the fundamental limitations. First, a generic formula is derived for quantifying the estimation error subject to sensor, model, and parameter uncertainties for the commonly used least-squares algorithm. Based on the formula, desirable data structures, which could minimize the errors caused by each uncertainty, are identified. These structures are then used as new criteria to formulate the novel data optimization framework. The proposed methodology is applied to the parameter estimation problem of a lithium-ion battery electrochemical model in simulation and experiments, showing up to two orders of magnitude improvement in estimation accuracy compared with the traditional Fisher-information-based approach and other baselines.

It is well recognized that voltage and current sensor noises are one of the major sources of battery state of charge (SOC) estimation errors. However, quantitative relationship between the estimation error and sensing noises remains to be established systematically. This paper is devoted to filling this critical gap by deriving the analytic expression of the SOC estimation error under sensing noises. The considered sensing noises include both bias and variance. The derivation is based on a Kalman filter and an equivalent circuit battery model. The obtained analytic expressions can be conveniently used to evaluate the robustness and reliability of SOC estimation as well as guide observer design.

Recognizing the important role of data in estimation, data optimization for offline and online battery state and parameter estimation has been receiving increasing attention recently. The main idea is to design/select data that are most sensitive to the target variables under estimation. However, due to the lack of efficient ways to compute sensitivity and heavy reliance on simulation, data optimization is often intractable because of the computational complexity. This issue is especially prominent for the states and parameters of the electrochemical battery models. In this work, we study the methodology for analytically deriving the sensitivity of battery electrochemical parameters. The derivation is based on a single particle model, and the results have been verified by comparing to the numerical simulation of a full order pseudo-2D electrochemical model. The obtained analytic results provide theoretic insight on the dynamic nature of the parameter sensitivity. The derived analytic expressions could enable fast sensitivity computation to serve the data optimization for offline system identification and online data selection/mining for real-time estimation.

To investigate the characteristics of lithium-ion battery packs under the condition that one cell is short-circuited when the whole battery pack is being discharged or charged, systematic battery external short circuit (ESC) experiments are conducted. Since not all battery cells are equipped with current sensors because of the space limitation and manufacturing cost, an artificial neural network (ANN)-based method is proposed to estimate the current of the short-circuited cell using only the voltage information, which is the feasible practice in electric vehicle application. Furthermore, the estimated current is used to predict maximum temperature increase as well as internal and surface temperature distribution of the ESC cell based on a 3D electro-thermal coupling model. Two experimental groups under constant current charging condition and constant power discharging condition are employed to validate the stability and accuracy of the proposed method. The results indicate that the root-mean-square-error between the estimated and measured current are 3.72 A and 6.61 A under the two validation experiments respectively, and the maximum estimation errors of temperature increase are 4.9 °C and 7.3 °C respectively.

Real-time estimation of battery internal states and parameters is a key task of battery management and has been studied extensively in literature. Estimation is usually performed by an algorithm using a model and the measured input-output data, such as current, voltage, and temperature. Traditionally, the estimation algorithms do not differentiate the data points and use all of them equally for estimating each variable. However, not all data points, but often only a small fraction of them, are sensitive to the target variables under estimation. Using insensitive data will induce significant estimation errors due to the commonly presented unknown disturbances such as measurement noise and model uncertainty. This paper studies the data selection mechanism to optimize and guarantee the accuracy of battery state and parameter estimation. A sensitivity-based data selection strategy is proposed, which automatically identifies the sensitive data in real-time and passes them to the observer for estimation. It is shown that the data selection strategy could improve the quality of estimation results significantly.

The Cramer-Rao bounds for battery state of charge, capacity and resistance estimation are derived in this paper. Independent of any specific form of observers, the bounds explore the quality of the data collected for estimation and indicate the best achievable accuracy of any (unbiased) estimator under measurement noise. The derivation is performed based on an equivalent circuit model. First, the Cramer-Rao bounds for standalone estimation, where only one state/parameter is estimated, is derived. The discussion is then extended to combined estimation where multiple state/parameters are estimated from the same data set. It is found that for current inputs that satisfy certain patterns, loss of accuracy in combined estimation can be avoided. The derived explicit analytic expressions are easy to use for improving the accuracy of both online and offline state/parameter estimation.