Online estimation of key battery states for management of vanadium redox flow battery
Date of Issue2017-05-29
School of Electrical and Electronic Engineering
Rechargeable battery systems are experiencing rapid advancement due to their pivotal roles in the renewables penetration, smart grid formation and transport electrification. Among various rechargeable battery systems, the all-vanadium redox flow battery (VRB) has received extensive attention especially in the field of large-scale energy storage. To enhance the reliability, efficiency, and longevity, the battery management system (BMS) has to be well-designed to supervise the operation of VRB. In this thesis, focuses are given to the investigation of online battery model identification and the associated estimation methods to keep track of the essential battery states including the peak power, the state of charge (SOC), and the instantaneous capacity. Reliable state estimation depends directly on an accurate battery model. However, the parameters of battery model are time varying with the operating condition variation and battery aging thus the common methods where model parameters are prescribed offline are not adequate. To address this issue, the first part of this thesis proposes to use an online adaptive battery model to reproduce the battery dynamics accurately. Two algorithms including the recursive least squares (RLS) and the extended Kalman filter (EKF) are compared on the ability of online model parameters identification and the RLS proves to be superior considering the modeling accuracy, convergence, and computational complexity. The peak power is an essential state defined as the maximum power that is possible to be accepted or delivered by the battery at a specific operating point without violating the Safe Operating Area (SOA). Based on the online identified battery model, an adaptive peak power estimator which incorporates the constraints of voltage limit, SOC limit and design current limit is proposed to fully exploit the potential of VRB. The proposed peak power estimator is verified with a “two-step verification” method. The influence of prediction time horizon on the peak power is also analyzed. The abovementioned model identification method relies on an open-circuit cell to accurately measure the open circuit voltage (OCV). Using such an additional cell, however, will increase the configuration complexity of the VRB system. To tackle this problem, a new online model identification method without requiring the open-circuit cell is proposed and based on which an online SOC estimation method is investigated in the second part of the thesis. Two main contributions have been drawn. Firstly, the estimations of model parameters and OCV are fully decoupled to rule out the possibility of cross interference between them. Secondly, multiple timescales are adopted for different estimators by analyzing the model sensitivity, stability, and accuracy. Based on the proposed multi-timescale estimator, the SOC is inferred directly from the SOC-OCV look-up table. Experimental results show that the proposed method is highly accurate and also robust to the initialization uncertainty and operating condition variation. The robustness to battery aging is also satisfactory within an aging degree of 14.78%. As another crucial index, the instantaneous capacity is a figure of merit describing the state of health (SOH) of the battery. The accurate update of capacity also contributes to improving the accuracy of the SOC estimate. Driven by this, the third part of the thesis focuses on the adaptive joint estimation of SOC and instantaneous capacity based on an online identified battery model. The model parameters are online adapted with the RLS method, based on which an EKF-based novel joint estimator is formulated to estimate the SOC and capacity concurrently. The proposed joint estimator compresses the filter order effectively which leads to a substantial improvement in the computational efficiency and numerical stability. Lab scale experiments are carried to show the fidelity of the proposed method. The comparison with other existing methods also suggests its superiority in terms of accuracy, convergence speed, and computational cost. The existence of noises is unavoidable in real applications and may decline the accuracy of both the model identification and state estimation. In the last part of this thesis, therefore, the effect of measurement noises is investigated. It is shown that the common methods for model identification are intrinsically biased if both the current and voltage sensors are corrupted with noises. The uncertainties in the battery model further degrade the accuracy of the SOC estimate. To address this problem, a novel technique which integrates the Frisch scheme based bias compensating recursive least squares (FBCRLS) with a SOC observer is proposed to enhance the model identification and SOC estimate. The proposed method estimates the noise statistics and compensates the noise effect so that the model parameters can be extracted without bias. The SOC is further estimated in real time with the online updated and unbiased battery model. Both simulation and experimental studies are carried out to show the validity of the proposed method. The thesis provides a set of data-driven based algorithms for the reliable and high-fidelity online estimation of key battery states. As only the onboard measured current and voltage signals are required, the proposed algorithms are also promising to be generalized to a broad range of other battery chemistries.