动力电池的模型参数辨识与状态估计是电动汽车电池管理的核心技术问题,对充分发挥动力电池的驱动效能、延长动力电池使用寿命以及保障电动汽车的整车安全等方面都具有重要意义。然而，由于电池系统是一个由多种电化学过程相互耦合形成的综合体系，内部机理复杂且可观测变量少，并易受多种环境因素影响，因此在实际应用中难以准确监测。如何在有限的测量信息下，利用电池模型的先验知识以及算法的自适应更新能力，对电池内部参数与状态进行综合估计，是目前理论研究与工程实践中的难点。为此，本文围绕动力电池的模型参数辨识、荷电状态估计以及健康状态估计三大主题，从以下几个方面开展研究： 其一，通过建立面向动力电池建模的测试平台，并设计相应的特性测试方案，对电池进行容量、直流等效内阻、动态工况等多项测试，为后续章节的模型参数辨识及状态估计提供数据支撑。同时，结合测试数据对动力电池进行建模，利用离线辨识方法获得不同环境条件下的模型参数，并对其进行误差分析。 其二，针对模型参数的实时更新需求，提出了基于预测误差最小化的系统辨识方法，利用在线监测获得的数据对模型参数进行辨识，解决了参数在动态工况下的自适应更新问题。此外，针对实际运行环境中的噪声干扰问题，提出了基于偏差补偿递推最小二乘法的辨识方法，能够有效减小由噪声导致的辨识误差，并结合动力电池不同电化学过程的时间特性，建立了模型参数辨识的多时间尺度运行框架，根据参数的优先级进行异步更新，增强递推辨识方法的稳定性。 其三，在动力电池电化学阻抗谱分析的基础上，研究了动力电池的分数阶特性，据此建立了分数阶等效电路模型，从结构上对传统模型进行优化。提出了基于分数阶无迹卡尔曼滤波的动力电池荷电状态估计方法，解决了在分数阶系统上实现状态估计的问题。同时，针对分数阶模型特有的模型阶次参数，设计了独立的滤波器对其进行估计，解决了分数阶模型的参数辨识问题，并实现电池荷电状态与模型阶次的同步更新。 其四，通过对动力电池进行长期的循环劣化测试，获得电池在不同温度、循环区间以及充放电倍率下的容量与内阻谱曲线的劣化规律。同时，对动力电池健康状态特征指标及其影响因素进行分析，并利用所得测试数据，结合学习算法建立了电池容量与直流等效内阻谱之间的关系模型，利用不同荷电状态下的内阻点信息，对电池当前实际容量进行预测，实现对动力电池的健康状态估计。 其五，作为模型参数辨识及状态估计的应用案例，针对退役电池梯次利用的性能检测与分选配组问题，建立了梯次利用批量测试平台，对同批次退役电池进行统一检测，并结合参数辨识与状态估计方法对检测数据进行分析，提出了基于充电电压曲线的快速分选方法，提高退役电池梯次利用的分选效率。 本论文的主要研究目的在于，通过充分利用动力电池的在线监测信息，以及结合电池模型的先验信息，实现电池模型参数及性能状态在动态工况下的实时准确估计。所提出的基于偏差补偿最小二乘法的参数辨识方法、基于分数阶双卡尔曼滤波的荷电状态估计方法、基于劣化模型的健康状态估计方法以及针对退役电池梯次利用的快速分选方法，解决了对动力电池隐含变量的在线估算问题，提高动力电池的使用效率与安全性，具有较强的现实意义与应用价值。 关键词：电动汽车；动力电池；参数辨识；荷电状态估计；分数阶模型；健康状态估计；梯次利用

 The parameter identification and state estimation of power battery is the core technical issue of electric vehicle, which plays an essential role in improving the power efficiency, prolonging the battery service life and ensuring the safety of the vehicle. However, since the battery is an integrated system which consists of various electrochemical processes, with complex internal mechanism and few observable variables, it is difficult to accurately measure its states in practical use. How to make full use of the prior knowledge of battery model as well as the adaptive updating capability of algorithms, and to estimate the internal parameters and states of the battery with limited information, is a difficult issue in theoretical research and engineering practice. Therefore, aiming at model parameter identification, state of charge estimation and state of health estimation of power battery, detailed studies have been carried out in this thesis from the following aspects: Firstly, through a modeling oriented testing platform, the battery testing schedule is designed, and the battery capacity, direct current equivalent resistance, and dynamic profile tests are performed. The test results are used to verify the effectiveness of methods in the following chapters. Meanwhile, the battery modeling is conducted with the testing data. The model parameters are acquired with offline identification method and the modeling error is also analyzed. Secondly, aiming at the requirement of real-time identification of battery model parameters, the error minimize based method is proposed and the parameters can be identified using the online data, which solves the problem of parameter adaptive updating under dynamic environment. Moreover, aiming at the problem of noise disturbance, the bias compensation recursive least square method is proposed and the noise caused estimation error can be eliminated effectively. In addition, the multi time scale identification mechanism is proposed, which has considered the time properties of different electrochemical processes. The model parameters can be asynchronously updated according to different priorities, making the recursive method more robust. Thirdly, based on the electrochemical impedance spectroscopy analysis, the fractional properties of power battery is studied and the fractional equivalent circuit model is established, which structurally improves the performance of classical models. The state of charge estimation method based on fractional unscented Kalman filter is proposed, which solves the problem of state estimation on fractional system. Meanwhile, aiming at the particularity of order parameter in fractional order model, an independent estimator is designed, and the system states and model orders can be estimated simultaneously. Fourthly, through the long term cycling tests, the degradation patterns of battery capacity and internal resistance curves under different temperature, different depth of discharge and different charge rate are investigated. The character indices and impact factors of battery state of health are analyzed. Meanwhile, the relation model of battery capacity and equivalent internal resistances is established with learning algorithm, based on the acquired testing data. The current actual capacity can be calculated by the input of several resistance points, and the state of health of the battery can then be estimated. Fifthly, as an application case using the proposed parameter identification and state estimation methods, aiming at the performance detection and classification of retired batteries, the batch testing platform for echelon use is designed and the cells in the same batch can be tested uniformly. By combining the testing data and the identification and estimation methods, the rapid classification method based on charge voltage curve is proposed, which can improve the classifying efficiency in the echelon use of retired batteries. The target of this thesis is that to make full use of the online measured data and the prior knowledge of the battery model, to achieve accurate estimation of model parameters and performance states under dynamic environment. The parameter identification method based on bias compensation recursive least square, the state of charge estimation method based on fractional dual Kalman filter, the state of health estimation method based on battery degradation model and the rapid classification method for the retired batteries are proposed in this thesis. Such models and methods are able to solve the practical problem of online estimation of battery internal variables, which can improve the utilization efficiency and safety of power batteries, and thus possesses significant applicable value in practical engineering. Key words: electric vehicle; power battery; parameter identification; state of charge estimation; fractional order model; state of health estimation; echelon use.