Research On Lithium Ion Battery SOC Estimation And RUL Prediction

With the rapid rise of the industrialized electric vehicle (EV), the energy issue hasbeen drawing wide attention from both academia and industry. The efficientmanagement of the lithium-ion battery pack becomes a key to ensure reliability andsecurity of EVs. As the core of battery management, the state of charge (SOC)estimation is one of the most difficult problems due to the nonlinear effects andreal-time requirement. On the other hand, the remaining useful life (RUL) is anothercritical performance indicator of battery management, which guarantees the timelyreplacement of batteries. However, the research on RUL prediction has not been starteduntil recent years, therefore there is little existing work can be found. Based on theabove considerations, this thesis conducts comprehensive study on the SOC and RULproblems and battery monitoring system design, major contributions are summarized asfollows:1. Lithium-ion batteries SOC estimation. Considering the external and internalnonlinear factors as well as the observation that the voltage remains mostly constant asSOC changes during the charge/discharge process of LiFePO4battery, the backpropagation (BP) neural network model is chosen as the training algorithm due to itsadaptive learning ability. Through the constant current discharge experiment, weexplore the relationship between the battery capacity and the charge/discharge rate.Based on the correlation between the battery parameters and the SOC, we choosecurrent and voltage as the inputs for the BP network to carry out the network trainingand SOC estimation. The correctness and accuracy of our algorithm are verified by theexperiment results.2. Lithium-ion batteries RUL prediction. First, three algorithms i.e. particlefiltering (PF), support vector machine (SVM), autoregressive moving average (ARMA)are applied for RUL prediction. Based on the advantages and disadvantages of theabove algorithms, a novel approach using improved autoregressive (AR) model withparticle swarm optimization (PSO) is proposed. Then, the root mean square error(RMSE) is used as the fitness function for AR model order determination. In addition, the information contained in the data is updated through metabolism at the predictionstage which makes the AR model order change adaptively. Finally, the experimentaldata are used to validate the proposed prognostic approach, and the results showaccurate RUL prediction trends and small errors.3. Lithium-ion batteries monitoring system development. Based on the algorithmsdeveloped above, a lithium-ion battery monitoring system is designed to meet therequirements of the lithium-ion battery monitoring. The system consists of the followingcomponents: signal acquisition for battery parameters, data transmission, computermonitoring, analysis and processing. Therefore, the SOC estimation and the RULprediction are achieved in a real time fashion.