State of Health Estimation for Lithium-Ion Batteries Based on Healthy Features and Long Short-Term Memory

Abstract

Precise estimation of state of health (SOH) are of great importance for proper operation of lithium-ion batteries equipped in electric vehicles. For real applications, it is however difficult to estimate battery SOH due to stochastic operation, which in turn speeds up aging process of the battery. To attain the precise SOH estimation, an efficient estimation manner based on machine learning is proposed in this study. Firstly, the voltage profile during charging and discharging process and incremental capacity variation are acquired through the cycle life test, and the healthy features correlating to battery degradation are extracted. Secondly, the grey relation analysis and entropy weight method are employed to analyze the healthy features. Finally, the long short-term memory is established to achieve the SOH estimation of battery. The experimental results highlight that the proposed method can effectively predict the battery SOH with preferable accuracy, stability and robustness. INDEX TERMS Healthy features (HFs), grey relational analysis (GRA), entropy weight method (EWM), long short-term memory (LSTM), state of health (SOH). NOMENCLATURE A. ABBREVIATIONS

Stage of Charge Estimation of Lithium-Ion Battery Packs Based on Improved Cubature Kalman Filter With Long Short-Term Memory Model

“…Moreover, with the accumulation of battery data and the improvement of computing capacity furnished by the graphics-processing unit (GPU), machine learning methods have progressively drawn wide attention. Among the black-box models, long-short term memory (LSTM) network, as an extended formation of recurrent neural network (RNN), can capture the long-term relationship of historical information [27]. Given the characteristics of long-term responses of battery voltage, LSTM network can be adopted as an ideal candidate to model the electrical performances of batteries.…”

“…illustrates the schematic diagram of a general LSTM network, where tanh denotes the hyperbolic function,  is the sigmoid activation function, whether a cell should remember or forget newly obtained information. More details about the LSTM can be referred to our earlier study in[27].…”

Capacity Prediction and Validation of Lithium-Ion Batteries Based on Long Short-Term Memory Recurrent Neural Network

“…Any read or modification operation can be achieved through controlling of these three gates. Additionally, the information selection of gate is mainly conducted by the sigmoid function, tanh function or matrix multiplication [24]. It can be seen from Fig.…”

“…In short, all of these data-driven methods need healthy features to establish a mapping relationship between SOH and feature variables. In other words, a reliable SOH estimation strongly requires proper feature extraction to perform qualified SOH diagnosis [24]. However, lithium-ion battery degradation is consecutive and generally involves hundreds to thousands of cycles, and the later degradation evolution is highly related with the former degradation information throughout these cycle operations.…”

A Dynamic Spatial-Temporal Attention-Based GRU Model With Healthy Features for State-of-Health Estimation of Lithium-Ion Batteries

“…Long short-term memory (LSTM) and Gated recurrent unit (GRU), as variants of RNN, were often employed to learn long-term dependency by using gating system. Wu et al [41] extracted several HFs during charging and discharging processes and used LSTM based method for SOH estimation. GRU which has fewer parameters than LSTM can reach or exceed the performance of LSTM in several areas, thus it has also been used to study battery degradation problems [42], [43].…”