Online remaining useful life prediction of lithium-ion batteries using bidirectional long short-term memory with attention mechanism

Critical Review on Improved Electrochemical Impedance Spectroscopy-Cuckoo Search-Elman Neural Network Modeling Methods for Whole-Life-Cycle Health State Estimation of Lithium-Ion Battery Energy Storage Systems

“…The DEGWO is used to optimize hyperparameters for enhanced model performance. Reference [151] constructs and compares the normal RNN [152], LSTM [98], [153], gated recurrent unit neural network (GRUNN) [154], opposite bidirectional structure [155] and other neural network models. The results demonstrate that the LSTM and the bidirectional LSTM (Bi-LSTM) have higher versatility and accuracy in estimating battery health states.…”

Remaining Useful Life Prediction of Lithium‐Ion Batteries Based on a Combination of Ensemble Empirical Mode Decomposition and Deep Belief Network–Long Short‐Term Memory

“…Additionally, a dilated convolutional operation is integrated with the conditional GCN to consider the temporal correlation among the aggregated features. Wang et al [ 25 ] proposed a bidirectional LSTM with attention mechanism model to predict online RUL by continuously updating the model parameters. One significant problem in data‐driven techniques is that insufficient or biased data might result in imprecise or wholly erroneous predictions.…”

A Temperature and Voltage Coupling Equivalent Electrical Behavior Model of Lithium‐Ion Battery Pack for Electric Unmanned Aerial Vehicle under Variable Load Conditions

“…A framework applying the Brownian motion‐based deterioration and particle filter under parameters optimization was proposed for online short‐term SOH evaluation and long‐term remaining useful life (RUL) forecasting by Dong et al [ 13 ] A method with Lebesgue sampling‐based extended Kalman filter was proposed for estimating the SOH and SOC of lithium‐ion battery by Yan et al [ 14 ] A deep full convolutional network model was proposed to estimate the state of charge of lithium‐ion battery from voltage, current and battery temperature values by Hannan et al [ 15 ] A nonlinear relationship between the open‐circuit voltage and the nominal charge state was established to estimate the lithium‐ion battery SOC by Sandoval et al [ 16 ] A linear regression model and a GPR model were developed to accurately predict lithium‐ion battery SOC using information measurements from electrochemical impedance spectroscopy by Babaeiyazdi et al [ 17 ] A stacked LSTM model combined with data from the constant current phase of the battery charging process was proposed to evaluate the battery SOC and SOH by Yayan et al [ 18 ] A probabilistic model with charge transfer resistance, temperature, and SOC as input variables was developed for lithium‐ion battery SOH estimation by Zhang et al [ 19 ] Two consecutive partial discharge intervals are used to estimate the battery equivalent circuit model parameters and the open‐circuit voltage to estimate the battery capacity, SOH, and SOC by Bavand et al [ 20 ] Battery life prediction not only understands the condition of the equipment but also avoids accidents caused by battery failure during operation. A method employing equivalent electrical behavior model was proposed for degradation identification of lithium‐ion battery by Xiong et al [ 21 ] A model employing KPCA and LS‐SVM was proposed for online life cycle health assessment for lithium‐ion battery in electric vehicles by Liu et al [ 22 ] A novel framework using enumerative method for optimization was proposed for lithium‐ion battery lifetime prediction by Astaneh et al [ 23 ] A new LMO‐NMC battery health state estimation algorithm based on enhanced single particle model (eSPM) parameter estimation was proposed for predicting SOH and RUL of battery in hybrid vehicles by Sadabadi et al [ 24 ] A bidirectional long short‐term memory with attention mechanism (Bi‐LSTM‐AM) model was proposed to predict SOH estimates in multiple steps in advance using the sliding window method by Wang et al [ 25 ] An improved feedforward‐long short‐term memory (FF‐LSTM) modeling method was proposed to achieve accurate whole‐life SOC prediction by effectively considering current, voltage, and temperature variations by Wang et al [ 26 ] Then, an improved noise‐resistant adaptive long‐ and short‐term memory (ANA‐LSTM) neural network with highly robust feature extraction and optimal parameter characterization was proposed for accurate RUL prediction based on an improved dual closed‐loop observation modeling strategy by Wang et al [ 27 ] With the development of data‐driven [ 26,28 ] methods, more and more methods have been used in battery hea...…”