Performance Comparison of Long Short-Term Memory and a Temporal Convolutional Network for State of Health Estimation of a Lithium-Ion Battery using Its Charging Characteristics

Abstract

The market for eco-friendly batteries is increasing owing to population growth, environmental pollution, and energy crises. The widespread application of lithium-ion batteries necessitates their state of health (SOH) estimation, which is a popular and difficult area of research. In general, the capacity of a battery is selected as a direct health factor to characterize the degradation state of the battery’s SOH. However, it is difficult to directly measure the actual capacity of a battery. Therefore, this study extracted three features from the current, voltage, and internal resistance of a lithium-ion battery during its charging–discharging process to estimate its SOH. A battery-accelerated aging test system was designed to obtain time series battery degradation data. A performance comparison of lithium-ion battery SOH fitting results was conducted for two different deep learning architectures, a long short-term memory (LSTM) network and temporal convolution network (TCN), which are time series deep learning networks based on a recurrent neural network (RNN) and convolutional neural network (CNN), respectively. The results showed that the proposed method has high prediction accuracy, while the performance of the TCN was 3% better than that of the LSTM regarding the average maximum relative error in SOH estimation of a lithium-ion battery.

On-Road Experimental Campaign for Machine Learning Based State of Health Estimation of High-Voltage Batteries in Electric Vehicles

“…Examples of data-driven methods include a support vector machine (SVM) [25][26][27][28][29], random forest (RF) [30,31], artificial neural networks (ANNs) [32][33][34][35][36][37], recurrent-neural networks (RNNs) [38], and variants such as long short-term memory (LSTM) [39][40][41][42][43][44] and a nonlinear autoregressive network with exogenous inputs (NARX) [45][46][47]. ANNs learn from historical data to predict future behavior, i.e., the learning procedure exploits a dataset representative of the battery behaviour to make predictions.…”

State of Health Prediction of Lithium-Ion Battery Based on Deep Dilated Convolution

“…At the same time, they used data enhancement technology to generate data for CNN training, which further improved the prediction accuracy of the CNN. Bi et al [ 28 ] used a time convolution network (TCN) based on RNN and CNN and deduced three groups of characteristic variables with obvious correlation with SOH through the battery charge–discharge data curve, obtaining high prediction accuracy.…”

“…With the deepening of ANN research and the continuous improvement of computing capacity, deep learning methods such as deep neural networks (DNN) [ 27 ], convolutional neural networks (CNN) [ 19 ], and recurrent neural networks (RNN) [ 28 ] have become widely used to predict SOH. By increasing the number of hidden layers, a deep learning model can obtain more nonlinear relationships, combine the features of lower layers to form very abstract high-level features, and then complete complex regression tasks through simple models.…”

State-of-Health Estimation of Lithium-ion Batteries Based on Singular Value Decomposition and an Improved Group Method of Data Handling

“…19 The data-driven method mainly extracts feature data that can characterize the battery SOH from the historical data of lithium-ion batteries, and then uses data-driven algorithms to estimate SOH based on the extracted feature data. [20][21][22] Peng et al proposed a lithium-ion battery health state estimation method based on multi-health feature extraction and improved long short-term memory (LSTM) neural network. At the same time, a quantum particle swarm optimization algorithm was introduced to solve the difficult problem of selecting hyperparameters of neural networks.…”