Estimation of Surface Temperature Distributions Across an Array of Lithium-Ion Battery Cells Using a Long Short-Term Memory Neural Network

Abstract

<div class="section abstract"><div class="htmlview paragraph">As electric vehicles are becoming increasingly popular and necessary for the future mobility needs of civilization, further effort is continually made to improve the efficiency, cost, and safety of the lithium-ion battery packs that power these vehicles. To facilitate these goals, this paper introduces a data driven model to predict a distribution of surface temperatures for a lithium-ion battery pack: a long short-term memory (LSTM) neural network. The LSTM model is trained and validated with lithium-ion cells electrically connected to form a battery pack. Voltage, current, state of charge (SOC), and cell surface temperature from two arrays are used as inputs from a wide range of high and low temperature drive cycles. Additionally, ambient temperature is added as an input to the LSTM model. In summary the LSTM model can accurately characterize and predict a distribution of lithium-ion cell surface temperatures arranged in a battery pack under extreme conditions to an accuracy of 1.53<b>°</b>C. Furthermore, making use of an external sensor to measure the ambient temperature of the battery pack further increases the accuracy of the LSTM model. With this data driven model, fewer testing data is required to validate a battery pack during development, less sensors are needed in production to monitor the health of the battery pack, and data can be generated for a wide variety of applications which may or may not be possible in a lab environment.</div></div>

An LSTM-PINN Hybrid Method to Estimate Lithium-Ion Battery Pack Temperature

“…This paper has developed a LSTM-PINN model to estimate the lithium-ion battery pack temperature. In [14], the same technical problem was reviewed, and a LSTM model was proposed to estimate the battery pack temperature. With the LSTM model which solely relied on data to train the prediction method, there is a higher prediction error where the geometric location of the prediction is different from the location of the input temperature in the LSTM model.…”

“…In [14], the temperature prediction made by the LSTM model showed good prediction accuracy when using the measured temperature in one module to predict the temperature of the same position but in a different module. However, the accuracy dropped when using the same input to predict the temperature of a different position in a different module.…”

“…The usage column indicates whether the data was used for training or testing the LSTM-PINN. See the appendix section of [14] for current traces of the drive cycles. Four aspects of the performance were investigated in four comparative studies.…”

“…In this comparative study, the prediction errors of the LSTM-PINN hybrid model were compared with the prediction errors of the LSTM model in [14]. Unlike our LSTM-PINN hybrid model, the LSTM model only utilizes the training data to learn the features.…”

“…This paper proposes a novel hybrid model in which a physics-based model is combined with two data-driven models (i.e., FNN and LSTM) to predict battery surface temperatures in separate locations within the pack. This work is an extension of existing research in [14]. Instead of using the LSTM alone, heat transfer equations are incorporated into the cost function of the FNN along with the outputs from LSTM to enable the hybrid model to converge faster and provide more accurate results.…”