Enhancing Battery Thermal Management With Virtual Temperature Sensor Using Hybrid CNN-LSTM

Temperature has a significant impact on lithium-ion batteries (LIBs) in terms of performance, safety, and longevity. Battery thermal management system is employed to ensure safe operation of the batteries, especially during fast charging, high power discharge, and extreme weather conditions, thus enhancing their performance and prolonging their lifespan. The thermal performance of batteries is typically monitored using temperature sensors, which directly measure their surface temperature (ST). But, as a battery pack's number of cells increases, so does its number of temperature sensors, which raises its cost and reduces its reliability. To address this problem, this paper introduces an innovative hybrid method leveraging deep learning algorithm, to accurately estimate the ST of lithium-ion batteries. The methodology integrates convolutional neural network (CNN), long-short term memory (LSTM), and deep neural network (DNN) components. Two distinctive CNN-LSTM configurations, series and parallel, are proposed for battery ST estimation. The effectiveness of the proposed approach is comprehensively validated using three distinct datasets with different chemistries and working operations. The validation process involves testing the model under two elevated ambient temperature of temperatures using constant current and Artemis urban drive profiles and on battery subjected to various dynamic driving profiles across a range of ambient temperatures (10°C ~ 25°C and -20°C ~ 10°C). The accuracy of the estimation is assessed through root mean square error (RMSE), revealing an error of less than 1.24°C and 1.30°C for fixed and varying ambient temperature conditions, respectively, which demonstrate the robustness and reliability of the proposed hybrid approaches in battery surface temperature.