The large-scale application of lithium-ion batteries in electric vehicles requires meticulous battery management to guarantee vehicular safety and performance. Temperatures play a significant role in the safety, performance, and lifetime of lithium-ion batteries. Therefore, the state of temperature (SOT) of batteries should be monitored timely by the battery management system. Due to limited onboard temperature sensors in electric vehicles, the SOT of most batteries must be estimated through other measured signals such as current and voltage. To this end, this paper develops an accurate method to estimate the surface temperature of batteries by combing the physics-based thermal model with machine learning. A lumped-mass thermal model is applied to provide prior knowledge of battery temperatures for machine learning. Temperature-related feature, such as internal resistance, is extracted in real-time and fed into the machine learning framework as supplementary inputs to enhance the accuracy of the estimation. A machine learning model, which combines a convolutional neural network with a long short-term memory neural network, is sequentially integrated with the thermal model to learn the mismatch between the model outputs and the real temperature values. The proposed method has been verified against experimental results, with accuracy improvement of 79.37% and 86.24% compared to conventional pure thermal model-based and pure data-driven approaches respectively.

State/temperature monitoring is one of the key requirements of battery management systems that facilitates efficient and intelligent management to ensure the safe operation of batteries in electrified transportation. This paper proposes an online end-to-end state monitoring method based on transferred multi-task learning. Measurement data is directly used for sharing information generation with the convolutional neural network. Then, the multiple task-specific layers are added for state/temperature monitoring. The transfer learning strategy is designed to improve accuracy further under various application scenarios. Experiments under different working profiles, temperatures, and aging conditions are conducted to evaluate the method, which covers the wide usage ranges in electric vehicles. Comparisons with several benchmarks illustrate the superiority of the proposed method with better accuracy and computational efficiency. The monitoring results under extremely current working profiles and variable internal and external conditions are evaluated. Results show that the mean absolute error and root mean square error of state of charge and state of energy estimation are less than 2.31% and 3.31%, respectively. The above errors in the prediction of future temperature five steps ahead are less than 0.89℃ and 1.29℃, respectively. The framework is also suitable for monitoring second-life batteries retired from electric vehicles. This paper illustrates the potential application of data-driven multi-state monitoring throughout the entire battery life.

To guarantee safe, efficient, and reliable operations of lithium-ion battery (LIB) systems, it is indispensable to monitor their state of temperature. However, subject to limited onboard temperature sensors and the challenges in measuring battery internal temperature, the directly available temperature information in a battery system is extremely insufficient. To this end, developing sensorless temperature estimation techniques to obtain the temperature of each cell is important. This article proposes an operando impedance-based method to estimate the volume-averaged temperature of LIB in real time under dynamic operating conditions. A generalized rule for selecting optimal impedance parameters has been revealed for the first time through a comprehensive analysis of the electrochemical impedance spectroscopy from different batteries. This rule can greatly reduce the time and effort to select optimal impedance parameters of the target cell for temperature estimations. The selected impedance parameters are then measured intermittently during battery operations via active pulse current injection, which allows the impedance acquisition under both loading and resting conditions. An estimation framework based on long short-term memory recurrent neural network has been proposed by taking advantage of the measured operando impedance, time-series current, and voltage data to achieve real-time temperature estimation, which distinguishes this work from existing impedance-based methods. The proposed methodology has been experimentally validated against different batteries and operating conditions, with the root mean square error of the estimations within 0.46 °C for all cases.