Battery temperature prediction using a hybrid machine learning and system identification technique

Abstract

Battery thermal management is crucial for electric vehicle technology. Maintaining the battery within a specific temperature range ensures safety and efficiency. Excessive heat can damage the battery and reduce its lifespan, while too low temperatures can also impair performance. Battery thermal management systems are responsible for maintaining the optimal temperature range for batteries. BTMSs work in conjunction with battery management systems to ensure that the battery's temperature is carefully monitored and controlled. BTMSs use a combination of active and passive methods to regulate the battery temperature, including cooling and heating systems, thermal insulation, and thermal energy recovery. Currently, the main function of Battery Thermal Management Systems is to monitor the battery temperature in real-time. What they lack is the ability to predict future temperatures. Introducing a temperature predictive system, which utilises input parameters like current battery temperature, ambient temperature, state of charge, and internal resistance and capacity, can enhance performance. The accuracy of the prediction model determines how well the system maintains the battery within the optimal temperature range. Machine learning is valuable for prediction applications due to its ability to analyse vast data sets, uncover intricate patterns, and establish correlations. However, it necessitates extensive data for training, encompassing diverse environmental conditions and usage scenarios. To address the challenges of accurately predicting battery temperature, this study proposes a new hybrid system, including an ML-based battery temperature prediction model along with an online battery parameter identification unit. The online battery parameter identification unit updates the battery's electrical parameters, which are used to calculate the battery's heat generation rate in real-time. That unit provides continuous updates to the prediction model, allowing it to dynamically adjust the battery's electrical and thermal parameters in response to any changes in the battery (i.e. changes in SoC, temperature and battery age). The system's ability to provide up-to-date information on the battery's state enhances the accuracy of the prediction model. The prediction model consists of an Adaptive Neuro- Fuzzy Inference System to analyse the battery's thermal behaviour based on various input parameters. The input parameters include the battery's current temperature, the ambient temperature, the battery's internal parameters and a short history of charge/discharge. The prediction model then uses that input data to predict the battery's future temperature in real-time. By incorporating the data from the battery online identification system, the prediction model can accurately adjust the thermal and electrical parameters of the battery, ensuring the accuracy of the temperature prediction. Experimental tests were conducted on mass-produced Samsung cylindrical cells for results validation, simulating real-world usage across a wide range of ambient temperatures (-10 °C to 40°C) and long-term ageing tests using the WLTP driving cycles. This generated a valuable dataset for thermal analysis. The proposed system accurately predicts cell temperature under various conditions, demonstrating robust performance against changes in state of charge, state of health, and ambient temperature. The model is designed for real-time industrial applications with minimised complexity.