Battery boost

An AI model that combines Long Short-Term Memory (LSTM) neural networks with Bayesian optimisation can improve both the accuracy and efficiency of electric vehicle battery state-of-health (SOH) estimates, a key measure used in battery management systems to track degradation over time. Details are provided in the International Journal of Vehicle Information and Communication Systems.

Lithium-ion batteries gradually lose capacity through repeated charging cycles. SOH expresses this decline as a percentage of the original charging capacity. Accurate SOH estimation is important for drivers charging the vehicles ahead of a road trip. If SOH has fallen, then the distance they will be able to travel will be less than when the vehicle’s battery was new. It is also a matter of safety, as degraded batteries are more vulnerable to overheating, electrical faults, and, in rare cases, thermal runaway, a self-reinforcing reaction that can lead to fire.

Electric vehicles have Battery Management Systems (BMS) to monitor voltage, current, and temperature. However, converting this data into a reliable SOH estimate is difficult because battery degradation is influenced by complex chemical processes, temperature changes, and driving behaviour.

The new model can retain earlier patterns in a sequence, helping capture long-term behaviour in battery performance. The model links “health features” extracted from the vehicle data to standardised battery capacity. By using the probabilistic statistical technique of Bayesian optimisation, the new model can home in on particular data points rather than processing all possibilities. This reduces unnecessary computation while maintaining performance and gives a useful improvement on accuracy and halves the average error rate.

By obtaining a more accurate SOH estimate, the vehicle can manage its battery better and indicate when maintenance and replacement are needed. The BMS system can thus operate closer to safe performance limits. There is also the potential for extending battery life by adjusting charging rates and extent as the battery ages.

Xiao, Z. (2026) ‘Bayesian optimised route and SOH estimation effect for Li-ion battery management system of electric vehicles based on LSTM’, Int. J. Vehicle Information and Communication Systems, Vol. 11, No. 2, pp.146–162.