IC2ML: Unified battery state-of-health, degradation trajectory and remaining useful life prediction via intra-cycle and inter-cycle enhanced machine learning

Strategic management of lithium-ion batteries (LIBs) depends on evaluating current health status and predicting future degradation paths. Despite extensive research on core management tasks like state of health (SOH) estimation, degradation trajectory prediction, and remaining useful life (RUL) prediction, these tasks remain isolated without leveraging their inherent connections. This work proposes an unified framework that enables joint battery SOH, degradation trajectory and RUL prediction via an intra-cycle and inter-cycle enhanced machine learning (IC 2 ML). IC 2 ML uses 1-D time-serials voltage data to implement SOH prediction, where inter-cycle embeddings are further self-attention for degradation trajectory prediction. The RUL is derived from degradation trajectory prediction based on anticipated SOH levels, enabled by cross attention between output embeddings and input inter-intra cycle embeddings. The results demonstrate that using 0.1V sampling interval data that can be extracted onsite, the average root mean square error for SOH, degradation trajectory, and RUL prediction is 1.85 %, 2.36 % and 23.90 cycles, respectively, validated on 121 batteries spanning 10 operation conditions. Sensitivity analysis shows that IC 2 ML can be adapted to scenarios where a few historical data is accessible. Broadly, this work highlights the potential of strategical battery algorithm co-design using intra-cycle and inter-cycle battery degradation information for various management tasks.