Intelligent Estimation of Lithium‐Ion Battery SoC and Temperature Under Uncertainty: A Hybrid Approach for High‐Performance Electric Vehicle Applications

Accurate estimation of lithium‐ion battery (LiB) state of charge (SoC) and core temperature under uncertain and extreme operating conditions is critical for electric vehicle (EV) battery management systems (BMS). Existing machine learning (ML) approaches often fail to simultaneously capture nonlinear battery dynamics, temporal dependencies, and parameter uncertainty, leading to degraded performance in real‐world scenarios. This paper proposes a novel hybrid framework, multi‐distance spatial‐temporal graph neural network‐stellar oscillation optimizer (MDSTGNN‐SOO), which reformulates SoC and temperature estimation as a joint spatiotemporal learning and adaptive parameter optimization problem. The MDSTGNN explicitly models multi‐hop spatial interactions among battery cells and long‐range temporal dependencies, enabling robust prediction under dynamic loads and noisy measurements. In parallel, the SOO continuously updates battery model parameters to compensate for aging and temperature variations, overcoming limitations of fixed‐parameter AI estimators. The proposed framework is validated on the NASA LiB dataset and implemented in MATLAB. Comparative results against artificial neural network (ANN), random forest (RF), and marine predators algorithm (MPA)‐based methods demonstrate superior estimation accuracy, achieving root mean square error values of 0.50%–0.56% for SoC and 0.05–1.10°C for core temperature, with minimal inference latency. The primary innovation lies in the tight integration of graph‐based spatiotemporal learning with adaptive parameter optimization, enabling reliable, SoC, and temperature estimation for high‐performance EV battery management.