Abstract:

Accurate estimation of the peak power state (State of Power, SOP) for power batteries is crucial to ensuring the safe operation and enhancing the driving range of new energy vehicles. To address the insufficient accuracy of existing SOP estimation methods, this study proposes a joint estimation model, MCC-SSA-ELM, which integrates multi-constraint conditions (MCC) with a Sparrow Search Algorithm (SSA)-optimized Extreme Learning Machine (ELM) for error prediction and correction. Using a single lithium manganese oxide (LMO) battery as the research subject, a second-order RC equivalent circuit model is first established. Online parameter identification is achieved through a forgetting factor-based recursive least squares method, and the battery's State of Charge (SOC) is dynamically estimated using an adaptive extended Kalman filter (AEKF) algorithm. Furthermore, three-level operational conditions (30 seconds, 2 minutes, and 5 minutes) are defined based on discharge durations to simulate real-world scenarios. A preliminary SOP estimation model under varying durations is developed by comprehensively considering constraints including SOC, voltage and the maximum allowable discharge current. Subsequently, absolute error datasets between SOP estimated values and measured values under multi-constraint conditions are utilized to train both ELM and SSA-ELM error prediction models, enabling dynamic compensation and correction of preliminary estimates. Experimental results demonstrate that the proposed MCC-SSA-ELM model significantly improves SOP estimation accuracy. Compared to the MCC and MCC-ELM models, the average relative errors of the MCC-SSA-ELM model under 30-second, 2-minute, and 5-minute durations are reduced by 0.382%, 6.215%, and 6.858%, respectively, with final errors consistently controlled within 0.15%. These findings validate the superiority and engineering practicality of the proposed method.

Key words: SOP, Sparrow Search Algorithm, Extreme Learning Machine, Multi-constraint conditions