SOH prediction for Lithium batteries using WPT and crested porcupine deep extreme learning machine under different temperatures

The state of health (SOH) of lithium batteries is crucial for safety assessment, yet the fluctuation of current, voltage, and temperature parameters poses a significant challenge to accurate estimation. This study introduces an enhanced approach for predicting lithium battery SOH across different temperatures by integrating Wavelet Packet Transform (WPT) with an improved Crested Porcupine Optimization (ICPO) algorithm. Initially, WPT is applied to denoise the data from various temperature groups. Subsequently, key SOH factors correlated with charging current and other metrics are extracted and their statistical correlation with SOH is analyzed. To further refine the process, the ICPO algorithm, which incorporates crossover and mutation operations from the Differential Evolution (DE) algorithm to bolster the global and local search capabilities of the original Crested Porcupine Optimization (CPO), is utilized. Specifically, the ICPO algorithm optimizes the critical parameters of a Deep Extreme Learning Machine (DELM), thus proposing the ICPO-DELM model. The proposed ICPO-DELM framework demonstrates high accuracy and robust generalization ability in predicting SOH across different temperature conditions, as evidenced by comparative experiments with other models in terms of accuracy and stability.