Lithium Battery Life Prediction Based on PSO-ELM

As the core component of energy storage and power systems, lithium batteries are widely used in consumer electronics, electric vehicles, aerospace and other fields. The life degradation caused by capacity attenuation directly affects the operational safety and reliability of equipment. Aiming at the engineering problems that lithium battery capacity is difficult to measure online directly and the prediction accuracy and stability are insufficient due to the random parameter generation of traditional extreme learning machine (ELM), this paper proposes a remaining useful life (RUL) prediction method for lithium batteries based on particle swarm optimization extreme learning machine (PSO ELM). Oriented to the application requirements of low computing power, real time online and easy engineering deployment, this paper selects four online measurable parameters (average discharge voltage, discharge time of equal voltage drop, time of the lowest discharge voltage, charging time of equal voltage rise) as indirect health indicators, and verifies their strong linear correlation with capacity by Pearson, Spearman, Kendall and first order partial correlation coefficients. The particle swarm optimization algorithm is used to optimize the input weights and hidden layer biases of ELM to solve the randomness defect of traditional ELM. Experiments are carried out based on the NASA lithium battery dataset. The results show that when 100 cycles are taken as the training set, the absolute errors of predicted failure cycles of B0005 and B0006 batteries are both less than 2, and the mean relative errors are as low as 0.009 and 0.0148 respectively. The prediction accuracy and stability are significantly better than those of traditional ELM. The proposed method provides a lightweight and highly reliable engineering application scheme for lithium battery health state assessment and remaining life prediction in resource constrained scenarios.