【摘要】 针对锂离子电池采用极限学习机（extreme learning machine, ELM）进行剩余使用寿命预测时，存在预测结果不稳定和预测准确度不高等问题，文章提出一种基于改进的麻雀搜索优化的极限学习机预测算法（Cubic sine-cosine firefly-enhanced sparrow search algorithm-ELM,CSFSSA-ELM）。该方法基于ELM并引入改进的麻雀搜索算法（CSFSSA）优化ELM。优化改进策略包含采用Cubic映射初始化优化麻雀群体、运用正余弦算法更新发现者位置，以及通过萤火虫扰动策略优化麻雀的位置3个部分。首先从锂离子电池的数据集中提取等压升充电时间（HI1）、等压降放电时间（HI2）、最低放电电压时间（HI3）、最高放电温度时间（HI4）及平均放电电压（HI5） 5个间接健康因子（HI）作为评估电池健康状态的指标；然后，使用NASA研究中心及CALCE的公开数据集对CSFSSA-ELM模型进行测试，证明了所提算法的有效性；最后分别与基于ELM、麻雀搜索算法优化的极限学习机（sparrow search algorithm-ELM,SSA-ELM）和基于Cubic映射的麻雀算法优化的极限学习机（Cubic-enhanced sparrow search algorithm-ELM,CSSA-ELM）模型针对2类数据集进行了预测结果对比。实验结果显示，所提CSFSSA-ELM模型对比ELM、SSA-ELM及CSSA-ELM模型在预测精度上有所提高，同时降低了预测误差，其中模型预测的均方根误差RMSE均在0.03以内，平均绝对误差MAE均在0.02以内，决定系数R~2均在0.98以上，证实了所提算法的有效性。更多还原

【Abstract】 Given the challenges in applying the extreme learning machine(ELM) to predict the remaining useful life(RUL) of lithium-ion batteries, such as unstable predictions and low accuracy, this paper proposes a Cubic sine-cosine firefly-enhanced sparrow search algorithm-extreme learning machine(CSFSSA-ELM) prediction algorithm. This method is based on the ELM optimized by an improved sparrow search algorithm(CSFSSA), incorporating three enhancement strategies: initializing the sparrow population with Cubic mapping, updating discoverer positions with the sine-cosine algorithm, and refining sparrow positions with a firefly disturbance strategy. Firstly, five indirect health indicators(HI) were extracted from the lithium-ion battery datasets for health status evaluation: the constant voltage charging time(HI1), the constant voltage discharging time(HI2), the lowest discharge voltage time(HI3), the highest discharge temperature time(HI4), and the average discharge voltage(HI5). Then, the CSFSSA-ELM model was tested using publicly available datasets from NASA’s research center and CALCE, demonstrating the effectiveness of the proposed algorithm. Finally, prediction results on these datasets were compared among the original ELM, sparrow search algorithm-ELM(SSA-ELM), and Cubic-enhanced sparrow search algorithm-ELM(CSSA-ELM) models. Experimental results showed that the proposed CSFSSA-ELM model improved prediction accuracy compared to the ELM, SSA-ELM, and CSSA-ELM models, while reducing prediction errors with RMSE below 0.03, MAE below 0.02, and R~2 above 0.98 for all predictions, confirming the effectiveness of the proposed algorithm.更多还原