基于VMD和Bat-KELM的仿真变电站蓄电池剩余寿命预测
Remaining Useful Life Prediction of Simulation Substation Batteries Based on VMD and Bat-KELM
任罡 1季宁 1胡晓丽 1李世倩 1张洁华 1吴祎2
作者信息
- 1. 国网江苏省电力有限公司技能培训中心,苏州 215004
- 2. 南京邮电大学自动化学院,南京 210023;南京邮电大学人工智能学院,南京 210023
- 折叠
摘要
仿真变电站蓄电池的工作模式呈现间歇非连续性,导致电池性能在退化过程中存在容量再生现象,退化规律具有非平稳性和随机性,增大了蓄电池精确剩余寿命RUL(remaining useful life)的难度.针对存在容量再生现象的蓄电池剩余寿命预测问题,提出了变分模态分解VMD(variational mode decomposition)和蝙蝠(Bat)优化核极限学习机 KELM(kernel extreme learning machine)组合的预测方法.基于 VMD 将蓄电池健康状态SOH(state of health)时间序列分解为整体退化分量和容量再生分量;利用Bat优化KELM构建各分量预测模型,以提高分量趋势预测精度;通过各分量独立预测结果的叠加,得到精确的蓄电池健康状态及剩余寿命预测值.将该方法应用于蓄电池退化数据实例分析中,结果表明该方法相较于KELM模型及VMD-KELM模型,预测精度更高,验证了该方法的优越性.
Abstract
Simulation substation batteries often work under discontinuous operation conditions,which will result in capacity regeneration of batteries during their performance degradation.The degradation of batteries shows nonstationary and random characteristics,leading to a low prediction accuracy for the remaining useful life(RUL).Aimed at the problem of RUL prediction of batteries with capacity regeneration,a prediction method is proposed based on variational mode decomposition(VMD)and bat optimized kernel extreme learning machine(Bat-KELM).First,VMD is employed to decompose the battery state-of-health(SOH)time series into overall degradation components and capacity regeneration components.Then,Bat-KELM is used to construct prediction models of each component,so that the prediction accuracy of component trend is improved.At last,the prediction results of all components are blended together to yield the accurate battery SOH prediction results as well as the RUL results.The proposed method is applied to the analysis of battery degradation instance data,and results show its superiority in terms of prediction accuracy compared with the KELM and VMD-KELM models.
关键词
仿真变电站/蓄电池/剩余寿命预测/变分模态分解/核极限学习机Key words
Simulation substation/battery/remaining useful life(RUL)prediction/variational mode decomposition(VMD)/kernel extreme learning machine(KELM)引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据