改进灰狼优化算法医疗锂电池剩余寿命预测

Improved Grey Wolf Optimization Algorithm for Life Prediction of Medical Lithium Batteries

【目的】通过改进灰狼优化算法对医疗锂电池进行剩余寿命预测,从而保障抢救时机并减少医疗事故的目的。【方法】运用小波核极限学习机(Wavelet kernel extreme learning machine,WKELM)与小生境灰狼算法(Niche grey wolf optimization,NGWO)相融合的算法形成改进灰狼优化算法WKELM-NGWO算法。采用NGWO算法对WKELM参数进行优化处理,并将最大化训练集的分类准确度作为目标函数,得到寻优过程的数学模型。采用差分方式对医疗电子设备锂电池容量的时间序列进行处理,得到多维时间序列特征向量,归一化处理获得特征向量,并将其分为训练集和测试集。计算得出每只灰狼个体的适应度值fi,并对适应度值fi进行排序,适应度值fi排在前三的个体位置分别记为Xα,Xβ,Xδ。选择最优的灰狼个体位置作为WKELM参数对数据进行训练后,对心脏起搏器用锂电池和心脏除颤仪用锂电池两种锂电池测试样本进行剩余寿命预测操作。【结果】在相同的预测起始点下,WKELM-NGWO算法的均方根误差(RMSE)误差低于WKELM和NGWO算法,基于融合算法WKELM-NGWO的医疗电子设备锂电池剩余寿命(Remaining useful life)预测曲线更接近电池的退化曲线。【结论】WKELM-NGWO融合算法增强了对不同数据的适应能力,既克服了小波核极限学习机(WKELM)学习速度慢、结构不稳定的问题,也克服了小生境灰狼算法(NGWO)求解精度低、收敛速度慢从而导致跳不出局部最优解的问题。

[Purposes]To improve the life expectancy of medical lithium batteries by improving the gray wolf optimization algorithm,so as to ensure the rescue timing and reduce the purpose of medical accidents.[Methods]The improved gray wolf optimization algorithm WKELM-NGWO algorithm was formed by the algorithm of wavelet kernel learning machine(WKELM)and niche grey wolf algorithm(NGWO).The NGWO algorithm is used to optimize the WKELM parameters,and the classification accuracy of the training set is maximized as the objective function to obtain the mathematical model of the optimization process.The time series of the lithium battery capacity of the medical electronic device is processed by differential method,and the multi-dimensional time series feature vector is obtained,and the feature vector is obtained by normalization,and is divided into a training set and a test set.The fitness value of each gray wolf individual is calculated,and the fitness value is sorted.The individual positions of the fitness value ranked in the first three are respectively recorded as.After selecting the optimal gray wolf individual position as the WKELM parameter to train the data,the remaining life prediction operation is performed on the lithium battery B1 of the cardiac pacemaker and the lithium battery test sample for the cardiac defibrillator.[Findings]Under the same prediction starting point,the root mean square error(RMSE)error of the WKELM-NGWO algorithm is lower than that of the WKELM and NGWO algorithms,and the remaining life of the medical electronic device based on the fusion algorithm WKELM-NGWO(Remaining Useful Life)The prediction curve is closer to the degradation curve of the battery.[Conclusions]The WKELM-NGWO fusion algorithm enhances the adaptability to different data,overcomes the problem of slow learning and structural instability of the wavelet kernel limit learning machine(WKELM),and overcomes the niche grey wolf algorithm(NGWO).Low precision and slow convergence result in the problem of not jumping out of the local optimal solution.