Extrapolating the Experimental Data to Predict the Longevity of LI-Battery


JiaGuo1, HeshengGong2 and Bo Gao3, 1Taizi Education and Technology, Co.Ltd., China, 2Eindhoven University of Technology, Netherland, 3Beijing Jiaotong University, China


In search of a particular lithium battery with reliable safety and high energy, quantities of research have been focused on the chemical substances for the Anode and Cathode, respectively. In Cui's laboratory, an efficiency of 98.54% for more than 600 cycles as well as long lifespan beyond 900h in a LiCu-Ag@Li cell can be realized. A high cyclability of 98% capacity can be achieved after 1000 cycles along with a long lifespan of 1500h in a SiOxCy@Li cell, which both prevents electrons from piercing through a separator, and leverages the efficacy of the lithium-ions via a binder. Thanks to Cui et al. and Severson et al., we either have got approved for or searched for the published data regarding the lithium-ion battery's lifespan and chart a series of diagrams that reveal the curve-shaped trendline and unexpected surges in the first, middle and last few cycles of a cell's life. The more a shocking cusp (outliers) surfaces, the more a decline steepens. We compare the data from the laboratory to on-board batteries and build a polynomial regression in order to predict the life end of those cells. While the non-linear regression is unable to best fit every moment of a cell's decrepitude, our team create a regression model to increase the accuracy of predication to an average of 97.693% in the primary test according to the first 30-225 cycles, then seek the optimization for longevity forecast by programming solver and hyperparameter, and finally find a (non-fixed) relationship between the speed and acceleration during the period of a cell's degradation. SVM model has also been created along with its corresponding 3D pattern with Temperature considered and so has the model Multiple Regression but the cost/benefit analysis will be continued in future study of relevant subject for prediction on newly-bonded cells or all-purpose commercial batteries.


prediction accuracy, non-linear regression, speed, acceleration, optimization

Full Text  Volume 13, Number 20