Premium
Predicting battery life with early cyclic data by machine learning
Author(s) -
Zhu Shan,
Zhao Naiqin,
Sha Junwei
Publication year - 2019
Publication title -
energy storage
Language(s) - English
Resource type - Journals
ISSN - 2578-4862
DOI - 10.1002/est2.98
Subject(s) - battery (electricity) , decision tree , battery capacity , computer science , internal resistance , feature (linguistics) , machine learning , work (physics) , artificial intelligence , tree (set theory) , engineering , power (physics) , mathematics , mechanical engineering , linguistics , physics , philosophy , quantum mechanics , mathematical analysis
This work applies machine learning tools to achieve the early prediction of commercial battery life. We compared the prediction accuracy of different machine learning algorithms to the battery database. Among various algorithms, the decision tree (DT) method exhibits the highest accuracy of 95.2% to predict whether the battery can maintain above 80% initial capacity after 550 cycles. Using the initial two cycles of data, DT proposes that the change of discharge capacity is the main feature to estimate the lifetime type of batteries. Given the first 100 cycles, the factor with the maximum weight turns to the internal resistance for estimating the battery lifetime.