TR2019-076

A Data-Driven Method for Predicting Capacity Degradation of Rechargeable Batteries


    •  Pajovic, M., Orlik, P.V., Wada, T., Takegami, T., "A Data-Driven Method for Predicting Capacity Degradation of Rechargeable Batteries", IEEE International Conference on Industrial Informatics (INDIN), July 2019.
      BibTeX Download PDF
      • @inproceedings{Pajovic2019jul2,
      • author = {Pajovic, Milutin and Orlik, Philip V. and Wada, Toshihiro and Takegami, Tomoki},
      • title = {A Data-Driven Method for Predicting Capacity Degradation of Rechargeable Batteries},
      • booktitle = {IEEE International Conference on Industrial Informatics (INDIN)},
      • year = 2019,
      • month = jul,
      • url = {https://www.merl.com/publications/TR2019-076}
      • }
  • MERL Contact:
  • Research Areas:

    Machine Learning, Signal Processing


Rechargeable batteries supply numerous devices with electric power and are critical part in a variety of applications. An accurate monitoring and prediction of capacity degradation is directly related to making timely decision as to when a battery should be replaced, so that power disruption of the system it supplies power to is avoided. We propose a methodology for predicting capacity of a battery over future time horizon. The proposed method is based on training data consisting of occasional measurements, taken under the same conditions, of capacity and charge/discharge voltage/current of a certain number of batteries sharing the same chemistry and manufacturer, that otherwise undergo different usage patterns. In the operational/online stage, capacity degradation over future time horizon of a test battery cell of unknown state of health and previous usage pattern is predicted based on its capacity and voltage/current measurements over one charge/discharge cycle and the training dataset. The experimental validation reveals that the proposed method predicts capacity of a test battery cell over prediction time horizon of few hundred days of battery’s operation with relative prediction error below 1%.