Cyclic Degradation Prediction of Lithium-Ion Batteries using Data-Driven Machine Learning
Accurately estimating the capacity degradation of lithium-ion (Li-ion) batteries is vital in ensuring their safety and reliability in electric vehicles and portable electronics. Future capacity estimation using machine learning (ML) models allow battery lifetime predictions with minimal cycling data...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
AIDIC Servizi S.r.l.
2022-09-01
|
| Series: | Chemical Engineering Transactions |
| Online Access: | https://www.cetjournal.it/index.php/cet/article/view/12688 |
| _version_ | 1828353291844583424 |
|---|---|
| author | Lerissah D. Lim Andrei Felix J. Tan Jan Goran T. Tomacruz Michael T. Castro Miguel Francisco M. Remolona Joey D. Ocon |
| author_facet | Lerissah D. Lim Andrei Felix J. Tan Jan Goran T. Tomacruz Michael T. Castro Miguel Francisco M. Remolona Joey D. Ocon |
| author_sort | Lerissah D. Lim |
| collection | DOAJ |
| description | Accurately estimating the capacity degradation of lithium-ion (Li-ion) batteries is vital in ensuring their safety and reliability in electric vehicles and portable electronics. Future capacity estimation using machine learning (ML) models allow battery lifetime predictions with minimal cycling data in the train set, well before capacity degradation occurs within the cell. The use of ML methods removes the need for prior knowledge of cell chemistry and the physical and chemical behaviors of batteries. In this paper, the data-driven ML models Gaussian process regression (GPR) and recurrent neural network – long short-term memory (RNN-LSTM) estimated the charge capacity of Li-ion batteries from the Oxford Battery Dataset, using only the battery's cycle index and capacity as input. With only 15 % of the battery’s lifetime as training data, the GPR model achieved a mean average percent error (MAPE) of 8.335 % and an R2 of 0.9755, while the LSTM model achieved a MAPE of 9.984 % and an R2 of 0.9898. These results indicate the goodness of fit and are comparable to results from similar models in the literature (MAPE = 9.1 to 15 %). The methodology may be applied to different features to help establish the relationship between health indicators and capacity fade and can be used in applications that require early capacity prediction such as in space technologies where lifetime and capacity are crucial in ensuring success and safety. This successful estimation highlights the promise and potential of accurately predicting Li-ion battery capacity degradation using a single-feature approach. |
| first_indexed | 2024-04-14T02:05:35Z |
| format | Article |
| id | doaj.art-37cef80980244df09b7854b69f0b3ca8 |
| institution | Directory Open Access Journal |
| issn | 2283-9216 |
| language | English |
| last_indexed | 2024-04-14T02:05:35Z |
| publishDate | 2022-09-01 |
| publisher | AIDIC Servizi S.r.l. |
| record_format | Article |
| series | Chemical Engineering Transactions |
| spelling | doaj.art-37cef80980244df09b7854b69f0b3ca82022-12-22T02:18:41ZengAIDIC Servizi S.r.l.Chemical Engineering Transactions2283-92162022-09-019410.3303/CET2294131Cyclic Degradation Prediction of Lithium-Ion Batteries using Data-Driven Machine LearningLerissah D. LimAndrei Felix J. TanJan Goran T. TomacruzMichael T. CastroMiguel Francisco M. RemolonaJoey D. OconAccurately estimating the capacity degradation of lithium-ion (Li-ion) batteries is vital in ensuring their safety and reliability in electric vehicles and portable electronics. Future capacity estimation using machine learning (ML) models allow battery lifetime predictions with minimal cycling data in the train set, well before capacity degradation occurs within the cell. The use of ML methods removes the need for prior knowledge of cell chemistry and the physical and chemical behaviors of batteries. In this paper, the data-driven ML models Gaussian process regression (GPR) and recurrent neural network – long short-term memory (RNN-LSTM) estimated the charge capacity of Li-ion batteries from the Oxford Battery Dataset, using only the battery's cycle index and capacity as input. With only 15 % of the battery’s lifetime as training data, the GPR model achieved a mean average percent error (MAPE) of 8.335 % and an R2 of 0.9755, while the LSTM model achieved a MAPE of 9.984 % and an R2 of 0.9898. These results indicate the goodness of fit and are comparable to results from similar models in the literature (MAPE = 9.1 to 15 %). The methodology may be applied to different features to help establish the relationship between health indicators and capacity fade and can be used in applications that require early capacity prediction such as in space technologies where lifetime and capacity are crucial in ensuring success and safety. This successful estimation highlights the promise and potential of accurately predicting Li-ion battery capacity degradation using a single-feature approach.https://www.cetjournal.it/index.php/cet/article/view/12688 |
| spellingShingle | Lerissah D. Lim Andrei Felix J. Tan Jan Goran T. Tomacruz Michael T. Castro Miguel Francisco M. Remolona Joey D. Ocon Cyclic Degradation Prediction of Lithium-Ion Batteries using Data-Driven Machine Learning Chemical Engineering Transactions |
| title | Cyclic Degradation Prediction of Lithium-Ion Batteries using Data-Driven Machine Learning |
| title_full | Cyclic Degradation Prediction of Lithium-Ion Batteries using Data-Driven Machine Learning |
| title_fullStr | Cyclic Degradation Prediction of Lithium-Ion Batteries using Data-Driven Machine Learning |
| title_full_unstemmed | Cyclic Degradation Prediction of Lithium-Ion Batteries using Data-Driven Machine Learning |
| title_short | Cyclic Degradation Prediction of Lithium-Ion Batteries using Data-Driven Machine Learning |
| title_sort | cyclic degradation prediction of lithium ion batteries using data driven machine learning |
| url | https://www.cetjournal.it/index.php/cet/article/view/12688 |
| work_keys_str_mv | AT lerissahdlim cyclicdegradationpredictionoflithiumionbatteriesusingdatadrivenmachinelearning AT andreifelixjtan cyclicdegradationpredictionoflithiumionbatteriesusingdatadrivenmachinelearning AT jangoranttomacruz cyclicdegradationpredictionoflithiumionbatteriesusingdatadrivenmachinelearning AT michaeltcastro cyclicdegradationpredictionoflithiumionbatteriesusingdatadrivenmachinelearning AT miguelfranciscomremolona cyclicdegradationpredictionoflithiumionbatteriesusingdatadrivenmachinelearning AT joeydocon cyclicdegradationpredictionoflithiumionbatteriesusingdatadrivenmachinelearning |