INFORMATION TECHNOLOGY

Deep learning frameworks predict lithium battery life


Recently, the research group of Luan Weiling, professor of the School of Mechanical and Power Engineering and Key Laboratory of Advanced Battery Systems and Safety of East China University of Science and Technology, cooperated with Chen Haofeng, a national high-level talent and chair professor of East China University of Science and Technology, to publish a paper in the global transportation science and technology journal “Transportation Electrification”, which proposed for the first time a interpretable deep learning framework for lithium battery life prediction.

Schematic diagram of an interpretable deep learning framework. Photo courtesy of East China Institute of Technology

In the field of lithium battery life prediction, establishing a comprehensive battery aging model is a difficult task. As a result, data-driven approaches are receiving increasing attention. Deep learning has proven to be a powerful data-driven fitting method in battery applications. However, interpretability remains a challenge in the field, limiting the practical application of deep learning methods.

With the development of interpretable techniques, deep learning can be used not only as a black-box tool, but also as a relationship between external battery data and internal electrochemical changes. The research team proposes an interpretable deep learning framework that utilizes gradient-weighted class activation mapping to explain the connection between the input and output of the trained convolutional neural network model.

The research team demonstrated an interpretable deep learning framework through the lithium battery capacity decay inflection point recognition task. It is found that the deep learning model can keenly capture the features related to the battery aging mechanism, including key features that are not fully understood by humans, on the basis of effectively predicting the inflection point of battery capacity decline. In addition, by validating the method in different prediction tasks, such as considering multiple battery systems, actual operating conditions and data sets, the excellent portability of the framework is demonstrated. Without prior knowledge, this explainable deep learning framework can provide researchers with new insights into complex battery aging mechanisms. The proposal of this interpretable deep learning method provides a new idea for data-driven research in battery-related fields, and will actively promote the wide application of artificial intelligence technology in the design, development and safe use of advanced batteries. (Source: Zhang Shuanghu, China Science News)

Related paper information:https://doi.org/10.1016/j.etran.2023.100281



Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button