학술논문

Lithium-Ion Battery Capacity Prediction Based on Partial Voltage Curve and Autoencoder
Document Type
Conference
Source
2024 IEEE 4th International Conference on Sustainable Energy and Future Electric Transportation (SEFET) Sustainable Energy and Future Electric Transportation (SEFET), 2024 IEEE 4th International Conference on. :1-6 Jul, 2024
Subject
Aerospace
Components, Circuits, Devices and Systems
Engineering Profession
Power, Energy and Industry Applications
Robotics and Control Systems
Transportation
Lithium-ion batteries
Support vector machines
Transportation
Voltage
Production
Predictive models
Feature extraction
Prediction algorithms
Vectors
Convolutional neural networks
autoencoder
convolutional neural network
long-short-term memory
dynamic feature
capacity prediction
support vector regression
Language
Abstract
In the manufacturing process of lithium-ion batteries, capacity grading is a crucial step. Precisely anticipating the capacity of lithium-ion batteries exhibits significant importance in saving production time, reducing production costs, and conserving energy. Therefore, this paper proposes a lithium-ion battery capacity prediction method based on partial voltage curve and autoencoder. Using the partial voltage curve of the grading process, multiple dynamic features (DF) are extracted based on the dynamic changes of the lithium-ion battery voltage. A CLAE feature extraction model based on autoencoder (AE) is proposed, which is composed of convolutional neural network (CNN), long-short-term memory (LSTM) and AE. It can extract spatial features and temporal features of voltage curves. Finally, a support vector regression (SVR) capacity prediction model optimized by differential evolution algorithm (DE) is established. Compared with other capacity prediction methods, the DF-CLAE-DE-SVR prediction model outlined in this study exhibits superior prediction performance for battery capacity, especially for low-capacity batteries.