Remaining useful life indirect prediction of lithium-ion batteries using CNN-BiGRU fusion model and TPE optimization
Xiaoyu Zheng,
Dewang Chen,
Yusheng Wang ,
Liping Zhuang
Affiliations
Xiaoyu Zheng
1. School of Transportation, Fujian University of Technology, 69-1 Xuefu South Road, Fuzhou, 350118, China 2. Intelligent Transportation System Research Center, Fujian University of Technology, 69‑1 Xuefu South Road, Fuzhou 350118, China
Dewang Chen
1. School of Transportation, Fujian University of Technology, 69-1 Xuefu South Road, Fuzhou, 350118, China 2. Intelligent Transportation System Research Center, Fujian University of Technology, 69‑1 Xuefu South Road, Fuzhou 350118, China 3. College of Mathematics and Computer Science, Fuzhou University, 2 Xueyuan Road, Fuzhou, 350108, China
Yusheng Wang
3. College of Mathematics and Computer Science, Fuzhou University, 2 Xueyuan Road, Fuzhou, 350108, China
Liping Zhuang
4. School of Electronics, Electrical Engineering and Physics, Fujian University of Technology, 69 Xuefu South Road, Fuzhou, 350118, China
The performance of lithium-ion batteries declines rapidly over time, inducing anxiety in their usage. Ascertaining the capacity of these batteries is difficult to measure directly during online remaining useful life (RUL) prediction, and a single deep learning model falls short of accuracy and applicability in RUL predictive analysis. Hence, this study proposes a lithium-ion battery RUL indirect prediction model, fusing convolutional neural networks and bidirectional gated recurrent units (CNN-BiGRU). The analysis of characteristic parameters of battery life status reveals the selection of pressure discharge time, average discharge voltage and average temperature as health factors of lithium-ion batteries. Following this, a CNN-BiGRU model for lithium-ion battery RUL indirect prediction is established, and the Tree-structured Parzen Estimator (TPE) adaptive hyperparameter optimization method is used for CNN-BiGRU model hyperparameter optimization. Overall, comparison experiments on single-model and other fusion models demonstrate our proposed model's superiority in the prediction of RUL in terms of stability and accuracy.
