Research on the impact of lithium battery ageing cycles on a data-driven lithium battery model
12月 6, 2024·,,,,,,·
0 分钟阅读时长
Haobin Cao
Guixiang Zhu
Huanhuan Chen
Zilong Su
Ruizhe Chen
Hongda An
Chen Wang
摘要
Although lithium-ion batteries offer significant potential in a wide variety
of applications, they also present safety risks that can harm the battery system
and lead to serious consequences. To ensure safer operation, it is crucial to develop
a mechanism for assessing battery health and estimating remaining service life,
enabling timely decisions on replacement or removal. For this purpose, this paper
develops a Long Short-Term Memory (LSTM) neural network model with a Multilayer
Perceptron (MLP) layer to predict the state of charge (SOC) of Li-ion batteries.
The model consists of three core modules: data processing module, neural network
module and analysis plotting module. In contrast to the traditional three-characteristic
input model of voltage, current, and battery temperature, this paper pioneers the
addition of a new characteristic: the number of charge/discharge cycles to enrich
the feature engineering of SOC prediction. In addition, our proposed model utilizes
the metric of Mean Square Error (MSE) as the loss function, which is very stable
for the scenario of SOC prediction. By processing the dataset and introducing a
new application scenario, the accuracy of the SOC prediction has substantially increased.
The average SOC forecast error is as low as 0.2% using the aforementioned four input
features for a double-layer LSTM and double-layer MLP model, regardless of the level
of battery aging.
类型
出版物
World Wide Web