4.7 Article

LSTM-based energy management algorithm for a vehicle power-split hybrid powertrain

期刊

ENERGY
卷 284, 期 -, 页码 -

出版社

PERGAMON-ELSEVIER SCIENCE LTD
DOI: 10.1016/j.energy.2023.129267

关键词

Hybrid electric vehicles; Energy management strategy; Recurrent neural network; Long-short term memory; Dynamic programming

向作者/读者索取更多资源

In this study, a neural network with two substructures was developed and LSTM was used as the core part for power-split hybrid powertrain energy management problem. The results showed that LSTM has good generalization ability and real-time control capability.
Recurrent neural networks (RNNs) have been used for vehicle speed prediction, trajectory prediction and state diagnosis. As a variant of RNNs, long short-term memory (LSTM) has better state memory ability to deal with problems that have sequential characteristics. In this study, a neural network with two substructures was developed and LSTM was used as the core part for a power-split hybrid powertrain energy management problem, and the optimal control data obtained from the dynamic programming (DP) algorithm was used as the training set. Then on a vehicle model established on coasting-down test data, LSTM conducted real-time control. This paper analyzes the battery status, fuel consumption, carbon dioxide emissions and operating condition of the engine and motor. The results showed that LSTM has good generalization ability and real-time control capability. In the test driving cycles of CHTC-LT and C-WTVC, the theoretical optimal scheme given by the DP algorithm is only 12.15% and 17.96% lower than the LSTM scheme in terms of fuel consumption. In addition, the influence of network size, training epochs and training set on energy-saving effect is compared in detail, it was found that the relatively optimal model required 128 LSTM layer units and 500 training epochs.

作者

我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。

评论

主要评分

4.7
评分不足

次要评分

新颖性
-
重要性
-
科学严谨性
-
评价这篇论文

推荐

暂无数据
暂无数据