| Description |
1 online resource (xi, 123 pages) : illustrations |
| Series |
Synthesis Lectures on Advances in Automotive Technology |
|
Synthesis lectures on advances in automotive technology.
|
| Contents |
1. Introduction -- 1.1. Motivation -- 1.2. HEV powertrain -- 1.3. Literature review |
|
2. Background : deep reinforcement learning -- 2.1. Reinforcement learning in energy management -- 2.2. The state space and action space in reinforcement learning -- 2.3. Literature review on deep reinforcement learning |
|
3. Learning of EMSs in continuous state space-discrete action space -- 3.1. Energy consumption model of a series hybrid electric vehicle -- 3.2. Energy management based on deep Q-learning method -- 3.3. Improvements for stable learning in discrete action space -- 3.4. The learning and evaluation of EMSs in continuous state space |
|
4. Learning of EMSs in continuous state-continuous action space -- 4.1. Energy management based on DDPG method -- 4.2. Continuous energy management enabled by trip information -- 4.3. Comparison between continuous DRL-based and MPC-based EMSs -- 4.4. Summary |
|
5. Learning of EMSs in discrete-continuous hybrid action space -- 5.1. Energy consumption model of a power-split HEB -- 5.2. Searching of optimal EMSs in hybrid action space by DDPG -- 5.3. EMSs considering terrain information under hybrid action space -- 5.4. Comparative case analysis -- 5.5. Summary |
|
6. An online integration scheme for DRL-based EMSs -- 6.1. Reconstruction of parameterized EMSs in Matlab/Simulink -- 6.2. Evaluation by HIL test -- 6.3. Summary -- 7. Conclusions |
| Summary |
The urgent need for vehicle electrification and improvement in fuel efficiency has gained increasing attention worldwide. Regarding this concern, the solution of hybrid vehicle systems has proven its value from academic research and industry applications, where energy management plays a key role in taking full advantage of hybrid electric vehicles (HEVs). There are many well-established energy management approaches, ranging from rules-based strategies to optimization-based methods, that can provide diverse options to achieve higher fuel economy performance. However, the research scope for energy management is still expanding with the development of intelligent transportation systems and the improvement in onboard sensing and computing resources. Owing to the boom in machine learning, especially deep learning and deep reinforcement learning (DRL), research on learning-based energy management strategies (EMSs) is gradually gaining more momentum. They have shown great promise in not only being capable of dealing with big data, but also in generalizing previously learned rules to new scenarios without complex manually tunning. Focusing on learning-based energy management with DRL as the core, this book begins with an introduction to the background of DRL in HEV energy management. The strengths and limitations of typical DRL-based EMSs are identified according to the types of state space and action space in energy management. Accordingly, value-based, policy gradient-based, and hybrid action space-oriented energy management methods via DRL are discussed, respectively. Finally, a general online integration scheme for DRL-based EMS is described to bridge the gap between strategy learning in the simulator and strategy deployment on the vehicle controller |
| Subject |
Hybrid electric cars.
|
|
Hybrid electric cars -- Power supply
|
|
Deep learning (Machine learning)
|
|
Deep learning (Machine learning)
|
|
Hybrid electric cars
|
| Form |
Electronic book
|
| Author |
He, Hongwen, author
|
| ISBN |
9783031792069 |
|
3031792068 |
|
9781636930251 |
|
1636930255 |
|
