基于强化学习的混合动力挖掘机实时能量管理控制器设计（英文）

发布时间：2018-05-10 19:13

本文选题：能量管理 + 实时性　；参考：《Journal of Zhejiang University-Science A(Applied Physics & Engineering)》2017年11期

【摘要】：目的:混合动力挖掘机的能量管理策略直接影响着系统的燃油经济性。本文旨在通过研究混合动力挖掘机能量管理系统,得到最优能量管理策略,并开发实时能量管理控制器,降低系统的燃油消耗。创新点:1.通过强化学习算法,设计时间无关的实时能量管理控制器;2.通过极大值原理求得最优能量管理问题的解析解,并用来辅助实时能量管理控制器设计。方法:1.建立负载的马尔科夫模型,运用强化学习算法,得到实时能量管理控制器;2.运用极大值原理,求得最优能量管理问题的解析解,并将其作为初始能量管理策略;3.通过仿真模拟和实验研究,验证所设计的实时能量控制器的性能。结论:1.基于强化学习的能量管理控制器是一个可以在线应用的与时间无关的实时能量管理控制器;2.基于强化学习的能量管理控制器优于广泛使用的恒温控制器和等效消耗最小化策略控制器;3.基于强化学习的能量管理控制器由于其闭环特性可适用于不同类型的作业工况。
[Abstract]:Objective: the energy management strategy of hybrid excavator directly affects the fuel economy of the system. The purpose of this paper is to obtain the optimal energy management strategy by studying the hybrid excavator energy management system, and to develop a real-time energy management controller to reduce the fuel consumption of the system. The innovation point is 1: 1. Through reinforcement learning algorithm, a time-independent real-time energy management controller is designed. The analytical solution of the optimal energy management problem is obtained by the maximum principle and is used to aid the design of the real time energy management controller. Method 1: 1. The Markov model of load is established and the real time energy management controller is obtained by using reinforcement learning algorithm. The analytical solution of the optimal energy management problem is obtained by using the maximum principle, which is regarded as the initial energy management strategy. The performance of the designed real-time energy controller is verified by simulation and experimental research. Conclusion 1. The energy management controller based on reinforcement learning is a time independent real-time energy management controller, which can be applied online. The energy management controller based on reinforcement learning is superior to the constant temperature controller and equivalent consumption minimization strategy controller. The energy management controller based on reinforcement learning can be used in different operating conditions because of its closed loop characteristics.
【作者单位】： State
【基金】：Project supported by the National Natural Science Foundation of China(No.51475414) the Science Fund for Creative Research Groups of National Natural Science Foundation of China(No.51521064)
【分类号】：TU621

【相似文献】