并行深度强化学习的柴油机动力系统VGT智能控制  

Intelligent control of diesel engine pressurization based on deep reinforcement learning in cloud computing framework

作　　者：赖晨光[1,2] 伍朝兵 李家曦 孙友长 胡博 LAI Chenguang;WU Chaobing;LI Jiaxi;SUN Youchang;HU Bo(Key Laboratory of Advanced Manufacturing Technology for Automobile Parts, Ministry of Education, Chongqing University of Technology, Chongqing 400054, China;Institute of Automotive Engineering, Chongqing University of Technology, Chongqing 400054, China)

机构地区：[1]重庆理工大学汽车零部件制造及检测技术教育部重点实验室,重庆400054 [2]重庆理工大学车辆工程学院,重庆400054

出　　处：《重庆理工大学学报（自然科学）》2022年第6期302-308,共7页Journal of Chongqing University of Technology：Natural Science

基　　金：国家自然科学基金项目(51905061)。

摘　　要：针对智能网联(ICV)在动力总成控制领域缺乏相关研究,传统动力总成控制既不智能也不网联的情况,采用最新的深度强化学习算法控制可变截面涡轮(VGT),促进传统内燃动力向智能网联发展。以某台可变几何截面涡轮柴油机为列,分别采用深度强化学习控制方法和PID控制方法进行仿真。结果表明:并行深度强化学习明显优于传统控制方法,最终收敛奖励值超过PID控制,4线程和8线程控制的绝对误差分别提升了37.87%和42.71%。As a new generation of automobile revolution technology,intelligent network connection(ICV)has developed rapidly in traffic and autopilot,but there is no related research in the field of powertrain control,and the traditional powertrain is neither intelligent nor connected.In this context,the latest depth reinforcement learning algorithm combined with a cloud computing framework to control the powertrain promotes the development of powertrain to intelligent networks.Taking a variable geometric section turbine diesel engine as a column,the traditional control method and the depth reinforcement learning control method are used to simulate,respectively.The results show that the depth reinforcement learning is superior to the traditional control method,and the absolute error is reduced.

关 键 词：深度强化学习 智能网联 并行计算 可变截面涡轮 

分 类 号：U461.6[机械工程—车辆工程]