检索结果-南通市图书馆

On large action space in EV charging scheduling optimization

Science China(Information Sciences) 2022年第2期65卷 45-60页

作者： Zhaoyu JIANG Qing-Shan JIA Xiaohong GUAN Center for Intelligent and Networked Systems Beijing National Research Center for Information Science and Technology Department of Automation Tsinghua University MOE KLINNS Lab Xi'an Jiaotong University

In order to reduce the air pollution by traditional fossil fuel and save the charging cost, much attention has been paid to the charging scheduling of electric vehicles(EVs) in the uncertain supplement of wind power and solar power. Owing to the randomness in the renewable power generation and the EV charging demand, simulation-based policy improvement(SBPI) has been well adopted for making charging decision. However, it is challenging to explore the large action space which grows exponentially with respect to the system scale. We consider this important problem in this paper and make the following contributions.First, we explore the structural property of the problem and develop an urgency index to rank the EVs.Second, we apply three methods to search in the action space. Third, we numerically demonstrate the performance of the urgency index and the search methods, and compare the SBPI methods with the CPLEXbased method. It shows that SBPI improves the base policies in all these cases.

关键词： EV simulation-based policy improvement large action space sampling

来源：

同方期刊数据库评论

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

A Batch reinforcement learning algorithm for by-product gas scheduling in steel industry

A Batch reinforcement learning algorithm for by-product gas ...

引用

第31届中国过程控制会议（CPCC2020）

作者： SU Lei HAN Zhongyang ZHAO Jun WANG Wei School of Control Science and Engineering Dalian University of Technology

Although Reinforcement Learning(RL) has been successfully applied in many fields, such as playing video games, chess and go, etc., its real practice in engineering is still limited, especially for by-product gas scheduling in steel plant which exhibits a large discrete action space. Driven by recent advances in batch RL, a novel algorithm Bayesian Networks-Neural Fitted Q-iteration(BN-NFQ) with action space decomposition is proposed in this study, in which the causal relationship in action space is qualitatively described as Bayesian Networks(BNs) and the scheduling amounts by adjustable units are quantitatively calculated by Maximum Possible Explanation(MPE). Furthermore, the BN structures considering different states are optimized by batch RL, so that the action space can be decomposed qualitatively and quantificationally. To verify the effectiveness of the proposed method, real data of the gas flow from a steel plant in China is employed. Comparing with standard NFQ and Q-learning in scheduling and learning performance, the result indicates that the proposed algorithm exhibits a significant advantage in by-product gas scheduling, which is beneficial for energy saving in steel industry.

关键词： large action space batch Reinforcement Learning Bayesian Networks by-product gas

来源： cnki会议评论

在线全文

cnki会议

学校读者我要写书评

暂无评论

欢迎您,

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

在线全文

在线全文

请选择保存的检索档案：

请选择收藏分类：

通借通还

欢迎您,

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

在线全文

在线全文

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：