用于多表连接优化的深度强化学习嵌入表示  

Deep reinforcement learning embedding representation for multi-relation join optimization

作　　者：王江晴[1,2] 王雪言 孙翀[1,2] 帖军[1,3] 尹帆[1,3] WANG Jiang-qing;WANG Xue-yan;SUN Chong;TIE Jun;YIN Fan(College of Computer Science,South-Central Minzu University,Wuhan 430074,China;Hubei Provincial Engineering Research Center for Intelligent Management of Manufacturing Enterprises,South-Central Minzu University,Wuhan 430074,China;Hubei Provincial Engineering Research Center of Agricultural Blockchain and Intelligent Managements,South-Central Minzu University,Wuhan 430074,China)

机构地区：[1]中南民族大学计算机科学学院,湖北武汉430074 [2]中南民族大学湖北省制造企业智能管理工程技术研究中心,湖北武汉430074 [3]中南民族大学农业区块链与智能管理湖北省工程研究中心,湖北武汉430074

出　　处：《计算机工程与设计》2023年第2期576-581,共6页Computer Engineering and Design

基　　金：教育部科技发展中心高校产学研创新基金-“青苔数智融合”协同创新基金项目(2020QT08);国家民委中青年英才培养计划基金项目(MZR20007);湖北省科技重大专项基金项目(2020AEA011);武汉市科技计划应用基础前沿基金项目(2020020601012267)。

摘　　要：针对数据库查询优化中多表连接优化问题,任务是找到一个合适的连接顺序使查询执行计划最优,为此提出一种查询语句的嵌入表示方法SmartEncoder。通过优化查询语句中多表连接的嵌入表示信息,得到更丰富的关于连接的信息,将多表连接顺序选择优化建模为深度强化学习问题,根据动作的概率分布选择连接,从过去的经验中学习,生成更好的查询执行计划。在Join Order Benchmark数据集上的实验结果表明,SmartEncoder能够有效提高查询的效率。For the multi-relation join optimization problem in database query optimization, the task is to find a suitable join order to make the query execution plan optimal. Therefore, an embedding representation of query statements, SmartEncoder, was proposed, which obtained richer information about joins by optimizing the embedding representation information of multi-relation join in query statements, and multi-relation join order selection optimization was modeled as a deep reinforcement learning problem, where joins were selected based on the probability distribution of actions, and past experiences were learned to generate better query execution plans. Experimental results on the Join Order Benchmark dataset show that SmartEncoder can effectively improve the efficiency of queries.

关 键 词：查询优化 多表连接优化 连接顺序 执行计划 嵌入表示 深度强化学习 查询效率 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]