基于多预测树组合算法的电力信息系统数据库缓存模型  

Electric Information System Database Cach Model Based on Predictive Forest Combination Algorithm

作　　者：曲朝阳[1] 刁赢龙 薄小永 朱莉[1] 任有学 颜佳[3] 

机构地区：[1]东北电力大学信息工程学院,吉林吉林132012 [2]吉林供电公司,吉林吉林132012 [3]吉林省电力有限公司信息通信公司,长春130062

出　　处：《电测与仪表》2014年第6期70-75,共6页Electrical Measurement & Instrumentation

基　　金：国家自然科学基金资助项目(51277023);国家自然科学基金资助项目(51077013);吉林省科技发展计划重点支撑项目(20120338)

摘　　要：为解决电力大数据时代背景下信息系统数据库体积大、并发量高导致的运行效率瓶颈,分析了当前几种缓存模型的基本架构和原理,综合电力内网中各业务系统的主要特点,构建了一个电力信息系统数据库缓存模型,在模型中设计了一种多预测树组合算法(预测树森林)。通过在应用层使用boosting方式对队列进行抽样得到具有时序性和业务逻辑特点的SQL语句构建预测树森林,智能预测可能发生的数据库操作并将结果集提前缓存到内存中,模型还可以根据系统的运行情况动态调整预测树森林的投票阈值,从而达到提高缓存效率和命中率的目的。并基于Oracle数据库对模型的可行性进行了讨论,在云平台下运行结果表明模型不仅算法收敛速度快而且能够实现同等内存容量下缓存效率的提高。To solve the efficiency bottleneck of electricity information system database with bulky size and high concurrency under the context of big data era, a power system database cache model is built and a multi-tree prediction combination algorithm （predictive forest） is designed, which is based on the analysis of basic structure and principles of current cache models and main characteristics of electricity network business system. With boosting method to sample the queue in the application layer and generate SQL statements with timing and business logic, the predictive forest is built, which allows intelligent forecasting of database operations and caching the result set beforehand. Moreover, the predictive forest voting threshold can be adjusted dynamically upon system operating status by cache mode to improve cache efficiency and hit rate. The feasibility of the model is discussed based on Oracle database, and the related experiment is conducted on a cloud computing platform. The results showed that the algorithm converges faster and cache more efficiently under the same amount of memory.

关 键 词：电力大数据 数据库 缓存模型 预测树森林 

分 类 号：TM93[电气工程—电力电子与电力传动]