基于随机森林算法的负载预警研究及并行化-《计算机技术与发展》

文章信息/Info

Title:: Research on Load Early Warning Based on Random Forest Algorithm and Parallel Method

Author(s):: WANG Cheng; TANG Zhen-kun; School of Telecommunications & Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China

Keywords:: power distribution monitoring; load warning; random forest algorithm; Spark; parallelization

摘要:: 近年来国内通信行业发展十分迅速,运营商通信网络的规模也随之壮大。在运营商通信网络中,数据中心机房是不可或缺的重要枢纽,承担着巨大的通信压力,数据中心机房的配电系统故障率和安全事故的风险也在不断提高,同时也导致机房运维难度和运维成本与日俱增。在现代电力系统中电力大数据的格局下,对高维海量数据进行深度挖掘,进而预测可能存在的告警,从而做到防患于未然,是一个值得研究的问题。针对电力大数据环境下高精度和实时性的负载预测展开了研究,提出了基于随机森林算法的负载预警,并基于 Spark 平台实现其并行化。结合某区域实际电力数据设计试验,进行模型训练和回归预测,通过试验证明,对同等的数据集,并行随机森林算法预测精度高于单机负载预测,为负载预测提供了一种新思路。

Abstract:: With the rapid development of domestic communication industry,the scale of operator communication network is also growing.In the communication network of operator, the data center room is an indispensable and important hub, bearing huge communicationpressure. The failure rate of the power distribution system and the risk of security accidents in the data center room are increasing,whichalso leads to the increasing difficulty and cost of operation and maintenance in the room. Under the pattern of big power data in modernpower systems,it is a problem worth studying to deeply mine high - dimensional massive data to predict the possible alarms,so as toprevent them from happening. Aiming at the high-precision and real-time load prediction in the power big data environment,we proposea load warning based on random forest algorithm and implement its parallelization based on Spark platform. Combining the actual powerdata in a certain area to design experiments,model training and regression prediction are carried out. It is proved that for the same dataset,the prediction accuracy of the parallel random forest algorithm is higher than that of the single machine load prediction, whichprovides a new idea for load prediction.