基于机器学习的中国移动业务监控方法改进研究

发布时间：2018-02-25 18:13

本文关键词： 聚类算法神经网络算法 Holt-Winters 文本分类数据分类　出处：《河北农业大学》2015年硕士论文　论文类型：学位论文

【摘要】：随着移动互联网的出现,手机游戏、手机视频、手机超市、无线音乐、手机阅读、手机动漫等业务也随之呈现。同时,中国移动的数据业务迅猛发展,使得日常业务管理的难度、复杂性越来越大。因错单、恶意消费等问题给公司造成损失也使得客户的满意度降低,因此需要一些强有力的手段来监控数据业务以便及时发现问题。目前实际运行的业务监控系统主要针对用户恶意消费、错单量大、SP平台利用系统漏洞非法获利等现象造成的数据量急剧增大现象进行监控分析,但对某些业务存在误报及漏报的情况。例如手机游戏、手机超市、手机阅读等包月业务大多在月初进行订购,会导致订购量大幅波动,属于正常现象,由于现有系统中算法的缺陷,未区分此类现象,产生大量错误的告警。本文针对现有业务监控系统中存在的误报漏报问题进行研究,引入新的算法解决误报漏报的问题,并通过机器学习算法训练智能告警过滤器,通过告警回复信息指导告警的过滤,以减轻工作人员的工作量。本文主要工作如下:(1)针对数据业务量波动较大的业务引入DBSCAN聚类算法,解决少量突发数据对算法结果的影响并且减少对波动很大但属于正常现象的数据的错误告警。同时,本文对该算法进行改进,提高了该算法的效率,使得该算法可以应用到时间粒度较精确的大数据业务。(2)针对周期性数据业务,本文提出应用神经网络算法与Holt-Winters组合模型进行监控的方法解决周期性业务异常数据漏报的问题。对于特殊业务进行特殊监控,提高了业务监控系统产生告警的准确率以及查全率。(3)业务监控系统自动产生的告警信息仍需交由工作人员进行处理,为减少人工工作量同时提高业务监控系统产生告警的准确度,本文提出利用告警回复信息,通过数据分类技术训练告警过滤器,指导告警的过滤。实验表明,利用告警过滤器可分离出无效的告警信息,减少人工工作量。本文对原有业务监控系统进行的算法改进可较好解决误报漏报问题,并通过加入告警过滤器分离无效告警,实现减少人工工作量的目的。下一步将对告警过滤系统进行深入研究,通过实现系统的自我学习、自我更新提高告警过滤效果。
[Abstract]:With the emergence of the mobile Internet, mobile phone games, mobile video, mobile supermarket, wireless music, mobile phone reading, mobile animation and other services are also emerging. At the same time, China Mobile's data business is developing rapidly. It makes the daily business management more and more difficult and complex. Because of the wrong order, malicious consumption and other problems to the company, it also makes the customer satisfaction decrease. Therefore, some powerful means are needed to monitor data services in order to detect problems in a timely manner. At present, the operational business monitoring system is mainly aimed at malicious consumption by users. The large amount of errors in SP platform makes use of system vulnerabilities and other phenomena such as illegal profit to monitor and analyze the phenomenon of a sharp increase in data, but some businesses exist false alarm and underreporting situation, such as mobile phone games, mobile supermarket, Most of the monthly services, such as mobile phone reading, order at the beginning of the month, which results in large fluctuations in the order volume, which is a normal phenomenon. Due to the defects of the existing system algorithms, there is no distinction between such phenomena. In this paper, a new algorithm is introduced to solve the problem of false false alarm, and the intelligent alarm filter is trained by machine learning algorithm. In order to reduce the workload of staff, the main work of this paper is as follows: 1) to introduce DBSCAN clustering algorithm for services with volatile data traffic. To solve the influence of a small amount of burst data on the result of the algorithm and to reduce the error alarm of the highly volatile but normal data. At the same time, this paper improves the algorithm to improve the efficiency of the algorithm. This algorithm can be applied to big data service with more accurate time granularity. In this paper, the neural network algorithm and Holt-Winters combined model are used to solve the problem of missing abnormal data of periodic services. It improves the accuracy of alarm generated by the service monitoring system and the recall rate. 3) the alarm information generated automatically by the business monitoring system still needs to be handled by the staff. In order to reduce the manual workload and improve the accuracy of alarm generated by the service monitoring system, this paper proposes to use the alarm response information and train the alarm filter through the data classification technology to guide the alarm filtering. The invalid alarm information can be separated by using the alarm filter, and the manual workload can be reduced. The algorithm improvement of the original service monitoring system can better solve the problem of false false alarm and separate the invalid alarm by adding the alarm filter. In the next step, the alarm filtering system will be deeply studied, and the alarm filtering effect will be improved through self-learning and self-updating.
【学位授予单位】：河北农业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP181;TP277

【参考文献】