基于EEMD技术在电力信息安全中的多步时间序列预测方法
2017-04-14于烨柴育峰康乐郭景维张波
于烨 柴育峰 康乐 郭景维 张波
摘 要: 针对用户访问轨迹的数据特征,提出一种基于EEMD技术的多步时间序列预测模型。该模型利用了集合经验模态分解EEMD结合极限学习机ELM模型,混合人工鱼群MAFA优化的方式,克服了算法中存在过拟合和多步时间序列预测的策略限制问题。通过该模型,实现了对访问轨迹时间序列多步预测,结合安全范围包络线,进而提前发现是否存在入侵行为。验证结果表明,优化后的EEMD?ELM模型比传统时间序列预测方法的迭代速率与精度得到了极大提高,泛化能力增强,说明了该方法的有效性、可行性。
关键词: 势态感知; 集合经验模态; 极限学习机; 混合人工鱼群; 多步时间序列预测
中图分类号: TN915.08?34; V249 文献标识码: A 文章编号: 1004?373X(2017)07?0159?04
Multi?step time series prediction method based on EEMD technology
in electric power information security
YU Ye, CHAI Yufeng, KANG Le, GUO Jingwei, ZHANG Bo
(Information and Communication Company, State Grid Ningxia Electric Power Company, Yinchuan 750000, China)
Abstract: According to the data characteristics of the user access path, a multi?step time series prediction model based on ensemble empirical mode decomposition (EEMD) technology is proposed. The model uses the EEMD combining with the extreme learning machine (ELM) model, and optimization method of the hybrid artificial fish swarm algorithm to overcome the constraint problems of the over?fitting and multi?step time series prediction strategy existing in the algorithm. The time series multi?step prediction of the access path was implemented with the model, and the intrusion behavior can be found in advance in combination with the envelope line of the safety range. The verification results show that the optimized EEMD?ELM model has higher iteration rate and accuracy than those of the traditional time series prediction methods, its generalization ability is enhanced, and the effectiveness and feasibility of this method was illustrated.
Keywords: situation awareness; ensemble empirical mode; extreme learning machine; hybrid artificial fish swarm; multi?step time series prediction
0 引 言
电力信息系统的安全性往往关系到企业的核心利益,不断发展与变化的网络信息技术和网络入侵攻击技术越来越表现出不确定性、复杂性、多样性等特点。
目前,国内外学者在时间序列预测的研究中,采用的都是单步时间序列预测ARIMA、直接策略、迭代策略、经验模态分解等[1],而目前还未能出现针对电力信息系统数据库的访问轨迹势态感知的多步时间序列预测方法。
本文基于集合经验模态分解EEMD技术引入极限学习机ELM模型,利用人工鱼群算法结合多模态函数优化算法建立了一个对访问轨迹的数据进行多步时间序列预测的模型。该算法在求解类似大规模访问轨迹数据的多步预测突破了传统算法策略的限制,具有更高的迭代效率和能力。……
