摘要
改进了异常点检测算法的复杂度和收敛速度慢的缺陷,采用加入动量项的改进BP神经网络算法,建立了算法的理论模型和学习过程。对随机抽样的145条记录测试试验中得到:算法大大缩短了训练时间,提高了训练速度,并且准确率也取得了令人比较满意的效果。算法应用实验表明BP具备一定的数据特征筛选的能力,并且能很好地实现异常点区分。上述结果表明改进的BP算法检测对数据集合分类具有可行性,设计成果对于社区安全技术的发展有着明显实际意义。
Improving anomaly detection algorithm complexity and slow convergence rate of defects,using a join the momentum of improved BP neural network algorithm,algorithm theory model and the learning process are built. Tests on a random samples of 145 record get： algorithm dramatically shorten training time,increased train speed and accuracy has yielded satisfactory results. Algorithm for feature selection experiments show that BP has a certain amount of data,and can be a very good realization of anomalies,distinguish between them. The results suggest that classification of improved BP algorithm to detect data collection is feasible, design security for data mining technology has obvious relevance for the development.
出处
《信息技术》
2016年第5期33-36,40,共5页
Information Technology
基金
唐山市科技局课题(12140201B-1)
关键词
异常点检测
复杂度
收敛速度
BP神经网络
anomaly detection
complexity
convergence speed
BP neural network
