摘要
为检测数据中的异常信息,提出基于高斯过程模型的异常检测算法。高斯过程可以根据训练样本从先验分布转到后验分布,对核函数的超参数进行推理,预测输出具有清晰的概率解释。对基于高斯过程模型的异常检测算法进行定义和描述,用Server Computers(电脑服务器)数据进行仿真实验,结合高斯过程先验和回归理论,在实验中选取RBF作为核函数,利用目标类数据的特性构造特征向量集,在TE工业过程时序数据集上验证了该算法的适用性和有效性。
To detect data anomalies,anomaly detection algorithm based on Gaussian process model was proposed.Gaussian process can transform the prior distribution into the posterior distribution,inference the hyper-parameters of kernel function and predict output with a clear explanation of probability.Anomaly detection algorithm based on Gaussian process model was described,and the simulation experiment was carried out on server computer feedback data.The priori and Gaussian process regression theory was combined,RBF was selected as the kernel function,and a set of feature vectors was constructed.The applicability and effectiveness of the algorithm are verified by applying the algorithm to TE industrial process data sets.
出处
《计算机工程与设计》
北大核心
2016年第4期914-920,953,共8页
Computer Engineering and Design
基金
中央高校基本科研业务费专项基金项目(2013QNB14)
国家自然科学基金项目(61402482)
关键词
高斯过程
协方差函数
异常检测
先验
回归
Gaussian process
covariance function
anomaly detection
priori
regression
