基于SBWS_GPR预测模型的不确定性多数据流异常检测方法

针对实际系统中采集的数据流的不确定性，给异常点检测与修正带来了现实挑战，论文根据滑动基本窗口采样算法（Sliding Basic Windows Sampling，SBWB），与高斯过程回归（Gaussian Process Regression, GPR）模型的特性，提出了基于SBWS_GPR预测模型的不确定性多数据流的异常检测方法。在基于时间序列采集的历史数据集中，引入索引号，对历史数据集进行聚类，分析数据集与索引号的映射关系，将实时获得的输入数据流通过滑动窗口匹配，实现对单数据流的异常点检测与修正。再利用输入输出数据间的相关性，基于GPR建立预测模型，比较实时观察的输出数据流与预测模型的输出数据流，最终从输入输出2种不同通道实现多数据流的异常检测与修正。

Outlier Detection of Uncertainty Multiple Data Stream Based on SBWS_GPR Prediction Model

The uncertainty of collecting data stream in the actual system, brings a serious challenge for outlier detection and correction. Based on the characteristic of Sliding Basic Windows Sampling (SBWS) and Gaussian Process Regression (GPR), outlier detection method of uncertainty multiple data stream based on SBWS_GPR prediction model is proposed. By collecting historical data set based on time series and introducing index number, cluster and analysis historical data set and get the mapping relation between the data set and index number. The real-time input data stream obtained is to realize outlier detection and correction by the sliding window pattern. And then based on the correlation between the input and output data and the GPR, set up prediction model and compare the real-time output data stream data with the prediction output data stream, to realize outlier detection and correction from two different input and output channels.