论文部分内容阅读
通过检测代表性能降质的异常点来实现故障的提前发现和快速恢复是提高通信网的可靠性的重要手段.采用基于统计假设检验的网络异常点检测方法,提出一种综合运用季节累积自回归滑动平均模型时间序列预测和置信区间计算来动态获取性能指标阈值的方法.利用累积自回归滑动平均模型在训练集上的拟合残差白噪声符合正态分布的假设,给出了一种通过构造满足t分布的随机变量来计算预测值在任意置信度1-α下置信区间的新算法.理论分析和实验结果表明,该阈值动态确定方法有效.
It is an important means to improve the reliability of communication network by detecting the abnormal points of degraded performance.This paper proposes a new method based on statistical anomaly detection to detect abnormal points in network, A method of dynamically obtaining the threshold of performance index by using moving average time series prediction and confidence interval calculation of sliding average model.Using the assumption that the fitting residual white noise of the cumulative autoregressive moving average model in the training set meets the normal distribution, A new random variable with t-distribution is constructed to compute a new confidence interval for predictive value under any confidence 1-α. Theoretical analysis and experimental results show that this method is effective.