基于变分自编码器的异常小区检测

An Abnormal Cell Detection Method Based on VAE

为了解决移动无线网高维度性能指标异常样本的标注问题,提出一种基于变分自编码器VAE的异常小区检测方法。该方法利用异常小区性能KPI数据在通过变分自编码器编码与解码过程中所产生的较大波动来实现异常检测。实验结果表明,该方法在样本不平衡的情况下,通过合理设置重构误差阈值,能准确地检测出异常小区,实现异常小区的机器判读。

In order to solve the problem of labeling abnormal samples of high-dimensional performance indicators in mobile wireless networks, this paper proposes an abnormal cell detection method based on variational auto-encoder(VAE). This method realizes the abnormal detection via the large fluctuation caused by the output from the VAE when the performance KPI data in abnormal cell pass through the encoding and decoding process. The experimental results show that the method can detect abnormal cells accurately and realize machine interpretation of abnormal cells by setting the reasonable reconstruction error threshold under the condition of unbalanced samples.