基于深度降噪自编码网络的监测数据修复方法

针对大规模监测系统中经常出现的监测点失效、数据异常等问题,提出基于深度降噪自编码网络的监测数据修复方法。首先,通过堆叠降噪自编码构造深度降噪自编码网络来提取监测点之间隐含的深层关联关系,进而,基于这种深层关联关系训练一种支持向量回归模型以预测待修复的监测数据。在某机场噪声实测数据上的实验表明,通过深度降噪自编码网络学到的深层关联关系能够有效地重构噪声监测数据;相比传统数据修复方法,所提出的数据修复方法具有更好的鲁棒性,数据的修复具有更高的精度。

Monitoring data repairing method based on deep denoising auto-encoder network

In view of the frequent failure of monitoring points and data anomalies, a monitoring data repairing method based on deep denoising auto encoder (DDAE) network is proposed. Firstly, the denoising auto-encoder (DAE) is used as the basic structural unit to construct the DDAE network. Then, the deep correlation of the data is extracted from the DDAE network, and the support vector regression (SVR) model is constructed to predict the monitoring data to be repaired. Experiments conducted on the airport noise data set demonstrate that the deep correlation can reconstruct the noise monitoring data very well. Compared with the traditional data repairing methods, the proposed data repairing method has better robustness and higher precision.