Temporal data-driven failure prognostics using BiGRU for optical networks
Chunyu Zhang, Danshi Wang, Lingling Wang, Jianan Song, Songlin Liu, Jin Li, Luyao Guan, Zhuo Liu, Min Zhang
Abstract
With a focus on service interruptions occurring in optical networks, we propose a failure prognostics scheme based on a bi-directional gated recurrent unit (BiGRU) from the perspective of time-series processing, which leverages actual datasets from the network operator. BiGRU neural networks can capture the temporal features of multi-sourced data and incorporate contextual information. A principal component analysis is introduced to reduce the data dimensionality. Experimental results show that the average accuracy of the prognostics, F1 score, false positive rate, and false negative rate of our method are 99.61%, 99.63%, 0.29%, and 0.84%, respectively, which proves the feasibility of the proposed scheme for failure prognostics of equipment used in optical networks.