DNN based Multi-Faults Localization for 5G Coexisting Radio and Optical Wireless Networks

Abstract

Multi-faults localization problem belongs to the Nondeterministic Polynomial Complete (NP-C) problem in optical networks. Optical networks have good performance for future high speed and large capacity service transmission. The 5G coexisting radio and optical wireless network that takes optical networks as backhaul becomes an emerging research direction. In this network, the number of network devices and links has greatly increased, which lead to a significance increase in the possibility of multiple faults. Moreover, the complexity of the fault localization algorithm increases linearly with the expansion of the network scale. This makes multi-faults localization more difficult. Lately, machine learning has been widely concerned and researched. As an important kind of machine learning technology, the deep neural network (DNN) has a natural advantage in processing massive data. Therefore, we apply two classical DNNs (i.e., the BP neural network and the LSTM neural network) into multi-faults localization in coexisting radio and optical wireless networks. The performance of two DNNs under different number of faults is evaluated. When there are 3 faults, the localization time and the localization accuracy of the BP neural network and the LSTM neural network are 3.9782s and 3.5728s, 87.95% and 89.65%. Namely, the performance of the LSTM neural network is better than that of the BP neural network.