Aplicação de redes neurais recorrentes e self-organizing maps em dados reais de operador de telecomunicações para predição de tráfego de interface de bng e análise de relatório de falha de rede GPON

Abstract

This work consists of the application of two Artificial Intelligence algorithms in two real situations of a telecommunications operator. The first application is the concept of recurrent neural network or recurrent neural network (RNN). A special type of neural network, which has great applicability in scenarios with temporal data and which was applied for traffic forecast in an interface of a broadband remote access server (also called broadband network gateway). The second situation of applied artificial intelligence was the evaluation of the use of the self-organizing maps (SOM) neural network model. This was used to classify and detect inconsistent data from a field failure report from a GPON broadband access network. The use of SOM for anomaly detection and classification was able to reduce the dimensionality of the data. It was possible to extract the list of these distant data (or outliers) and make an observation. The study seeks to demonstrate that neural network applications can be used as a tool for automating network analysis in telecommunications, generating benefits such as cost reduction and greater agility, which provides better quality to users, more optimized networks and evolution of telecommunications services.