ecurrent Neural Network-based Base Transceiver Station Power System Failure Prediction
No Thumbnail Available
Date
2019-12
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Global network infrastructures are increasing with the development of new technologies and
growth in Internet traffic. As network infrastructures increases, maintaining and monitoring them
will become very challenging since thousands of alarms are generated every day. Clearing those
alarms by corrective maintenance activities require considerable effort and resources (car, labor,
and budget).
In mobile networks, a Base Transceiver Station (BTS) is one key infrastructure element
performing the task of connecting customer equipment with the cellular network. BTS services
may be interrupted due to transmission, optical fiber cut, power system failure, natural disaster or
many more. In the case of Ethio Telecom (ET), the sole telecom service provider in Ethiopia,
power system failure takes the biggest share for interruption of BTS services. Minimizing power
system failure will reduce downtime of the BTS thereby, guarantee customer satisfaction and
maximize revenue. Recently, machine learning algorithms are used to predict failure in various
areas like power distribution, hydropower generation plants, solar power generation plants, high
voltage transmission grid and many more.
This thesis investigates predicting BTSs power system failure using a recurrent neural network
(RNN) types namely, long short term memory (LSTM) and gated recurrent unit (GRU) with linear
and sigmoid activation function applied for the output. In parallel, the prediction performance of
LSTM and GRU has been compared. Data collected from five BTS sites for twenty weeks of
observations are used to train and test the model. The data are prepared with two different data
arrangements, which are a single site and multiple sites. The relevance of using different data size
is, to check the impact of increasing data size with different arrangements on the prediction results.
Mean squared error (MSE) and number of epoch are used to evaluate the performance of the
models with different configurations. Based on the results found, GRU using sigmoid activation
function with feature reduction achieves better performance than LSTM. In addition, both LSTM
and GRU can be used for predicting BTS power system failure.
Description
Keywords
Base Transceiver Station, Gated Recurrent Unit, Long Short Term Memory, Recurrent Neural Network