AUTOREGRESSIVE INTEGRATED MOVING AVERAGE-BASED PREDICTIVE MODEL FOR BASE STATION AVAILABILITY OF TELECOMMUNICATION NETWORKS IN MINNA

Abstract

There is a standard of 99.999% (five ‘nines’) availability for telecommunication hardware and software. This is to guarantee the high level of service required by the Mobile Network Operator (MNO) for service delivery. MNOs in Nigeria and most sub-Saharan Africa countries are, however, not being able to meet up with the expected base station availability mainly due to high restoration time after the outage. In this thesis, the historical Base Transceiver Station (BTS) Availability reports of a thousand data points each for four MNOs were used. The MNOs (MNO W, MNO X, MNO Y and MNO Z in Minna) data were acquired from 1st of January 2018 to 26th September 2020. The first 73% of the data was partitioned into the Training period and the remaining 27% was set for Validation. The data is in the form of Time Series (TS) and was modelled using Autoregressive Integrated Moving Average (ARIMA) prediction. Correlation plots of the data were done and the ARIMA (p,d,q) parameters were got with the aid of the Autocorrelation Function (ACF) and the Partial Autocorrelation Function (PACF). The ARIMA-Based models for the MNOs are ARIMA (0,1,3), ARIMA (1,0,1), ARIMA (2,0,4) and ARIMA (0,1,1) for MNO W, MNO X, MNO Y and MNO Z, respectively. The predictive models were used to predict BTS Availability for the MNOs from 27th September 2020 to 20th December 2020. The performance of the models was evaluated with data in the validation period for Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). The MAEs for the respective MNOs are: 1.3959, 0.6602, 1.5666 and 0.6177; while their MAPE are: 0.0150, 0.0068, 0.0176 and 0. 0063. The long short-term (LSTM) model was used for comparison with the ARIMA model for the same MNOs and their MAE and MAPE are 2.8397, 0.8894, 2.8223, and 1.1245; 0.0322, 0.0092, 0.0349 and 0.0118 for MNO W, MNO X, MNO Y and MNO Z respectively. From the results, it is observed that the LSTM models have higher MAE values than the ARIMA models by 51%, 26%, 44% and 45% for MNO W, MNO X, MNO Y and MNO Z respectively. Similarly, for MAPE, the LSTM models have 53%, 26%, 50% and 47% higher values than the ARIMA models for the respective MNOs. These indicate that the ARIMA models have performed better than the LSTM models in all the MNOs. The values of the MAE and MAPE for the predictive models are very low which implies that the predicted Availability data is close to the actual values and can be used for proper planning and decision-making. MNOs can proactively schedule Predictive Maintenance (PdM) with the PdM algorithm developed in this work. Using the 95% availability threshold of this algorithm, MNO W and MNO Y have no savings in maintenance count, while MNO X and MNO Z have savings of 33 and 32 respectively.