Abstract/Summary

The telecommunications network is a crucial part of the national infrastructure of the United Kingdom, with its reliable service becoming an increasingly vital feature of commercial activities and daily life. As with other national infrastructure, weather poses a risk to its operation by caus­ing disruptions in service through faults in the physical network, though the precise relationships between weather and faults are poorly understood. This thesis explores the effect that weather has on the network and demonstrates that by using modem weather forecasting techniques, the nega­tive effects of weather on the network can be minimised. The organisation used for this study is Openreach, a subsidiary of BT and the operator of 87% per cent of UK broadband infrastructure. Using the ERA-Interim reanalysis product, statistical relationships between weather and faults in the telecommunications network are identified in the construction of an impact model, with pre­cipitation identified as the primary historical weather driver for faults. The effects of weather on the network are quantified for the first time using long-term weather data (36 years) and the sta­tistical impact model to produce a consistent "synthetic" fault dataset, demonstrating that weather has a wide range of effects on the network. A multiple linear regression model explains 62% of variability in the fault data, highlighting the importance of weather. Until now, only short-term records (less than 5 years), have been used to quantify weather variability, leading to an under­estimation of the variability. Specifically, a clear seasonality is observed in the fault variability, with winter being the worst performing season, leading to a focus on forecasting this season in this study. Forecasting of faults is performed for weeks two to four in the business planning cycle, aligning with decision timescales in network operations. Large-scale atmospheric patterns are used in the analysis, namely; the North Atlantic Oscillation, as it is shown to have forecast skill at the required time frame. Skill was demonstrated in both deterministic and probabilistic methods of forecasting faults using the North Atlantic Oscillation, with the probabilistic methods demonstrating smaller errors than the deterministic. To assess the impact that using forecast information would have on the telecommunications network, the decision-making processes concerning fault repairs at Openreach are modelled. Using forecasting information, especially probabilistic information, leads to improvements in the key performance indicators used to assess network performance by the regulatory body Ofcom. Importantly, the performance increase is dependent on the nature of the system and the level of flexibility within it. The reactivity or ability of the system to rapidly respond to forecasts changes the usefulness of forecasts to the system, with more value being placed on the forecast in less reactive systems.