An investigation into an expert system for telecommunication network design
Many telephone companies, especially in Eastern-Europe and the 'third world', are developing new telephone networks. In such situations the network design engineer needs computer based tools that not only supplement his own knowledge but also help him to cope with situations where not all the information necessary for the design is available. Often traditional network design tools are somewhat removed from the practical world for which they were developed. They often ignore the significant uncertain and statistical nature of the input data. They use data taken from a fixed point in time to solve a time variable problem, and the cost formulae tend to be on an average per line or port rather than the specific case. Indeed, data is often not available or just plainly unreliable. The engineer has to rely on rules of thumb honed over many years of experience in designing networks and be able to cope with missing data. The complexity of telecommunication networks and the rarity of specialists in this area often makes the network design process very difficult for a company. It is therefore an important area for the application of expert systems. Designs resulting from the use of expert systems will have a measure of uncertainty in their solution and adequate account must be made of the risk involved in implementing its design recommendations. The thesis reviews the status of expert systems as used for telecommunication network design. It further shows that such an expert system needs to reduce a large network problem into its component parts, use different modules to solve them and then combine these results to create a total solution. It shows how the various sub-division problems are integrated to solve the general network design problem. This thesis further presents details of such an expert system and the databases necessary for network design: three new algorithms are invented for traffic analysis, node locations and network design and these produce results that have close correlation with designs taken from BT Consultancy archives. It was initially supposed that an efficient combination of existing techniques for dealing with uncertainty within expert systems would suffice for the basis of the new system. It soon became apparent, however, that to allow for the differing attributes of facts, rules and data and the varying degrees of importance or rank within each area, a new and radically different method would be needed. Having investigated the existing uncertainty problem it is believed that a new more rational method has been found. The work has involved the invention of the 'Uncertainty Window' technique and its testing on various aspects of network design, including demand forecast, network dimensioning, node and link system sizing, etc. using a selection of networks that have been designed by BT Consultancy staff. From the results of the analysis, modifications to the technique have been incorporated with the aim of optimising the heuristics and procedures, so that the structure gives an accurate solution as early as possible. The essence of the process is one of associating the uncertainty windows with their relevant rules, data and facts, which results in providing the network designer with an insight into the uncertainties that have helped produce the overall system design: it indicates which sources of uncertainty and which assumptions are were critical for further investigation to improve upon the confidence of the overall design. The windowing technique works by virtue of its ability to retain the composition of the uncertainty and its associated values, assumption, etc. and allows for better solutions to be attained.