Explainable Strategic Optimisation of Grand Scale Problems

Explainable Strategic Optimisation of grand scale problems aims to identify solutions that provide long term planning advantages to problems that cannot undergo traditional optimisation techniques due to their level of complexity. Usually, optimisation tasks focus on improving a limited number of objectives in the pursuit of obviously immediate target. However, this methodology, when applied to grand scale problems is found to be insufficient; a major reason for this is the inherent complexities typical of problems such as utility optimisation and massive logistical operations. One approach to these problems is Generational Expansion Planning that typically addresses long-term planning of country/county-wide utility problems. This thesis draws influence from the Generational Expansion Planning field; a significant field in relation to this work as it typically focuses on large scale optimisation problems. Problems such as the improvement and maintenance of national utility operations. However, this thesis takes a novel approach that places empathises on an abstract strategic planning method that concerns itself with the extraction of design insights that can guide an experts understanding of an unrelentingly complex problem. The proposed system was developed with data from British Telecom (BT) and was developed within their organisation in which its deployment is being planned. The techniques behind the proposed systems presented in this thesis are shown to improve the popular many-objective Non-Dominated Sorting Genetic Algorithm II in a series of experiments in which the improved Type-2 dominance method outperformed the traditional dominance method by 59%. Several component parts are brought together within this thesis so that the unique optimisation of varied regions that exist inside the United Kingdom’s Access Network can be explored. The proposed system places great import on the interpretability of the system and the solutions that it produces. As such, an Explainable Artificial Intelligent (XAI) system has been implemented in the hope that with greater interpretability, AI systems will be able to provide solutions with greater context, nuance, and confidence, particularly when the decision of an AI model has a direct impact on a person or business. This thesis will explore the related material and will explore the proposed framework; which brings together a multitude of technologies, such as, novel fuzzy many-objective optimisation, fuzzy explainable artificial intelligence, and strategic analysis. These technologies have been approached and combined in order to develop a novel system capable of dealing with complex grand scale problems, which traditionally are tackled as piecemeal optimisation problems. The proposed systems were shown to improve the optimisation of focused scenarios; in these experiments the proposed system was able to provide solutions for the optimisation of telecommunication networks that outperformed the current methodology for the planning/upgrading of the access network. The proposed systems were tested on rural, mixed, and urban regions of a simulated United Kingdom; it was observed that when the proposed systems were used the network solutions produced were 51.99% cheaper for rural regions, in which a combination of technologies were used as opposed to only FTTP. It was also observed that solutions produced by the proposed system in mixed regions were 54.16% cheaper while still providing the customer broadband requirements. These results identify how an expansive system such as the novel system proposed in this thesis is able to provide sound business solutions to complex real-world problems that consists of an ever growing number of variables, constraints, and objectives. Additionally, the proposed systems are capable of producing greater understanding of design principles/choices in network solutions, which in turn provides BT and users with a greater level of trust in the solutions and the system. This is a major obstacle that must be overcome when the problem domain that is being considered is incredible vast, uncertain, and extremely vital to the success of a company. The results of this thesis identify how the proposed systems can be developed and implemented to provide an insight into the planning and execution of an access network not required for decades to come. This is a significant change from the current reactive approach to a proactive approach that provides insight into the ever changing variables and needs of the network. The proposed systems are able to instil the confidence that allow a more thoughtful approach to be taken that is beneficial to both company and customer.