Conference Agenda

The deployment of Fibre to the Home (FttH) networks represents a critical advancement in the telecommunications sector, enabling the provision of high-speed internet access to fixed-line customers. Although the implementation of FttH has been successful in urban areas, the expansion into less dense regions remains a financial challenge.

In alignment with the "France Très Haut Débit" initiative, our study introduces a novel approach utilizing HDBSCAN clustering to optimize FttH deployment strategies in zones with uneven population densities. This methodology leverages geospatial data for the precise design and cost estimation of network infrastructures, facilitating effective and efficient resource allocation.

Additionally, this work examines the dynamics of public-private partnerships (PPPs) under incomplete contract theory. We analyze regulatory adjustments and strategic participation needed to sustain private investment in public services without complete financial guarantees. Through this dual approach, our research contributes to the broader discourse on telecommunications infrastructure development, offering scalable solutions and policy recommendations for the deployment of FttH networks in diverse demographic settings.

Ranked lists are often found in applications, such as comparing the results of a query sent to two search engines. In our use case, participants ranked a list of ideas by assigning a value (the “benefit”) to each idea – the higher the benefit, the better the idea. Now, questions arise such as the similarity of rankings by different users or different groups of users, which naturally leads to the need for a measure of ranking similarity. We start with a brief discussion of Kendall's tau, to show its limitations with respect to situations where (i) differences in the top ranks are considered more important than differences in the tail (the top-weighted aspect), (ii) it is not feasible or useful to compare the full ranking (the completeness aspect), (iii) two rankings may have different entries for reasons related to the evaluation process (the nonconjointness aspect), and (iv) the correlations of the benefits should be utilized to account for the benefits’ statistical uncertainty (the correlations aspect). We then present a new algorithm that combines Carterette’s statistical approach with Webber et al.’s rank-biased overlap (RBO) method to satisfy all requirements (i) – (iv). Finally, we present results from simulation studies that demonstrate the practicability of the new algorithm.

Transportation is one of the largest contributors to carbon emissions. In railway transportation, optimizing train schedules, circulation plans, and locomotive assignments to minimize energy consumption offers great potential for energy savings. In the railway literature, the most preferred approach to modeling energy or fuel consumption is deterministic modeling, particularly using the Davis model, which calculates the resistance force required to keep the train moving at a constant speed. However, these models were developed many years ago through experimental work on old types of diesel locomotives and require parameter adjustment to be used today. In recent years, electric locomotives are generally used in rail transportation, and the energy consumption of locomotives can be recorded at certain time intervals by using some sensors. In this study, to predict the energy consumption of electrical locomotives using data-driven methods, first train time schedules, locomotive energy consumption data (by the Austrian Federal Railways, ¨ OBB), and some infrastructure information are merged. Then, the most significant parameters affecting energy consumption are found by performing feature analysis on the acquired dataset. The effective parameters obtained are utilized as input for machine learning models, including XGBoost (eXtreme Gradient Boosting) and Random Forest, which are trained to predict the energy consumption of locomotives. These models, which have demonstrated success in regression predictive modeling applications, predict energy consumption with an R2 value of 0.88. Finally, the trained models are integrated into a meta-heuristic algorithm and agent-based simulation model developed for the locomotive assignment problem.