Advanced optimization models and methods for electrical mobility

Electric vehicle (EV) adoption is rapidly increasing, yet the development of charging infrastructure still faces strategic and operational challenges. Identifying profitable locations for new charging stations and designing effective pricing policies are central issues for both private operators and public planners. Existing works formalize the problem employing different methodologies, e.g., flow capturing model with capacity constraints, or pricing policy decisions. Nevertheless, to the best of our knowledge, none of them combine the aforementioned approaches with multi-period pricing and heterogeneous user preferences at both spatial and price levels. In addition, it is crucial to incorporate user behaviour derived from real charging session data rather than relying solely on declared preferences from surveys. Therefore, the contribution of this dissertation is two-fold. As the first contribution, we constructed a binary classification model to predict whether candidate locations are likely to exhibit high daily charging activity and determine which features have a major impact on station usage. The analysis of EV charging demand was carried out using a realworld dataset integrating spatial, temporal, and infrastructural attributes. Results show that the predictive model achieves an accuracy of 70% in identifying high-demand areas, providing an interpretable and computationally efficient tool for preliminary siting decisions. As the second and main contribution, leveraging this analysis, we propose a bilevel capacitated charging station location and pricing model. In the upper level, a charging point operator determines where to build new stations, how many connectors to install, and how to set multi-period prices. In the lower level, users select stations by minimizing an individual utility function that reflects price sensitivity, distance costs, and locationspecific attractiveness. The attractiveness component is informed by the empirical insights derived from the aforementioned machine learning analysis, thereby capturing demand response and the impact of pricing on customer station choice. Computational experiments were conducted both on randomly generated and realistic instances that demonstrated the viability of the approach and highlighted the impact of temporal demand distribution and charging duration on instance hardness. Overall, the combination of the machine learning techniques and optimization components provides a comprehensive decision-support framework for EV infrastructure planning, enabling operators to (i) anticipate demand patterns and (ii) design profitable, user-oriented charging networks.