| Abstract [eng] |
The global transition towards electric vehicles requires an optimal solution for charging infrastructure planning. By 2030, the global stock of electric vehicles will rise to 250 million; however, the existing charging station placement method relies on heuristic approaches that fail to incorporate clustering, demand forecasting, site classification, and optimization in a signal framework. This thesis develops a multi-stage machine learning framework for optimal electric vehicle charging station placement in Vilnius, Lithuania. The proposed framework incorporates seniūnija-level population data (636,023 residents, 21 districts), traffic congestion events, and points of interest from OpenStreetMap. Firstly, DBSCAN clustering with convex hull geometry for candidate zone identification, second, an ensemble of Ridge Regression and random forest for demand prediction, third, a support vector machine used for site suitability classification on revealed preference, lastly, a Genetic algorithm is used for optimal station site selection respecting distance constraints. The framework identified 201 candidate zones. An ensemble achieved an R² of 0.96. The support vector machine achieved an accuracy of 87%. The generic algorithm selected 20 charging station locations with a coverage of 95%, covering 20 districts out of 21, with the fitness of o.4387 compared to the greedy baseline. This research provides a validated data-driven framework for electric vehicle infrastructure planning, providing a comprehensive guideline for urban planners for the fair distribution of charging infrastructure deployment. |
