| Abstract [eng] |
The improvement of fast charging station layout and capacity planning for urban electric vehicle networks in Bengaluru, India was investigated using data-driven machine learning and multi-criteria decision analysis methods. A synthetic electric vehicle session dataset of 456,744 records was generated for the period 2023 to 2024, with generation parameters calibrated from four peer-reviewed sources. Three machine learning models were trained, validated, and compared on an hourly demand time series: XGBoost with Tweedie loss, a two-layer Long Short-Term Memory network, and Facebook Prophet. XGBoost was selected as the production forecasting model based on a test set Root Mean Square Error of 132.27 kWh and a coefficient of determination of 0.9411. A four-criteria Multi-Criteria Decision Analysis framework was formulated using the Analytic Hierarchy Process, incorporating demand intensity, grid connection capacity, user accessibility, and land use compatibility. The framework was applied to 20 candidate sites across Bengaluru, and a ranked list of ten optimal deployment locations was produced, with Whitefield IT Corridor identified as the highest-scoring site. Performance was evaluated using the M/M/c queuing model under two comparative scenarios. Average user waiting time was reduced from 13.23 minutes to 3.46 minutes, representing a 73.8 percent reduction exceeding the project target of 15 percent. The probability of waiting was reduced from 31.7 percent to 10.6 percent. Spatial coverage was doubled from 10.1 percent to 20.2 percent of the study area. A total capital expenditure of Indian Rupee 76.5 million was established for 17 new Direct current(DC) fast chargers, with a payback period of 7.4 years under standard conditions and 4.4 years under the PM E-DRIVE subsidy scheme. |
