Research Report on Electric Vehicle Charging
Infrastructure Optimization
This report investigates and optimizes the performance of electric vehicle (EV) charging
stations using simulations in the urban context of Nha Trang. It focuses on accessibility,
station utilization, energy efficiency, and customer satisfaction. The aim is to improve the
existing charging infrastructure while proposing a data-driven approach to expand and
manage future EV charging networks.
We define four primary evaluation criteria:
Convenience Factors reflect how easily users can access charging services. This includes
three elements: reroute distance (how far vehicles need to deviate from their original path
to charge), waiting time (how long they have to queue), and charging time (how long it takes
to complete charging). Higher values in these indicators typically indicate lower user
satisfaction. A vehicle experiencing long delays or detours implies that the infrastructure is
under strain.
Daily Power Usage helps assess how much energy is being drawn from the grid each day
by the entire charging network. Monitoring this value allows us to determine peak load
conditions and evaluate sustainability. Simulations are constrained to ensure energy
usage does not exceed realistic values (e.g., 3000–4000 kWh per day, based on global
benchmarks for hybrid stations).
Vehicles Unable to Reach Charging Stations indicate infrastructure failure. These are
EVs that deplete their battery before accessing a charger. By plotting these failure points
on heatmaps, we identify underserved regions and prioritize them for station expansion.
We also simulate different battery consumption rates (0.1%–0.3% SoC/min) to reflect
driving conditions.
Station Utilization Rate is the percentage of time stations are actively serving vehicles.
Over-utilization suggests overloading, while under-utilization indicates inefficient
allocation. We also analyze queue lengths and unmet demand at each location.
Simulation Setup and Parameters
The simulation accounts for various urban dynamics:
• Vehicle types (private EVs and electric taxis).
• Home charging availability (60% of residential buildings have chargers).
• Time-of-day variations (peak hour multipliers from 0% to 80%).
• Weekday vs. weekend behavior.
• Taxis are modeled to charge predominantly at night if SoC < 90%.
• Travel speed defaults to 20 km/h.
We simulate traffic based on two categories:
• Commuter Vehicles start from home and travel to workplaces. At noon, 50% visit
food locations and return; the rest stay. In the evening, half go out before returning
home.
• Leisure Vehicles move randomly between entertainment and dining zones. They
dwell 1–3 hours at destinations before returning home.
Spatial data is used to identify residential, commercial, and recreational areas. We
assume 70% of buildings are homes. Among non-residential zones: 50% are workplaces,
30% food-related, and 20% leisure.
Optimization and Evaluation
We use several scenarios to test the effectiveness of infrastructure modifications:
1. Identify zones where vehicles die before reaching a charger.
2. Add fixed stations in these zones and observe changes in service rates and wait
times.
3. Repeat with mobile charging units and compare effectiveness.
4. Evaluate impacts under three vehicle densities: 3000, 5000, and 8000.
5. Quantify changes using the four metrics and present results as heatmaps and
charts.
We establish decision rules:
• If >10 vehicles are unserved in a region → add new station.
• If >10 vehicles deplete battery within 5 km radius → add new station.
• If >15% of vehicles report poor experience → prioritize those zones for upgrade.
Technical Aspects and Algorithms
Routing is handled using the Floyd-Warshall algorithm to find shortest reroute distances.
Traffic density adjusts vehicle speed using a coefficient:
speed_coeff = 1.0 - (nb_vehicles / road_capacity)
road_capacity = 1 + (num_lanes * [Link] / 3)
Power consumption rates, waiting times (randomized 10–15 minutes), and travel behavior
are parameterized to reflect real-world conditions. Clustered stations are merged for more
accurate analysis.
Reward-Based Optimization
We incorporate reward functions to evaluate potential new station locations:
• Substation Proximity Reward: Higher if closer to substations.
• Waiting Time Reward: Inverse of average wait time.
• Population Density Reward: Derived from spatial raster data.
• Distance from Existing Stations: Rewards sparsely covered zones.
• Charging Time Improvement: Evaluates the gain from faster sessions.
These scores guide machine learning models and heuristic evaluations of where to best
locate additional infrastructure.
Summary and Future Directions
Simulation results provide a roadmap for EV infrastructure growth in Nha Trang. Evaluating
multiple densities, charging behaviors, and infrastructure types reveals that mobile
chargers can complement fixed stations but are best used strategically. Future work
includes:
• Calibrating models with real-world EV data (e.g., from VinFast).
• Integrating live traffic data for dynamic simulation.
• Comparing with real-world data to prove the simulation's credibility.
