The growing adoption of electric vehicles in urban logistics has increased the need for routing models that jointly address operational efficiency and environmental impact. This paper studies a Sustainable Electric Vehicle Routing Problem with Time Windows, which extends the classical EVRPTW by integrating mixed time windows with penalties and carbon-emission costs, while restricting each charging station to at most one visit per route and prohibiting depot-to-station and station-to-station movements. To solve this NP-hard problem, two greedy-based hybrid metaheuristics are proposed, Greedy Simulated Annealing (GSSA) and Greedy Variable Neighborhood Search (GSVNS). Computational experiments on 76 benchmark instances show that greedy-based hybridization significantly improves the baseline methods. GSSA achieves an average total-cost reduction of 49.48% over SA, while GSVNS improves VNS by 11.49% in total cost, 16.89% in distance, and 19.52% in fleet size. In addition, GSVNS attains the best total cost on 80.26% of the tested instances, confirming its effectiveness, particularly on large-scale instances.