EVRPTW-ARC

Overview EVRPTW-ARC (Arc-Based Realistic Conditions) is a multi-scenario dataset providing realistic travel information for research on the Electric Vehicle Routing Problem with Time Windows (EVRPTW). The dataset is built from the 92 benchmark instances proposed by Schneider et al. (2014). The original routing information is preserved (depots, customers, charging stations, demands and time windows), while the benchmark nodes are mapped onto the real road network of Lille, France, replacing the abstract spatial representation with geographically referenced road information. For every directed connection between two nodes, the dataset provides road-network characteristics such as distance, travel time, speed limits, elevation changes, road slope and road type. Each arc is additionally associated with a WLTC-derived speed profile at one-second resolution. The dataset also represents variations in vehicle operating conditions through: three driving styles: passif, modere, agressif; two traffic conditions: fluide, congestionne; three temperature scenarios: hiver, tempere, ete. Traffic and temperature conditions are crossed, producing 6 records per directed arc. Motivation EVRPTW benchmark instances are widely used to evaluate routing algorithms, but their spatial representation does not describe the physical characteristics of travel on a real road network — distances come from synthetic coordinates, and information such as road gradients, speed characteristics and driving dynamics is unavailable. This matters for electric vehicles because energy consumption depends not only on distance travelled but on the conditions under which it is travelled. Driving-cycle and vehicle-operation datasets, conversely, generally describe detailed vehicle motion without representing the combinatorial structure of a routing problem, they lack the customers, demands, time windows and charging stations an EVRPTW formulation needs. EVRPTW-ARC combines both: the Schneider instances supply the routing problem, and the arc-level enrichment supplies the physical and environmental conditions of travelling between nodes, forming an intermediate layer between EVRPTW optimization and detailed energy-consumption modelling. Researchers can use the provided road and driving attributes as inputs to physical, empirical, or machine-learning-based energy models. Source instances Each of the EVRPTW benchmark instances introduced by Schneider et al. (2014) contains a depot, customers and charging stations together with the information an EVRPTW requires: node identifier, node type, spatial coordinates, customer demand, ready time, due date, service time, and vehicle/battery parameters. Nodes follow the original Schneider notation: d (depot), c (customer), f (charging station). These routing-related node attributes are retained unchanged through the geographical enrichment process described below. Geographical enrichment Coordinates. Each instance's synthetic coordinates are linearly projected, independently per instance, onto a bounding box over Lille, France, preserving the relative spatial arrangement of its nodes. The resulting points are matched to the real road network extracted from OpenStreetMap. Charging stations. Each benchmark charging-station point is matched to the nearest real EV charging location retrieved from OpenStreetMap via the Overpass API. The original station identity and routing role are retained; only its physical location is replaced. Road network. The Lille drivable-road network is extracted with OSMnx, keeping only the largest strongly connected component to guarantee connectivity between mapped nodes. The network is projected into a metric CRS and enriched with road speeds and travel times. Node elevations come from OpenTopoData (EU-DEM 25 m) and are used to derive road gradients. Arc construction The network is represented as a directed graph. For each ordered node pair (i, j), a shortest path is computed with Dijkstra's algorithm using road distance as weight. Because the network is directed, the arc i → j can differ from its reverse j → i. Each resulting path is summarized by: Road distance (distance_km) : total path length. Travel time (time_h) : estimated free-flow travel time. Speed characteristics : speed_mean_kmh and speed_max_kmh, describing the road infrastructure rather than instantaneous vehicle speed. Terrain characteristics : length-weighted mean slope, maximum absolute slope, cumulative elevation gain and loss. Road structure : dominant road type and number of OSM edges composing the path. WLTC driving-profile library Road-network information alone does not describe the vehicle's instantaneous motion. To add that, the dataset draws on the WLTC Class 3b driving cycle, sliced into a library of representative micro-segments. Each segment is characterized by mean speed, speed standard deviation, mean acceleration, mean speed–acceleration interaction, percentage of time accelerating, stop ratio, travelled distance, and the complete second-by-second speed profile (stored as JSON in v_profile, 1 s resolution). Driving styles. Three styles (passif, modere, agressif) are derived by K-Means on standardized kinematic features of the segment library. Assignment to arcs. Each arc is matched to the WLTC segment whose mean speed is closest to the arc's own mean road speed, then that segment's profile is cyclically repeated and truncated to exactly cover the arc's real travel duration. Operating-condition scenarios Traffic. Two conditions, fluide and congestionne, each defined by a sampled reduction factor τ (stored in traffic_pct): fluide: τ in range [0.05–0.20] congestionne: τ in range [0.40–0.75] Temperature. Three seasonal scenarios derived from ERA5 hourly reanalysis data for Lille (2019–2023, Open-Meteo archive), using the 5th–95th percentile of each seasonal group as the sampling interval (temp_C): hiver : [−1.7 to 11.5 ]°C tempere : [3.1 to 20.3]°C ete : [12.0 to 26.4]°C Crossing 2 traffic conditions × 3 temperature conditions yields the 6 records per arc noted above; each retains the same underlying road and routing characteristics while representing a different operating condition. Data dictionary Column Description Unit from Origin node identifier – to Destination node identifier – lat_i, lon_i Coordinates of origin node degrees lat_j, lon_j Coordinates of destination node degrees distance_km Length of the shortest road path km time_h Free-flow travel time of the road path h speed_mean_kmh Mean speed limit along the path km/h speed_max_kmh Maximum speed limit along the path km/h slope_mean_pct Length-weighted mean road slope % slope_max_pct Maximum absolute slope % elevation_gain_km Cumulative elevation gain km elevation_loss_km Cumulative elevation loss km road_type Dominant OSM road class – n_edges Number of OSM edges forming the path – demand Demand at the destination node original units ready Opening time of destination time window original units due Closing time of destination time window original units seg_id Assigned WLTC segment identifier – window_size Duration of the final extended profile s v_moy_seg_kmh Mean speed of the extended profile km/h a_moy_seg Mean acceleration of the extended profile m/s² stop_ratio Fraction of the profile at standstill % dist_seg_km Distance covered by the extended profile km v_profile Second-by-second speed profile (JSON) m/s style_conduite Assigned driving style – congestion Traffic scenario – traffic_pct Sampled traffic reduction factor τ – speed_corrected_kmh Speed after traffic adjustment km/h temperature_scenario Temperature scenario – temp_C Sampled ambient temperature °C Acknowledgements The authors gratefully acknowledge the financial support of the CPER MANIFEST program, funded by the French State and the Région Hauts-de-France. This work was conducted within the GREENROUTE project, led by LTI (Université de Picardie Jules Verne) in collaboration with LGCgE (Université d'Artois). References Schneider, M., Stenger, A., & Goeke, D. (2014). The electric vehicle-routing problem with time windows and recharging stations. Transportation Science, 48(4), 500–520. Jiang, Y., Guo, J., Zhao, D., & Li, Y. (2024). Intelligent energy consumption prediction for battery electric vehicles: A hybrid approach integrating driving behavior and environmental factors. Energy, 308, 132774. Al-Wreikat, Y., Serrano, C., & Sodré, J. R. (2021). Driving behaviour and trip condition effects on the energy consumption of an electric vehicle under real-world driving. Applied Energy, 297, 117096. Cheng, R., Zhang, W., Yang, J., Wang, S., & Li, L. (2025). Analysis of the effects of different driving cycles on the driving range and energy consumption of BEVs. World Electric Vehicle Journal, 16(3), 124.