Instance-Aware Parameter Tuning for Electric Vehicle Routing
A recent study introduces an instance-aware parameter configuration for Bilevel Late Acceptance Hill Climbing (BLAHC) aimed at addressing the Electric Capacitated Vehicle Routing Problem (ECVRP). This method employs offline tuning to create specific parameter labels for each instance, which are then predicted from instance characteristics using a regression model, allowing for parameter forecasting for new instances prior to execution. Evaluated against the IEEE WCCI 2020 benchmark and its extensions, the approach realizes an average reduction of 0.28% in objective value. This research tackles the issue that parameters tuned globally do not effectively utilize the diverse characteristics present in ECVRP instances, which differ in structure, demand patterns, and energy limitations.
Key facts
- Instance-aware parameter configuration is proposed for Bilevel Late Acceptance Hill Climbing.
- The method targets the Electric Capacitated Vehicle Routing Problem.
- Offline tuning obtains instance-specific parameter labels.
- A regression model maps instance features to parameter labels.
- Parameter prediction occurs for unseen instances prior to execution.
- Tested on IEEE WCCI 2020 benchmark and extensions.
- Average objective value reduction of 0.28% is achieved.
- Globally tuned parameters fail to exploit instance heterogeneity.
Entities
Institutions
- arXiv
- IEEE WCCI