Data and code for "Optimization of Charging Scheduling for Battery Electric Buses Considering SoC-Dependent Charging Power: A Lagrangian Matheuristic Approach"

Abstract

This record contains the data, code and results of the paper "Optimization of Charging Scheduling for Battery Electric Buses Considering SoC-Dependent Charging Power: A Lagrangian Matheuristic Approach" (J. Leal Enciso and H. Wang, submitted to the International Journal of Systems Science: Operations & Logistics). The paper decides when each battery electric bus charges at a single depot and how many chargers to install, over a cyclic 24-hour day, under a time-of-use tariff, a depot power limit and a charging power that decreases as the state of charge (SoC) increases. The problem is formulated as a mixed-integer linear program (MILP) and solved with D-LAGR, a deterministic Lagrangian matheuristic that returns a feasible schedule and a certified optimality gap, and with an exact benchmark (AMPL + Gurobi). The instances are based on the TransMilenio system of Bogotá: 25 instances of 30 buses and the full fleet of 84 buses. Files:- README.md: description of every file and step-by-step instructions to run the code and reproduce the results.- code.zip: Python code of D-LAGR (HiGHS through highspy) and of the exact benchmark (AMPL + Gurobi), the instance generator, requirements.txt and LICENSE (MIT).- data.zip: scenario manifest, timetables of the 30-bus sample and of the full fleet, and the 26 instance files used in the paper.- results.zip: for each instance, the D-LAGR run log, the summary (cost, certified bound and gap, comparison with the exact solver), schedules, bounds by charger count, figures and the validation report (twelve checks), and the schedules of the exact solver.- paper_numbers.zip: the table behind the results figure of the paper and the scripts that build it. Notes:- Extract code.zip, data.zip and results.zip into the same folder to rebuild the project layout.- Instance ids sc25, sc26 and sc27 in the files correspond to sc23, sc24 and sc25 in the paper (id_map.csv).- D-LAGR needs only Python and open-source packages; the exact benchmark needs an AMPL licence with Gurobi.- The raw logs of the exact solver are available from the corresponding author upon reasonable request.- Licences: data and results CC BY 4.0; code MIT.