Abstract:

Annual municipal waste collection schedules determine daily workloads and constrain subsequent vehicle routing by assigning districts and waste fractions to collection days. However, minimising peak daily workload alone can produce uneven workloads throughout the year, arbitrary holiday replacement assignments and geographically inefficient schedules. This study develops and evaluates two extensions of a mixed-integer programming model for annual waste collection scheduling implemented in the PLANIQA decision-support system. The baseline model, M0, minimises the maximum daily collected waste mass. Model M1 introduces a second optimisation stage that minimises the sum of weekly peak workloads and penalises less preferred replacement days while preserving the peak workload obtained in the first stage. Model M2 incorporates geographical information through a capacitated vehicle routing relaxation evaluated over a representative four-week window. Computational experiments use data from a Polish municipality comprising 450 scheduling items, 35 districts and five waste fractions. Compared with M0, M1 reduces the mean weekly peak by 7.1% and increases the share of first-choice replacement days from approximately 11% to 61%, with virtually unchanged estimated travel distance. Compared with M1, M2 reduces estimated travel distance by 22.6–23.4% and the number of trips by 8–10%, at the cost of a 2.9–4.9% increase in peak daily workload and less balanced workloads. Computation times remain suitable for annual planning: approximately 26 seconds for M1 and five to six minutes for M2. The findings demonstrate the value of considering workload distribution, replacement-day preferences and geography jointly. Distance savings remain relaxation-based estimates requiring validation through feasible vehicle routes over the full annual horizon.