Hierarchical clustering-based algorithms for optimal waste collection point locations in large-scale problems

• Development of a robust generalized decision-support tool for waste collection bin location and allocation. • Hierarchical clustering applied to simplify the large-scale and complex problem. • Criteria-based clustering applied for the bin location-allocation problem. • An implementation and comparison of two clustering strategies for waste bin allocation. • Sub-problem definition and representative selection identified as suitable approaches. The cities face the challenge of optimizing investments in waste management to meet EU standards while maintaining economic affordability. One of the issues is the optimal location for specialized waste collection points. The main target is to find the lowest number of collection points that would still attain waste production, and the average walking distance to the waste container would be kept beneath the tolerable limit for citizens. The population density and waste production vary over city parts; thus, the need for specialized containers in more populated city centers, industrial zones, or household streets differs. This paper develops a new computational approach providing a robust generalized decision-support tool for waste collection bin location and allocation. This task leads to a mixed-integer linear program which is not solvable for larger cities in a reasonable time. Therefore, hierarchical clustering is applied to simplify the model. Two strategies for solving waste bin allocation (for multiple variants of the model formulation) are implemented and compared – sub-problem definition and representative selection approaches. The resulting framework is tested on the artificial instance and a few case studies where the structure and properties of results are discussed. The combination of presented approaches proved to be appropriate for large-scale instances. The representative selection approach leads to a better distribution of containers within the area in the single-objective model formulation.