Learning-guided bi-objective evolutionary optimization for green municipal waste collection vehicle routing

Abstract

Waste management has emerged as a critical issue in modern society, where vehicles are scheduled to visit multiple locations for waste collection and transport. This study focuses on a key problem in waste management: route optimization of waste collection vehicles, and formulate it as a bi-objective vehicle routing problem with stochastic demand (VRPSD), aiming to minimizing both total costs and carbon emissions. Although previous studies have significantly advanced our understanding of solving similar problems, the lack of real-world data and limited problem-solving capabilities still restrict the practical applicability of existing methods. To bridge this research gap, this study designed a regression model using nighttime light data to efficiently and accurately generate two real-case instances in Beijing. Furthermore, a multi-objective evolutionary algorithm integrates Efficient Non-dominated Sorting with Sequential Search and a one-dimensional convolutional neural network (MEAE1C) is proposed to solve the VRPSD problem. MEAE1C integrates a CNN evolver to leverage knowledge from current high-quality solutions to guide subsequent population evolution. Experimental results confirm the superior accuracy in estimates of waste generation, and extensive simulations on benchmark datasets and real-case scenarios consistently demonstrate the superiority of MEAE1C over existing methods. The above results highlight the practical feasibility of the proposed methods in addressing real-world municipal waste management challenges.