A time-space modeling and hierarchically progressive ALNS for recyclables collection scheduling under spatiotemporal coupling

Improving recyclable recovery is essential for advancing the circular economy, yet collection inefficiencies continue to limit system performance. Smart recycling bins simplify disposal behavior and provide high-resolution accumulation data, but they also introduce new complexities for collection scheduling. The resulting recyclables collection scheduling problem (RCSP) extends the multi-period vehicle routing problem by incorporating fine-grained intra-period time granularity, time-varying disposal rates, and tightly coupled interactions among vehicle capacity, bin inventory dynamics, and overflow penalties, creating a high-dimensional and strongly coupled decision space. To represent these dynamics, we construct a time-space network that captures bin accumulation, remaining capacity, and overflow evolution, forming the basis of a mixed-integer programming model. Due to the computational intractability, we develop a Hierarchically Progressive Adaptive Large Neighborhood Search algorithm that decomposes decision making into three coordinated stages: collection frequency estimation, route construction, and time-detailed scheduling. Inter-stage information flow and feedback maintain solution coherence and mitigate quality loss typically associated with decomposition. Experiments on IRP benchmarks, randomly generated RCSP instances, and a real-scale case show that the proposed method achieves solution quality comparable to commercial solvers while reducing computation time from over three hours to under one minute on medium-scale instances, and produces solutions over 10% better on large-scale instances. The results also yield practical insights regarding period configuration, fleet capacity allocation, and hybrid core-reserve fleet strategies for enhancing operational efficiency and reducing system costs.