Generating Realistic Benchmarks for Dynamic Truck and Trailer Scheduling using Gaussian Copulas

Academic research in dynamic optimisation uses benchmark generators to artificially simulate controlled and reproducible changing-environments to systematically compare algorithmic performance under uncertainty. However, due to the scarcity or difficulty in acquiring real-world data, benchmarks often fail to incorporate real-world features, such as problem constraints or the time-linkage property, where previously made decisions influence future events. This study introduces a Gaussian Copula-based real-world data-driven synthetic data generation model for Dynamic Truck and Trailer Scheduling Problem (DTTSP). The model offers a realistic, privacy-preserving DTTSP benchmark instance generator, which can be used to recreate the dynamism, constraints, heterogeneity, and time-linkage of logistics and supply chain operations. This work examines the utility, fidelity, and privacy of the suggested model in four workday case studies from a local transportation company. The conducted experiments demonstrate the systematical application of Gaussian Copulas to produce accurate, useful, and secure DTTSP benchmark instances that capture the statistical properties and correlation of variables, as well as the temporal patterns, in the original annual data. Nevertheless, the utility analysis of the conditional sampling indicates that there is still room for improvement in the modelling process.