Selecting hybrids of metaheuristics for resource-constraint project scheduling problems with discounted cashflows

Abstract

Solving resource-constrained project scheduling problems with discounted cash flows (RCPSP-DC) is a critical challenge for project and finance managers, as efficient resource allocation can significantly impact a company’s financial success. While prior research addresses this NP-hard problem, most approaches depend on hybrid metaheuristics requiring expertise in hybridisation and lack systematic methods for architecture selection, often relying on a single criterion with trial-and-error parameter tuning. In this paper, we propose a novel collaborative parallel hybridisation framework that integrates Thompson sampling and multicriteria decision analysis (MCDA) to holistically evaluate and identify the best hybrid architecture from a diverse set of options. Unlike conventional approaches, our method employs onboard Taguchi design of experiments (DOE) for structured and efficient parameter tuning. Additionally, Thompson sampling, applied in the form of Bayesian learning, mitigates the stochastic nature of metaheuristics through multiple experiments. This framework was used to select the best architecture from 57 hybrid combinations of six metaheuristics for solving RCPSP-DC. Extensive experiments using standard datasets demonstrate that the proposed framework achieves statistically significant performance improvements, selecting a hybrid architecture that outperforms state-of-the-art methods. The selected architecture’s competitiveness is validated through a Z-test of proportions, underscoring its effectiveness in solving RCPSP-DC problems.