Model and Data Driven Hyper-Heuristics for Combinatorial Optimization and Their Applications in Port Operation Integrated Scheduling

This project focuses on intelligent decision marking optimization in complex systems such as those in intelligent manufacturing and modern services. The project shall investigate theories, methodologies as well as applications for intelligent optimization under complex and dynamic environment. The current operations research methods based on mathematical models are too rigid to adapt to the complexity and dynamic nature of real-world systems. Meanwhile, the machine learning methods driven by massive data training lack necessary reliability and interpretability. There is a real technological bottleneck preventing the advancement of the current single node level intelligence to an integrated system level intelligence. This project proposes to hybridize the above two types of methodologies into a hyper-heuristic framework. The project shall research on advanced problem-specific features extracted from the mathematical formulations (objectives and constraints) and their duals, and utilize historical data and high quality simulation data to train deep reinforcement learning for optimal algorithm combinations for different scenarios. As a generic algorithms’ selection framework, hyper-heuristic is an ideal choice for this hybridization. The project shall take the real-life container terminal yard allocation and truck dispatching as the main testing problems to evaluate the feasibility and efficacy of the uncertainty handling by the proposed theories and methodologies.