This paper deals with a reentrant hybrid flow shop problem with sequence-dependent setup time and limited buffers where there are multiple unrelated parallel machines at each stage. A mathematical model with the minimization of total weighted completion time is constructed for this problem. Considering the complexity of the problem at hand, an effective cooperative adaptive genetic algorithm (CAGA) is proposed. In the algorithm, a dual chain coding scheme and a staged-hierarchical decoding approach are, respectively, designed to encode and decode each solution. Six dispatch heuristics and a dynamic adjustment method are introduced to define initial population. To balance the exploration and exploitation abilities, three effective operations are implemented: (1) two new crossover and mutation operators with collaborative mechanism are imposed on genetic algorithm; (2) an adaptive adjustment strategy is introduced to re-optimize better solutions after mutation operations, where ant colony search algorithm and modified greedy heuristic are intelligently switched; (3) a reset strategy with dynamic variable step strategy is embedded to re-generate some non-improved solutions. A Taguchi method of design of experiment is adopted to calibrate the parameter values in the presented algorithm. Comparison experiments are executed on test instances with different scale problems. Computational results show that the proposed CAGA is more effective and efficient than several well-known algorithms in solving the studied problem.

Motivated by the operational scenarios of lock scheduling, we propose a serial-lock chain navigation problem (SLCNP) modeled on the Three Gorges-Gezhouba Dam (TGGD) for the first time. Ship grouping, synchronized moving, and grouped waiting operations are integrated into the ship navigation process. A mixed integer programming (MIP) model that incorporates real-world constraints such as ship priority, service fairness, traffic flow equilibrium, and phased ship placement is presented to optimize ship throughput and ship stay time. To solve the SLCNP, a sort-pick strategy-based swarm intelligence algorithm (SPSSIA) framework is developed that integrates the characteristics of SLCNP through a hybrid multi-section encoding method and a two-stage heuristic decoding approach. A swarm intelligence evolution mechanism is used to improve the search ability and robustness of the framework. Several instances are generated based on real data to verify the correctness and effectiveness of the model and algorithm. Computational results demonstrate the applicability and effectiveness of the proposed SPSSIA. Further analysis of the experimental results indicates that the key impact factors significantly influence the navigational performance of the TGGD system. The results of this study will provide practical guidance for the operational processes of inland river hubs with comparable characteristics.

Approach channel, ship lock and ship lift are critical resources of Three Gorges-Gezhou Dams (TGGD) along the Yangzi River. Encouraged by the operational scenarios of TGGD, this paper investigates a multi-resource co-scheduling problem of ship-lock-channel, where the synchronous velocities of vessels are influenced by freeboard height and steering skill. The vessel’ sailing interval, sailing sequence and sailing point are closely related with the water discharge in flood season. A mixed integer programming model is established to simultaneously minimize the water resource usage, average waiting time and carbon emissions over all vessels. The vessel grouping, synchronous moving and waiting process are incorporated into the scheduling model. Thereafter, a problem-specific Q-learning-based multi-objective heuristic algorithm, named QLMHA, is proposed to optimize the multi-resource co-schedule. With a well-tailored bi-level encoding scheme, QLMHA makes use of hybrid multi-stage decoding mechanism and fuzzy correlation entropy analysis to respectively decode and assess each candidate solution. To enhance the searchability of the QLMHA, Q-learning technique is coupled to select the optimal action for solution set obtained by external archive. Numerical experiments are implemented by using the practical data from TGGD. The statistical result corroborates the higher effectiveness and practicality of the QLMHA in terms of solving the concerned problem. In addition, our findings demonstrate that QLMHA is able to generate more satisfactory schemes, offering TGGD and shipping managers a broader array of preferred choice.