An enhanced decomposition-based multi-objective evolutionary algorithm with neighborhood search for multi-resource constrained job shop scheduling problem

Abstract

Today’s industry 5.0 emphasizes the synergy between humans and equipment to raise productivity. This paper investigates a multi-resource constrained job shop scheduling problem, aiming to minimize both the makespan, the total energy consumption of automated guided vehicles (AGVs), and the total workload of workers. To address the problem, we apply an extended disjunctive graph and establish a multi-objective mixed integer linear programming model based on it. Afterward, we develop an enhanced decomposition-based multi-objective evolutionary algorithm with neighborhood search (EMOEA/D-NS) to efficiently solve this problem. In this algorithm, we design a three-layer solution representation and propose a hybrid heuristic based on priority weights to yield high-quality individuals, which comprises a congestion alleviation assignment rule for AGVs and a shortest-earliest rule for allocating workers. Furthermore, three lemmas for determining non-critical tasks are given and six neighborhood search approaches are designed to improve the quality of solutions. To enhance the exploration and exploitation capabilities of the EMOEA/D-NS, we propose a multi-rank individual-driven evolutionary mechanism that classifies the individuals into guiding, working, and following groups. For the individuals within the guiding group, we propose a self-evolution strategy that allows themselves to evolve in the way of utilizing their own experiences. For the individuals of the working group, we design a collaborative evolutionary strategy, consisting of a priority weights-based crossover and mutation operators, to evolve them with other individuals to explore promising space and exploit known space. The individuals of the following group are evolved toward the direction of those within the guiding group by an oriented evolutionary strategy, which aims to improve the quality of population and accelerate the convergence of the algorithm. Numerical experiments are carried out on 40 modified benchmarks to highlight the efficiency of the EMOEA/D-NS. Lastly, we conclude our work and outline further research directions.