A network memetic algorithm for energy and labor-aware distributed heterogeneous hybrid flow shop scheduling problem

With the penetration of decentralization into factories, the production scheduling among non-identical factories has emerged as a hot issue for industrial demand response. However, little attention has been paid to differentiated production resources for distributed production scheduling. This paper studies an energy and laboraware distributed heterogeneous hybrid flow shop scheduling problem (ELDHHFSP). The heterogeneous factors, i.e. different numbers and capabilities of machines, time-of-use electricity price, labor shift, salary and bonus of workers, are included in the ELDHHFSP. The mixed-integer linear programming (MILP) model of ELDHHFSP is presented with the objectives of minimization of total tardiness, total production cost, and total carbon emission. A network memetic algorithm (NMA) is proposed for the ELDHHFSP from the solution and strategy space. The NMA mainly includes two critical parts, i.e., probability network model and learning-based local search, corresponding to global and local search. A probability network model is established to guide the search towards the optimal regions from different directions by employing multiple objective weight vectors. The sampling part samples from the probability matrix of each node in the network to construct new solutions. The learning-based local search integrates several local search operators and a reward learning mechanism is designed to find suitable local search operators when evolutionary. A comprehensive experiment on extensive testing instances is conducted to investigate the effectiveness of the probability network and local-learning strategy. The comparison of NMA and other related algorithms shows the effectiveness and efficiency of NMA.

Automated summary

This paper investigates the distributed heterogeneous hybrid flow shop scheduling problem and proposes a solution based on a network memetic algorithm. The algorithm includes a probability network model and a learning-based local search, guiding the search using multiple objective weight vectors and effectively solving the production scheduling problem.