Learning-based simulated annealing algorithm for unequal area facility layout problem

This paper proposes a learning-based simulated annealing (LSA) algorithm to tackle the NP-hard unequal area facility layout problem (UA-FLP). The goal of UA-FLP is to optimize the material flow between facilities of different sizes to enhance manufacturing efficiency. The LSA algorithm incorporates a novel solution representation, an improved penalty function and a diverse set of neighborhood operators to refine the search space. By utilizing a reinforcement learning-based controller, LSA enables a flexible and efficient exploration through state detection and fast feedback. A two-stage greedy local search is employed to further exploit the search space and enhance solution quality. Additional features include temperature sampling generation to minimize parameter settings, a greedy initial solution production to relax infeasible restrictions. Experimental results on 16 well-known instances validate LSA's high proficiency compared to several state-of-the-art algorithms, and it exceeds 7 best-known solutions within a comparable time, particularly its excellent performance in large instances within a short execution time.

Automated summary

This paper proposes a learning-based simulated annealing algorithm to tackle the NP-hard unequal area facility layout problem. The algorithm incorporates a novel solution representation, an improved penalty function, and a diverse set of neighborhood operators to refine the search space. By utilizing a reinforcement learning-based controller, the algorithm enables a flexible and efficient exploration, further exploiting the search space and enhancing solution quality.