Kernkonzepte
Integrating constraint programming with deep learning improves FJSSP solutions.
Zusammenfassung
Recent advancements in solving the flexible job-shop scheduling problem (FJSSP) have favored deep reinforcement learning (DRL). However, DRL approaches face challenges in finding optimal solutions efficiently. This paper proposes a hybrid approach, BCxCP, combining constraint programming (CP) with deep learning to enhance solution quality and performance. The CP capability predictor accurately forecasts when instances can be solved by CP in real-time. Experimental results show that BCxCP outperforms DRL methods and achieves competitive results with meta-heuristic algorithms across various benchmarks.
Statistiken
Recent advancements favor DRL for FJSSP.
BC outperforms DRL methods.
OR-Tools yields poor results except in less complex instances.
BCxCP achieves better results and accurately predicts CP solvability.