Exact solving machine scheduling problems assisted by graph neural networks

Raspoređivanje na strojevima jest jedan od najopćenitijih i najtežih tradicionalnih kombinatornih problema. Egzaktne općenite metode za rješavanje takvih problema dijele se na pristupe bazirane na matematičkom programiranju i na one bazirane na zadovoljenju ograničenja. U ovom radu koristit će se grafovska neuronska mreža za ubrzanje izvođenja izračuna u rješavaču baziranom na matematičkom programiranju. Mreža će se utrenirati imitacijskim učenjem nad odabranim skupom problema raspoređivanja na strojevima.Job-shop scheduling problem is one of the most common and most difficult traditional combinatorial problems. Exact general methods for solving such problems are divided between mathematical programming and those based on meeting constraints. In this paper, graph-convolution neural network will be used for performance acceleration of solver based on mathematical programming. The network will be trained by imitation learning over a selected set of instances of Job-shop scheduling problem