Learning to schedule new orders in batch plants using aproximate dynamic programming

ARREDONDO, FACUNDO; MARTÍNEZ, ERNESTO

Bucarest, Rumania

Simposio; 17th European Symposium on Computer Aided Process Engineesing – ESCAPE17; 2007

Production scheduling in a wide range of batch plants involves minimizing tardiness of batches already scheduled when inserting new orders. This problem is addressed here as learning an “order insertion policy” using intensive simulations in the framework of approximate dynamic programming (ADP). Simple insertion operators are defined and the values of choosing them at different schedule states found by the incoming order are learnt using a Q-learning algorithm. To generalize values of insertion operators across schedule states a locally weighting regression technique is used. Results obtained highlight that simulation-based heuristic learning is very appealing to increase responsivenes of scheduling and planning systems in disruptive event handling.