Economic lot-size using machine learning, parallelism, metaheuristic and simulation

Abstract

The use of discrete event simulation optimisation methods is a tool commonly used as a decisionmaking support system in industrial problems, concerning management and resource allocation in order to maximise a set of values regarding costs, revenues and other enterprise interests. The present study has proposed and tested an optimisation algorithm developed on Python, with different wall clock time reduction strategies including parallelism, the Greedy Randomized Adaptive Search Procedure (GRASP) population-based metaheuristic, and ten machine learning methods. With the selected best machine learning method (Decision Trees Regressor) 6 optimisation scenarios were generated and then applied to an economic lot-size problem for a theoretical shop floor. The results showed improvements in the reduction of the processing time of 95.0% comparing the serial GRASP with the parallel machine learning GRASP, obtaining a solution of 94.0% of the best local optimum.