Simulation-based parallel direct search optimization for revenue maximization of coalmines

Addressing revenue maximization problems using detailed coalmine simulation models is computationally expensive because of large numbers of model parameters, decision and random variables. To rectify this challenge, a simulation-based parallel Direct search optimization is proposed in this paper using a real surface coalmine as case study. To this end, a simulation model of truck-shovel-crusher system in Arena is used and Direct search toolbox in MATLAB is applied to optimize the simulation model. Firstly, the simulation optimization process is initiated on two local computers in parallel. When a local computer achieves a threshold improvement, the optimization routine is stopped and its current best decision is delivered to the central computer. This asynchronous interaction also takes place when there is no improvement for a certain number of iterations. In both cases, the central computer uses Direct search along with the list of solutions collected to search for the decision to initiate next round of optimization on the local computers. The whole process is continued for a preset total number of iterations. Through comparison with optimization results obtained on a single computer using OptQuest, it is demonstrated that the total time involved in the proposed procedure is reduced without sacrificing quality of solution.