Parallelization of industrial process control program based on the technique of differential evolution using multi-threading

Differential Evolution is a population based optimization technique which discretizes the sample space of solution based on generations of population members and performs a fitness evaluation to determine the best population members. The fitness evaluation process involves lengthy calculations, thus becoming the most redundant and time consuming aspect of the program. The aim of the proposed work is to parallelize an industrial process control program which is based on differential evolution. The parallelization of program has been carried out using a multi-threading approach, in which the independent iterations of the fitness evaluation of Differential Evolution have been distributed uniformly into separate worker threads, which are then simultaneously executed on a multi core architecture GPU machine. By parallelizing the fitness evaluation of the program over the population members in a particular generation, a speed up of 3.99 times in the overall execution time has been observed.