Implementation of Parallel Computation for Identification of Interval Models based on Multi-core Parallelism and CUDA Technology

The NP-complex computational problem has been considered, which involves finding the minimum of the quality assessment function of discrete models of a complex object using interval data analysis. Such computational problems relate to the identification of model parameters based on a sample of imprecise data presented in interval form. The complexity of this problem lies in the utilization of an interval criterion function for the optimization task. To solve it, a well-known method involving parallelization of computational processes through the application of a multi-agent system in the form of a colony of honeybee agents has been employed. For the implementation of a parallel computational scheme, two technologies are suggested: multi-core parallelism and CUDA technology. The comparison was conducted using examples of constructing a complex and a simple model in terms of the dimensionality of the identification task. It has been established that the use of CUDA technology for parallelizing the computational task of identifying parameters of interval discrete models in the form of difference equations provides higher efficiency for complex tasks and in case of increasing the number of computational agents (bees in the colony) in the computational algorithm. Additionally, in the process of experimental studies for these examples, the optimal number of computational agents was determined.