Possibilities of parallelism under identifying a quasi-linear recurrent equation

Time series analysis and forecasting are one of the widely researched areas nowadays. Identification using various statistical methods, neural networks or mathematical models has long been used in various fields of research from industry, to medicine, the social sphere, and the agricultural researches. The article considers a parallel version of the algorithm for identifying the parameters of a quasi-linear recurrent equation for solving the task of regression analysis with interdependent observable variables, based on the generalized least modules method (GLDM). Unlike neural networks, which are widely used nowadays in various forecasting systems, this approach allows us to explicitly obtain qualitative quasi-linear difference equations that adequately describe the considered process. This makes it possible to improve the quality of the studied processes analysis. A significant advantage of the model using the generalized least deviation method, in comparison with numerous neural network approaches, is the possibility of interpreting the coefficients of the model from the point of view of the research task and using the resulting equation as a model of a dynamic process. The conducted computational experiments using time series show that the maximum acceleration of the algorithm occurs when using the number of threads equal to half of the possible threads for a given device.