Построение систем прогнозных моделей многомерных взаимосвязанных процессов

Abstract

Разработаны теоретические основы рекуррентно-параллельных вычислений в комбинаторном алгоритме МГУА для моделирования и прогнозирования сложных многомерных взаимосвязанных процессов в классе моделей векторной авторегрессии. Продемонстрирована эффективность разработанного алгоритма.

Розроблено теоретичні основи рекурентно-паралельних обчислень у комбінаторному алгоритмі МГУА для моделювання та прогнозування складних багатовимірних взаємозв’язаних процесів у класі моделей векторної авторегресії. Продемонстровано ефективність розробленого алгоритму.

Introduction. The problem of the mathematical modelling and prediction of the multidimensional interrelated time series is considered. It is used in economy, ecology and sociology. While many scientific proceedings are dedicated to modelling of one-dimensional time series, the experience of multidimensional time series modelling is insufficient. Methods. An approach to the structural and parameters identification of the multidimensional time series is considered when parameters for every model is estimated independently. An algorithm with selecting of more than one best model for every process is used. The purpose is to combine all possible variants of system models and to select the best one by additional criterion. Results. Theoretical grounds of recurrent-and-parallel computing in combinatorial GMDH algorithm and software for modeling and prediction of complex multidimensional interrelated processes in the class of vector autoregression models are developed. Conclusion. The scheme of paralleling for recurrent COMBI algorithm allows to solve the problem when arguments amount exceeds capability of scheme with the exhaustive search. The effectiveness of the constructed algorithm is demonstrated by prediction of the interrelated processes in the field of investment activity of Ukraine with the purpose of information support of administrative decisions.