СРАВНИТЕЛЬНЫЙ АНАЛИЗ РЕКУРРЕНТНЫХ НЕЙРОННЫХ СЕТЕЙ И МОДЕЛИ АВТОРЕГРЕССИИ ARIMA ПРИ ПРОГНОЗИРОВАНИИ НЕСТАЦИОНАРНЫХ ВРЕМЕННЫХ РЯДОВ

Abstract

А comparative analysis of the recurrent neural network model and the ARIMA autoregression model was carried out to predict light fractions output produced by the crude and vacuum distillation unit. There is a mathematical description of these models. There are the implementation of models using Keras and Pmdarima libraries in Python. A series of experiments was carried out, values of K-2 column cube temperature, crude oil consumption and gasoline fraction consumption were used as input data. The conclusion is made about the superiority of prediction quality of neural networks over ARIMA.