Predição de Tempo de Ciclo Utilizando abordagens de Séries Temporais

Abstract

The development of a prioritization agent and statistical analysis play fundamental roles in the search for optimizing cycle time and improve production efficiency, making it possible to better anticipate deviations in cycle time and identify intervention opportunities in advance. Analyzing the cycle time problem by a time series approach, prediction models were developed that would allow predicting the cycle time value at a future instant. Three approaches were implemented in this study, based on related work: an ARIMA model, a hybrid model consisting of ARIMA + MLP, and an LSTM recurrent network. The results suggest that the three approaches have good results when applied to the cycle time series and more accurate results are found when approaches that model the linear and nonlinear components of the time series are combined.