First-Order Fuzzy Time Series based on Frequency Density Partitioning for Forecasting Production of Petroleum

Abstract

Forecasting method based on fuzzy time series has been widely developed in recent years. In this paper, we propose a new improvement at determining universe of discourse, variation historical data and partitioning stage. At early stage, we define the universe of discourse then calculate the basis value to find out how much interval should be used with variatin historical data. Secondly, we are partitioning the main intervals into several numbers of sub-intervals. The empirical analysis shows that sub-interval caused the fuzzy number getting closer to crisp value. It causes the better forecasting value. We use the data of yearly production petrolium Indonesia for simulation. We compare the forecasting results and error value of the method with previous existing methods The modifications give better forecasting results than previous methods indicated with smaller The Means Squared Error (MSE) and Average Forecasting Error (AFER).