Time Series Forecasting: Applications to the Upstream Oil and Gas Supply Chain

Abstract

This paper describes different models which are used for forecasting in the time series context of petroleum engineering. The objective is to reproduce and further predict future oil production in different scenarios in an adjustable time window. Such time series are very similar to those from the sequential manufacturing processes which are usual in many areas of manufacturing industries. We mainly focus on a feedforward neural network model and a Gamma classifier and compare them both on a benchmark and real industrial data under univariate and multivariate settings. While the former model has become recently a standard tool for modeling and prediction, time series forecasting is not the kind of tasks envisioned while designing and developing the Gamma model. The Gamma classifier is inspired on the Alpha-Beta associative memories, taking the alpha and beta operators as basis for the gamma operator. As experimental results show, pattern recognition based classifier shows very competitive performance. The advantages and limitations of each model are discussed.