A Comparative Analysis of Multivariate and Univariate Time-Series Forecasting for Oil Production

Production networks and processing facilities can be very complex, with multiple interactions and constraints. It is become a problem whenever the "helicopter view" of the entire production profile, or composite information of the field is required, moreover to see what will happen in the next upcoming time frame. This research presents a study of forecasting techniques for oil production. The goal of the study is to evaluate the performance of different forecasting methods and identify the most effective approaches for predicting future total oil production. The study includes a review of relevant literature, a description of the data and methods used, and an analysis of the results. This study compared two methods for forecasting time-series data: Simple Moving Average (SMA), Seasonal Autoregressive Integrated Moving Average with exogenous variable (SARIMAX), and multivariate time-series forecasting using Light Gradient Boosting Machine (LightGBM). Multivariate time series forecasting, especially with LightGBM, was found to be highly effective for predicting oil production data. LightGBM showed similar performance to SMA in terms of Mean Absolute Error (MAE) score, with MAE of 3748.71, compared to SMA's MAE of 3853.04. This research also revealed that SARIMAX has the highest MAE value 5280.83. The oilfield total production chart signature, and along with its forecast, seems producing the oil asymptotically since 2021, unless there is any significant intervention, such as discovery of huge oil reserve or on the other hand, a notable fundamental event that accelerate the decline to another low level.