Petroleum Production Forecasting Using Machine Learning Algorithms

Abstract

Field forecasting and predictive production analysis has tremendous effects on assets planning and allocation and it cannot be over emphasized. For accurate actions to be taken, appropriate forecasts need to be made on each asset, hence the need to develop methods to aid the process. As a tilt from the conventional methodology of forecasting involving use of curve fitting techniques, and multi-level computational analysis, data driven approaches can be employed. This study presents the applications of data driven approaches to forecast production. Deep learning neural network algorithm and statistics- based data driven approach were considered. An LSTM model was developed and for the statistical algorithms, an ARIMA model, and a Holt Winters model was developed. The models were deployed, and the performance of the models were checked to determine more accurate approach for forecasting. Error analysis on the results form the models showed that the deep learning neural network model provided better results in comparison to the statistical models with an MAE of 0.0328. Based on the model performances, LSTM model can be considered for use in forecasting petroleum production overcoming effects of seasonal changes, and production anomalies in the life of the reservoir.

Mathematical Modeling of Motion of the Solid Particles in the Filter-Coalescer Wastewater Treatment

“…In most cases, the operation wastewater form lowconcentration suspension containing fine particles with dimensions of 0.1-15 micrometer. It may be said that the wastewater of many enterprises forms an aggregative stable system [10][11][12][13][14][15]. The particles with dimensions of 0.1-5 micrometers and less filtrate with little to no and remaining in suspension state are taken by water flow to extended distance, they are deposited on the internal surface of the aggregates, and it has a negative impact on their working capacity.…”