A novel approach to sand production prediction using artificial intelligence

Highlights

• A comprehensive statistical approach is presented for sanding onset prediction.

• Four parameters are identified as the most affecting parameters on the onset of sanding.

• MLR model is simple and GA doesn’t significantly improve the performance.

• ANN detects the complex relationships between CTD and all affecting parameters.

• PSO algorithm improves the performance of ANN in comparison with BP algorithm.

Abstract

Over the years, accurate and early prediction of oil or gas well sanding potential has been of great importance in order to design an effective sand control management strategy. Significant technical and economic benefits can be achieved if the correct and early design of sand control method is considered. In this study, critical total drawdown (CTD) as an index of sand production onset is aimed to be estimated through 4 proposed methods. A total of 23 field data sets collected from problematic wells of North Adriatic Sea were used to develop these models. First, simple regression analysis was performed to recognize the statistically important affecting parameters. Using these variables, multiple linear regression (MLR) and genetic algorithm evolved MLR (GA-MLR) were developed for estimation of CTD. Two artificial neural networks (ANN) with back propagation (BP) and particle swarm optimization (PSO) algorithms were constructed to correlate CTD to all affecting parameters extracted from the literature. The performance comparison showed that the artificial intelligent system could be employed successfully in sanding onset prediction and minimizing the uncertainties. More accurate results were obtained when PSO algorithm was applied to optimize the weights and thresholds of neural network.