Evaluation of chemical composition of waters associated with petroleum production using Kohonen neural networks

Abstract

Chemical composition assessment of produced water in oil wells from the same producing zones was performed using Kohonen neural networks, a pattern recognition technique based on neural networks. Produced water samples from three production zones were characterized according to their levels of salinity and calcium, magnesium, strontium, barium and sulphate (mg/L) contents. Data were normalized by variance before analysis, and Kohonen maps were generated using hexagonal structure, planar shape, and training algorithms for each batch.The analysis with Kohonen neural networks allowed assessing the chemical profile of each production zone, and identifying the formation of clusters related to the individual oil wells, as well as patterns related to seasonality. Production Zone 1 revealed the presence of two distinct sample populations associated to the different oil wells from which samples originated as from two different reservoirs. Production Zone 2 presented a homogeneous cluster of samples from the same oil well, and Production Zone 3 revealed five samples clusters constituted by samples from five different oil wells from the same reservoir. It was also possible to identify samples with anomalous behavior and characterize them according to the contents of the variables involved.