Inferência do teor de óleos e graxas em água produzida via rede convolutional LSTM

Abstract

Offshore oil and gas production units face the challenge of treating and disposing of produced water in accordance with regulationsand established limits. The criticalparameter in this treatment process is the Oil and Grease Content (TOG) present in the water to be discharged. However, the official TOG value is only made available approximately 20 days after the discharge. Therefore, this study focuses on evaluating two neural network models that utilize Convolutional Long ShortTerm Memory (ConvLSTM) to estimate the TOG value based on process variables, laboratory analyses, and other relevant data. Due to the dynamic nature of the process, data windows of the previous 48 hours are considered for estimating the TOG value. The results obtained demonstrate that the proposed models perform comparably to the models mentioned in the literature and outperform simpler models in TOG estimation. These results reinforce the feasibility of using recurrent neural network methods in the industry, enabling the implementation of an online sensor capable of estimating TOG and assisting platform operators in making decisions regarding the continued disposal of water into the sea.