-
Karina dos Reis Teixeira
Petróleo Brasileiro S.A., ES; Departamento de Engenharia Elétrica, Universidade Federal do Espírito Santo, ES
-
José Marques Oliveira Júnior
Petróleo Brasileiro S.A., ES
-
Ricardo Emanuel Vaz Vargas
Petróleo Brasileiro S.A., ES
-
Patrick Marques Ciarelli
Departamento de Engenharia Elétrica, Universidade Federal do Espírito Santo, ES
-
Celso J. Munaro
Departamento de Engenharia Elétrica, Universidade Federal do Espírito Santo, ES
Keywords:
Total oil and grease, Recurrent neural networks, ConvLSTM, Soft sensor, Fault detection, Process monitoring
Abstract
Offshore oil and gas production units need to treat and dispose their produced water respecting the limits and reference measurement methods imposed by current laws. The main index to be controlled of this treatment process is the Total Oil and Grease (TOG) in the water to be discarded, which has its official value available for only about 20 days after the disposal is carried out. This work evaluates two neural network models with Long Short-Term Memory convolutions to estimate the TOG value from process variables, laboratory analysis and other data. Due to the dynamics of the process, data windows of 48 hours are used before the TOG value to be estimated. The results obtained indicate that the proposed models are as good as those presented in the literature, and better than simpler models to estimate the TOG value. These results corroborate the feasibility of using methods based on recurrent neural networks in the industry as a means to implement an online sensor capable of estimating the TOG and assisting in the decision-making of a platform operators regarding the continuity of water disposal to the sea.
