ARTIFICIAL NEURAL NETWORK BASED IDENTIFICATION OF THE GAS VOLUME FRACTION IN AN ELECTRICAL SUBMERSIBLE PUMP

摘要

The electrical submersible pump (ESP) under multiphasernflow is very common in the oil industry. These pumpsrnpresent frequent premature failures when the gas flow isrnhigh. In addition to this, a further increase of the gas mayrnfill most of the pump impeller, making the flow rate to decreaserndown to zero, known as gas locking. Due to lack ofrninformation and mathematical models that can be used inrnreal time for this type of pumps, experimental studies arernusual in this area. This paper applies artificial neural networkrn(ANN) modeling for the volume gas fraction identificationrnin ESP. The algorithm uses experimental data collectedrndirectly from the system for different gas fractions,rnsuch as pressure, flow rate, mechanical torque, elevation,rnetc. This model uses a back propagation learning algorithmrnand multi-layer perceptron neural network, where differentrnstructures are analyzed to find the optimal number of hiddenrnlayers. Results show that the system is able to identifyrnthe volume gas fraction in the pump with a very good accuracy.