Prediction of Maximum Oil Production by Gas Lift in an Iranian Field Using Auto-Designed Neural Network

Artificial neural networks have been becoming increasingly popular in oil industry over the last decades. But there was not a specific framework and procedure to design appropriate networks in respect to the problem. One of drawbacks of neural network application is its dependence on designer’s experience. In this work we proposed a method in which we design an artificial neural network coupling it with a genetic algorithm to not only optimize weights and biases but also number of neurons and connections. This method can be used to design complex systems in which time and simplicity are important factors as we used it in predicting gas lift aided recovery to obviate the need to run simulation software which is expensive and time consuming. First we create a network with neural network toolbox of MATLAB. This network was built fully-connected. Then we start our program with this initial guess and compare the final structure and mean square error (MSE) with the network created by MATLAB. The network obtained by our program was simpler and also it has lower MSE indicating a network that is simpler and more accurate.