چكيده به لاتين
Measuring multi phase flow is a very important parameter in the oil industry. Therefore, various methods have been proposed to determine the flow and measure the flow of fluids, including laboratory methods and experimental approaches. Due to the high operating costs of such methods, researchers have always been looking for methods with lower operating costs and acceptable computational accuracy. Nowadays, because of the development of computing software and smart tools, this tool is used to model complex engineering systems. Furthermore, these can also be used to predict the measurement of multi-phase currents. In this work, we tried to use smart models and deep learning networks that have been proposed recently as a tool to predict the measurement of multiphase flows in the oil industry. The used data include two categories of data, the first and the second, related to one of iran's oil fields and one of the australian oil fields respectively. After collecting the data, first the preprocessing involves normalizing the data and also removing the outlier data on them, and then randomly, in all developed models, 70% of the data is selected as the training data and they are used in the model development process. The rest are used to evaluate the models as test data. To develop all the models, several parameters have been studied. In order to optimize some of the model parameters, adaptive gradient optimization methods such as adam's algorithm, which has recently been used in deep networks, have been used and a comparison has been made between them. Finally, given the data of iran's oil field, a deep correction model is presented, which estimates the total flow of 8 oil wells with high accuracy. Also in the second series of data, a deep neural network model was developed in which the oil, water, and gas flows of four oil wells were estimated separately with high accuracy.
