Loading...

Accelerating Flash Calculations in Compositional Simulators Using Machine Learning Algorithms

Asadian, Amir Hossein | 2022

97 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55687 (06)
  4. University: Sharif University of Technology
  5. Department: Chemical and Petroleum Engineering
  6. Advisor(s): Pishvaie, Mahmoud Reza
  7. Abstract:
  8. Flash calculations using equations of state are the basis of compositional simulators and inaccuracy in these calculations leads to errors in some time steps or even complete failure of a simulation. One of the problems of this approach is the use of iterative loops, which leads to a high computational cost. The purpose of this research is to use deep machine learning methods, namely multi-layer perceptrons and convolutional neural networks to solve this problem. At first, we defined two 5-component and 10-component hydrocarbon fluid models and used one of the conventional methods of flash calculations (Modified Successive Substitution method) in order to generate the data required for neural network training. In the following, we trained neural networks for both fluid models and also investigated the effect of different parameters on network training and obtained networks with acceptable accuracy, in such a way that the coefficient of determination (R-Squared) for 5-components model in multilayer perceptron networks and convolutional neural networks is equal to 0.98637 and 0.98432 and for 10 component model it is equal to 0.97705 and 0.9817, respectively. Also, by increasing the training data for 10-components model, we achieved higher accuracy. In this case, the coefficient of determination is 0.98209 for the multilayer perceptron method and 0.98515 for the convolutional neural network. Finally, the results of neural networks for 10-components model were used as the initial guess of the conventional flash calculation and it was observed that by using the results of multi-layer perceptron and convolutional neural network, the speed of calculations increased by 1.29 and 2.55 times, respectively
  9. Keywords:
  10. Flash Calculation ; Machine Learning ; Artificial Neural Network ; Convolutional Neural Network ; Multi-Layer Perceptron (MLP)

 Digital Object List

 Bookmark

No TOC