Neural Network as a Tool for Predicting and Controlling the Technological Regime of Production Wells

Abstract

Abstract The article states the possibility of forecasting the operating modes of production wells through artificial neural networks simulation. The main difficulty in medium- and long-term forecasting is to take into account the influence of adjoining wells and the performance pattern of formations. The flow simulation considers these factors and can be used to teach the artificial neural networks together with the real field data on well operation. The present research paper dwells upon the ways of forming training samples for the artificial neural networks as well as the quality of forecasting operating modes of wells. Training and forecasting of modes was carried out on the data of computational fluid dynamics for the patch with three production and one injection gas well. We compared two types of artificial neural network forecasting capabilities: MLP and LSTM architecture of the recurrent type. We illustrated how qualitative training sample allows forecasting operating modes of well operation with high accuracy over 10 years.