production forecasting of oil wells based on production history using multi-scale time averaging and lstm networks

One of the most important issues in the oil and gas industry is predicting the future production of a reservoir. although the numerical simulation of the reservoir is the most effective method to predict the production, but implementation this requires a static and dynamic model of the reservoir, which requires a lot of cost, energy, and time. the aim of this paper is to provide a method for predicting the oil production of a reservoir using machine learning and time series forecasting algorithms. these methods have a good yield for time series data and whereas the production from oil reservoirs is somehow dependent on time, these methods can be used to predict the future production of wells. nowadays, due to the ever-increasing growth of computer software and significant progress in the measurement of pressure inside the well using permanent pressure gauges that are installed inside the well, many amounts of flow rate, pressure and temperature data are obtained that can be used to predict the future of production. in this research, we have used lstm neural network model and two different sub-models (basic model and multi time scale model). in this method, at first a part of the past production data was used to train the artificial intelligence model, then it started to predict and by using other real data, it checked the accuracy of the obtained data and if the prediction was not accurate enough, it was optimized and adjusted artificial intelligence algorithms. finally, the prediction, compare and analyze of results obtained from the trained artificial intelligence models was done, and their accuracy and performance were also verified; we found that the multi time scale model is a more suitable method for predicting the production process in the long term with less error.