ABSTRACT:

With the increasing demand for oil and gas resources, the depth of some development wells has exceeded 8000m. The mechanical properties of reservoir rocks have been changing from elasticity to Plasticity under high temperature and high stress, and the existing constitutive equations no longer can meet the needs of modeling. As a new sequence data analysis technology, Recurrent Neural Network (RNN) can fit the complex logical and nonlinear relations in the sequence data, and can predict the next data. In this study, 72 groups of rock stress-strain curve data in the target block are taken as the training data of Long Short-Term Memory (LSTM), and the final model after training can predict the stress-strain curve under the condition where the temperature is 25-200°C and closed stress is 0-11000psi. 9 groups of new curves are used as test data, and the validity of LSTM model is proved by comparing the predicted data with the real data. In this paper, Recurrent Neural Network is applied to the experiment of rock mechanics, and a new method for predicting the constitutive relation of rock in target area is obtained, which has high application value.

1. Introduction

With the increasing demand for oil and gas resources, as well as the development and progress of oil and gas exploration, the depth of wells in China has exceeded 8,000 m. It's necessary to make rock mechanical analysis for any wells. However, due to the depth of such wells, the bottom stress, temperature, porosity, permeability and saturation are also different from ordinary wells. Under the high temperature and high stress, the mechanical properties of reservoir rocks have transformed from elasticity into plasticity and the traditional constitutive equations has no longer meet the needs of this model establishment.

As a kind of geological material, heterogeneity is one of the basic characteristics of rock. The heterogeneity has a great influence on the mechanical properties of rock and its mechanical behavior under load. In the experimental study of this paper, it is also found that the heterogeneity of rock has a significant influence on the rock mechanics under high temperature and high pressure, which makes the regularity obtained in the analysis of rock constitutive relations not obvious. Recurrent Neural Network (RNN), as an emerging technology of sequence data analysis, is capable of fitting the complex logical and nonlinear relationships in sequence data and making data prediction at the next moment. Through theory and analysis of experimental data, this paper deeply studies the influence of rock heterogeneity on rock mechanical properties under high temperature and high stress, and uses recurrent neural network to analyze the influence of rock heterogeneity on rock mechanical properties under external conditions (temperature, stress and pore pressure) This is beneficial to the development of ultra-deep wells in Tarim oilfield.

This content is only available via PDF.
You can access this article if you purchase or spend a download.