The Study Of Sand Prediction Method For Ultra-deep Fractured Sandstone Reservoir In Keshen Block

The Keshen block is one of the important gas sources of the West-East Gas Transmission,and it is also an important production area of ultra-deep and tight sandstone gas in China.Ultra-deep reservoirs are dense,with low matrix permeability and developed filled fractures.In some wells,sand production was serious during the production process,and even sand burying or even scrapping occurred,which severely restricts safe and efficient production.Unlike traditional weakly cemented reservoirs,the Keshen block reservoir rocks are highly intensified by high temperature,high stress,and high pore pressures,so conventional sand production prediction models can't explain sand production in the reservoirs.There is less research on sand production prediction methods for associated ultra-deep and tight reservoirs.Therefore,it is important to systematically carry out research on sand production prediction models for the reservoirs.This paper identifies the key factors of sand production,and establishes a database of sand influencing factors by analyzing the engineering geological characteristics of reservoir.Based on data mining and analysis,a sand prediction model was established through BP neural network and field verification was carried out.The main research content is as follows:(1)Through rock mechanics experiments and logging data calculations,the basic parameters of rock mechanics parameters and ground stress in the Keshen block were obtained.It was found that the rock failure in the formation was dominated by shear failure,and a rock damage correction suitable for the area was established.Guidelines;(2)Through statistical analysis and theoretical calculation,we analyzed the influence of various factors on sand production,and established a database of sand influencing factors.According to the characteristics of sand production in Keshen area,we selected the BP neural network prediction method for sand production;(3)The sand prediction network with an error of 0.013957 was obtained through engineering data training.The predicted results are in good agreement with actual sand production.And on-site verification,the prediction results show that the sand prediction effect is good.