Research On Machine Learning Based Prediction Algorithm Of Reservoir And Production For Wells In SD Gas Field

Oil-gas related researchers have proposed plentiful methods to extract useful information from production data of oil and gas field.These methods are generally based on reservoir engineering principles and have achived high accuracy,but most of them require a large amount of man-hours,economical expenses,and strict requirements on the type and accuracy of input data.Thanks to the rapid development of computer speed and the surprising advancement of machine learning technology,big data-based solutions and predictive models have been widely recognized and applied in the industry.In today's era of automation and artificial intelligence,some problems in oil and gas fields can be solved by machine learning models derived from extensive data training.Although machine learning technology has been widely used in many fields of petroleum engineering,few studies have focused on gas well production data,most of which use logging geologic parameters and fracturing construction parameters to give reserves or capacity forecasts.This paper attempts to combine the gas field production data that has not been effectively utilized with the machine learning method to achieve automated and accurate calculation of dynamic reserves and future daily production.Through the research in this paper,the following research work and results is mainly completed and acquired:1.Established the use of shut-in wellhead pressure to calculate the static pressure at the bottom of the well,to draw the balance curve of the converted material,and proposed a method for transforming and optimizing the nonlinear conversion material balance curve.2.Using machine learning technology and computer programming,a linear/nonlinear classifier for calculating the balance curve of the material is established.Based on this,a fully automatic process and method for predicting dynamic reserves using the closed-circuit conversion material balance curve in the SD gas field is established,and Python is used.The programming language has been implemented.The autocorrelation of the time series of gas production in the gas wells of SD gas field was studied by using the autoregressive moving average model.4.Using the deep-circulating neural network based on long-term and short-term memory to realize the accurate prediction of the gas production in the future,combined with the autoregressive moving average model to help it adjust the super-parameters,establish a prediction system for the gas production of open wells,and use The Python programming language has been implemented.