The Study Of Fractured Horizontal Well Performance Of Volcanic Reservoir Based On Machine Learning

Recently the newly discovered reverse of volcanic reservoir is increasing rapidly within the worldwide and volcanic reservoirs have grown in China.The exploration and exploitation of volcanic reservoir have presented a bright future.However,the key factors that control the well performance are not identified and there is no good method to forecast well performance due to the double porosity.The lack of the guiding theory leads to an unsatisfactory development effect in oilfield.It is necessary to clarify the controlling factors and establish a set of method to predict well performance for volcanic reservoir.This study investigates the fractured well performance of volcanic reservoir from data perspective where the machine learning,numerical simulation,and hydrocarbon seepage mechanism are combined.Firstly,the basic knowledge of machine learning is first introduced.The development features of volcanic reservoir are summarized and the numerical simulation model is established on the basis of the data from studied area.Then the sensitivity studies are made to investigate the influence on the well performance of single factor.The controlling factors of fractured horizontal well for volcanic reservoir are identified using the Grey Correlation Analysis according to orthogonal experiment data set.Next the initial production models are proposed with Multiple Linear Regression,Random Forest,and Support Vector Machine.Then the comparison is made with the new designed testing set.Finally,time-series well performance forecasting models are proposed with decline curve analysis and neural networks for the production curve generated from numerical simulation and production curve from practical oilfield respectively.The results show that:(1)The well performance of volcanic reservoir decide on the initial production and the capacity to keep the production stable.A high initial production and good capacity to keep production stable give an excellent well performance.The key factors that influence the initial production are the density of natural fracture,well length,the stages of fracturing,the height from horizontal well to bottom water,and the inclination of natural fracture.The ability to keep production stable relies on the porosity of matrix and fracture,net to gross ratio,and the scale of the bottom water;(2)The initial production forecasting model based on the RF and SVM are better than the conventional MLR and the SVM model is the with the highest accuracy.The well performance can be basically evaluated with the combination of the predicted initial production and the decline regulation of the development area;(3)The time sequence forecasting method ARIMA and LSTM both give reliable results for the ideal production variation.When it come to the production with complex variation mode,the LSMT model outperform others.