Productivity Regularity Research Of Fractured Well In Tight Gas Reservoir

With the rapid development of the economy,demand for natural gas resources is increasing,conventional natural gas have been behind the pace of economic development,more and more countries pay attention to the tight gas.However,the characteristics of tight gas are different from conventional gas,which brings a lot of problems to mining and using.For difficulties and deficiencies in tight gas reservoir fracturing well productivity predicting on site,the article researches the productivity prediction method of tight gas reservoir fractured well.Based on the study of geological features and stimulation of tight gas reservoirs,this article establishes a sample set of tight gas reservoir fracturing well productivity influence factors.This is mainly based on the screening principles and determines the effect degree of factors by qualitative and quantitative analysis.Adopt the way of numerical simulation to take a qualitative analysis for reducing the complexity of quantitative analysis.Quantitative analysis adopts orthogonal experiment method and gray relative analysis method to obtain the specific impact of each factor.Based on this set of indicators,the article improves the production predicting model of the tight gas reservoir fractured wells.Adopt the BP neural networks and support vector machine to predict tight gas reservoir fractured well,and compare them with the analytical formula.Then adopt the two methods and GM(1,1)model to predict the yield variation of a fractured gas well after the production.Compile the tight gas reservoir fracturing well productivity prediction software,and analyze a well of the gas reservoir.By the analysis of a well of the gas reservoir,the accuracy rate of mathematical and statistical methods is better than formula methods when predicting tight gas reservoir fractured well productivity,and the support vector machines has the highest.When predicting the yield variation of a fractured gas well after the production,the support vector machine has the highest accuracy rate.Support vector machine does not require large amounts of sample data,but with a better prediction.