Recognition Of High Production Area Of Coalbed Methane Based On Symbolization And Deep Neural Network

A large number of coalbed methane wells have entered the stage of large-scale development.But in the actual production process,many wells have the problems of low gas production,fast attenuation and difficult to establish a stable drainage and production mechanism,which poses a great challenge to the process of coalbed methane industrialization and affects the economic benefits of the whole industry.However,due to the soft rock characteristics of coal reservoir,it is sensitive to effective stress.Even similar construction and production conditions will appear great differences in coalbed methane production performance in adjacent areas.In order to discover the causes of large differences in gas production performance and explore its formation law and genetic mechanism,it is necessary to analyze the important processes involved in the whole life cycle of coalbed methane development engineering,such as spatial location selection,drilling and completion,fracturing and production.At present,most of the researchers’ studies on the characteristics of drainage and mining are based on single-stage and fewer geological parameters.On the one hand,It’s unable to effectively obtain and share data because there are many parameters caused by the development of multi-stage characteristics and complex geological causes.On the other hand,due to the sensitivity of the current traditional mathematical methods to parameter dimensions,it is unable to effectively analyze all the data in multi-stage development.However,with the development of big data and artificial intelligence,deep learning has a more mature technology accumulation in the field of computer vision,and is accelerating the landing of integration with various industries.Combined with deep learning,the study of coalbed methane development multi-stage project based on geographic information will break the limitation of parameter dimension and realize the treatment of coalbed methane engineering with strong heterogeneity.This paper combines parameter symbolization,deep network and the integration of various stages of the project parameters for the analysis of the data.As a result,the variables are symbolized to form a template to change the utilization of parameter data.Then establish and optimize deep network to identify the high yield of coalbed gas wells and explore the formation law and parameter sensitivity of different gas production forms.And explore the application of deep learning in coalbed methane engineering.The main contents of this paper are as follows:1.Firstly,this paper summarizes the progress of current research on gas production law and mechanism,and also summarizes the problem that multi-stage characteristics and high-dimensional parameters can’t be taken into account in the current research on gas production difference characteristics and mechanism of coalbed gas wells.The current mathematical methods are sensitive to parameter dimensions and can’t make full use of and share data in different formats.Few studies use deep learning combined with symbolic image template to explore the influencing factors and parameter sensitivity of coalbed methane production.As a result,the learned features are used to identify and analyze high coalbed methane production areas.2.Secondly,this paper forms a complete symbolization process from the numerical analysis of symbolic phase space to the construction of symbolic database,then to rapid mapping and access display based on template.Symbolic mapping rules for two kinds of data are established.The first category is discrete data symbolization,which can be divided into discrete spatial data symbolization and discrete attribute data symbolization.The second category is continuous time series data symbolization.Based on previous studies on various stages of coalbed methane engineering,a total of15 kinds of discrete parameter data(including spatial location information and geological structure information)and a kind of continuous time series parameter data are selected and preprocessed as the symbolic data basis.All data are automatically cleaned and processed by Arc GIS and Python programming.A unique symbolic image template is formed,which takes into account not only the spatial location attributes of coalbed gas wells and geological structures,but also the influence factors of gas production characteristics in each stage.This not only makes up for the lack of previous studies focusing on the single stage of coalbed methane engineering,but also alleviates the lack of sensitivity of traditional mathematical methods to parameter dimensions.At the same time,a set of convenient methods for parameter information summary and data sharing in different formats are established.To some extent,the symbolic image template can reduce the resolution of the sensitive data involved,and share and spread after decryption.3.Finally,on the basis of established sharing database of the selected coalbed methane engineering parameter,the fast automatic symbolic mapping based on template rules is realized and the data can be shared.Fanzhuang-Zhengzhuang block is selected as the study area to establish the recognition model of high coalbed methane producing areas.The symbolized mapping results are marked with high,middle and low gas production,and the model is trained and verified.By means of data enhancement,Dropout and regularization strategy,and model super-parameter adjustment such as learning rate and batch size,the performance and accuracy of network discrimination are maximized.The experimental results show that when the initial learning rate is 0.001 and dropout is set to 0.5,the adaptive algorithm Adam will perform better.The accuracy of Top1 can reach 70.84%.The specific recognition process of the model is analyzed through the visualization of feature images.By using the method of generating class activation response degree diagram,the identification response and sensitivity of coalbed methane production form to engineering multi-stage parameters are analyzed.To a certain extent,it can provide help for adjusting and dispatching later construction schemes and methods when some parameters in the early stage of the project are known.In the end,the application analysis software for automatic identification of high production areas is established,which provides basic GIS functions,access and modification of parameter symbolization shared database.Through the GUI interface,it is convenient to use the coalbed methane engineering identification,analysis and application scheme established.