Research On Drill Bit Selection Based On Deep Learning

In the field of petroleum drilling,as an indispensable tool,the use effect of drilling bits in the process of drilling directly affects the efficiency and economic benefits of petroleum exploitation.At present,the selection of drill bit mostly depends on experience,and the quality of drilling staff is very high,and the selection effect is not guaranteed.With the emergence of logging while drilling technology,artificial intelligence technology has gradually entered the field of drilling.Among them,in the selection of drill bit,the shallow learning algorithm in artificial intelligence is used to study,which has some problems,such as poor learning ability,easy to over-fit,gradient easy to diverge and so on.The emergence of in-depth learning can solve these problems very well.At present,the application of deep learning network in data prediction and classification is quite mature,and bit selection can be considered as a classification problem.Therefore,it is necessary and feasible to introduce deep learning into drilling field and establish a mathematical model for bit selection.In this paper,the mathematical relationship between rock mechanics parameters and bit type is established by using field logging data.Rock drillability is an important parameter in rock mechanics parameters,so this paper first studies rock drillability.Through the analysis of previous studies,it can be found that there is a certain relationship between acoustic time difference,density,gamma value,resistivity and rock drillability,and most of them are studied by shallow neural network.For the non-linear and multi-dimensional problems,it is difficult to obtain accurate results by using shallow neural networks when the amount of data increases.Therefore,this paper is based on the drilling data obtained.With acoustic time difference,density,gamma value and resistivity as input and rock drillability as output,the key parameters of the network are determined.Finally,the depth learning model of rock drillability is established.Compared with the neural network,the performance of the network is better,and the prediction accuracy can reach more than 90%,which is about 10%higher than that of the neural network.Before drill bit selection,in order to ensure the reliability of field data,the data are optimized.Firstly,the strata are numbered according to the bit usage.The rock similarity is evaluated by rock shear strength,compressive strength,internal friction angle,shale content and rock drillability in rock mechanics parameters.The similarity of each numbered strata is compared by using the fuzzy similarity theory.Then,the strata with more than 80%similarity are taken as a class,and the mechanical specific energy theory of Fanhonghai is used to calculate the mechanical specific energy between different similar strata,and the strata with smaller mechanical specific energy are selected as the learning data of bit selection,so as to achieve the purpose of data screening.The screened data were sorted out into database files,and 36 strata were screened out by processing 126 strata.Using the selected data,taking rock mechanics characteristic parameters as input and bit type as output,a mathematical model of bit selection based on depth learning is established,and compared with the model established by neural network,its prediction accuracy is about 13%higher than that of neural network,and its network performance is also better.Using Python language based on Tensorflow to compile the mathematical model of Rock Drillability Prediction and bit selection is difficult to operate and understand.Therefore,this paper combines the development of PyQt and Qt designer software to develop a set of bit selection software.Combining the functions of bit drillability prediction,data screening and bit selection,the software is developed by connecting Excel,Spyder,MATLAB and other software,as well as a variety of internal GUI packages.The software has the ability to use database files to train models,obtain mathematical models and predict data.Moreover,it can set the parameters of the model,obtain better structural parameters of the mathematical model,and visualize the performance of the model,which makes it easier for operators to see the reliability of the obtained model.In order to verify the reliability of the obtained mathematical model,this paper uses data outside the database to verify the model.The results of rock drillability and bit selection obtained by the software are in good agreement with the actual situation in the field,which indicates that the model has certain engineering significance and can be applied to practice.