Research And Implementation Of Logging Curve Completion Algorithm Based On Deep Learning

The study of logging curve completion is one of the fundamental studies in the field of petroleum engineering.The complete and effective logging curves can reflect physical parameters such as lithology,porosity,saturation and permeability of the strata under the oil and gas field.However,inevitable situations such as logging instrument failure can occur during actual well logging,resulting in the loss or distortion of some logging curves,and it is difficult to re-logging the well.Therefore,it is especially critical to study how to complete the missing logging curves without consuming additional economic.With the rapid development of deep learning,using deep learning algorithms to complete the missing logging curves has become a hot research topic.In this thesis,the CNN-AB-BiGRU and RES-AB-BiGRU logging curve completion algorithms are proposed to solve the problem that the GRU model cannot consider both the local details of the logging curves and the trend of the logging curves with depth,and does not consider the interaction between the logging curves data points.The main contents of this thesis are as follows.(1)The logging curve completion algorithm based on GRU is designed and implemented.There are many algorithms for completing the logging curve.First,based on the characteristics of the logging curves,study the basic model in traditional machine learning and deep learning algorithms.After the theoretical analysis,GRU is the most suitable model,it can fully use the timing information of the logging curves.Experimental compared with RF,SVR,BPNN,1DCNN,RNN,LSTM model and quantitative evaluation.The completion effect of the GRU model is the best,so the GRU model is selected as the basic model for the subsequent study.(2)The logging curve completion algorithm based on CNN-AB-BiGRU is studied and implemented.The basic GRU model has certain shortcomings,it can only use the characteristics of the unidirectional effects between the data points of the logging curves,and it does not fully use the characteristics of the logging curves.Therefore,the CNN-AB-BiGRU model is constructed.This model combines the local feature extraction property of CNN,the feature weight assignment property of Attention Mechanism,and the long-term information memory property and the bidirectional information capture property of BiGRU.Experimental compared with GRU,BiGRU,CNN-BiGRU model and quantitative evaluation.The completion effect of the CNN-AB-BiGRU model is the best.In addition,the comparison experiments of the four models have proved that the CNN,the BiGRU,and the Attention Mechanism can further improve the completion accuracy of the model.(3)The logging curve completion algorithm based on RES-AB-BiGRU is studied and implemented.Considering that only the basic CNN is used in the CNN-AB-BiGRU model.Res Net is a kind of CNN,which can avoid gradient disappearing to a certain extent.In addition,the direct mapping structure of Res Net can ensure the logging curve information does not lose with the increase of convolution.To a certain extent,it ensures the integrity of the logging curve information,which is more conducive to the subsequent mapping process.Therefore,Res Net is used to extract the local detail features of the logging curves and construct the RES-AB-BiGRU model.This model combines the local feature extraction property of Res Net,the feature weight assignment property of Attention Mechanism,and the long-term information memory property and bidirectional information capture property of BiGRU.The experimental results show that compared with the RF,the CNN-AB-BiGRU and the TCN models,the RES-AB-BiGRU model can most accurately complete the different logging curves of different wells.The comprehensive experimental results show that,for the problem of logging curve completion,compared with other methods,the logging curve completion algorithm based on CNN-AB-BiGRU and the logging curve completion algorithm based on RES-AB-BiGRU proposed in this thesis can complete the missing logging curves more accurately on the one hand;on the other hand,the completion cost is greatly reduced while the completion accuracy is guaranteed.It can provide a new idea for oil and gas field researchers to complete the logging curves.