Research On Well Test And Logging Data Interpretation Method Based On Deep Learning

In the oilfield development,well testing and logging are the important basis to grasp the oil and gas development performance.Well testing understand the wellbore and formation parameters through flow characteristics,while well logging obtain the information around the wellbore.The detailed interpretation of Well testing and logging data is one of the important basis for the adjustment and optimization of reservoir development plan.Traditional manual interpretation method has strong subjectivity,high cost and low efficiency.Therefore,this dissertation applies the deep learning method to well test analysis and logging interpretation in oilfield,aiming to realize automatic interpretation of well test and logging data,improve the efficiency of well test analysis and the accuracy of logging interpretation,and promote the development of intelligent oilfield.The innovation points and main research work of this dissertation include two parts:1.Based on radial basis function(RBF)neural network,a well test analysis method of automatic parameter inversion is proposed.Compared with the traditional polynomial fitting method,this method has better fitting results and higher precision.Even when the polynomial fitting method fails,this method still works well.It successfully solves the problem of needing too many trial calculation cases,and has a good application prospect.In this method,the RBF neural network model is constructed by using the bottom hole pressure(BHP)data generated by sampling,and the objective function is defined by the deviation between the predicted value of the RBF neural network and the actual observed value,and then the particle swarm optimization(PSO)algorithm is used to optimize the objective function.Finally,the optimal solution of the uncertain parameters is obtained and the inversion parameters are obtained.This method greatly improves the efficiency of well test analysis.2.Based on convolutional neural network(CNN),a new method to identify the type and depth of downhole tools automatically by using magnetic positioning logging curve is proposed.This method realizes the automation of downhole tool identifying,effectively avoids the subjectivity caused by manual identifying and other problems,and the identifying accuracy is high,which greatly improves the efficiency of logging interpretation.Oil field data show that this method has a good application prospect.In this method,the tool images at the corresponding positions are automatically intercepted according to manual annotation to build data sets,and the CNN is established in imitation of Alexnet network structure for training.Dropout and L2 regularization are added in the training process to prevent overfitting,so as to improve the generalization ability of the model.Finally,a CNN model to realize downhole tool classification is obtained.The peaks and troughs of the magnetic positioning curve to be identified are combined to obtain the magnetic positioning curve segment to be identified,which is then input into the trained CNN model to obtain its category,and the final identification result is obtained by combining with the depth position of the curve segment.This method realizes the automation of downhole tools identification.