Research On Drag And Torque Prediction Of Drill String Based On Neural Network

With the development of drilling technology,the well depth,the length of horizontal section and the curvature of borehole trajectory has been increasing,and the drag and torque approaches or even exceeds the bearing limit of equipment sometimes.Therefore,a higher calculation accuracy of drag and torque is necessary.In order to solve the problem that the designed well path is too smooth and the calculated drag and torque is too small,this paper adds tortuosity to the designed path to simulate the tortuous change of practical borehole trajectory.By analyzing the tortuosity distribution law of practical borehole trajectory,the probability density function of borehole inclination and azimuth of practical borehole trajectory is obtained,and the variation rate deviation of the known tortuosity distribution law is generated randomly.The variation rate deviation is added to the designed well path,the values of azimuth and inclination simulating the practical borehole trajectory are obtained,and the accuracy of the prediction results of drag and torque is improved.Accurate friction coefficient is the prerequisite for the drag and torque prediction.There are many factors affecting the friction coefficient,and there are some uncertainties,so it is difficult to use a universal mathematical equation to clearly describe the relationship between the factors and the friction coefficient.In this paper,the characteristics of friction are analyzed comprehensively,and the improved BP neural network is used to analyze the friction coefficient.Firstly,the basic theory and mathematical expression of BP algorithm are deeply analyzed.On this basis,adaptive learning rate,momentum term and other methods are introduced to optimize.Finally,based on the optimized BP neural network,the friction coefficient prediction model is established.It can be seen from the experimental results that the optimized BP neural network can predict the friction coefficient with high precision.On the basis of the predicted friction coefficient,combined with the well path and related parameters simulating the practical borehole trajectory,the BP,LSTM and LSTM-BP neural network models were compared,and the LSTM-BP neural network based friction torque prediction model was proposed.The reliability and rationality of LSTM-BP neural network in the prediction of drag and torque are verified.The network model can learn different types of data adaptively,so as to further improve the prediction ability of the model.Finally,the method used in this paper is verified by a field case.In the experiment,the root-mean-square error of the hook load is 38 k N,the average absolute percentage error is 0.04725,the average absolute percentage error of the traditional prediction is 0.076,and the accuracy is reduced by2.9%.The root mean square error of torque is 1.26 k N·m,the average absolute percentage error is 0.13505,the average absolute percentage error of traditional prediction is 0.2726,and the accuracy is reduced by 13.7%.The predicted values has a better match with the practical values.