The Study Of Predieting Pipe Sticking Based On Neural Network

Pipe sticking accident has a great probability of occurrence and it takes a long time tounfreezing. After statistics from a drilling company，Time to deal with sticking accidents isover60%of the total drilling accident treatment time. So with the purpose of improving thedrilling efficiency，reducing sticking accident，this paper studied the predicting pipe stickinkin-depth.Firstly，BP neural network which has a strong nonlinear mapping ability to forecast thesticking was selected based on the analysis of a large number of drilling data，and summing updifferent kinds of factors influencing sticking accident causes，symptoms and solution.This paper analysed mechanism of sticking deeply，and established sticking predictionmodel based on BP neural network. At the same time, a large number of drilling data has beencollected，and composed of training sample and testing sample of the neural network. Thentrain the sticking prediction and verify the sticking prediction ability. Validation results showthat the training error and testing error of the sticking prediction model within the scope of thepermit.In the view of the defects of BP learning algorithm，which include low learning efficiencyand slow convergence speed，this paper improves learning algorithm of the network by usingadaptive learning rate，so as to improve the prediction precision of the network，and reduce thetraining time. By verifying the model’s prediction accuracy has been raised and the time oftraining has been reduced.Lastly，this paper sets up loop sticking prediction model in order to predict highfrequency loop sticking，and verifies the predictive and generalization ability of the loopsticking prediction model. The verification result shows that the model can predict loopsticking accidents for regional area exactly，and also has good generalization ability as well.