Intelligent Prediction Method Of Sticking Based On Coupling Mechanism And Data Model

Stuck pipe phenomena can have disastrous effects on drilling performance,with outcomes that can range from time delays to loss of expensive machinery.This paper focuses on the key problems of accurate prediction and classification of sticking during drilling.The main work results are as follows:The comprehensive mud logging of more than 100 wells in three blocks from the site are sorted and cleaned,the basic data preprocessing process of unified sampling frequency,data extraction and data splicing is formed,and the mud logging sequence database that can be used for neural network training is established to provide data samples for the establishment of subsequent intelligent models.According to the sequential and non-sequential characteristics of logging data,BP and LSTM networks are optimized,BP-LSTM network architecture is designed,and the intelligent prediction model of hook load and torque based on BP-LSTM is established.The intelligent real-time calculation model of weight on bit and torque on bit is established based on BP-LSTM;Coupling the torque and drag model,a real-time inversion method of the friction coefficient is established to characterize the sticking trend;Based on the stuck sample data set,an intelligent prediction model of stuck probability is established by using LSTM network.The comprehensive evaluation method of sticking is established by coupling the friction coefficient and sticking probability.The results show that the accuracy is 97%.Finally,BP-LSTM network is used to predict the main control parameters of sticking(hook load,torque and standpipe pressure).Combined with the historical main control parameter data,the specific sticking type is determined by analyzing the change trend of main control parameters.