Real-time Prediction Of ROP Based On Deep Learning Method

ROP has always been an important indicator of the efficiency and cost of drilling activities.For many years,domestic and foreign scholars have used various methods to build ROP prediction models,but these models generally rely on traditional statistical and mathematical methods,making them usually characterized by insufficient data usage,failure to take into account the interactions between parameters affecting ROP,and low accuracy of ROP prediction.With the development of artificial intelligence and deep learning in recent years,drilling practitioners have started to try to use neural networks in deep learning to build ROP prediction models and circumvent the drawbacks of traditional ROP prediction models,which makes ROP prediction models change from traditional experience-driven to current data-driven.This paper proposes a real-time prediction of ROP based on deep learning method for the development trend.The paper collects real-time drilling data and pre-processes the data including data cleaning,data normalization and data correlation analysis,and selects the parameters to build the ROP prediction model firstly.Secondly,we analyze the feasibility of deep learning,recurrent neural network and convolutional neural network to build ROP prediction model,and then set the parameters and hyperparameters of neural network to build ROP prediction model.Then,in order to improve the robustness and generalizability of the three neural network models,a structural optimization mechanism is used for the deep neural network ROP prediction model,an attention mechanism is added to the recurrent neural network ROP prediction model,and a continuous learning architecture is used for the convolutional neural network ROP prediction model.Finally,a real-time ROP prediction system based on the deep learning method is established based on the three ROP training models.The research results show that the ROP prediction models built by the three neural networks have high accuracy rates.The accurate prediction model of ROP studied in this paper can be further applied to the planning stage and real-time operational optimization.