Accurate rate of penetration (ROP) prediction is of great significance for safe and efficient drilling in oil and gas engineering. Due to the complexity and uncertainty of formation conditions, the traditional prediction of ROP based on physical model can only be used as a reference before drilling. Long and short-term memory artificial neural network LSTM is a kind of cyclic neural network, which is often used for the prediction of various time series problems. Deep series and time series are essentially the same, that is, the previous data point and the next data point have potential connection. This paper innovatively applies LSTM to the prediction of depth series problems. Based on the ROP data set in the drilling integrated log, the LSTM prediction model for predicting ROP is trained and tested, which can predict the ROP ahead of the bit according to the ROP in the drilled section. The optimal model for different footage to be drilled is explored, and the recommended size of the input layer, the number of layers of neural network and the number of units per layer are given. The model has been applied in the field. The effect is good and the minimum value of absolute error is only 2.11 m/h.
Safe and efficient drilling is the eternal pursuit of drilling engineers. The important parameter that can reflect the drilling efficiency is the rate of penetration. The rate of penetration (ROP) is defined as the bit footage drilled per unit time, which directly reflects the speed of drilling. Engineers often need to know how much of a ROP can be generated by current construction parameters in order to guide subsequent construction. Traditional physics-based models use empirical coefficients that are highly lithologic dependent and constantly changing due to calibration, limiting their applicability. Therefore, the accurate prediction of ROP is still a problem to be solved.