Trajectory parameter prediction of directional drilling in underground coal mine based on convolutional neural network

Abstract

A predictive model is established to determine parameters of the drilling trajectory based on one-dimensional convolutional neural network（1DCNN）, which can obtain the actual path parameters at the drill bit during directional drilling to improve the effect of directional drilling borehole tracing control and ensure the accuracy of the actual drilling path. There are 502 groups training data and each contains 12 parameters such as dip angle, azimuth and angle of measurement while drilling system. The inclination angle and azimuth angle obtained in the last measurement of each group were taken as output values, and other parameters were taken as input values to train the model. Other 12 groups drilling path parameters are used to test the model and the results are compared with the predicted results based on experience of the technical staff. This study result shows that: the one-dimensional convolutional neural network is feasible to predict the trajectory parameters, showing a slight advantage in the two evaluation indexes of root mean square error and determination coefficient, and the absolute errors of dip angle and azimuth predicted by one-dimensional convolutional neural network model are 0.69° and 0.70° which are 22.5% and 18.1% lower on average than the forecasts of technical staff.

Keywords

• coal mine engineering geology
• directional drilling
• drilling path parameters
• one-dimensional convolutional neural network
• predictive model