Intelligent wellbore trajectory control method for multiple obstacle-well constraints based on DeepSets-TD3

Abstract:

Aiming at the challenges in borehole trajectory control under dense well patterns， including constraints from multiple adjacent wells， sensitivity to the input order of obstacles， and stable control of continuous actions， this paper proposes an intelligent control method for multi-obstacle well constrained borehole trajectory named DeepSets-TD3. The method recursively calculates the borehole trajectory using the minimum curvature method， and takes the increments of inclination and azimuth as continuous control actions. The basic state consists of the relative displacement of the target， inclination angle， as well as the sine and cosine values of azimuth angle. Each element in the set is constructed by the surface clear distance of obstacle wells and the sine and cosine values of relative azimuth. A permutation-invariant representation is formed via a shared encoder network and mean-max symmetric pooling. In the test with four fixed obstacles， the DeepSets-TD3 achieves an average target error of 7.52 m and zero physical collision rate. In the random four-obstacle test， the average target error is 11.2 m， the physical collision rate is 1.2%， the near-miss buffer rate is 1.9%， and the compliance rate of dogleg severity reaches 99.6%. All four indicators outperform the conventional TD3. The results demonstrate that the set representation can mitigate policy degradation caused by changes in the arrangement of obstacle wells， and improve the accuracy， safety and generalization capacity of trajectory control under complex constraints of adjacent wells.