Wellbore Trajectory Generation With Genetic Algorithm‐Assisted Generative Adversarial Network

ABSTRACT The generation of three‐dimensional double‐build wellbore trajectories for horizontal wells is a critical task in petroleum engineering, constrained by complex underground obstacles and trajectory target requirements. Existing frameworks rely on random exploration for new solutions, unable to learn from the fitness landscape, thus failing to proactively move toward high‐quality regions. Consequently, many generated solutions are discarded as subpar, significantly raising computational costs from invalid attempts. Neural networks, while capable of learning solution distributions, lack targeted local search and easily get trapped in local optima. To address these issues, this paper proposes a novel framework named GA‐optiGAN, which integrates the Generative Adversarial Network and Genetic Algorithm. First, we construct a dynamically maintained candidate solution set, where an exponential decay contraction factor is adopted for dynamic scale adjustment. This encourages our framework to focus on high‐quality regions in the later stage while maintaining a consistent exploration‐exploitation balance throughout the process. Meanwhile, Genetic Algorithm performs parameter‐level fine‐tuning within this set, which allows for refined local search and prevents entrapment in local optima. Second, we design a tanh‐based range scaling module, embedding the constraint ranges of wellbore trajectory parameters into GA‐optiGAN's generator to ensure engineering‐compliant parameter outputs and avoid invalid solutions. Third, we design an objective function centered on minimizing trajectory length, and enforce critical constraints via a two‐stage penalty mechanism that eliminates solutions failing to satisfy these constraints, including obstacle avoidance, target‐hitting accuracy, and parametric compatibility. Experimental results demonstrate that GA‐optiGAN achieves superior performance over state‐of‐the‐art frameworks in generating three‐dimensional double‐build wellbore trajectories.