Enhanced high-resolution prediction modeling of working face reconstruction using multi-scale grid pruning and dynamic multi-population genetic algorithms

This study develops an advanced reconstruction framework combining search space pruning and dynamic multi-population genetic algorithm (DMPGA) to address the challenges of working face computed tomography (CT) reconstruction under incomplete projection conditions. The proposed method first employs prior knowledge-guided adaptive meshing to effectively constrain the solution space, while simultaneously reformulating and optimizing the working face reconstruction objective function, which effectively reduces computational complexity while preserving reconstruction fidelity. Subsequently, a DMPGA model is developed to dynamically adjust population diversity and convergence strategies, enabling efficient optimization of the reconstruction objective function's search process. Experimental results demonstrate the framework's superior performance, achieving high reconstruction accuracy and stability while reducing computational costs by approximately 40%–60% compared to standard approaches. The proposed method yields particularly strong results at the optimal 3.125 m scale, balancing reconstruction accuracy (0.9612 dB/m), computational efficiency (219.4341 s), and stability (S = 0.1919 dB/m). Field validation confirms the practical effectiveness, with high boundary matching accuracy to the truth data. These advancements establish a new paradigm for high-fidelity industrial CT reconstruction, offering substantial improvements in both imaging quality and computational performance for complex underground engineering applications where complete projection data acquisition is challenging.