ABSTRACT China holds abundant deep reservoirs, where temporary plugging and diversion fracturing (TPDF) is a key technology for efficient development. Effective transport and plugging of agents (TPAs) within fractures to boost net pressure are key to the success of this technology. Moreover, optimising the transport parameters can provide practical data support for field operations. While numerical simulation is an important optimization tool, it often yields single‐point parameter solutions and requires lengthy computation, limiting its real‐time applicability. In contrast, surrogate modelling enables fast target prediction and enhances optimization efficiency. Among various multi‐objective optimization algorithms, the non‐dominated sorting genetic algorithm II (NSGA‐II) demonstrates strong performance in both speed and convergence. In this study, a temporal convolutional network (TCN) was ultimately selected to construct the surrogate model for rapid prediction, which was coupled with NSGA‐II for optimization. The main findings are: (1) The TCN model achieved prediction errors below 5%, with transport pattern predictions showing over 80% agreement with simulations; (2) Cost was highly sensitive to the mass concentration of TPAs and injection velocity, minimising cost reduced their optimal ranges; (3) The mass concentration had little influence on most objectives except cost, resulting in minimal variation in its optimal range across different criteria; (4) Injection velocity strongly influenced the dimensionless average velocity of TPAs, minimising it required lower injection velocity; (5) Carrying liquid viscosity significantly impacted dimensionless inlet pressure, leading to higher optimal ranges when high dimensionless inlet pressure was desired.
Wu et al. (2026) studied this question.