Abstract: Curing-induced deformation frequently occurs in resin-based composites due to the coupling effects of thermal, chemical, and mechanical fields during the curing process. The rational design of curing process parameters is therefore crucial for reducing deformation, improving forming accuracy, and ensuring production efficiency. However, thermo–chemo–mechanical coupled numerical simulations are computationally expensive, which limits their direct use in multi-objective optimization. To overcome this limitation, a curing process parameter optimization framework combining a feedforward neural network (FNN) surrogate model with the NSGA-II multi-objective optimization algorithm is proposed. Sample data are generated through thermo–chemo–mechanical finite element simulations. The heating rate, cooling rate, holding time, and holding temperature are used as input variables to construct the FNN surrogate model, enabling rapid prediction of key responses including spring-in angle, maximum deformation, and final degree of cure. The training and validation results demonstrate that the developed FNN model achieves high prediction accuracy and good generalization capability. Subsequently, NSGA-II is employed to perform multi-objective optimization of the curing parameters, resulting in a set of non-dominated solutions forming a Pareto front. The optimization results reveal a significant negative correlation between the spring-in angle and the total process time, indicating an inherent trade-off between forming accuracy and production efficiency. The obtained Pareto front is continuous and well distributed, and the influence of different process parameters on curing deformation is clarified, providing multiple feasible parameter combinations for various engineering requirements. The proposed FNN–NSGA-II framework can effectively guide the design of composite curing processes while significantly reducing computational cost.