<p>Asphalt mixture is indispensable in road construction, and its quality—a key factor in pavement durability and resistance—heavily depends on the design of component materials and mineral aggregate gradation. However, traditional methods for determining mix ratios often suffer from large errors and poor flexibility. To address these issues, we formulate asphalt mix design as a nonlinear optimization problem and develop a projection-based neural network model to solve it. Our approach achieves significantly faster convergence to the precise local optimum compared to conventional techniques. Numerical simulations demonstrate its effectiveness, providing clear evidence of improved accuracy and computational efficiency.</p>