Ship propulsion shafting system is the core component for power transmission, and its whirling vibration characteristics directly affect navigational safety and passenger comfort. Traditional shafting design relies on finite element numerical simulations and a trial-and-error process involving multiple iterations, which suffers from complex modeling, time-consuming calculations, and long design cycles. Unlike the conventional "trial-and-error" paradigm, forward design is a performance-oriented methodology that directly derives key design parameters meeting specified requirements. To address the aforementioned challenges, this paper proposes a forward design method for shafting systems. First, a parametric finite element model of a typical propulsion shafting was established using ANSYS APDL. The accuracy of the finite element model and its calculations was validated by comparing results with literature data. Based on this, a dataset containing 1000 sets of design parameters and corresponding first three orders of whirling natural frequencies was generated in batches. Second, a deep neural network prediction model was constructed, enabling high-precision and rapid prediction from shafting design parameters to whirling natural frequencies. The coefficient of determination R² on the test set reached 0.981, with an average error of 1.41%. Finally, a forward design model comprising a prediction network cascaded with a generative network was developed to achieve the inverse mapping from target vibration frequencies to key design parameters. The first-order fundamental frequency error of the design results was less than 5.5%. The results demonstrate that the proposed method achieves high accuracy and can effectively shorten the shafting design cycle, contributing to the intelligent and efficient design of ship propulsion shafting.