To address the limited focus on actuator mechanisms in current variable rate fertilization (VRF) research, this study presents a novel segmented-shaft actuator for outer-grooved wheel fertilizer applicators, enabling dual-variable control of rotor speed and opening size for real-time adjustment of fertilizer dosage. A discrete element model was developed using EDEM 2.2 to simulate fertilizer discharge behavior under different structural configurations and control strategies. Simulation results were validated through bench and field experiments, demonstrating that the improved device reduced discharge variability by 9.9%, with the optimal variation coefficient reaching 1.85% under speed-priority control. To enhance control precision and real-time responsiveness, a hybrid predictive model based on a multilayer perceptron (MLP) neural network was constructed and optimized using the Levy Flight Algorithm (LFA) and Particle Swarm Optimization (PSO), forming the LFA-PSO-MLP (LMP) model. Additionally, an inverse LMP (ILMP) model was established for closed-loop computation of actuator parameters from target fertilizer rates. The LMP model achieved rapid convergence (≈50 iterations), high fitting accuracy (R2 = 0.999), and low mean absolute percentage error (1.83%). Field trials further confirmed a mean application accuracy of 93.92%. These results demonstrate the feasibility and effectiveness of the proposed actuator design and intelligent modeling approach for improving precision fertilization in rice-wheat cropping systems, contributing to sustainable and intelligent agricultural machinery development.