To address output stability and transmission efficiency degradation in WPT systems caused by coupling misalignment and load variations, this paper introduces a Predictive Neural Network-based Deadbeat Predictive Control (PNN-DPC) approach. Initially, a transfer function model is constructed via the T-type equivalent circuit of the dual-LCC resonant topology, and the mathematical link between the phase-shift angle and output voltage is derived through output performance parametric analysis. Subsequently, the model is reformulated to reduce traditional model predictive control's computational complexity and parameter reliance. A neural network estimator is incorporated to identify system uncertainties and external disturbances, boosting controller robustness. Finally, a predictive controller merging deadbeat control with neural network-estimated model data is designed, and Lyapunov stability theory confirms the error system's asymptotic stability under input constraints. Experiments show that PNN-DPC maintains constant voltage output across diverse coupling coefficients and loads, outperforming conventional methods in dynamic response and disturbance rejection.