Amorphous tungsten trioxide (a-WO₃) has emerged as an ideal candidate material for non-volatile memristors, attributed to its high concentration of oxygen vacancies, moderate band gap, and compatibility with CMOS technology. This renders it broad application prospects in neuromorphic computing systems. However, its poor analog switching performance severely hinders its practical application in high-efficiency artificial intelligence data processing. To enhance the analog switching performance of WO₃ memristors, this study adopts radio frequency (RF) magnetron sputtering technology to deposit a five-layer amorphous tungsten trioxide (a-WO₃) thin film with a gradient distribution of oxygen vacancy concentration on a platinum/silicon (Pt/Si) substrate. X-ray Photoelectron Spectroscopy (XPS) analysis confirms that the oxygen vacancy (Vo) concentration decreases gradually from the bottom to the top layer,verifying the successful fabrication of the five-layer a-WO₃ thin film with a gradient distribution of oxygen vacancies. Compared with a-WO₃ memristors with a uniform Vo concentration, the device with the Vo gradient distribution exhibits highly reliable analog switching characteristics (low cycle-to-cycle variability, high linearity in potentiation/depression processes), ultra-long data retention (>10⁴ s), and self-current-limiting behavior. An artificial neural network (ANN) based on this structured memristor achieves a handwritten digit recognition accuracy of 97.64%. The RS essence of a-WO₃ memristors with Vo concentration gradient distribution lies in the formation/rupture of VOdominated conductive filaments (CFs). The Vo gradient distribution enables controllable evolution of CFs by modifying the electric field and ion migration rules. During CF formation, oxygen ions migrate toward the top electrode, and Vo accumulates gradually first in the bottom electrode region; meanwhile, the electric field induced by Vo gradient suppresses the local abrupt growth of CFs, leading to the formation of uniform nonconical structures and avoiding resistance mutation. During CF rupture, ions migrate toward the bottom electrode, and non-conical CFs can rupture synchronously and progressively, ultimately achieving precise regulation of multi-level conductance. The conduction mechanism shows that the low-voltage region of the high-resistance state (HRS) exhibits an I-V linear relationship, corresponding to the ohmic conduction mechanism. In thehigh-voltage region of HRS, I has a linear relationship with both V² and V^2.5, which conforms to the space-charge-limited current (SCLC) theory. The gradient distribution of oxygen vacancies (VO) regulates the formation and rupture of conductive filaments (CFs), thereby solving the core issue of poor analog switching performance in traditional WO₃ memristors. This provides a critical “Vo gradient regulation” design strategy for highdensity neuromorphic computing. It is expected to play a significant role in fields such as image recognition, speech recognition, and intelligent robots.