Abstract: To address the limitations of traditional edge computing and caching technologies in handling computationally intensive and latency-sensitive tasks, we propose an active edge computing and caching optimization scheme centered on unmanned aerial vehicles (UAVs). The scheme leverages UAVs to actively sense vehicle demands, enhancing the accuracy of road vehicle demand prediction by integrating binary classification mathematical models with Hawkes processes. The problem is formulated as a Markov decision process, and to optimize edge caching and task offloading, an uncertainty-aware exploration proximal policy optimization (UAE-PPO) algorithm is introduced, building upon improvements to the proximal policy optimization (PPO) algorithm. The UAE-PPO algorithm enhances model stability and generalization by incorporating uncertainty-aware exploration and dynamically adjusting exploration strategies within the actor network. Additionally, it integrates adaptive attenuation of the clip parameter and L2 regularization techniques. Simulation results demonstrate that, compared to the traditional PPO algorithm, the proposed UAE-PPO algorithm improves reward convergence speed by 28.6% and increases the reward value by 6.3%.