Space-Air-Ground Integrated Networks (SAGIN) is considered an important area for future communication research. SAGIN can serve a wide variety of scenarios,such as edge and coverage limitation, mobile edge computing, communication recover after sudden disaster. In this paper, the problem of low service quality for users on the edge of cities by SAGIN was solved.We propose an approach based on dynamically adjusting the trajectory of UAV and optimizing the joint offloading of UAV, Satellites, and Base Station to reduce UAV energy consumption and user service latency in urban edge communication scenarios. We introduce the Line-of-Sight (LoS) probability model into our SAGIN to better adapt to the urban edge environment. To address the dynamics and complexity of the system, we propose a Deep Deterministic Policy Gradient with Improved Parameter Replay Buffer algorithm (DIPRB). The algorithm learns the optimal offloading and trajectory strategy by trained parameters and exploring more action space. Through different experiments, we demonstrate the effectiveness of DIPRB algorithm, the impact of UAV altitude on service quality, and the superiority of our approach over other approaches in different task sizes.