Abstract: The present disclosure provides a positioning method and system for autonomous driving through Long Short-Term Memory (LSTM)-based Deep Reinforcement Learning (DRL). The method includes: performing normalization preprocessing on a complex environment of autonomous driving based on a Partially Observable Markov Decision Process (POMDP), to acquire a real-time kinematic (RTK) positioning result; inputting the RTK positioning result into an LSTM-based DRL model for correction to acquire an evaluated value of a position correction action; and performing position correction on an autonomous vehicle based on the evaluated value of the position correction action. The system includes a prediction module, a correction module, and an application module. The present disclosure considers that autonomous driving is highly dynamic, temporal, and complex in a complex environment, and generates a more accurate satellite positioning position.