To enhance the accuracy, robustness, and real-time performance of RFID positioning technology in complex environments, this paper proposes a multi-node edge computing collaborative RFID phased array intelligent positioning method. The proposed method employs phased array antennas for dynamic beam control and utilizes multi-node edge computing to process large-scale tag data, effectively mitigating the impact of multipath effects and signal attenuation. Additionally, the system integrates the Asynchronous Advantage Actor-Critic reinforcement learning algorithm to dynamically optimize positioning parameters in response to environmental changes, further improving adaptability and stability. Experiments were conducted in both standard and complex environments, with the latter simulating extensive metallic shelving, multipath effects, and dynamic interference sources to evaluate positioning error and accuracy in comparison with RSSI and TDOA methods. Experimental results show that in the standard environment, the proposed method achieves positioning errors of 0.8~0.9 meters and an accuracy of 92%; in the complex environment, errors remain within 1 meter, with accuracy exceeding 90%, significantly outperforming traditional methods. Furthermore, practical deployment in an intelligent warehouse asset management system demonstrates the high precision and robustness of the proposed method, improving inventory accuracy from 85% to 96% while reducing the misjudgment rate to 1.5%. This research provides reliable technical support for the application of RFID positioning technology in smart cities, power grid asset management, and logistics warehousing, demonstrating excellent environmental adaptability and high-efficiency positioning capabilities.