Abstract: five-layer collaborative architecture intelligent logistics system based on digital twin technology. This architecture integrates the physical perception and execution layer，the data transmission and processing layer，the virtual-real mapping and model layer，the business logic and application layer，and the simulation decision-making and optimization layer. It combines 3D modeling，IoT data collection and bidirectional interaction technologies，achieving a full-process closed-loop optimization from data collection， real-time mapping to simulation rehearsal and dynamic decision-making. Taking the collaborative distribution scenario of electric logistics vehicles and AGVs in an intelligent logistics park as an example for empirical research，the results show that compared with the traditional scheduling mode，this system effectively resolves operational conflicts through simulation pre-performance， increasing the delivery efficiency by 23.2%，reducing the average cost by 58.6%，and compressing the empty running rate to 4.2%. At the same time，the congestion frequency is reduced by 96.8%，the single vehicle energy consumption is decreased by 66.6%， and the global optimization of efficiency, cost, robustness and energy efficiency is achieved. This research not only quantitatively evaluates the actual effectiveness of digital twin in path planning, cost control and resource scheduling，but also provides theoretical basis and practical paradigms for the architecture design and large-scale implementation of intelligent logistics systems in complex urban distribution scenarios.