With the development of Industry 5.0, the logistics industry, serving as a bridge between production and consumption, is undergoing profound changes. However, this transformation faces challenges such as data fragmentation, difficult system integration, and insufficient real-time monitoring capabilities. Consequently, the modern logistics system demands higher standards for the prediction and management of transportation behavior. To address these challenges, this paper introduces Digital Twin (DT) technology and proposes a research methodology for DT-driven management strategies. DT technology constructs virtual models of physical objects to enable real-time monitoring and data analysis of unmanned vehicle states, effectively resolving the identified issues. Specifically, the proposed method leverages DT to integrate multi-source heterogeneous data and establishes a digital model of unmanned vehicles. Furthermore, it combines the LSTM neural network algorithm to design a predictive model for time-series forecasting of transportation behaviors. The digital model is dynamically adjusted based on prediction results, further optimizing the management strategy. Finally, the effectiveness of the proposed method is validated through a case study on unmanned vehicle transportation behavior. Experimental results demonstrate that the DT-based management strategy significantly improves the accuracy of predicting unmanned vehicle transportation behaviors and exhibits superior performance in decision aid and fault tolerance. Additionally, simulation tests confirm the reliability and efficiency of the improved algorithm in practical applications, providing an important reference for the intelligent development of modern logistics systems.