Mining co-movement patterns from trajectory streams refers to discovering groups of moving objects with same behaviors at the same time， which is essential for transportation logistics， epidemic prevention and control and so on. However， the existing research faces difficulties in responding quickly to large-scale trajectory data streams. Therefore， this paper proposes a novel distributed sliding window algorithm for mining co-movement patterns from spatio-temporal trajectory streams. The algorithm employs a sliding window computing model instead of a snapshot computing model， and utilizes incremental updates instead of re-computing， making it more suitable for handling unbounded and rapidly arriving trajectory data streams. The proposed algorithm demonstrates superior performance in terms of efficiency and effectiveness. Secondly， to address the issue of load imbalance in distributed stream processing systems， this paper proposes an adaptive multi-level dynamic data partitioning strategy. This strategy can adapt to the dynamic changes in trajectory stream data， continuously monitor the system load in real-time， and make appropriate adjustments based on the degree of load imbalance. Finally， this paper implements the above functions on the Flink distributed big data processing platform and uses real data sets for experiments. Comprehensive empirical study demonstrates that the proposed algorithm has faster response speed and lower delay than the baseline method.