Efficient and safe motion planning for intelligent connected vehicles in complex traffic scenarios remains a pivotal challenge in the field of autonomous driving. This research proposed ST-Trans traffic prediction model based on the Transformer architecture and developed a predictive motion planner for intelligent connected vehicles leveraging ST-Trans. The ST-Trans model utilizes the Transformer architecture to mine spatial-temporal evolution patterns from real-time vehicle data and lane segment structural information provided by dynamic highdefinition maps, thereby predicting future traffic states of lane segments. It further enhances prediction accuracy by incorporating lane segment connectivity and intersection signal phase information. The model adopts an encoderdecoder framework, where a lane encoder fuses vehicle and lane features, a road encoder models dynamic topological relationships, and a decoder iteratively generates future traffic state sequences. Experimental results demonstrate that ST-Trans outperforms the optimal baseline model by 12.2%, 12.1%, and 3.55 percentage points in terms of mean absolute error (MAE), root mean square error (RMSE), and accuracy, respectively. Based on the predictions from ST-Trans, the proposed predictive motion planner employs a two-layer structure. The lower-layer path planner dynamically selects target points and integrates dynamic programming with quadratic programming to generate smooth paths. The upper-layer speed planner constructs spatio-temporal corridors to compress the solution space and similarly combines dynamic programming and quadratic programming to generate safe efficient, and comfortable speed profiles. This structure significantly reduces the computational complexity of the motion planning task. Simulation experiments were conducted using SUMO and CARLA to evaluate the predictive motion planner.The results indicate that the ST-Trans-based predictive motion planner successfully implements predictive path and speed planning, and outperforms traditional motion planners in terms of safety, efficiency, comfort, and computational speed. The experiments verify that the proposed method effectively shortens the duration of high-risk states, improves traffic throughput and maintains low computational latency.