Aimed at the problem of vehicle lane change phase recognition affected by the vehicle trajectory characteristics in the merging area of expressways, this study conducted a multi-dimensional analysis of vehicle trajectory characteristics, proposed a force model between vehicles based on the Lennard-Jones potential energy function, and constructed a vehicle lane change phase recognition model based on Particle Swarm Optimization eXtreme Gradient Boosting (PSO-XGBoost).By analyzing the vehicle trajectory data in the merging area of Mengshan Avenue in Linyi City, it was found that the lane-changing points of ramp vehicles followed a Gaussian distribution in space, the lane-changing conditions in the middle of the acceleration lane were better than those at both ends, and most drivers chose to change lanes in the middle. Most vehicles were in the speed range of 45-75 km/h when changing lanes, with 55-60 km/h being the optimal speed range. When driving to the end of the acceleration lane, drivers adopted aggressive driving strategies to force lane changes, resulting in increased risk levels.The intermolecular force model was introduced, and a force model between vehicles based on the Lennard-Jones potential energy function was constructed. The interaction between front and rear vehicles was quantified, and the correlation mechanism between vehicle dynamic game behavior and lane-changing decisions was revealed. On this basis, a vehicle phase recognition model based on PSO-XGBoost was developed, achieving accurate recognition of vehicle lane changes in each phase. The results showed that the classification accuracy of the model was 0.92, and the area under the ROC curve was 0.987. Compared with the LightGBM (LGB), Support Vector Machine (SVM), and Random Forest (RF) , the accuracy improved by 0.03, 0.06, and 0.04, respectively, demonstrating the model's excellent recognition capability.The characteristics of forced lane changing are analyzed based on vehicle trajectory, and the Lennard-Jones potential energy function is introduced to quantify the interaction strength between vehicles. A recognition model of vehicle lane changing phase based on PSO-XGBoost is established. The model is optimized by particle swarm optimization algorithm, which improves the accuracy of the model, and provides a reference for the study of forced lane changing in the merging area.