The autonomous selection of obstacle avoidance speed and power source torque distribution mode for intelligent connected vehicles is affected by the original lane-changing trajectory of the side vehicle, which limits the initiative of collision avoidance and has a low safety factor. Therefore, a prediction model for lane-changing trajectory of the side vehicle based on a long short-term memory network is proposed. A bidirectional long short-term memory network was applied to predict the lane-change trajectory of the side vehicle, and a feedforward and feedback control structure was introduced to control the deviation between the vehicle and the expected path. Controlling the directional deviation Δθ at the vehicle center of mass point and the lateral deviation at the preview point, we endow the obstacle avoidance process with active attributes, and coordinate the control of the collision avoidance path of intelligent connected vehicles. The simulation experiment results show that the sidecar lane-change trajectory prediction error is relatively low at different vehicle speeds, especially at medium and high speeds, with significant advantages. Compared with other methods, the prediction accuracy is higher and the error growth is slower. In addition, comparing the obstacle avoidance effects of three obstacle avoidance methods by setting obstacle points on the designated lane, the proposed method has the shortest obstacle avoidance path distance, no unnecessary obstacle avoidance actions, and a high safety factor, indicating that it can effectively achieve active coordinated collision avoidance control for intelligent connected vehicles and improve driving safety and adaptability.