The lane change decision making module plays a crucial role in autonomous driving systems, facing the challenge of balancing collaborative traffic operation. Modeling complex interactions among multiple autonomous vehicles in coexisting environments poses significant challenges. This study focuses on collaborative lane change decision making for multiple autonomous vehicles by employing deep graph convolutional neural networks. These networks effectively model the interaction and collaboration among vehicles, while reinforcement learning facilitates the iterative evolution of decision-making. To evaluate the performance of the proposed Graph Reinforcement Learning (GRL) method, an interactive driving scenario with two ramps on a highway was developed. Simulation experiments were conducted on the SUMO platform to compare different GRL methods. Results were analyzed from multiple perspectives and dimensions to compare the characteristics of different GRL methods in the scenario of highway merging traffic. The findings demonstrate that the utilization of deep graph convolutional neural network can effectively model the complex interactions among vehicles and the combination of graph convolution and reinforcement learning can significantly improve the performance of lane-changing behaviors in terms of both efficiency and safety.