The Intelligent Transportation System (ITS) plays a critical role in the development of smart cities, and the traffic control capability of ITS requires accurate traffic flow prediction as the footstone. Due to the neglect of the dissimilarities between lanes, existing traffic flow prediction methods like T-GCN are intractable to capture the dynamic changes of traffic flow direction, lane relationship, etc. Therefore, studies are shifting their concern to lane-level traffic flow prediction for tackling with more complicated scenarios. However, how to handle temporal-spatial correlation and model heterogeneous data effectively is still a challenge. In this paper, we propose a heterogeneous graph convolution model based on dynamic graph generation to address the issue. The model consists of three components. The internal graph convolution network captures the real-time spatial dependency between lanes in terms of generated dynamic graphs. The external heterogeneous data fusion network comprehensively considers other parameters such as lane speed, lane occupancy, and weather conditions. The codec neural network utilizes a temporal attention mechanism to capture the deep temporal dependency. Substantial experiments based on two real-world datasets indicate that the proposed heterogeneous graph convolution model can improve prediction accuracy.