Dynamic gesture recognition based on hand skeleton is a research hotspot in the field of computer vision and human-computer interaction. The joints involved in gestures are more closely distributed in space and have stronger correlation. Aiming at the problems of complex spatial features and slow recognition calculation speed in skeleton-based dynamic gesture recognition, an attention-guided spatial graph convolution simple recurrent units(ASGC-SRU) network is proposed. Firstly, the spatial graph convolution is embedded into the gate structure of the SRU, so that the SRU with high-speed parallel computing ability can model the temporal and spatial information of complex gestures. Then, an attention-guided module of finger joints is introduced to make more important finger joints have higher attention. Finally, an attention-enhanced spatial dropgraph(ASD) regularization is introduced to alleviate the effect of the over-fitting problem. To verify the effectiveness of the proposed method, a series of experiments are conducted on the recognized dynamic gesture datasets SHREC'17 and DHG14/28. The results show that the proposed method has achieved high recognition accuracy and simultaneously maintains excellent calculation efficiency.