In order to improve the prediction accuracy of short-term traffic flow, a short-term traffic flow prediction method based on phase space reconstruction and improved BAS-CRJ is proposed．First, the chaotic analysis of the traffic flow speed sequence is carried out, the phase space of the sequence is reconstructed, and the research on the traffic flow speed sequence is mapped into the phase space where it is located. Then, the variable step size factor and simulated annealing technology are introduced to improve Beetle Antennae Search Algorithm(BAS), and the improved Beetle Antennae Search Algorithm is used to optimize the parameters of the cycle reservoir with regular jumps network (CRJ). A forecast model is developed. Finally, the effectiveness of the model is analyzed through an example. The results show that the internal dynamic characteristics of the traffic flow speed sequence are tapped through phase space reconstruction， which is more suitable for network modeling. And the prediction result of the proposed model are better than comparison models, in which the average absolute percentage error is decreased by 1.05%～6.04%. The prediction accuracy of short-term traffic flow speed is effectively improved.