Abstract: To improve the prediction precision of short-term traffic flow, this paper proposes a nonlinear short-term traffic flow prediction model based on parameters joint optimization algorithm, which uses the relationship between the phase space reconstruction and prediction model parameters. The phase space reconstruction and prediction model parameters are taken as particle of Improved Particle Swarm Optimization algorithm（IPSO） while the prediction accuracy of short-term traffic flow as the evaluation function of CPSO. The optimization parameters are obtained by collaboration among particles. The performance of  proposed model is tested by short-term traffic low data. The results show that the proposed method has improved the prediction precision compared with the traditional parameters optimization algorithm; it is a new way for the nonlinear prediction problem.

Key words: short-term traffic flow, phase space reconstruction, particle swarm optimization, neural network

摘要： 为了提高短时交通流量预测精度，利用相空间重构和预测模型参数间的相互联系，提出一种粒子群优化神经网络的短时交通流量预测模型。将相空间重构和预测模型参数编码为粒子群的粒子，短时交通流量预测精度作为粒子群的适应度函数，通过粒子之间协作获得预测模型全局最优参数，通过BP神经网络建立预测模型，利用短时交通流量数据对模型性能进行测试。结果表明，相对于传统参数优化方法，粒子群优化神经网络提高了短时交通流量的预测精度，为非线性预测问题提供了一种新的研究思路。

关键词: 短时交通流量, 相空间重构, 粒子群优化算法, 神经网络