采用MPSO优化SVR的短时交通流预测方法-《计算机技术与发展》

文章信息/Info

Title:: Short-term Traffic Flow Prediction Method of Optimized SVR Based on MPSO

Author(s):: YAN Yu-chan; WU Qi-sheng; BAI Lin; XI Wei; School of Electronics and Control Engineering,Chang’an University,Xi’an 710064,China

Keywords:: traffic flow prediction; linear decreasing inertial weight; particle swarm optimization ( PSO); support vector regression (SVR); parameter optimization

摘要:: 为了提高高速公路交通流量预测精度以及预测方法的稳定性,降低预测用时,提出了一种后期随机惯性权重粒子群算法与支持向量回归机相结合的短时交通流预测模型(MPSO-SVR)。该预测模型用均匀分布的随机惯性权重替代标准PSO算法中不变的惯性权重ω,使算法中粒子在搜索后期拥有较大的ω,从而有效地避免算法陷入局部最优解,加快了算法的寻优速度。最后,通过不断更新惯性权重来更新粒子的速度与位置。算法不仅对支持向量回归中的惩罚因子c和核函数参数g进行寻优,而且能很好地平衡算法全局搜索与局部搜索能力,提高了算法的性能。实验结果表明,MPSO-SVR方法在沪宁高速交通流数据中比PSO-SVR方法预测精度更高、稳定性更强、耗时更短,且均方误差和平均百分比误差分别降低到28.689和12.952%。

Abstract:: In order to improve the prediction accuracy and stability of road traffic flow and reduce the predictive consuming-time,we propose a short-term traffic flow prediction model based on random inertial weight particle swarm optimization combined with support vectorregression machine (MPSO-SVR). In this model,the uniform distributed random inertial weight is substituted for the invariant inertialweight ω of the standard PSO algorithm,so that the particles in the algorithm have lager ω in the late search,thus effectively avoiding falling into the local optimal solution for the algorithm and accelerating optimization search. Finally,we update the velocity and position of the particle by updating the inertial weight constantly. The algorithm not only optimizes the penalty factor c and the kernel function parameter g in the support vector regression machine,but also balances the global search and local search of the algorithm well with the improvement of algorithm performance. Experiment shows that the MPSO-SVR has higher prediction accuracy,stronger stability and shorter time consumption than standard PSO-SVR in Shanghai-Nanjing expressway traffic flow data,and the mean square error and the average percentage error can decrease to 28. 689 and 12. 952% respectively.