基于相空间重构的GQPSO-WNN短时交通流预测

GQPSO-WNN SHORT-TERM TRAFFIC FLOW FORECASTING BASED ON PHASE SPACE RECONSTRUCTION

交通流预测有助于减少交通拥堵和交通事故发生。在定量分析交通流变化过程的混沌特性以及可预测性基础上,提出一种基于相空间的GQPSO-WNN的混合预测模型。引入遗传算法,使用混合优化后的量子粒子群算法初始化小波神经网络的各项参数,克服网络因初始值设置不当造成无法收敛或陷入多个局部极小值的问题。由于神经网络输入的随机性,采用重新构建交通流时间序列的相空间技术,用重构后的数据作为输入样本。实验结果表明,与WNN、PSO-WNN预测模型相比,该模型可以更加准确预测交通流,算法收敛性也有明显提高。

Traffic flow prediction plays an important role in traffic congestion and accident control. The paper quantitatively analyzed the chaotic characteristics and predictability of traffic flow changes,and proposed a hybrid prediction model based on phase space GQPSO-WNN. We introduced genetic algorithm to initialize the parameters of the wavelet neural network by using the hybrid optimized quantum particle swarm algorithm,and it could overcome the problem that the network could not converge or fall into multiple local minimum values due to inappropriate initial settings. Due to the randomness of the neural network input,we used the phase space technique of reconstructing the traffic flow time series,and the reconstructed data was used as the input samples. The experimental results show that,compared with the WNN and PSO-WNN prediction models,the model can predict traffic flow more accurately,and the convergence of the algorithm is also significantly improved.