基于改进花朵授粉算法的BP神经网络短时交通流预测

Short-term traffic flow prediction based on BP neural network optimized by modified flower pollination algorithm

国内道路交通拥堵问题日益严重,及时和准确的短时交通流预测是实现智能交通管控、减轻道路拥堵的关键基础,因此设计了一种基于惯性权重改进花朵授粉算法(MFPA)和误差逆向传播(BP)神经网络结合的MFPA-BP短时交通流预测模型。首先通过引入惯性权重和Mantegna方法改进花朵授粉算法,形成MFPA优化算法,并将其应用到BP神经网络进行初始权值和阈值的优化,使模型的收敛速度和效率得到提升。选用实际路段处理过的交通流数据对MFPA-BP模型进行训练并预测,与BP神经网络和RBF神经网络相比平均绝对误差(MAE)分别减少了20.01%和27.89%,均方根误差(RMSE)分别减少了18.25%和21.73%,同时MFPA-BP模型可以有效地减少迭代次数,提高道路交通流量预测的准确性,更好地应用于智能交通系统中。

The traffic congestion in China is becoming more and more serious. Timely and accurate short-term traffic flow prediction is the key for realizing intelligent traffic control and management and reducing road congestion. Therefore,an MFPA-BP short-term traffic flow prediction model based on both the inertial weight modified flower pollination algorithm(MFPA)and the error back-propagation(BP) neural network is designed. The flower pollination algorithm is modified by inertia weight and Mantegna method to form the MFPA optimization algorithm,which is applied to the BP neural network to optimize the initial weight value and threshold value to increase the convergence speed and improve the efficiency of the improved model. The MFPA-BP model is trained and predicted by the processed traffic flow data of actual road sections. In comparison with the basic BP neural network and RBF(radial basis function)neural network,the mean absolute error(MAE)of the MFPA-BP model is reduced by20.01% and 27.89%,and the root mean square error(RMSE) of the MFPA-BP model is reduced by 18.25% and 21.73%,respectively. At the same time,the MFPA-BP model can effectively reduce the iterations and improve the prediction accuracy of road traffic flow,so it can be applied to intelligent transportation systems.