基于遗传算法优化BP神经网络的交通流预测

Prediction of Traffic Flow Based on BP Neural Network Optimized by Genetic Algorithm

为了使传统的BP神经网络算法能够适用于中时交通流预测,提出一种基于遗传算法优化深层BP神经网络算法。将传统遗传算法优化的BP神经网络进行了优化和调整,分别在不同的隐含层数量、输入节点数量以及隐含层节点数量的条件下进行多次实验,从预测精度和运算效率两个方面综合考虑得到了针对中时交通流预测的最优神经网络结构。以此结构通过MATLAB R2016b进行仿真实验,精度指标采用平均相对误差(Mean Relative Error, MRE),准确率及均等系数(Equality Coefficient, EC)进行综合判断。结果表明,在30 min内,交通流预测的MRE低于3%,准确率和EC则分别高于95%和0.98,而预测延长至60 min内时,MRE仍然能够保持在低于7%的水平,准确率和EC则分别保持在80%和0.95以上。

In order to make the traditional BP neural network adaptable to the prediction of mid-term traffic flow and improve the prediction accuracy, a method of deep BP neural network optimized by genetic algorithm was proposed. The traditional BP neural network was optimized and adjusted through multiple experiments with different numbers of hidden layers, input nodes and hidden nodes, and finally, the optimal neural network structure for the prediction of mid-term traffic flow was obtained. The simulation was carried out through MATLAB2016b, which accuracy indexes were based on Mean Relative Error（MRE）, Accuracy and Equality Coefficient （EC）. The results show that in 30 minutes, the accuracy indexes are less than 3%（MRE）, over 95%（Accuracy） and over 0.98（EC） respectively, while in 60 minutes, the accuracy indexes are less than 7%（MRE）, over 80%（Accuracy） and over 0.95（EC）.