基于遗传算法改进的卷积神经网络短时交通流预测  

Short-Term Traffic Flow Prediction Based on Convolution Neural Network Improved by Genetic Algorithms

作　　者：谭锦新 秦斐燕 任斌 TAN Jinxin;QIN Feiyan;REN Bin(School of Electrical Engineering&Intelligentization,Dongguan University of Technology,Dongguan 523808,China)

机构地区：[1]东莞理工学院电子工程与智能化学院,广东东莞523808

出　　处：《东莞理工学院学报》2021年第5期31-37,共7页Journal of Dongguan University of Technology

基　　金：广东省基础与应用基础研究基金(2019A1515110802);广东省企业重点实验室项目(2020B1212020012019);广东省基础与应用基础研究基金区域联合基金重点项目(2020B1515120095)。

摘　　要：交通流预测对于减少拥堵、节能减排具有重要意义。基于卷积神经网络的预测方法普遍采用梯度下降法训练神经网络,缺点在于预测对网络初始参数敏感。本文采用遗传算法对卷积神经网络的网络参数进行确定从而对短时交通流进行预测。首先,根据交通流数据的特点,设计了适用于交通流预测的卷积神经网络结构;然后,确定卷积神经网络的卷积核与全连接层参数的解空间;随后,采用遗传算法对卷积神经网络参数在可行域中通过选择、交叉、变异三种遗传操作不断迭代搜索得到最优参数解。仿真结果表明,与梯度下降法训练的卷积神经网络相比,该方法拥有更高的预测精度。Traffic flow forecasting is of great significance for reducing congestion,energy saving and emission reduction.The prediction method based on the convolutional neural network generally uses the gradient descent method to train the neural network.The disadvantage is that the prediction is sensitive to the initial parameters of the network.In this paper,genetic algorithm is used to determine the network parameters of the convolutional neural network to predict the short-term traffic flow.First,according to the characteristics of traffic flow data,a convolutional neural network structure suitable for traffic flow prediction is designed.Then,the solution space of the convolution kernel and the fully connected layer parameters of the convolutional neural network is determined.Subsequently,the genetic algorithm is used to continuously search for the optimal parameter solution through three genetic operations of selection,crossover,and mutation in the feasible region of the convolutional neural network parameters.The simulation results show that this method has higher prediction accuracy than the convolutional neural network trained by the gradient descent method.

关 键 词：短时交通流预测 卷积神经网络 遗传算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]