Research On Highway Traffic Flow Forecasting Algorithms Based On Improved IndRNN

With the rapid development of urbanization,the supply and demand contradiction of urban transportation has become increasingly intensified and it has seriously affected people's quality of life.In order to ensure the safety of driving and the road unobstructed,we are committed to doing research about traffic flow prediction technology to obtain real-time accurate prediction values,providing guiding suggestions for people's travel,and achieving harmonious operation of people and vehicles.Due to the complexity of traffic flow,there are many problems in current forecasting methods,such as inadequate extraction of hidden features and lack of time series information learning.In order to overcome the above problems,this paper studies the traffic flow prediction of expressway,and carries out the following research work:Firstly,aiming at the network model based on deep convolution neural network and independent recurrent neural network,the traffic flow prediction is carried out.Three kinds of deep learning methods are explored and constructed.They are 1)ResNet-based traffic flow forecasting algorithm,2)DenseNet-based traffic flow forecasting algorithm and 3)IndRNN-based traffic flow forecasting algorithm.ResNet takes residual module as the basic feature extraction module,which can effectively extract the hidden features of traffic flow data;DenseNet,on the basis of ResNet,uses DenseBlock to reuse multi-layer features to enhance the generality and effectiveness of feature extraction;IndRNN,compared with traditional RNN,can better deal with the problems of gradient disappearance explosion,and combines relu activation function.Stacked multi-layer network,build a deeper network,effective learning of timing information.In the above explorations,we find that IndRNN network has good performance in the temporal learning of traffic flow prediction,but there are still some problems that can not effectively learn the high-level hidden features of the original data.To solve this problem,this paper focuses on improving IndRNN.First,DenseBlock is used as the hidden feature extractor.Then the extracted feature map information is pooled globally.Finally,the prediction head composed of IndRNNCells is connected to learn the timing information.In view of DenseBlock's strong feature extraction ability and IndRNNCells' excellent prediction ability,a traffic flow prediction algorithm model based on improved IndRNN is proposed.In order to verify the validity of the above methods,this paper validates them based on the measured data of Expressway provided by PeMS platform of California Transportation Administration.By comparing the improved IndRNN model with 1)2)3)and the industry's top SAE method,it is proved that the improved IndRNN model proposed in this paper has a good effect in traffic flow forecasting.