Research On Short-term Vehicle Speed Prediction Based On Bidirectional Long Short-term Memory Networ

As an important part of intelligent transportation system,short-term traffic speed prediction can dynamically grasp the development trend of traffic speed,which is the premise and basis for route planning and traffic control,and plays an important role in urban management and resource allocation.Bidirectional Long Short-Term Memory Network(Bi LSTM)can not only keep the memory of the past state,but also have the characteristics of dependence on the future state,so it is widely used in short-term traffic speed prediction.However,the short-term traffic speed prediction model based on the traditional Bi LSTM has the following defects:Bi LSTM is prone to fall into the local optimum,resulting in over-fitting in the prediction;The influence of spatiotemporal dynamic characteristics of noise and traffic speed on the prediction results is not fully considered;The underlying relationship between traffic speed data is ignored.In order to address the above shortcomings and improve the accuracy of short-term traffic speed prediction,this paper improved the short-term traffic speed prediction model of the traditional Bi LSTM.The main research contents and innovations are as follows:(1)In order to solve the over-fitting phenomenon caused by deep learning in the training process,Bi LSTM is used to extract the traffic speed data features as the feature nodes of the Broad Learning System(BLS),and BLS is used to replace the traditional full-connection layer as the predictor of the model,so as to build the Bi LSTM-BLS short-term traffic speed prediction model.By using a public dataset for instance analysis,the experimental results show that compared with the baseline model and existing models,the Bi LSTM-BLS model performs best on the three selected evaluation indicators.(2)In order to extract the spatiotemporal characteristics of the road network,the graph convolution neural network is combined with Bi LSTM.At the same time,in order to reduce the interference of noise,the Variational Mode Decomposition(VMD)is introduced for noise reduction,and a spatiotemporal convolution short-term traffic speed prediction model is built based on the variational mode decomposition.Through validation on real datasets both domestically and internationally,the spatiotemporal convolutional short-term traffic speed prediction model incorporating VMD not only reduces the influence of noise but also fully extracts spatiotemporal features,resulting in further improved prediction accuracy.Particularly,it significantly improves the fitting effect at peak and valley moments.(3)In order to excavate the hidden deep relationship between the nonlinear and non-stationary traffic speed data,a short-term traffic speed prediction model based on the Generative Adversarial Network(GAN)is constructed.The generator of the model is composed of VMD,Bi LSTM and residual network,and the discriminator is composed of convolutional neural network.Simulation results show that on the selected multiple datasets,the R~2 of the short-term traffic speed prediction model based on GAN are 0.9635,0.9624,and 0.9748,respectively,and the predicted values are closer to the true values.The short-term traffic speed prediction model proposed in this paper can accurately predict traffic speed at future moments and has certain applicability to different datasets,providing guiding suggestions for traffic planning.