Research On Short-Term Traffic Flow Prediction Based On Fusing Neural Network

Short-term Traffic Flow Prediction(STFP)plays an important role in the rapid development of Intelligent Transportation systems(ITS).Accurate and efficient prediction of future traffic flow signals in a short period of time can not only provide valuable real-time data for traffic planning and supervision departments,but also help travelers plan more efficient travel routes,improve travel efficiency.Compared with the traditional method,the STFP method based on deep learning does not need to fully grasp the geographical relationship of each node in the highway network,and can train high-precision and high-efficiency prediction model only by historical data,has the high research value.In this paper,an IVGAT-CLS prediction model is proposed to predict three kinds of traffic flow signals in highway network with complex topology.Firstly,the traffic signal is highly non-stationary,non-linear and easily influenced by uncontrollable factors.If the traffic signal is modeled and analyzed directly,the prediction model will have low precision and weak anti-jamming ability,poor generalization ability.Based on the method of contrast energy difference,an Improved Variational Mode decomposition(IVMD)algorithm is proposed in this paper.The IVMD can decompose the original signal into several sub-Mode components adaptively,each sub-mode has a certain stability and regularity,which can reflect the periodic characteristics of the original signal.The prediction accuracy and anti-jamming ability of STFP model can be greatly improved by directly modeling and analyzing each sub-mode,and then linear superimposing the prediction results of each sub-model.Secondly,this paper uses the fusion neural network to extract the spatiotemporal features of the highway network.Graph Attention Networks(GAT)are used to extract the spatial features of each node in the road network,and one-Dimensional Convolution Neural Network(1DCNN)is combined with Long Short-Term Memory(LSTM),at the same time,a Self Attention Model(SAM)is added,which adaptively assigns a weighting factor to each sequence neuron,it makes the model focus on the more important time points,and improves the identification ability and prediction precision of the model to the time series.Finally,taking Pe MS04 and Pe MS08 road network data sets as examples,using IVGAT-CLS algorithm,the short-term prediction of three kinds of traffic flow signals:traffic flow,vehicle speed,Lane occupancy is carried out,and compared with 4competition models.The experimental results show that the R-score of IVGAT-CLS algorithm proposed in this paper is more than 0.9 in both data sets,which is 7% and9% higher than ASTGCN and TGCN.It verifies the effectiveness and reliability of the algorithm in STFP.