Research On Traffic Flow Prediction Model Based On Attention Mechanism And Neural Network

With the rapid development of economy and society,the country urbanization process has been significantly accelerated and the pressure on transportation is increasing.Traffic congestion has become an annoying problem in life.With the rapid development of technologies such as the Internet of Things and artificial intelligence,new engine points have been added for the development of intelligent transportation.The intelligent transportation system using artificial intelligence technology as a tool is becoming a research hotspot of scholars.With new theories and new methods,the intelligent transportation system efficiently solves the problem of urban road traffic congestion.The intelligent transportation system provides decision-making for people’s daily travel and good suggestions for urban road planning and construction.Because the traditional traffic flow forecasting methods only focus on the temporal characteristics of traffic flow without considering the spatial characteristics.And most of the traffic flow forecasting methods are mostly short-term forecasts without considering medium and long-term traffic flow forecasts.Therefore,this paper studies the above problems and the main contributions are reflected in two aspects.Firstly,according to the structure of urban elevated roads,a combined prediction model(ACBi GRU)of convolutional neural network and bidirectional gated recurrent unit with the introduction of attention mechanism is proposed.The combined model first utilizes the convolutional neural network that introduces the attention mechanism to dig out the spatial correlation of adjacent road traffic flow.The attention mechanism is embedded into the convolutional neural network.The attention mechanism pays attention to the results of the convolutional layer with different weights to effectively extract the spatial features of traffic flow;then the time series features of traffic flow are extracted through the Bi GRU model.And the extracted time and space features are fused to complete short-term traffic flow prediction.Finally,building a prediction model and studying the performance of the model in multi-step prediction and ablation experiments by means of experimental comparison.The results showed that the combined model can reduce RMSE by 7.09%,MAE by25.5% and MAPE by 80.2%,which effectively improves the traffic flow prediction accuracy.Secondly,in order to complement the long-term forecast and short-term forecast of traffic flow,and improve the forecast performance of the medium-long term model,this paper proposes a combined forecast model ACNN-Trans based on CNN and Transformer network.The ACNN-Trans model uses an improved CNN to dig out the spatial correlation of urban road structures and uses the attention mechanism to pay attention to the feature results with different weights.The combined model fully digs out the spatial features of the node traffic flow.The relative position weight of the data between the traffic flow sequence is calculated through the Transformer network.The time correlation of long-term series data are captured by the Transformer network.Finally the extracted time and space features are respectively used to model the traffic flow prediction of 30 minutes and 60 minutes.The experimental results show that the prediction results of the ACNN-Trans combination model are better than other models on two different real data sets.Compared with the baseline model,the RMSE index of the prediction results is reduced by an average of 34.58%,which verifies the effectiveness of the medium and long-term traffic flow.