Research And Realization Of Short-term Traffic Flow Forecast And Route Guidance Based On Deep Learning

With the rapid development of Chinese cities and the increase in car ownership in recent decades,urban traffic management is facing huge challenges.Among them,solving the congestion during the peak period of traffic and realizing reasonable traffic planning have become problems that need to be overcome in the modern transportation system.Intelligent Transportation System(ITS)combines various advanced information technologies to improve the utilization rate of roads while ensuring the traffic quality of the road network.Short-term traffic flow prediction and optimal route guidance have always been the focus of ITS research.Through the deep learning method,the historical and real-time short-term traffic flow with rich characteristics are predicted to realize the analysis of the vehicle traffic in the road network.Excavate the capacity of different intersections in depth.On the basis of making full use of the prediction results,an optimal guidance path is searched out at a faster speed.It provides a reliable basis and method for formulating the guidance plan for avoiding congestion and realizing the management plan of urban traffic flow.It is precisely under the requirements of the above-mentioned background that this article mainly conducts research work on the following contents:(1)Considering that short-term traffic flow has characteristics of characteristic tendency,long-term relevance and periodicity.After analyzing and summarizing various prediction methods,a convolution-bidirectional long-short-term memory(CNN-BiLSTM)hybrid neural network prediction model is proposed.First,clean and format the massive short-term traffic flow data set to improve the fault tolerance and compatibility of the data set.Then the deep features of the local space are extracted through the CNN layer,and the repeated features are randomly discarded through the Dropout operation.Combining the temporal memory function of the BiLSTM layer and the characteristics of the two-way transfer structure to predict short-term traffic flow.Finally,experiments prove that the accuracy of the CNN-BiLSTM hybrid neural network for short-term traffic flow prediction reaches 87.64%.(2)Combining the accurate short-term traffic flow prediction results and the optimal route search method,it is possible to mine the guidance route plan with strong traffic capacity in the future road network.The adaptive variable neighborhood search(AVNS)algorithm is improved here.First,through the K-mediods analysis method,the predicted short-term traffic flow is divided and clustered in the urban area.The neighborhood can be quickly constructed based on the subject area that is easy to congestion and the subject area that is not easy to be congested.Then,a probabilistic search is performed on the designed neighborhood structure according to the adaptive update method.This reduces the search time for long-term unimproved solutions in some neighborhoods,and improves the efficiency of the algorithm.A quick search for the optimal guidance path is realized.Finally,through the comparative experimental analysis of various types of areas in Yangzhou,the superiority of AVNS algorithm in the quality and speed of searching for induced path solutions is demonstrated.(3)Based on the above research,this paper designs a route guidance system based on short-term traffic flow prediction.The system uses the configuration and cooperation of the detector module,database module,traffic flow prediction module and regional command center module.A guide route to avoid future congested road sections can be drawn up based on the reserved driving time and OD points(Origin and Destination).According to the simulation experiment of the induced path and the shortest path in Yangzhou city.Through comparative analysis,it is found that the average travel time of the guidance route during peak traffic flow,urban long guidance and road section restrictions are 2.67%,6.65%and 2.89%lower than that of the shortest route.It is proved that the guidance path generated by the system greatly improves the driving quality of the driver and the traffic capacity of the road network.